优化动态递归小波神经网络短期负荷预测模型

Short-term Load Forecasting Model Based on Optimized Dynamic Recurrent Wavelet Neural Network

提出了优化动态递归小波神经网络(dynamic recurrent wavelet neural network,DRWNN)短期负荷预测模型。与常规小波神经网络相比,DRWNN有两个关联层,关联层节点起存储网络内部状态的作用;模型构造过程中增强了网络的前馈与反馈联接,形成多层次的网络递归。采用分布估计算法和遗传算法相融合对DRWNN进行优化,融合实质是在解空间"宏观"和"微观"两个层面进行寻优,可克服DRWNN陷入局部最小,提高DRWNN的泛化能力。对两类不同负荷系统日、周预测仿真测试,验证了模型能有效提高预测精度。

An optimized DRWNN （dynamic recurrent wavelet neural network） model for STLF （short--term load forecasting） is constructed in this paper. Compared with conventional wavelet neural network, DRWNN owns two context layers, nodes of which can save internal state of network; The feed--forward connection and feedback connection are increased, which forms recursion from multi-level. The DRWNN is optimized by the combining estimation of distribution algorithm with genetic algorithm, essence of which is searching the opti- mal solution from microscopic and macroscopic level, and it can avoid DRWNN immersing in the local minimal points and improve generalization ability. Two kinds of load systems are used in case study, and the testing results show the proposed model can effectively improve the precision of STLF.