傳統上使用序率動態規劃分析多水庫聯合操作時，且每一狀態變數之分類等級多，則在計算上很容易超出電腦的容量，造成所謂「維度障礙」問題。而多水庫系統聯合操作即常面臨此問題，因此本研究將遺傳算法引入序率動態規劃，不僅可解決傳統序率動態規劃的維度障礙問題，並可增加最佳化搜尋速度，同時將序率動態規劃實際應用於多水庫聯合最佳化操作模式，使水資源有效利用，本研究將以翡翠、石門兩水庫之並聯操作為案例研究。 Dynamic programming (DP) based on Bellman's Principle of Optimality is used extensively in the optimization of water resources system. However, the larger the number of state variables, the more combinations of discrete states that have to be surveyed at each stage. This may cause the problem of "curse of dimensionality" on a computer while optimizing a large-scale system. As compared to conventional optimization models, GA can handle complex problems with relative ease and can obtain appropriate solutions within reasonably low computation time. The main purpose of this study is to propose a decomposition SDP-based approach to handle the problem of "curse of dimensionality". A multiple-reservoir case would be selected to prove the applicability of the approach based on the combination of SDP and GA.
Multiple reservior system