神经网络在短期电力负荷预测中的应用

Applications of neural network to short-term electric load forecasting

给出了改进的BP网络和RBF网络的构造过程和训练方法.在改进的BP网络中不仅加入了动量项和变步长法,而且在模型中合理地考虑了影响负荷变化的主要气象因素,使其能够适应天气的变化.在RBF网络中,为了克服传统K均值聚类法局部寻优的缺陷,采用了正交最小二乘法选取RBF中心.利用改进的BP网络和RBF网络进行了短期电力负荷预测,并对训练的收敛速度和预测精度进行了分析.比较两种模型,RBF网络比BP网络更具有实用性和可开发性.

The constituting process and training method of the improved BP and RBF neural networks are put forward. In the improved BP network, the momentum item and the algorithm using variable step length are employed. Furthermore, main meteorological factors influencing load changes are included in proposed mathematical model to meet weather variations. In the RBF network, to overcome the defects of traditional K-means scheme with local search, an orthogonal least square algorithm is used to select RBF center. By the improved BP and RBF neural networks, short-term electric load is forecast and training convergence rate and forecasting precision are analyzed. Comparing with BP network, the RBF has more advantages in practical applications.