广义回归神经网络模型在短期电力负荷预测中的应用研究  

Research on application of general regression neural network in short-term load forecasting

作　　者：谷志红[1] 牛东晓[1] 王会青[1] 

机构地区：[1]华北电力大学工商管理学院,河北保定071003

出　　处：《中国电力》2006年第4期11-14,共4页Electric Power

基　　金：高等学校博士点专项基金资助项目(20040079008);河北省自然科学基金资助项目(G2005000584);华北电力大学重大预研基金资助项目

摘　　要：介绍了广义回归神经网络(GRNN)的基本理论,指出其回归的实质就是对平滑参数的优化。考虑到常规差分进化算法容易“早熟,”全局寻优效率偏低,提出了基于优进策略的差分进化算法,利用种群繁衍的有用信息改进子代分布,并引入确定性寻优操作,实现了高效全局搜优。以推广能力作为优化目标,所建的GRNN有很强的非线性拟合能力和优良的预报性能,将其成功地为短期电力负荷预测建模,获得了满意的预测结果。The basic theory of general regression neural network （GRNN） was introduced. It is pointed out that the essential of GRNN is optimizing of smooth parameter. Considering that the common differential evolution algorithm （DE） is easy to ＂early mature＂ and the global optimizing efficiency is lower, a modified evolution algorithm （MDE） is proposed. The modified evolution algorithm advances the performance of global optimization through collecting population information during evolution and at the same time introducing deterministic operation, amending distribution of individuals adaptively. The GRNN-MDE, which is based on DE and provides powerful capacity in non-linear modeling and predicting, is applied to modeling short-term power load forecasting, and the result is satisfied.

关 键 词：负荷预测 广义回归神经网络 差分进化算法 优进策略 

分 类 号：TM715[电气工程—电力系统及自动化]