平行演算於多水庫系統序率最佳操作之應用

Abstract

本研究的主要目的在發展一個適合於平行計算的水庫序率操作優選模式。 其主要的精神為將一長時列的序率最佳控制問題，分解(decompose)成幾 個較短時列的次問題，因此就解題的觀點而言，可分成兩個層次(two- level)，上層(high-level)為如何適當的將原問題分解為次問題，其下 層(low-level)為如何計算各個分解後之次問題，因此上層是一個參數最 佳化的問題，而下 層的次問題和原問題相似，同為最佳控制問題，只是 操作時間較短。 經分解後各次問題最佳解的總和，應等於原問題未分解 時之最佳解。演算的方法上層採用二次規劃，下層則採用限制型序率微分 動態規劃法。本研究目前在個人電腦上發展，為了模擬平行演算的環境， 各次問題並不依其時間先後次序處理，而採交錯的方式計算，以驗證各次 問題的獨立性， 以便往後能於平行電腦上執行。 This research develop a parallel computation model for the stochastic optimalreservoirs operation. The basic prnciple of the algorithm is to decompose the originalstochastic optimal control problem into several subproblems with the same problemstructure but less time horizon. A hierarchical two level optimization algorithmis applied to solve the decomposed problem. The top level problem is a parameteroptimization problem and is solved by the Quadratic Programming. The low- levelproblem is a stochastic optimal control problem and is solved by a ConstrainedStochastic Differential Dynamic Programming (SCDDP). The model is implementedon a personal computer. The computational environment is not capable to developa true parallel computational. To verify the algorithm is ready for a parallel computation,it is necessary to prove that all the subproblem are computationally independent.This study has demonstrated the computational independence by computing the optimalsolution for each subproblem with random sequence.