Optimization of CC & HPV power plants with HD-MPC

Size: px

Start display at page:

Download "Optimization of CC & HPV power plants with HD-MPC"

Willis Ball
5 years ago
Views:

1 Optimization of CC & HPV power plants with HD-MPC D. Faille, F. Davelaar (EDF) A. Tica, H Guéguen (SUPELEC) HD-MPC Workshop,Leuven,24/06/2011

2 Agenda Introduction New Challenges for power generation Motivation for HD-MPC solution Optimal control of CC & HPV Operation Process Description & Control Objectives HD-MPC solutions Results Conclusion, Lessons Learned 2

3 Introduction Driving factors for Power Plant Optimization Grid requirements HD-MPC solutions for power plant optimization 3

Drivers for Power Plant Optimization World Demand Electricity Increases Raw Material Prices Increases Climate Changes Electricity Market

4 Drivers for Power Plant Optimization World Demand Electricity Increases Raw Material Prices Increases Climate Changes Electricity Market Liberalization Grid Stability Environment (air, water, waste) Power Plants Optimizations Grid (Voltage, frequency) Cost (Fuel, Maintenance) 4

5 Ex: Grid Requirements. (UCTE) Po : Daily program k F : Primary. freq. response 100 Dégagement de la réserve primaire pour un DF de 200 mhz P = Po + k F + N pr DP en % de RPM t en s ec 100 % in 30s. Active Power Set-point : N Pr : Secondary. freq. response 1.5 Rampe de téléréglage 1 DP en % de P R *Pr in 800 s t e n s e c 5

6 HD-MPC for Power Plant optimization Goal : Find new solutions for Power Plant Operations adapted to : Large Interconnected Systems Multiple objectives Non linear dynamics Different Time Scales 6

7 Hydro Power Valley Process Description & Control Objectives Modeling & HD-MPC solutions Results 7

8 Process & Control Description 4 run-of-river power plants 5 Reservoirs Environment Level, flow rate constraints HPV Optimizations Grid Po + kdf + NPr Cost (Limited Storage ) 8

9 Process & Control Description Control Room Po TSO rank 3 Control (PHV) Plant 1 Plant 2 Plant 3 Plant 4 Plant 5 Plant 6 Plant 7 rank 2 Control (PA) Group Control Dam Control rank 1 Control (APG, APS) Rank 3 : Alert, Supervision, Dispatching, Valley Control Room, Communication with TSO Rank 2 : Level/flow control, Machine Commitment, RSFP, RSFT Rank 1 : Safety functions, group start-up & shutdown 9

10 Modeling of a River Reach x : longitudinal coordinate [m] t : time [s] S(x,t) : wetted section [m 2 ] P(x,t) : wetted perimeter [m] Q(x,t) : volumetric flow [m 3 /s] q lat : lateral inflow per unit length [m 2 /s] z(x,t) : height [m] K : Strickler Coefficient [SI] k : lateral inflow coefficient abscisse X abscisse z h Q I Momentum conservation equation x S P Mass Conservation equation Q t + x Q S 2 + z gs x + J = k. q lat. v S t + Q x = q lat J = S 2 Q Q R 4/3 h K 2 R h = S( x, t) P( x, t) 10

11 Control Models Spatial Discretisation of Saint-Venant Equations 11

12 Simulation platform Platform made of : - OPC server - HMI (editor and supervisor) - Alarms - Archived data measurements - Matlab + OPC Toolbox Alarms Historical Data Process Data [OPC] Supervision HMI HMI Editor OPC server Controller Simulation Model OPC client HPV CONTROL OPC client HPV MODEL OPC client HMI 12

13 HD-MPC solutions Methods are under development 1) Centralized MPC controller Reference 2) Two level Hierarchical MPC solution 3) Sensitivity-driven Gradient method (RWTH) 4) Multiple shooting Method (KUL) 5) Game theory (USE, UNC) 6) 13

14 2-Level MPC solution Pref Upper Level HPV Control Xobs, Qlatobs Uref,Yref Plant 4 Control Plant 5 Control Plant 6 Control Plant 7 Control Hs, Pe Qt Lower Level 14

15 Results (Two level hierarchical MPC) Perturbation is correctly rejected when no constraints are active. Control performance decreases when constraints are active. Load following Load following with perturbation 15

16 Combined Cycle Start-up Process Description & Control Objectives Modeling & HD-MPC solutions Results 16

17 Process Description & Control 1 GT-1 HRSG -1ST 1 Pressure Level Environment NOx, CO2 CC SU Optimizations Grid Startup Time Cost Fuel, Water, Life Consumption 17

18 Process & Control Description Unit : General Sequence & Control Group : Sequence & Control Subsystem (Feedwater, Fuel,.) Rank 1 : Sequence & Control of components (Pump, ) 18

19 CCPP modeling: Modelica CCPP model: developed in Modelica language using Modelica simulation environment, Dymola Modelica: modeling language for large, complex and heterogeneous physical systems: Free language Object-oriented Equation based Multi-domain 19 19

CCPP model - Dymola 1-1-1 configuration (GT-HRSG-ST) One single level of pressure (HP) Equations derived from first principles (mass, energy and momentum balances) Model complex: ~ 1500 equations

20 CCPP model - Dymola configuration (GT-HRSG-ST) One single level of pressure (HP) Equations derived from first principles (mass, energy and momentum balances) Model complex: ~ 1500 equations Validated against experimental data Includes a model of thermal and mechanical stress in critical components (superheated steam header and ST rotor) Smooth model have been developed for optimization 20 20

21 CCPP - Plant inputs/outputs load of GT Inputs: feed water flow rate for the HRSG circuit power of GT fuel flow in GT drum level Outputs: water flow feeding the desuperheater position of bypass valve position of throttle valve generator grid breaker of the steam turbine temperature of steam in the superheater steam flow header stress temperature of steam in the SL power of ST frequency of ST rotor stress of ST 21 21

22 GT Load profile Optimization CCPP non-linear model: x &(t ) = f( x(t ),u,l(t )) x = model states (temperatures, pressures, enthalpy...) u = model inputs (feed water, feed DSH...) 1.2 L = GT load 1.0 initial load L(t0) = Lm full load L(tf ) = LM 0.8 h=2, k=400, p=10, r=3000, Li=0.3 h=2, k=400, p=10, r=3000, Li=0.5 h=4, k=400, p=20, r=3000, Li=0.5 h=2, k=200, p=10, r=2000, Li=0.5 Assumption: GT Load is a parameterized function: L(t,q) sum of 2 Hill functions 2-Hill functions L( t, q) t h = L + ( L L ) + ( L L ) m i m h h M i t + k t p t p + r p Time [s] q=[h, k, p, r, L i ] 22 22

23 GT Load profile optimization Minimum time problem definition: min q,t f subject to the constraints: J t = f dt t 0 Parameters to be optimized: tf, q=[h,k,p,r,li] x(t & ) = f ( x(t ),u,l(t,q )) L(t f,q ) L M ε1 f ( x(t f ),u,l(t f,q )) ε2 h( x(t )) 0, t 0 t t f constraints on the state variables (temperatures, pressures, stresses) 23 23

24 Optimization Results Optimal start-up is 20 minutes faster than a classical start-up (with a constant GT load ramp rate satisfying the stress limit) standard GT Load profile optimal GT Load profile 1.0 standard consumption optimal consumption GT Load [normalized] Fuel consumption [normalized] Time [s] Time [s] reduction of the operating costs due to lower fuel consumption ( 20 %) 24 24

25 Optimization Results Rotor stress: standard start-up optimal start-up 110 Header stress: standard start-up optimal start-up Rotor Stress [MPa] Header Stress [MPa] Time [s] Time [s] 25 25

26 Distributed Approach Model decomposition GT Load p,h Ot GT+ HRSG w ST S1 S2 2 profiles optimization (GT Load, Ot) Physical decomposition: 2 Dymola models GT Load Ot S1 w S2 p,h 26

27 Hierarchical Approach High level: Minimal time: uh* = L1(q,t) Th HL sample time t f min J = dt q, t f t0 Optimization high level Optimization low level Plant uh* ul* x(t0+th) Low level: Quadratic: min u ( q, t ) J = t 0 + T h ( u t 0 ( q, t ) u * h ( q, t 0 + T c )) 2 dt ul* = L2(q,t), Tc Th Tl LL sample time 27

28 Summary A Modelica model based on the ThermoPower library is developed and used for start-up optimization The start-up optimization method is based on parameterized functions This approach is used to minimize the start-up time, while keeping the constraints within their limits The idea of solving the start-up problem by optimizing a parameterized function can be expanded in many ways (e.g. by considering other types of functions, by using as optimization signals other model inputs (ST throttle) Further work: use this optimization procedure in hierarchical and distributed feedback control algorithms (HD-MPC) 28 28

General conclusion & lessons learned Solutions are still under development, HD-MPC solutions can be applied naturally in the HPV case. HD-MPC Optimization with Modelica CCPP models is difficult.

29 General conclusion & lessons learned Solutions are still under development, HD-MPC solutions can be applied naturally in the HPV case. HD-MPC Optimization with Modelica CCPP models is difficult. Library developed for simulation are not suited for optimization (discontinuities) Initialization problems with Dymola Analytical Jacobian is not available for complex Dymola models Unfeasible simulation occurs (water-steam table) 29

Modelling of Primary Frequency Control and Effect Analyses of Governing System Parameters on the Grid Frequency. Zhixin Sun

Modelling of Primary Frequency Control and Effect Analyses of Governing System Parameters on the Grid Frequency Zhixin Sun Outline Introduction Mathematical Model of Primary Frequency Control Dynamic Test