The Optimizing-Simulator: Merging Optimization and Simulation Using Approximate Dynamic Programming
|
|
- Nathaniel Miles
- 5 years ago
- Views:
Transcription
1 The Optimizing-Simulator: Merging Optimization and Simulation Using Approximate Dynamic Programming Winter Simulation Conference December 5, 2005 Warren Powell CASTLE Laboratory Princeton University Warren B. Powell, Princeton University 2005 Warren B. Powell Slide 1
2 Yellow Freight System 2004 Warren B. Powell, Princeton University 2005 Warren B. Powell Slide 2
3 Yellow Freight System 2004 Warren B. Powell, Princeton University 2005 Warren B. Powell Slide 3
4 The fractional jet ownership industry 2005 Warren B. Powell Slide 4
5 NetJets Inc Warren B. Powell Slide 5
6 2005 Warren B. Powell Slide 6
7 2005 Warren B. Powell Slide 7
8 Schneider National 2005 Warren B. Powell Slide 8
9 Schneider National 2005 Warren B. Powell Slide 9
10 2005 Warren B. Powell Slide 10
11 Air Mobility Command Cargo Holding Fuel Maintenance Air Mobility Command Ramp Space Cargo Handling 2005 Warren B. Powell Slide 11
12
13 The challenges Needs for simulation:» Are we using the right mix of people and equipment?» What is the effect of new policies regarding the management of people and equipment?» What is the marginal contribution from serving customers?» What is the effect of last-minute demands on the system? 2005 Warren B. Powell Slide 13
14 The challenges We need simulation technology that accomplishes the following:» Decisions have to handle high dimensional states and actions (assigning different types of resources to different types of tasks).» The simulator has to capture behaviors that produce good behaviors not just at a point in time, but over time (decisions have to think about the future).» Performance statistics must match historical performance Warren B. Powell Slide 14
15 Outline Modeling and problem representation 2005 Warren B. Powell Slide 15
16 Modeling Resources can have a number of attributes: a = Location Equipment type Location ETA Equipment type Train priority Pool Due for maint Home shop Location ETA Bus. segment Single/team Domicile Drive hours Duty hours 8 day history Days from home Location ETA A/C type Fuel level Home shop Crew Eqpt1 Eqpt Warren B. Powell Slide 16
17 Modeling The attribute vector a t a1 a an 2 = The resource state variable R = Number of resources with attribute a at time t. ta ( ) Rt Rta a A = = Resource state variable 2005 Warren B. Powell Slide 17
18 Modeling Decision set function: D( a) = Set of decision types we can use to act on a resource with attribute a. a t a a a 1 2 = n d a t + 1 Modified resource label 2005 Warren B. Powell Slide 18
19 Modeling The modify function M ( a, W, d ) ( a, c) t 1 t t = t The information process Wt = Vector of information arriving during time interval t. Ex: new customer requests, equipment failures, weather delays Warren B. Powell Slide 19
20 Modeling Decisions x tad = Number of resources with attribute a that we can act on with decision d using the information available at time t. ( ), xt = xtad a A d D The decision function π x = X ( I ) t t t π Π =Set of decision functions (policies) Information available for making a decision 2005 Warren B. Powell Slide 20
21 Approximate dynamic programming Information and decision processes: Exogenous information process W 1 W W W W W Time x 0 x x x x x x Decisions determined by a policy 2005 Warren B. Powell Slide 21
22 Modeling System dynamics (classical view): π Given a decision function (policy) X ( S ) and exogenous information process, we can model the evolution of the state of our system using: W t t t S = f( S, X ( S ), W ) π t+ 1 t t t t Warren B. Powell Slide 22
23 Modeling X π t ( S ) t S t + t 1 x S t W t Warren B. Powell Slide 23
24 Modeling User provides: Model of physical system Our research goal: The decision function Data: Resource vector Information process Software: Decision set function ( ) Modify function M( a, d, W + 1) R t W t D a t t t Decision functions X π t ( I ) t 2005 Warren B. Powell Slide 24
25 Outline The optimizing simulator 2005 Warren B. Powell Slide 25
26 Optimizing over time Resources 2005 Warren B. Powell Slide 26
27 Optimizing over time Tasks 2005 Warren B. Powell Slide 27
28 Optimizing over time Optimizing over time t t+1 t+2 Optimizing at a point in time 2005 Warren B. Powell Slide 28
29 The optimizing simulator Classical simulation:»simple» Extremely flexible But...» Limited solution quality» Often requires extensive user defined tables to guide the simulation.» Can respond to changes in inputs in an unpredictable way. t < T??? t = 0 t = t + 1 Make decision at time t Update system state at t Warren B. Powell Slide 29
30 The optimizing simulator Optimization» Intelligent» Responds naturally to new datasets. But...» Struggles to handle complexity of real operations.» Does not model evolution of information.» Might be too intelligent? t min t t t t 1 t 1 t t t t t cx Ax B x = b Dx x u t 0 t 2005 Warren B. Powell Slide 30
31 Multicommodity flow Time Space Type 2005 Warren B. Powell Slide 31
32 The optimizing simulator To simulate or to optimize... Simulation» Strengths Extremely flexible High level of detail» Weaknesses Low level of intelligence Lower solution quality May have difficulty behaving properly with new scenarios. Difficulty adapting to random outcomes. Optimization» Strengths High level of intelligence System behaves optimally even with new datasets Reduces data set preparation.» Weaknesses Strict rules on problem structure Low level of detail Inflexible!... Why are we asking this question? 2005 Warren B. Powell Slide 32
33 Decision-making technologies Cost-based» The standard assumption of math programming.» Easily handles tradeoffs.» Easily handles high dimensions.» Can be difficult to tune to get the right behavior. Rule-based» Typically associated with AI.» Very flexible.» Difficult coding tradeoffs.» Struggles with higher dimensional states Warren B. Powell Slide 33
34 The four information classes Knowledge Kt Forecasts of exogenous events Forecasts of impacts on others Expert knowledge Ω t Vt ρ 2005 Warren B. Powell Slide 34
35 The four information classes Knowledge Kt 2005 Warren B. Powell Slide 35
36 Knowledge Rule-based: one aircraft and one requirement Aircraft Requirements California Taiwan Germany England New Jersey New Jersey Colorado 2005 Warren B. Powell Slide 36
37 Knowledge Cost based: one requirement and multiple aircraft Aircraft Requirements California Taiwan Germany England New Jersey New Jersey Colorado 2005 Warren B. Powell Slide 37
38 Knowledge Costs allow you to make tradeoffs: California Germany Issue Repositioning cost Appropriate a/c type Utilization Requires modifications Special maintenance at airbase Total cost cost / bonus -$17, Warren B. Powell Slide 38
39 Knowledge Cost based: multiple requirements and aircraft Aircraft Requirements California Taiwan Germany England New Jersey New Jersey Colorado 2005 Warren B. Powell Slide 39
40 The information classes Knowledge Kt Forecasts of exogenous events Ω t 2005 Warren B. Powell Slide 40
41 Forecasts of exogenous information Resources that are known now Aircraft Requirements California Taiwan Germany England New Jersey New Jersey Colorado X π ( I) involves solving a linear program/network model Warren B. Powell Slide 41
42 Forecasts of exogenous information Resources that are known now California California Taiwan England Taiwan New Jersey England Aircraft Aircraft Requirements Germany New Germany Jersey Colorado New Jersey New Jersey Colorado X π ( I) involves solving a linear program/network model Warren B. Powell Slide 42
43 Forecasts of exogenous information ( ) R t R > = t ' t' t and are forecasted for the future. = California Taiwan England New Jersey California Taiwan England New Jersey Aircraft Requirements Germany New Jersey Colorado Germany New Jersey Colorado 2005 Warren B. Powell Slide 43
44 The information classes The Information classes Knowledge Kt Forecasts of exogenous events Forecasts of impacts on others Ω t Vt 2005 Warren B. Powell Slide 44
45 Approximate dynamic programming Decisions now may need to know the impact on future decisions:» What is the cost of assigning this type of aircraft to move a requirement?» What is the value of having a certain number of aircraft in a region?» Should this requirement be satisfied now? Later? Never? For these questions, it is important that we optimize over time Warren B. Powell Slide 45
46 Time t V(a ) V(a ) a
47 Time t V( a ) ' 1 V( a ) ' 2 a 1 a 2
48 The optimization challenge? 2005 Warren B. Powell Slide 48
49 State variables Systems evolve through a cycle of exogenous and endogenous information ω = ˆR 1 ˆR ˆR ˆR ˆR ˆR Time x 0 x x x x x x Warren B. Powell Slide 49
50 State variables Systems evolve through a cycle of exogenous and endogenous information ˆR 1 ˆR ˆR ˆR ˆR ˆR Time x 0 x x x x x x R 0 R R R R R R Warren B. Powell Slide 50
51 Approximate dynamic programming Using this state variable, we obtain the optimality equations: { } V ( R ) = max C ( R, x ) + E V ( R ) R t t t t t t+ 1 t+ 1 t x X Three curses Problem: Curse of dimensionality State space Outcome space Action space (feasible region) 2005 Warren B. Powell Slide 51
52 Approximate dynamic programming The computational challenge: { } V ( R ) = max C ( R, x ) + E V ( R ) R t t t t t t+ 1 t+ 1 t x X How do we find V ( R )? t+ 1 t+ 1 How do we compute the expectation? How do we find the optimal solution? 2005 Warren B. Powell Slide 52
53 Approximate dynamic programming A possible approximation strategy: We start with: { ( ) } 1 1 V ( R ) = max C ( R, x ) + E V R R t t t t t t+ t+ t x t We solve this for a sample realization: Can t compute this!!! ( ω ) V ( R, ω) = max C ( R, x ) + V R ( ) x t t t t t t+ 1 t+ 1 t Now substitute in function approximations: V ( R, ω) = max C ( R, x ) + V R ( ) x ( ω ) t t t t t t+ 1 t+ 1 t Need to approximate V Don t know what this is! 2005 Warren B. Powell Slide 53
54 Approximate dynamic programming One big problem. V ( R, ω) = max C ( R, x ) + V R ( ) x ( ω ) t t t t t t+ 1 t+ 1 t Seeing is cheating! R t Warren B. Powell Slide 54
55 Approximate dynamic programming Alternative: Change the definition of the state variable: ˆR 1 ˆR ˆR ˆR ˆR ˆR Time x 0 x x x x x x R R R R R R R R R R R3 R R R R6 6 R R Warren B. Powell Slide 55
56 Approximate dynamic programming Now our optimality equation looks like: Expectation outside of the max operator. { } ω V ( R ) = E max C ( R, x ) + V ( R ( x, )) R x x x t 1, t t 1 x X t t t t t t t 1 t We drop the expectation and solve the conditional problem: Finally, we substitute in our approximation: ( ) V ( R, Rˆ ( ω)) = max C ( R ( ω), x ( ω)) + V R x, ω ( x ) x t 1 t 1 t x( ω) X ( ω) t t t t t t Post-decision state variable ( ) V ( R, Rˆ ( ω)) = max C ( R ( ω), x ( ω)) + V R x, ω ( x ) x t 1 t 1 t x( ω) X ( ω) t t t t t t Convenient value function approximation Warren B. Powell Slide 56
57 Approximate dynamic programming Approximating the value function:» We choose approximations of the form: Linear (in the resource state): V ( R ) = v R t t ta ta a A Piecewise linear, separable: V ( R ) = V ( R ) t t ta ta a A Best when assets are complex, R ta which means that is small (typically 0 or 1). Best when assets are simple, which means that may be larger. R ta 2005 Warren B. Powell Slide 57
58 Approximate dynamic programming A myopic decision rule (policy): x = arg max C ( R ( ω), x ( ω)) n t x( ω) X ( ω) t t t A decision rule that looks into the future: ( ) ( x ) x = arg max C ( R ( ω), x ( ω)) + V R x, ω n t x( ω) X ( ω) t t t t t t 2005 Warren B. Powell Slide 58
59 Approximate dynamic programming Simulating a myopic policy: t t+1 t Warren B. Powell Slide 59
60 Approximate dynamic programming A myopic decision rule (policy): x = arg max C ( R ( ω), x ( ω)) n t x( ω) X ( ω) t t t A decision rule that looks into the future: ( ) ( x ) x = arg max C ( R ( ω), x ( ω)) + V R x, ω n t x( ω) X ( ω) t t t t t t 2005 Warren B. Powell Slide 60
61 Approximate dynamic programming V( a ) ' 1 a 1 a 2 V( a ) ' Warren B. Powell Slide 61
62 2005 Warren B. Powell Slide 62
63 Classification yards Option 1: Send directly to customers Option 2: Send to regional depots Option 3: Send to classification yards
64 Approximate dynamic programming Two-stage resource allocation under uncertainty 2005 Warren B. Powell Slide 64
65 Approximate dynamic programming We obtain piecewise linear recourse functions for each regions Warren B. Powell Slide 65
66 Approximate dynamic programming The function is piecewise linear on the integers. Profits We approximate the value of cars in the future using a separable approximation Number of vehicles at a location 2005 Warren B. Powell Slide 66
67 Approximate dynamic programming To capture nonlinear behavior: Each link captures the marginal reward of an additional car Warren B. Powell Slide 67
68 Approximate dynamic programming 2005 Warren B. Powell Slide 68
69 Approximate dynamic programming 2005 Warren B. Powell Slide 69
70 Approximate dynamic programming n R1 n R2 n R3 n R4 n R Warren B. Powell Slide 70
71 Approximate dynamic programming We estimate the functions by sampling from our distributions. Marginal value: ( n v ) 1 ω ( n v ) 2 ω ( n v ) 3 ω n R1 n R2 n R3 ( n D ) 1 ω n D3 ( ω ) ( n D ) 2 ω ( n v ) 4 ω ( n v ) 5 ω n R4 n R5 DC n ( ω ) 2005 Warren B. Powell Slide 71
72 Approximate dynamic programming The time t subproblem: V ( R, R, R ) n ta t1 t 2 t3 t (i-1,t+3) Gradients: ( vˆ, vˆ ) n n+ t1 t1 ( vˆ, vˆ ) n n+ t2 t2 ( vˆ, vˆ ) n n+ t3 t3 R t1 R t 2 R t3 (i,t+1) (i+1,t+5) 2005 Warren B. Powell Slide 72
73 Approximate dynamic programming Left and right gradients are found by solving flow augmenting path problems. V ( R, R, R ) n ta t1 t 2 t3 Gradients: t (i-1,t+3) i ( ˆ ) n v + t3 R t3 The The right right derivative derivative (the (the value value of of one one more more unit unit of of that that resource) resource) is is a a flow flow augmenting augmenting path path from from that that node node to to the the supersink. supersink Warren B. Powell Slide 73
74 Approximate dynamic programming Left and right derivatives are used to build up a nonlinear approximation of the subproblem. V k it ( R ) 1t k R 1t R 1t 2005 Warren B. Powell Slide 74
75 Approximate dynamic programming Left and right derivatives are used to build up a nonlinear approximation of the subproblem. V k it ( R ) 1t Left derivative k v t Right derivative k v + t k R 1t R 1t 2005 Warren B. Powell Slide 75
76 Approximate dynamic programming Each iteration adds new segments, as well as refining old ones. V k it ( R ) 1t ( k 1) v + t ( k 1) v + + t k+1 R 1t R 1t 2005 Warren B. Powell Slide 76
77 Approximate dynamic programming 2.5 Approximate value function Functional Value, f(s) = ln(1+s) Exact 1 Iter 2 Iter 5 Iter 10 Iter 15 Iter 20 Iter Number Variable of Value, resources s 2005 Warren B. Powell Slide 77
78 Approximate dynamic programming Simulating a myopic policy t 2005 Warren B. Powell Slide 78
79 Approximate dynamic programming Simulating a myopic policy 2005 Warren B. Powell Slide 79
80 Approximate dynamic programming Using value functions to anticipate the future t Here and now Downstream impacts 2005 Warren B. Powell Slide 80
81 Approximate dynamic programming Using value functions to anticipate the future 2005 Warren B. Powell Slide 81
82 Approximate dynamic programming Using value functions to anticipate the future 2005 Warren B. Powell Slide 82
83 Approximate dynamic programming Using value functions to anticipate the future 2005 Warren B. Powell Slide 83
84 2005 Warren B. Powell Slide 84
85 2005 Warren B. Powell Slide 85
86 2005 Warren B. Powell Slide 86
87 2005 Warren B. Powell Slide 87
88 Approximate DP vs. LP Approximate dynamic programming % of Objective Upperbound The mathematical optimum Agg_PWLinear_1 Agg_PWLinear_2 Agg_PWLinear_3 DisAgg_Linear DisAgg_PWLinear Decomp_Location Iteration No Warren B. Powell Slide 88
89 Downloadable at Warren B. Powell Slide 89
90 The information classes Knowledge Kt Forecasts of exogenous events Forecasts of impacts on others Expert knowledge Ω t Vt ρ 2005 Warren B. Powell Slide 90
91 Low dimensional patterns Old modeling approach: Engineering costs Behavior Objectives x * = arg min cx Subject to : Ax = b, x 0 Physics 2005 Warren B. Powell Slide 91
92 Flows from history 2005 Warren B. Powell Slide 92
93 Flows from history Flows from the model 2005 Warren B. Powell Slide 93
94 Low dimensional patterns Bottom up/top down modeling: Patterns Specify the behaviors you want at a general level. Specify costs, driver availability, work rules, routing preferences, load avail. Engineering 2005 Warren B. Powell Slide 94
95 Low dimensional patterns Pattern matching Behavior Cost function * x cx H x = arg min +θ (, ρ) The happiness function measures the degree to which model behavior agrees with a knowledgeable expert. Hx (, ρ) = Gx ( ) ρ where Gx ( ) is an aggregation function 2005 Warren B. Powell Slide 95
96 Low dimensional patterns Patterns and aggregation:» What we do: We define patterns based on an aggregation of the attributes of a single vehicle. Patterns indicate the desirability of a single decision.» Patterns can be expressed at different levels of aggregation, simultaneously. Don t send C-5 s into Saudi Arabia Don t send C-5 s needing maintenance into Saudi Arabia Don t send C-5 s needing maintenance loaded with freight to southeast Asia into Saudi Arabia.» Patterns are not hard rules they express desirable or undesirable patterns of behavior Warren B. Powell Slide 96
97 Flows from history Flows from the model 2005 Warren B. Powell Slide 97
98 Flows from history Flows from the model 2005 Warren B. Powell Slide 98
99 Low dimensional patterns Length of haul calibration-teams With pattern Min Solo w/ pattern Solo w/o pattern Max 650 Without pattern Iteration 2005 Warren B. Powell Slide 99
100 Low dimensional patterns Patterns can come from history: 2005 Warren B. Powell Slide 100
101 Low dimensional patterns or an expert: 2005 Warren B. Powell Slide 101
102 The information classes Knowledge Kt Forecasts of exogenous events Forecasts of impacts on others Expert knowledge Ω t Vt ρ 2005 Warren B. Powell Slide 102
103 The military airlift problem 2005 Warren B. Powell Slide 103
104 Optimizing simulator Increasing information sets Policy Rule-based Myopic cost-based, one requirement to a list of aircraft, known now and actionable now Myopic cost-based, one requirement to a list of aircraft, known now and actionable in the future Myopic cost-based, a list of requirements to a list of aircraft, known now and actionable now Myopic cost-based, a list of requirements to a list of aircraft, known now and actionable in the future Rolling horizon Approximate Dynamic Programming Expert knowledge Information Classes I t = R tt I t = ( Rtt, ct ) I t = (( Rtt ) t t, ct ) I t = ( Rtt, ct ) I I I I t = (( Rtt ) t t, ct ) t = {( Rt ' t'' ) t'' t', ct t t = {( R ), c, V tt t t t tt t T t T = {( R ), c, V, ρ t T tt t t t tt ph t ph t ph t } } } Decision Functions (RB:R-A) (MP:R- AL/KNAN) (MP:R- AL/KNAF) (MP:RL- AL/KNAN) (MP:RL- AL/KNAF) (RH) (ADP) (EK) 2005 Warren B. Powell Slide 104
105 Optimizing simulator Costs of different policies Million Dollors Total cost Transportation cost Late delivery cost Repair cost 50 0 Rule Based (RB:R-A) Choice of aircraft Actionable Now Policies Actionable future (MP:RL-AL/ KNAN) Value functions (ADP) Increasing information sets 2005 Warren B. Powell Slide 105
106 Optimizing simulator Millions Throughput curves of policies Increasing information sets Pounds Cumulative expected thruput (RB:R-A) (MP:R-AL/KNAN) (MP:RL-AL/KNAN) (MP:RL-AL/KNAF) (ADP) Time periods 2005 Warren B. Powell Slide 106
107 Optimizing simulator Millions Throughput curves of policies Pounds Cumulative expected thruput (RB:R-A) (MP:R-AL/KNAN) (MP:RL-AL/KNAN) (MP:RL-AL/KNAF) (ADP) Time periods 2005 Warren B. Powell Slide 107
108 Optimizing simulator Areas between the cumulative expected thruput curve and different policy thruput curves Millions Pound * days (RB:R-A) (MP:R-AL/KNAN) (ADP) (MP:RL-AL/ KNAF) (MP:RL-AL/ KNAN) Policy Increasing information sets 2005 Warren B. Powell Slide 108
109 Outline Recent experiments with modeling airlift operations 2005 Warren B. Powell Slide 109
110 Random demands and equipment failures 2005 Warren B. Powell Slide 110
111 Pilots Aircraft Customers 2005 Warren B. Powell Slide 111
112 Case study Questions:» What is the effect of uncertain demands on a military airlift schedule?» What is the effect of equipment failures?» How does adaptive learning change the effect of randomness on the performance of the simulation?» What is the effect of advance information? 2005 Warren B. Powell Slide 112
113 Total contribution Iterative learning Determ demand No Break Learn Determ demand Break Learn Random demand No break Learn Determ demand No Break No learn Random demand Break Learn Determ demand Break No learn Random demand No Break No learn Random demand Break No learn 2005 Warren B. Powell Slide 113
114 Deterministic demands, no failures With learning Without learning Determ demand No Break Learn Determ demand Break Learn Random demand No break Learn Determ demand No Break No learn Random demand Break Learn Determ demand Break No learn Random demand No Break No learn Random demand Break No learn 2005 Warren B. Powell Slide 114
115 Deterministic demands, with failures With learning Without learning Determ demand No Break Learn Determ demand Break Learn Random demand No break Learn Determ demand No Break No learn Random demand Break Learn Determ demand Break No learn Random demand No Break No learn Random demand Break No learn 2005 Warren B. Powell Slide 115
116 Random demands, no failures With learning Without learning Determ demand No Break Learn Determ demand Break Learn Random demand No break Learn Determ demand No Break No learn Random demand Break Learn Determ demand Break No learn Random demand No Break No learn Random demand Break No learn Warren B. Powell Slide 116
117 Random demands, with failures With learning Without learning Determ demand No Break Learn Determ demand Break Learn Random demand No break Learn Determ demand No Break No learn Random demand Break Learn Determ demand Break No learn Random demand No Break No learn Random demand Break No learn 2005 Warren B. Powell Slide 117
118 Effect of advance booking Effect of advance notice Percent coverage Without learning Prebook 0 hours Prebook 2 hours Prebook 6 hours 2005 Warren B. Powell Slide 118
119 Effect of advance booking Effect of advance notice Percent coverage With learning Without learning Prebook 0 hours Prebook 2 hours Prebook 6 hours 2005 Warren B. Powell Slide 119
120 Midair refueling: initial solution 2005 Warren B. Powell Slide 120
121 Midair refueling: initial solution Path followed by tanker (moves up and down Atlantic) Warren B. Powell Slide 121
122 Midair refueling: initial solution First plane refuels Second plane crashes Green: full of fuel Yellow to red: nearing empty Black: empty (plane crashes) 2005 Warren B. Powell Slide 122
123 Midair refueling: exploration Learning over many iterations Warren B. Powell Slide 123
124 Midair refueling: final solution Planes learn to meet in the middle so both can refuel Warren B. Powell Slide 124
125 Outline Calibrating a model for a major truckload motor carrier 2005 Warren B. Powell Slide 125
126 Schneider National 2005 Warren B. Powell Slide 126
127 Schneider National 2005 Warren B. Powell Slide 127
128 2005 Warren B. Powell Slide 128
129 Truckload trucking Questions for the model:» What types of drivers should they hire? Domicile? Single drivers vs. teams?» What is the value of knowing about customer requests farther in the future?» What is the profitability of different customers?» What is the value of increasing terminal capacity? 2005 Warren B. Powell Slide 129
130 Truckload trucking LOH LOH Historical maximum Simulation Historical minimum US_SOLO US_IC US_TEAM Capacity category 2005 Warren B. Powell Slide 130
131 Truckload trucking Revenue per WU Historical maximum Simulation Historical minimum 1200 Utilization US_SOLO US_IC US_TEAM Capacity category Revenue per WU Utilization Historical maximum Simulation Historical minimum US_SOLO US_IC US_TEAM Capacity category 2005 Warren B. Powell Slide 131
132 Truckload trucking Challenge» We want to know the marginal value of each type of driver.» A driver type is determined by: Location 100 a = Domicile 100 = Driver type 3» There are 30,000 driver types!!!» We need to take the derivative of our simulation for each type Warren B. Powell Slide 132
133 Multistage problems vˆn t 1 t X π t ( R ) t t +1 t + 2 Time ˆn2 v t Resource State-Type ˆn3 v t 2005 Warren B. Powell Slide 133
134 Multistage problems v ˆn t + 1,1 t +1 X π ( R ) t+ 1 t+ 1 t + 2 Time v ˆn t + 1,2 Resource State-Type v ˆn t + 1, Warren B. Powell Slide 134
135 Multistage problems v ˆn t + 2,1 t + 2 X π ( R ) t+ 2 t+ 2 Time v ˆn t + 2,2 Resource State-Type v ˆn t + 2, Warren B. Powell Slide 135
136 Multistage problems vˆn t 1 t X π t ( R ) t t +1 t + 2 Time ˆn2 v t Resource State-Type ˆn3 v t 2005 Warren B. Powell Slide 136
137 Multistage problems v ˆn t + 1,1 t +1 X π ( R ) t+ 1 t+ 1 t + 2 Time v ˆn t + 1,2 Resource State-Type v ˆn t + 1, Warren B. Powell Slide 137
138 Multistage problems v ˆn t + 2,1 t + 2 X π ( R ) t+ 2 t+ 2 Time v ˆn t + 2,2 Resource State-Type v ˆn t + 2, Warren B. Powell Slide 138
139 Backward pass X π t ( R ) t X π ( R ) X ( R ) t+ 1 t+ 1 t+ 2 t+ 2 π 2005 Warren B. Powell Slide 139
140 Backward pass t + 2 Time v ˆn t + 2,1 Resource State-Type 2005 Warren B. Powell Slide 140
141 Backward pass t +1 t + 2 Time v ˆn t + 1,2 Resource State-Type 2005 Warren B. Powell Slide 141
142 Backward pass t t +1 t + 2 Time Resource State-Type ˆn3 v t 2005 Warren B. Powell Slide 142
143 Backward pass t t +1 t + 2 Time Resource State-Type ˆn3 v t 2005 Warren B. Powell Slide 143
144 Driver fleet optimization simulation objective function resources +40 resources +60 resources +50 resources +20 resources +10 resources Base +5 case resources # of drivers s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 avg pred 2005 Warren B. Powell Slide 144
145 Driver fleet optimization simulation objective function # of drivers s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 avg pred 2005 Warren B. Powell Slide 145
146 Driver fleet optimization simulation objective function v a # of drivers s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 avg pred 2005 Warren B. Powell Slide 146
147 Driver fleet optimization Driver types 2005 Warren B. Powell Slide 147
148 Add drivers 2005 Warren B. Powell Slide 148
149 Reduce drivers 2005 Warren B. Powell Slide 149
150 2005 Warren B. Powell Slide 150
151 Questions?
Approximate Dynamic Programming in Rail Operations
Approximate Dynamic Programming in Rail Operations June, 2007 Tristan VI Phuket Island, Thailand Warren Powell Belgacem Bouzaiene-Ayari CASTLE Laboratory Princeton University http://www.castlelab.princeton.edu
More informationApproximate Dynamic Programming: Solving the curses of dimensionality
Approximate Dynamic Programming: Solving the curses of dimensionality Informs Computing Society Tutorial October, 2008 Warren Powell CASTLE Laboratory Princeton University http://www.castlelab.princeton.edu
More informationThe Dynamic Energy Resource Model
The Dynamic Energy Resource Model Group Peer Review Committee Lawrence Livermore National Laboratories July, 2007 Warren Powell Alan Lamont Jeffrey Stewart Abraham George 2007 Warren B. Powell, Princeton
More informationApproximate Dynamic Programming for High Dimensional Resource Allocation Problems
Approximate Dynamic Programming for High Dimensional Resource Allocation Problems Warren B. Powell Abraham George Belgacem Bouzaiene-Ayari Hugo P. Simao Department of Operations Research and Financial
More informationStochastic Programming in Transportation and Logistics
Stochastic Programming in Transportation and Logistics Warren B. Powell and Huseyin Topaloglu Department of Operations Research and Financial Engineering, Princeton University, Princeton, NJ 08544 Abstract
More informationAn Integrated Optimizing-Simulator for the Military Airlift Problem
An Integrated Optimizing-Simulator for the Military Airlift Problem Tongqiang Tony Wu Department of Operations Research and Financial Engineering Princeton University Princeton, NJ 08544 Warren B. Powell
More informationThe Optimizing-Simulator: An Illustration using the Military Airlift Problem
The Optimizing-Simulator: An Illustration using the Military Airlift Problem Tongqiang Tony Wu Warren B. Powell Princeton University and Alan Whisman Air Mobility Command There have been two primary modeling
More informationUsing Static Flow Patterns in Time-Staged Resource Allocation Problems
Using Static Flow Patterns in Time-Staged Resource Allocation Problems Arun Marar Warren B. Powell Hugo P. Simão Department of Operations Research and Financial Engineering, Princeton University, Princeton,
More informationDynamic Models for Freight Transportation
Dynamic Models for Freight Transportation Warren B. Powell Belgacem Bouzaiene-Ayari Hugo P. Simao Department of Operations Research and Financial Engineering Princeton University August 13, 2003 Abstract
More informationDynamic Models for Freight Transportation
Dynamic Models for Freight Transportation Warren B. Powell Belgacem Bouzaiene-Ayari Hugo P. Simao Department of Operations Research and Financial Engineering Princeton University September 14, 2005 Abstract
More informationApproximate Dynamic Programming in Transportation and Logistics: A Unified Framework
Approximate Dynamic Programming in Transportation and Logistics: A Unified Framework Warren B. Powell, Hugo P. Simao and Belgacem Bouzaiene-Ayari Department of Operations Research and Financial Engineering
More informationThere has been considerable recent interest in the dynamic vehicle routing problem, but the complexities of
TRANSPORTATION SCIENCE Vol. 38, No. 4, November 2004, pp. 399 419 issn 0041-1655 eissn 1526-5447 04 3804 0399 informs doi 10.1287/trsc.1030.0073 2004 INFORMS The Dynamic Assignment Problem Michael Z. Spivey,
More informationMODELING DYNAMIC PROGRAMS
CHAPTER 5 MODELING DYNAMIC PROGRAMS Perhaps one of the most important skills to develop in approximate dynamic programming is the ability to write down a model of the problem. Everyone who wants to solve
More informationApproximate Dynamic Programming in Transportation and Logistics: A Unified Framework
Approximate Dynamic Programming in Transportation and Logistics: A Unified Framework Warren B. Powell, Hugo P. Simao and Belgacem Bouzaiene-Ayari Department of Operations Research and Financial Engineering
More informationSome Fixed-Point Results for the Dynamic Assignment Problem
Some Fixed-Point Results for the Dynamic Assignment Problem Michael Z. Spivey Department of Mathematics and Computer Science Samford University, Birmingham, AL 35229 Warren B. Powell Department of Operations
More informationApproximate Dynamic Programming for High-Dimensional Resource Allocation Problems
Approximate Dynamic Programming for High-Dimensional Resource Allocation Problems Warren B. Powell Department of Operations Research and Financial Engineering Princeton University Benjamin Van Roy Departments
More informationHandout 1: Introduction to Dynamic Programming. 1 Dynamic Programming: Introduction and Examples
SEEM 3470: Dynamic Optimization and Applications 2013 14 Second Term Handout 1: Introduction to Dynamic Programming Instructor: Shiqian Ma January 6, 2014 Suggested Reading: Sections 1.1 1.5 of Chapter
More informationTutorial: Stochastic Optimization in Energy
Tutorial: Stochastic Optimization in Energy FERC, Washington, D.C. August 6, 2014 Warren B. Powell CASTLE Labs Princeton University http://www.castlelab.princeton.edu Warren B. Powell, 2014 Slide 1 Mission
More informationWe present a general optimization framework for locomotive models that captures different levels of detail,
Articles in Advance, pp. 1 24 ISSN 0041-1655 (print) ISSN 1526-5447 (online) http://dx.doi.org/10.1287/trsc.2014.0536 2014 INFORMS From Single Commodity to Multiattribute Models for Locomotive Optimization:
More informationOn-line supplement to: SMART: A Stochastic Multiscale Model for the Analysis of Energy Resources, Technology
On-line supplement to: SMART: A Stochastic Multiscale Model for e Analysis of Energy Resources, Technology and Policy This online supplement provides a more detailed version of e model, followed by derivations
More informationAcquisition of Multi-Function Equipment at DIA: Conditions, Factors, Considerations & Integration. Presented by Mike Carlson September 20, 2012
Acquisition of Multi-Function Equipment at DIA: Conditions, Factors, Considerations & Integration Presented by Mike Carlson September 20, 2012 1 Denver International Airport 5 Runways 12,000 /. (3,658m)
More informationFacility Location and Distribution System Planning. Thomas L. Magnanti
Facility Location and Distribution System Planning Thomas L. Magnanti Today s Agenda Why study facility location? Issues to be modeled Basic models Fixed charge problems Core uncapacitated and capacitated
More informationDeparture time choice equilibrium problem with partial implementation of congestion pricing
Departure time choice equilibrium problem with partial implementation of congestion pricing Tokyo Institute of Technology Postdoctoral researcher Katsuya Sakai 1 Contents 1. Introduction 2. Method/Tool
More informationCHAPTER 11 Integer Programming, Goal Programming, and Nonlinear Programming
Integer Programming, Goal Programming, and Nonlinear Programming CHAPTER 11 253 CHAPTER 11 Integer Programming, Goal Programming, and Nonlinear Programming TRUE/FALSE 11.1 If conditions require that all
More informationMARKOV DECISION PROCESSES (MDP) AND REINFORCEMENT LEARNING (RL) Versione originale delle slide fornita dal Prof. Francesco Lo Presti
1 MARKOV DECISION PROCESSES (MDP) AND REINFORCEMENT LEARNING (RL) Versione originale delle slide fornita dal Prof. Francesco Lo Presti Historical background 2 Original motivation: animal learning Early
More informationStudy Guide - Part 2
Math 116 Spring 2015 Study Guide - Part 2 1. Which of the following describes the derivative function f (x) of a quadratic function f(x)? (A) Cubic (B) Quadratic (C) Linear (D) Constant 2. Find the derivative
More informationIncorporating Demand Response with Load Shifting into Stochastic Unit Commitment
Incorporating Demand Response with Load Shifting into Stochastic Unit Commitment Frank Schneider Supply Chain Management & Management Science, University of Cologne, Cologne, Germany, frank.schneider@uni-koeln.de
More informationSurge Pricing and Labor Supply in the Ride- Sourcing Market
Surge Pricing and Labor Supply in the Ride- Sourcing Market Yafeng Yin Professor Department of Civil and Environmental Engineering University of Michigan, Ann Arbor *Joint work with Liteng Zha (@Amazon)
More informationMath 381 Midterm Practice Problem Solutions
Math 381 Midterm Practice Problem Solutions Notes: -Many of the exercises below are adapted from Operations Research: Applications and Algorithms by Winston. -I have included a list of topics covered on
More informationSupplementary Technical Details and Results
Supplementary Technical Details and Results April 6, 2016 1 Introduction This document provides additional details to augment the paper Efficient Calibration Techniques for Large-scale Traffic Simulators.
More informationApproximate dynamic programming in transportation and logistics: a unified framework
EURO J Transp Logist (2012) 1:237 284 DOI 10.1007/s13676-012-0015-8 TUTORIAL Approximate dynamic programming in transportation and logistics: a unified framework Warren B. Powell Hugo P. Simao Belgacem
More informationMulti-Area Stochastic Unit Commitment for High Wind Penetration
Multi-Area Stochastic Unit Commitment for High Wind Penetration Workshop on Optimization in an Uncertain Environment Anthony Papavasiliou, UC Berkeley Shmuel S. Oren, UC Berkeley March 25th, 2011 Outline
More informationA Distributed Decision Making Structure for Dynamic Resource Allocation Using Nonlinear Functional Approximations
A Distributed Decision Making Structure for Dynamic Resource Allocation Using Nonlinear Functional Approximations Huseyin Topaloglu School of Operations Research and Industrial Engineering Cornell University,
More informationProper Security Criteria Determination in a Power System with High Penetration of Renewable Resources
Proper Security Criteria Determination in a Power System with High Penetration of Renewable Resources Mojgan Hedayati, Kory Hedman, and Junshan Zhang School of Electrical, Computer, and Energy Engineering
More informationCost-Benefit Analysis of the Pooled- Fund Maintenance Decision Support System: Case Study
Cost-Benefit Analysis of the Pooled- Fund Maintenance Decision Support System: Case Study Zhirui Ye (WTI) Xianming Shi (WTI) Christopher K. Strong (City of Oshkosh) 12 th AASHTO-TRB TRB Maintenance Management
More informationPerhaps one of the most widely used and poorly understood terms in dynamic programming is policy. A simple definition of a policy is:
CHAPTER 6 POLICIES Perhaps one of the most widely used and poorly understood terms in dynamic programming is policy. A simple definition of a policy is: Definition 6.0.1 A policy is a rule (or function)
More informationImprovements to Benders' decomposition: systematic classification and performance comparison in a Transmission Expansion Planning problem
Improvements to Benders' decomposition: systematic classification and performance comparison in a Transmission Expansion Planning problem Sara Lumbreras & Andrés Ramos July 2013 Agenda Motivation improvement
More informationManagement of intermodal container terminals using feedback control
Management of intermodal container terminals using feedback control A. Alessandri, S. Sacone $, S.Siri $ Institute of Intelligent Systems for Automation ISSIA-CNR National Research Council of Italy Via
More informationA Parallelizable and Approximate Dynamic Programming-Based Dynamic Fleet Management Model with Random Travel Times and Multiple Vehicle Types
A Parallelizable and Approximate Dynamic Programming-Based Dynamic Fleet Management Model with Random Travel Times and Multiple Vehicle Types Huseyin Topaloglu School of Operations Research and Industrial
More informationCE 191: Civil & Environmental Engineering Systems Analysis. LEC 17 : Final Review
CE 191: Civil & Environmental Engineering Systems Analysis LEC 17 : Final Review Professor Scott Moura Civil & Environmental Engineering University of California, Berkeley Fall 2014 Prof. Moura UC Berkeley
More informationWeatherCloud Hyper-Local Global Forecasting All rights reserved. Fathym, Inc.
WeatherCloud Hyper-Local Global Forecasting based on current forecast techniques EVOLVING FORECASTING TECHNOLOGY 1) The WeatherCloud backend forecast system allows for routing around hazardous weather
More informationSCHEDULING PROBLEMS FOR FRACTIONAL AIRLINES
SCHEDULING PROBLEMS FOR FRACTIONAL AIRLINES A Thesis Presented to The Academic Faculty by Fei Qian In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the H. Milton Stewart
More informationRegularized optimization techniques for multistage stochastic programming
Regularized optimization techniques for multistage stochastic programming Felipe Beltrán 1, Welington de Oliveira 2, Guilherme Fredo 1, Erlon Finardi 1 1 UFSC/LabPlan Universidade Federal de Santa Catarina
More informationInteger Linear Programming Modeling
DM554/DM545 Linear and Lecture 9 Integer Linear Programming Marco Chiarandini Department of Mathematics & Computer Science University of Southern Denmark Outline 1. 2. Assignment Problem Knapsack Problem
More informationRobust Optimization for Empty Repositioning Problems
Robust Optimization for Empty Repositioning Problems Alan L. Erera, Juan C. Morales and Martin Savelsbergh The Logistics Institute School of Industrial and Systems Engineering Georgia Institute of Technology
More informationTRUCK DISPATCHING AND FIXED DRIVER REST LOCATIONS
TRUCK DISPATCHING AND FIXED DRIVER REST LOCATIONS A Thesis Presented to The Academic Faculty by Steven M. Morris In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in Industrial
More informationA Benders Algorithm for Two-Stage Stochastic Optimization Problems With Mixed Integer Recourse
A Benders Algorithm for Two-Stage Stochastic Optimization Problems With Mixed Integer Recourse Ted Ralphs 1 Joint work with Menal Güzelsoy 2 and Anahita Hassanzadeh 1 1 COR@L Lab, Department of Industrial
More informationMarkov Decision Processes and Dynamic Programming
Markov Decision Processes and Dynamic Programming A. LAZARIC (SequeL Team @INRIA-Lille) Ecole Centrale - Option DAD SequeL INRIA Lille EC-RL Course In This Lecture A. LAZARIC Markov Decision Processes
More informationINTRODUCTION TO TRANSPORTATION SYSTEMS
INTRODUCTION TO TRANSPORTATION SYSTEMS Lectures 5/6: Modeling/Equilibrium/Demand 1 OUTLINE 1. Conceptual view of TSA 2. Models: different roles and different types 3. Equilibrium 4. Demand Modeling References:
More informationRecoverable Robustness in Scheduling Problems
Master Thesis Computing Science Recoverable Robustness in Scheduling Problems Author: J.M.J. Stoef (3470997) J.M.J.Stoef@uu.nl Supervisors: dr. J.A. Hoogeveen J.A.Hoogeveen@uu.nl dr. ir. J.M. van den Akker
More informationPlanning in Markov Decision Processes
Carnegie Mellon School of Computer Science Deep Reinforcement Learning and Control Planning in Markov Decision Processes Lecture 3, CMU 10703 Katerina Fragkiadaki Markov Decision Process (MDP) A Markov
More informationLessons Learned Using ESRI s Network Analyst to Optimize Snow Treatment Routes in Kentucky
Lessons Learned Using ESRI s Network Analyst to Optimize Snow Treatment Routes in Kentucky Ben Blandford, PhD Eric Green, PE, PhD Candidate Kentucky Transportation Center University of Kentucky Overview
More informationTechnology and Network Design Issues. Anna Nagurney Isenberg School of Management University of Massachusetts Amherst, MA 01003
Technology and Network Design Issues Anna Nagurney Isenberg School of Management University of Massachusetts Amherst, MA 01003 c 2002 Introduction In this lecture, I explore technology and network design
More informationDiscrete-event simulations
Discrete-event simulations Lecturer: Dmitri A. Moltchanov E-mail: moltchan@cs.tut.fi http://www.cs.tut.fi/kurssit/elt-53606/ OUTLINE: Why do we need simulations? Step-by-step simulations; Classifications;
More informationAnticipatory Freight Selection in Intermodal Long-haul Round-trips
Anticipatory Freight Selection in Intermodal Long-haul Round-trips A.E. Pérez Rivera and M.R.K. Mes Department of Industrial Engineering and Business Information Systems, University of Twente, P.O. Box
More informationBayesian Active Learning With Basis Functions
Bayesian Active Learning With Basis Functions Ilya O. Ryzhov Warren B. Powell Operations Research and Financial Engineering Princeton University Princeton, NJ 08544, USA IEEE ADPRL April 13, 2011 1 / 29
More informationAPPLICATION OF RECURRENT NEURAL NETWORK USING MATLAB SIMULINK IN MEDICINE
ITALIAN JOURNAL OF PURE AND APPLIED MATHEMATICS N. 39 2018 (23 30) 23 APPLICATION OF RECURRENT NEURAL NETWORK USING MATLAB SIMULINK IN MEDICINE Raja Das Madhu Sudan Reddy VIT Unversity Vellore, Tamil Nadu
More informationCoordinated Aggregation of Distributed Resources
Coordinated Aggregation of Distributed Resources UC Berkeley October 13, 2011 Coordinated Aggregation 1 of 39 Co-conspirators Anand Subramanian, Manuel Garcia, Josh Taylor [Berkeley Students] Duncan Callaway,
More informationA Hierarchy of Suboptimal Policies for the Multi-period, Multi-echelon, Robust Inventory Problem
A Hierarchy of Suboptimal Policies for the Multi-period, Multi-echelon, Robust Inventory Problem Dimitris J. Bertsimas Dan A. Iancu Pablo A. Parrilo Sloan School of Management and Operations Research Center,
More informationIntegrated schedule planning with supply-demand interactions for a new generation of aircrafts
Integrated schedule planning with supply-demand interactions for a new generation of aircrafts Bilge Atasoy, Matteo Salani and Michel Bierlaire Abstract We present an integrated schedule planning model
More informationLogic, Optimization and Data Analytics
Logic, Optimization and Data Analytics John Hooker Carnegie Mellon University United Technologies Research Center, Cork, Ireland August 2015 Thesis Logic and optimization have an underlying unity. Ideas
More informationOptimum Repartition of Transport Capacities in the Logistic System using Dynamic Programming
Theoretical and Applied Economics Volume XVIII (011), No. 8(561), pp. 17-0 Optimum Repartition of Transport Capacities in the Logistic System using Dynamic Programming Gheorghe BĂŞANU Bucharest Academy
More informationAbstract. 1. Introduction
Abstract Repairable system reliability: recent developments in CBM optimization A.K.S. Jardine, D. Banjevic, N. Montgomery, A. Pak Department of Mechanical and Industrial Engineering, University of Toronto,
More informationReinforcement Learning
Reinforcement Learning Model-Based Reinforcement Learning Model-based, PAC-MDP, sample complexity, exploration/exploitation, RMAX, E3, Bayes-optimal, Bayesian RL, model learning Vien Ngo MLR, University
More informationCIV3703 Transport Engineering. Module 2 Transport Modelling
CIV3703 Transport Engineering Module Transport Modelling Objectives Upon successful completion of this module you should be able to: carry out trip generation calculations using linear regression and category
More informationUncertainty in energy system models
Uncertainty in energy system models Amy Wilson Durham University May 2015 Table of Contents 1 Model uncertainty 2 3 Example - generation investment 4 Conclusion Model uncertainty Contents 1 Model uncertainty
More informationLINEAR PROGRAMMING MODULE Part 1 - Model Formulation INTRODUCTION
Name: LINEAR PROGRAMMING MODULE Part 1 - Model Formulation INTRODUCTION In general, a mathematical model is either deterministic or probabilistic. For example, the models and algorithms shown in the Graph-Optimization
More informationTown of Barnstable. Department of Public Work. Snow and Ice Control Operations Plan
Town of Barnstable Department of Public Work Snow and Ice Control Operations Plan I. Mission: The mission of the Department of Public Works is to remove accumulations of snow and ice from town, county
More informationA Joint Tour-Based Model of Vehicle Type Choice and Tour Length
A Joint Tour-Based Model of Vehicle Type Choice and Tour Length Ram M. Pendyala School of Sustainable Engineering & the Built Environment Arizona State University Tempe, AZ Northwestern University, Evanston,
More informationAdministration. CSCI567 Machine Learning (Fall 2018) Outline. Outline. HW5 is available, due on 11/18. Practice final will also be available soon.
Administration CSCI567 Machine Learning Fall 2018 Prof. Haipeng Luo U of Southern California Nov 7, 2018 HW5 is available, due on 11/18. Practice final will also be available soon. Remaining weeks: 11/14,
More informationTOWN OF GRAND FALLS WINDSOR
TOWN OF GRAND FALLS WINDSOR DEPARTMENT OF ENGINEERING AND WORKS FREQUENTLY ASKED QUESTIONS FOR WINTER OPERATIONS... What streets are the first to be plowed & salted/sanded? The roads crews are each assigned
More informationR O B U S T E N E R G Y M AN AG E M E N T S Y S T E M F O R I S O L AT E D M I C R O G R I D S
ROBUST ENERGY MANAGEMENT SYSTEM FOR ISOLATED MICROGRIDS Jose Daniel La r a Claudio Cañizares Ka nka r Bhattacharya D e p a r t m e n t o f E l e c t r i c a l a n d C o m p u t e r E n g i n e e r i n
More informationA Representational Paradigm for Dynamic Resource Transformation Problems
A Representational Paradigm for Dynamic Resource Transformation Problems Warren B. Powell, Joel A. Shapiro and Hugo P. Simao January, 2003 Department of Operations Research and Financial Engineering, Princeton
More informationLecture 15. Dynamic Stochastic General Equilibrium Model. Randall Romero Aguilar, PhD I Semestre 2017 Last updated: July 3, 2017
Lecture 15 Dynamic Stochastic General Equilibrium Model Randall Romero Aguilar, PhD I Semestre 2017 Last updated: July 3, 2017 Universidad de Costa Rica EC3201 - Teoría Macroeconómica 2 Table of contents
More informationCS 188: Artificial Intelligence Spring Today
CS 188: Artificial Intelligence Spring 2006 Lecture 9: Naïve Bayes 2/14/2006 Dan Klein UC Berkeley Many slides from either Stuart Russell or Andrew Moore Bayes rule Today Expectations and utilities Naïve
More informationVehicle Routing with Traffic Congestion and Drivers Driving and Working Rules
Vehicle Routing with Traffic Congestion and Drivers Driving and Working Rules A.L. Kok, E.W. Hans, J.M.J. Schutten, W.H.M. Zijm Operational Methods for Production and Logistics, University of Twente, P.O.
More informationA Progressive Hedging Approach to Multistage Stochastic Generation and Transmission Investment Planning
A Progressive Hedging Approach to Multistage Stochastic Generation and Transmission Investment Planning Yixian Liu Ramteen Sioshansi Integrated Systems Engineering Department The Ohio State University
More informationFORECASTING STANDARDS CHECKLIST
FORECASTING STANDARDS CHECKLIST An electronic version of this checklist is available on the Forecasting Principles Web site. PROBLEM 1. Setting Objectives 1.1. Describe decisions that might be affected
More informationArtificial Intelligence & Sequential Decision Problems
Artificial Intelligence & Sequential Decision Problems (CIV6540 - Machine Learning for Civil Engineers) Professor: James-A. Goulet Département des génies civil, géologique et des mines Chapter 15 Goulet
More information10/18/2016 The Hoosier Co. Inc W. 86th Street, Indianapolis, IN
10/18/2016 The Hoosier Co. Inc. 5421 W. 86th Street, Indianapolis, IN 46268 1 Today s Topics Weather impacts What is RWIS? How old is this tool? Key RWIS Technology/Innovation Key RWIS sensing parameters
More informationIntegrating advanced discrete choice models in mixed integer linear optimization
Integrating advanced discrete choice models in mixed integer linear optimization Meritxell Pacheco Shadi Sharif Azadeh, Michel Bierlaire, Bernard Gendron Transport and Mobility Laboratory (TRANSP-OR) École
More informationThe Knowledge Gradient for Sequential Decision Making with Stochastic Binary Feedbacks
The Knowledge Gradient for Sequential Decision Making with Stochastic Binary Feedbacks Yingfei Wang, Chu Wang and Warren B. Powell Princeton University Yingfei Wang Optimal Learning Methods June 22, 2016
More informationChanges in the Spatial Distribution of Mobile Source Emissions due to the Interactions between Land-use and Regional Transportation Systems
Changes in the Spatial Distribution of Mobile Source Emissions due to the Interactions between Land-use and Regional Transportation Systems A Framework for Analysis Urban Transportation Center University
More informationStochastic Integer Programming
IE 495 Lecture 20 Stochastic Integer Programming Prof. Jeff Linderoth April 14, 2003 April 14, 2002 Stochastic Programming Lecture 20 Slide 1 Outline Stochastic Integer Programming Integer LShaped Method
More informationAESO Load Forecast Application for Demand Side Participation. Eligibility Working Group September 26, 2017
AESO Load Forecast Application for Demand Side Participation Eligibility Working Group September 26, 2017 Load forecasting for the Capacity Market Demand Considerations Provide further information on forecasting
More informationChapter 6 Queueing Models. Banks, Carson, Nelson & Nicol Discrete-Event System Simulation
Chapter 6 Queueing Models Banks, Carson, Nelson & Nicol Discrete-Event System Simulation Purpose Simulation is often used in the analysis of queueing models. A simple but typical queueing model: Queueing
More informationStochastic prediction of train delays with dynamic Bayesian networks. Author(s): Kecman, Pavle; Corman, Francesco; Peterson, Anders; Joborn, Martin
Research Collection Other Conference Item Stochastic prediction of train delays with dynamic Bayesian networks Author(s): Kecman, Pavle; Corman, Francesco; Peterson, Anders; Joborn, Martin Publication
More informationLinear Programming. H. R. Alvarez A., Ph. D. 1
Linear Programming H. R. Alvarez A., Ph. D. 1 Introduction It is a mathematical technique that allows the selection of the best course of action defining a program of feasible actions. The objective of
More informationTown of Oconomowoc Snow & Ice Control Policy
Town of Oconomowoc Snow & Ice Control Policy Introduction The purpose of this policy is to provide a detailed overview of the Town s snow & ice control operations including its goals and objectives. All
More informationChapter 4. Greedy Algorithms. Slides by Kevin Wayne. Copyright 2005 Pearson-Addison Wesley. All rights reserved.
Chapter 4 Greedy Algorithms Slides by Kevin Wayne. Copyright 2005 Pearson-Addison Wesley. All rights reserved. 1 4.1 Interval Scheduling Interval Scheduling Interval scheduling. Job j starts at s j and
More informationProbabilistic Planning. George Konidaris
Probabilistic Planning George Konidaris gdk@cs.brown.edu Fall 2017 The Planning Problem Finding a sequence of actions to achieve some goal. Plans It s great when a plan just works but the world doesn t
More informationOverfitting, Bias / Variance Analysis
Overfitting, Bias / Variance Analysis Professor Ameet Talwalkar Professor Ameet Talwalkar CS260 Machine Learning Algorithms February 8, 207 / 40 Outline Administration 2 Review of last lecture 3 Basic
More informationFederal Aviation Administration Optimal Aircraft Rerouting During Commercial Space Launches
Federal Aviation Administration Optimal Aircraft Rerouting During Commercial Space Launches Rachael Tompa Mykel Kochenderfer Stanford University Oct 28, 2015 1 Motivation! Problem: Launch vehicle anomaly
More informationTypical information required from the data collection can be grouped into four categories, enumerated as below.
Chapter 6 Data Collection 6.1 Overview The four-stage modeling, an important tool for forecasting future demand and performance of a transportation system, was developed for evaluating large-scale infrastructure
More informationThe Road to Improving your GIS Data. An ebook by Geo-Comm, Inc.
The Road to Improving your GIS Data An ebook by Geo-Comm, Inc. An individual observes another person that appears to be in need of emergency assistance and makes the decision to place a call to 9-1-1.
More informationCITY OF EAST PEORIA SNOW AND ICE CONTROL PROGRAM INTRODUCTION
CITY OF EAST PEORIA SNOW AND ICE CONTROL PROGRAM INTRODUCTION The responsibility for providing snow and ice control on East Peoria s 240 lane miles of streets and alleys rests with the Street Divisions
More information1.225 Transportation Flow Systems Quiz (December 17, 2001; Duration: 3 hours)
1.225 Transportation Flow Systems Quiz (December 17, 2001; Duration: 3 hours) Student Name: Alias: Instructions: 1. This exam is open-book 2. No cooperation is permitted 3. Please write down your name
More informationA Benders decomposition method for locating stations in a one-way electric car sharing system under demand uncertainty
A Benders decomposition method for locating stations in a one-way electric car sharing system under demand uncertainty Hatice Calik 1,2 and Bernard Fortz 1,3 1 Department of Computer Science, Université
More informationLandmark LRM. MM and RefPts can be same or different
1 2 This project began in the fall of 2005. At the time CDOT had three different business units maintaining different LRS s. That is, the beginning and ending reference points and the measured length of
More information