基于神经网络矫正的非线性短时负荷预测模型

为了解决传统神经网络负荷预测模型中,当预测日天气出现快速变化时预测误差随之增加的问题,提出了一种改进的未来一小时实时负荷预测模型。在该模型中,预测负荷通过对预测日的类似日负荷数据加一个矫正值来获得,矫正值从神经网络产生,网络结构得到简化。由于采用在线实时学习方式,该模型可以学习快速的天气变化和预测误差之间的关系,减小预测误差。仿真结果验证了该模型的有效性。

Nonlinear short-term load forecasting model based on neural networks correction

In traditional load forecasting model of neural networks, the complicated network structure and overfull input variables affect the forecasting effect. Moreover, while weather varies rapidly on the forecasting day, the load curve changes greatly, and the forecasting error will increase evidently. In order to overcome the shortcomings, an improved one-hour-ahead load forecasting model of real time is proposed. In the model, the forecasting load can be got by adding a correction value to similar day load data of forecasting day. The correction values are obtained by neural networks, and the network structure becomes simple. Owing to the manner of online learning of real time, the model can learn the relation between rapid variety of weather and forecasting errors, and forecasting errors are decreased. The simulation results show the validity of the forecasting model.