The total current injected in the system must be zero, therefore the sum of vector entries B j j is zero, è1;:::1èb j j =: Excluding j by means of è1è

Size: px

Start display at page:

Download "The total current injected in the system must be zero, therefore the sum of vector entries B j j is zero, è1;:::1èb j j =: Excluding j by means of è1è"

Ferdinand Snow
5 years ago
Views:

1 1 Optimization of networks 1.1 Description of electrical networks Consider an electrical network of N nodes n 1 ;:::n N and M links that join them together. Each linkl pq = l k joints the notes numbered p and q and has the the resistivity R pq = R k. The potential diæerences v on some nodes are prescribed that is batteries are added to the networkatsomelinks. The current sources are added at some nodes as well. The potentials u n and currents j k in the network are found from the following linear system. The potential diæerence u pq = u k in each linkis e k = u p, u q + v k or e = Au + Bv è1è where u =èu 1 ;:::u N è is the vector of potentials at the nodes, v =èv 1 ;:::v M è is the vector of added potential diæerences èbatteriesè, A is the matrix of the structure that shows whether two nodes are connected. It has entries equal to zero, one or negative one, and each row has only two nonzero entries equal to one and to negative one, respectively. ç ç ::: 1 :::,1 A = ::: ::: ::: ::: ::: ::: ::: Matrix B is the matrix of sources that shows to which link the battaries are attached It structure is similar to the structure of A. One potential is assumed to be zero, u 1 =. The currents j k in the links are equal to j k = 1 R k e k ; or j = R,1 e = R,1 èau + Bvè è2è where R is the M æ M diagonal matrix of resistivity. Its entries represent the resistances in the links, R k = ç 1 k, where ç k is the conductance of the link. Finally, the Kirchoæ law states that the sum of all current at each node equals zero, or A T j, B j j = è3è where j are the added independent external currents and B j is the matrix of current sources that shows where they are attached and how they are distributed. For example, if two equal positive currents are injected in the nodes numbered one and two, and the negative current is injected into note four, the matrix B j is 1=21 1=2 B j = C,1 A 1

2 The total current injected in the system must be zero, therefore the sum of vector entries B j j is zero, è1;:::1èb j j =: Excluding j by means of è1è and è2è we ænd the equation for the potentials u in the nodes of the network A T R,1 Au = A T R,1 Bv + B j j = b è4è where b = A T R,1 Bv + B j j is the vector or external excitation. that combines both batteries and the current sources. The potentials u at the nodes are u =èa T R,1 Aè,1 b assuming that the matrix A T R,1 A is nondegenerative that is that the all nodes are really connected in one system and the potentials can be deæned. The currents j in the links are j = R,1 A, èa T R,1 Aè,1 b + Bv æ In optimization problem, we refer to equations è1è- è3è or è4è as ëdirect system". 1.2 Optimization of resistances Consider the problems: Choose the resistors R k in the network subject to constraints ç R min ç R k ç R max ç1 to minimize or maximize the potential diæerence between two nodes, or to minimize or maximize the total current at selected links. The minimizing function for these problems can be represented as F = ç T j + ç T u è5è where ç 2 R M and ç 2 R N are given vectors. Using Lagrange multipliers ç 2 R N and ç 2 R M to account for the equations è2è and è3è of the network, respectively, we rewrite the problem as where min min R max J;u ç;ç LèR; J;u;ç;çè LèR; J;u;ç;çè=ç T j + ç T u + ç T èa T j, B j j è+ç T èj, R,1 èau + Bvè: è6è 2

3 is the extended Lagrangian. The stationary of u and corresponds to the conditions èadjoint systemè for ç and ç: Excluding ç from these conditions, we =è ç + A T R,1 ç =; ç + Aç + ç = è7è èa T R,1 Aèç =,ç + A T R,1 ç è8è This equation shows that ç can be interpreted as a potentials in the nodes of the same network excited by the batteries,ç and the current sources,ç. Vector ç is the corresponding current. Notice that the sources in the adjoint system correspond to the goal function è5è and are independent of real sources. The stationarity conditions with respect to the design variables í conductivities ç k = R 1 k are easily derived since R,1 is a diagonal matrix. The optimality of an intermediate value of ç k corresponds to the ç T R,1 èau + Bvè èç + Açè T R,1 èau + Bvè k Since the dependence on ç k is linear, this conditions cannot be satisæed unless the potential diæerence in the corresponding link l k in either the direct or adjoint problem is zero. The last condition would mean that the function F is invariant to this resistance. We conclude that the resistance take only the limiting values R min or R max. The choice depends only on the sign of the product èç + Açè k èau + Bvè k of elements èç + Açè k and èau + Bvè k which represent the currents in the direct and adjoint systems through the kth link. R k = ç Rmax if èç + Açè k èau + Bvè k é R min if èç + Açè k èau + Bvè k é è9è If these current are codirected, the resistance is minimal, and if they are opposite directed, the resistance is maximal. Minimal networks Assume that R max = 1 or that we can remove alink from the system. Then the optimal network consists of only links in which the currents are codirected. How many links survive? 1.3 Energy minimization The special case is minimization or maximization of the total energy of the network or its total resistance. Assume for clearness that there is no battaries v = and the equation for the network is èa T R,1 Aèu = B j j 3

4 and minimize the work of external currents F = u T B j j equal to the energy of the network. The corresponding objective function è5è correspond to the vectors ç =; ç = B j j and the adjoint system becomes èa T R,1 Aèç =,ç =,B j j : Its solution is negative to the solution u of the direct system, ç =,u. The optimality conditions è9è for the resistances showthatr k = R min for all k which is physically evident: To minimize the total resistivity, use the smaller resistances everywhere. Additional constraints on the resistors To make the problem nontrivial, let us add an additional constraint: Assume that the number of each type of resistors is prescribed: There are T resistors with resistivity R min and the rest have the resistivity R max. The stationarity conditions are correspondingly changed: In this case, the smaller current is put through the larger resister and larger current í through smaller resistance. To ænd the layout of the resistors, we need to range the currents and renumber them jj 1 jçjj 2 jç:::çjj M j Then we assign the resistors R min to the links with currents jj M,T jç:::çjj M j and resistors R max to the rest of links. 1.4 Optimization of sources The sources j and v can be optimized as well. Assume that these vectors are free but two linear constraints are prescribed ç T j =1 and è T v =1 è1è where ç and è are given vectors, together with inequalities j ç v ç è11è èthese mean that each element of the corresponding vector is nonnegativeè. Adding these two constraints with the Lagrange multipliers æ and æ, respectively, to the extended Lagrangian and computing the stationarity conditions. The conditions for stationarity of u and j are given by è7è. Stationarity with 4

5 respect to j and v we obtain conditions for the potential and currents in the adjoined system at the points of excitation ç = æèb j è T ç; R,1 ç =,æèbè T è è12è Here, we face the problem: the potentials ç and the currents ç in the adjointed system are completely deæned from the system è7è or è8è and the extra conditions è12è cannot be satisæed. The only remaining possibility is that the optimal excitation corresponds to the boundary of the set of currents set by inequalities è11è. This means that all independent sources j but one are zero èj è opt =è;:::;1=ç k ;:::;è and similarly, èvè opt =è;:::;1=è k ;:::;è 1.5 Optimization of sources and resistances When the resistances are optimized along with the sources, the rule for optimal resistances must be added to the equations, this rule determines the matrix R. Here me may expect that the resistivity layout is chosen to equalize the impact of several diæerent current sources. 1.6 Numerical solution The optimization problems are solved using iterative scheme. Consider optimization of the resistances. Assigning the design parameters R k, one solves the direct and adjoint problem and checks the optimality conditions è9è for R k. If some of them are not satisæed, the design changes; then the iteration repeats until all conditions are satisæed. Similar scheme is used for optimization of sources. 5

also has x æ as a local imizer. Of course, æ is typically not known, but an algorithm can approximate æ as it approximates x æ èas the augmented Lagra

also has x æ as a local imizer. Of course, æ is typically not known, but an algorithm can approximate æ as it approximates x æ èas the augmented Lagra Introduction to sequential quadratic programg Mark S. Gockenbach Introduction Sequential quadratic programg èsqpè methods attempt to solve a nonlinear program directly rather than convert it to a sequence