短期负荷预测Volterra滤波器间隔采样模型

Interval Sampling Short-Term Load Forecasting Model Based on Volterra Filters

针对现有Volterra滤波器模型按混沌轨道逐点训练的模式易发生训练不充分或过拟合现象并最终影响短期负荷预测效果的问题,提出了依据相空间邻近轨道演化相似性特点,建立基于高阶非线性Volterra滤波器(HONFIR)的短期负荷预测多步预测模型(MSF-HONFIR)。通过定义距离相似度、趋势相似度来衡量轨道演化相似性,提出了负荷吸引子邻近轨道判别的新方法。在MSF-HONFIR模型基础上将原始负荷序列分解为多个子序列并分别对各个子序列建立预测模型,显著削弱了系统累积误差。短期负荷预测仿真结果表明MSF-HONFIR模型的多步预测性能优于原有HONFIR模型。

To avoid the disadvantage of the existing Volterra filters model, which prediction performance are easily affected because the point-by-point training pattern along the chaotic orbit are prone to inadequate training or over training, the short-term electrical load multi-step-forecasting model （MSF-HONFIR） based on the higher-order nonlinear Volterra filter（HONFIR） is first constructed on account of the similarity of the evolve tendency of the neighbor orbits in this paper. Then the method of choosing neighbor orbits in the phase space is presented by considering the Euclidean distance and the evolve tendency. In addition, the original load data is divided into several sub-series and the MSF-HONFIR is then applied to the sub-series, by which the system accumulation error is greatly decreased. The simulation results of practical load forecasting demonstrate that the performances are improved compared to the HONFIR method.