基于相空间重构和随机配置网络的电力负荷短期预测

Short-Term Power Load Forecasting Based on Phase Space Reconstruction and Stochastic Configuration Networks

针对配电网负荷随时间空间变化的非线性特征导致短期负荷预测精度低和模型训练时间成本高的问题,设计了一种基于相空间重构(phase space reconstruction,PSR)和随机配置网络(stochastic configuration networks,SCN)的电力负荷短期预测模型.首先将配电网数据中与负荷相关的气象数据通过主元分析法(principal component analysis,PCA)进行数据降维,并与负荷序列组合成多变量的时间序列,运用混沌时间序列理论,通过互信息法和虚假近邻法求取参数并重构相空间,最后使用随机配置网络预测电力负荷.采用欧洲电网公开数据集的历史负荷和气象数据验证所提方法,结果表明,与网格搜索法优化的支持向量机(support vector machines,SVM)、反向传播神经网络(back propagation neural networks,BP)、长短期记忆网络(long short-term memory network,LSTM)和整合移动平均自回归(autoregressive integrated moving average,ARIMA)相比,所设计方法具有智能化水平高、运算高效的特点,有一定的实用价值.

Aiming at the problem of the non-linear characteristics of the distribution network load changing with time and space,the short-term load forecasting accuracy is insufficient and the model training time cost is high.In this paper,a shortterm load forecasting model based on phase space reconstruction(PSR)and stochastic configuration network(SCN)is designed.Firstly,the meteorological data related to the load in the distribution network data is reduced by principal component analysis(PCA),and is combined with the load sequence to form a multivariable time series.Using chaotic time series theory,the paper reconstructs the phase space through mutual information method and false nearest neighbor method,and finally uses stochastic configuration network to predict power load.The proposed method is verified with historical load and meteorological data of public data sets of European power grid.The results show that,compared with the support vector machine(SVM)optimized by the grid search method,backpropagation neural network(BP),long and short-term memory network(LSTM),and autoregressive integrated moving average(ARIMA),the proposed method can complete load forecasting relatively,accurately and efficiently.The analysis of the calculation example verifies that the proposed method has the characteristics of high level of intelligence and efficient operation,and has certain practical value.