針對BP 算法易陷入局部最優(yōu),提出將一種新的混沌遺傳算法(CGA) 用于全局優(yōu)化給水管網狀態(tài)神經網絡模型的初始權閾值. 該算法將混沌搜索與自適應遺傳算法相結合,根據(jù)混沌運動的初值敏感性、內在隨機性以及遍歷性的特點,通過混沌映射搜索自適應遺傳算法的較優(yōu)初始種群,并利用自適應遺傳算法進一步尋優(yōu),對混沌映射和遺傳進化進行循環(huán)計算直至達到最大進化代數(shù),最終獲得BP 模型的較優(yōu)權閾值. 實例分析結果表明,與自適應遺傳算法(A GA) 相比,該算法搜索穩(wěn)健,全局搜索能力強,并且新算法優(yōu)化模型具有更高的預測性能.

As back2propagation (BP) neural network suffer s f rom the existence of many local minima , a
novel chaos genetic algorit hm ( CGA) combining chaotic search wit h self2adaptive genetic algorit hm(A GA) was proposed for globally optimizing t he initial weight and t hreshold of neural network2based state model for pipe network. Wit h the characteristics of sensitive dependence on initial conditions , int rinsic stochastic property and ergodicity of chaotic motion , chaotic map was used to search the optimal initial pop ulation of A GA , t hen A GA was employed to optimize t he weight and threshold. The operations of chaotic map and A GA evolution were performed until CGA reached t he maximum generation , and t he optimal weight and t hreshold of BP neural network was achieved. The case analysis shows t hat CGA has bet ter global convergence and st ronger running robustness t han A GA , and t hat the model optimized by CGA has higher forecasting performance.