平行運算於地下水管理模式之應用

Abstract

本研究應用平行演算於提高非拘限含水層之水量及水質管理模式之計算效率。由於所應用之地下水管理模式整合了兩個優選模式，分別為遺傳演算法和限制型可微分動態規劃，以及地下水數值模擬模式，使得此一整合模式在運作時之計算量十分的龐大，因此有必要發展平行運算以增進計算效率，使得模式能應用於大型之地下水管理規劃中。由於遺傳演算法本身即具有『隱平行』的特性，因此平行演算乃以遺傳演算法作為平行運算的主軸，並以國家高速電腦中心之平行電腦作為硬體計算平台配合使用MPI平行運算軟體，研究成果顯示平行效率十分優良。由結果顯示，若採用50顆CPU參與平行計算，則平行化之非拘限含水層水量管理模式之計算時間約為未平行化之循序模式之1/22，證明了平行運算確實可有效節省計算時間。

This study attempted to solve a groundwater optimal management problem for an unconfined aquifer by performing parallel computation to increase the computational capacity of a hybrid algorithm. The proposed hybrid optimization algorithm integrates a genetic algorithm (GA) and constrained differential dynamics programming (CDDP). Since the groundwater management problem considers fixed and operating costs simultaneously, the computational complexity of the problem is significantly increased. However, the integrated algorithm can be effectively parrallelized since the GA is an ”implicit parallel” algorithm. This study also implements the parallel computation using MPI. Computational results demonstrate that the computational time using fifty CPUs is only 1/22 of the one using a single CPU, demonstrating the high efficiency of the parallel computation.