用相似性原理及人工神经网络预测电价

提出了用相似性原理和BP神经网络来预测日前市场出清电价的新方法,该法尤其适用于只能获得有限原始数据的情况。运用相似性原理对人工神经网络的训练模型进行选择,使其有与预测日相似的负荷特征。用选择出的相似训练模式对选定的BP神经网络进行训练,通过BP神经网络的反向传播过程不断修正模型中的神经元连接权值和阀值,实现对未来24 h市场出清电价的有效预测。对周末和节假日采用了峰值处理步骤后,此方法更加完整。最后以美国宾西法尼亚州、新泽西州和马里兰州公布的2002年数据进行了模型训练和预测,结果表明该方法所建立的预测模型具有较高的预测精度。

New Method Based on Analogous Theory and Neural Network for Market Clearing Price Forecasting

In this paper,a new method to forecast the short-term market clearing prices based on analogous and BP neural network is presented.It provides more accurate results relatively especially when the data source is very limited.To gradually improve the forecasting accuracy,finding out a exact learning pattern which have the analogous characteristic as the forecasting day is necessary,and then,using the pattern which has been picked as learning sample to train BP neural network.As for some adjustment of the ANN itself, the day-ahead MCPs come to be forecasted effectively.The peak price value adjustment process for weekend and festival days makes the method more complete.