应用EBP神经网络进行预测的研究

Study of Forecast by Applying EBP Neural Network

针对低阶多层、变拓扑结构的人工神经网络，运用误差反向传播算法，对离散非线性系统的拟合逼近进行了研究．对一个工业实际现象（某地区电力系统负荷情况）使用神经网络动态建模，结果表明，所提出的模型能自动训练学习修正误差，较全面地反映了影响负荷变化的各种因素．并以此模型对电网负荷进行预测计算，预测精度高，收敛速度快，适用于对各种在线离散型非线性系统的预测．

The approximation and fitting of discrete time nonlinear system is studied by error back propagation (EBP) algorithm in accordance with artificial neural networks of low order multilayered, changing topological structure. A dynamic model is constructed in the light of an industrial practical phenomena (load circumstance of power system in some areas). The result indicates that the model above-mentioned may be used to learn automatically and to correct errors. It is used to reflect totally all factors that affect the changing of loads. The model is also used to compute for forecasting. It has a high accuracy and a rapid convergence rate, and so is suitable for all on-line discrete types of nonlinear system.