一种多变量时间序列的短期负荷预测方法研究

针对短期负荷影响因素多的特点提出了电力短期负荷的多变量时间序列预测方法,并根据单变量时间序列的延时重构对由历史负荷序列及其相关因素序列所构成的多变量时间序列进行了相空间重构,采用互信息法计算了各子序列的延迟时间,各子序列的嵌入维数则运用平均一步绝对误差和最小一步绝对误差进行选取,然后通过RBF神经网络的非线性映射能力进行电力短期负荷预测.研究结果表明多变量时间序列的预测效果相对于单变量序列有较大提高.

Method of Multivariate Time Series of Short-Term Load Forecasting

To counter the characteristics of the influencing factors of the short-term electric load, a model for forecasting the short-term load is presented by multivariate chaotic time series. The model first draws on the phase space reconstruction of univariate chaotic time series, in which phase space of multivariate time series for electric load time series and its influencing factors time series is reconstructed. The good time delay is chosen for each scalar time series by mutual information. The method to obtain the minimum embedding dimension is based on minimum forecasting errors. Then, the non-linear approaching capacity of the Radial Basis Function (RBF) neural network is applied to forecast the load on a day ahead and a week ahead. As the results of an example of factual forecasting show the method presented in this paper can work more effectively.