基于聚类分析和PSO优化的神经网络短期负荷预测研究

Study on Short-term Load Forecasting Based on Clustering Analysis and Optimized Neural Network by PSO

为了提高电力系统短期负荷预测的精度,本文在分析传统的负荷预测模型在实际应用中存在问题的基础上,提出了一种新的预测模型:基于聚类分析和粒子群优化的BP神经网络模型。由于负荷具有波动性大、日周期性强等特点,对初始负荷数据进行预处理,按时段对数据空间进行划分,对每个子空间的数据分别建模,可以大幅度的提高神经网络的预测精度和泛化能力,同时利用惯性权重线性微分递减的粒子群算法优化神经网络的连接权值和阀值,可以提高神经网络的全局搜索能力和收敛速度。以某市公布的全网负荷数据进行预测验证,证明了此方法所建立的模型的合理性和有效性。

In order to improve the accuracy of short-term load forecasting,this paper proposes a new forecasting model on the basis of analyzing the problems of the traditional load forecasting methods in practical applications,which is a BP neural network model based on cluster analysis and particle swarm optimization.Because of the strong volatility and periodicity about power load,we first preprocess the initial load dates,and divide the date space by period and modeling each subspace date,thus significantly improving the neural network prediction accuracy and generalization ability.And optimizing neural network connection weights and thresholds with particle swarm optimization can improve the global search ability and convergence speed of the neural network.With the whole network load dates of some city,the rationality and effectiveness of the model established by this method be well proved.