决策树和粒子群算法在日前电价预测中的应用

Day-Ahead Electricity Price Forecasting Using Decision Tree and Particle Swarm Optimization

由于日前24点电价特性差异较大,采用单一模型很难描述,提出了一种基于相似点的日前电价预测新方法。将数据空间按时点划分成子24个子空间,并定量考虑对电价造成影响的相关因素,利用改进决策树技术对子空间的历史数据进行自动聚类,再通过粒子群算法训练各相关因素的最优权值,大大增加了选择相似点的可信度,仿真结果表明该方法能有效提高预测精度。

The features of electricity prices differ greatly in day-ahead 24 points in time, and it is hard to describe the features by a single model, so a new algorithm to forecast day-ahead electricity price is proposed. In accordance with points in time the data space is divided into 24 sub-spaces and the factors related to electricity price are quantitatively considered. The data in sub-space are automatically clustered by means of improved decision tree, and then the best connection weight of every factor is trained by particle swarm optimization （PSO）. The reliability of select the similar points is greatly improved. Foresting results show that the proposed method can improve forecasting accuracy effectively.