用奇异性的短期负荷预测混沌方法优化参数

为优化电力系统短期负荷预测的混沌相空间重构的线性方法中的3个参数，以多元线性回归分析和矩阵计算的奇异性理论为基础，通过数值实验得到了优化的参数．发现首先应该根据取样序列的“平稳性”和“奇异性”，特别是避免“奇异性”来优选延迟时间；其次，根据嵌入窗长为24h来优选嵌入相空间的维数；最后，按照嵌入相空间维数的3～5倍来选择邻近矢量的数目，而不是按照固定距离来选择邻近矢量数目．

Parameter Optimization Based on Singularity in Chaotic Forecasting of Short Term Load

Based on the theories of multivariate linear regressive analysis and singularity in matrix calculation, three parameters in linear regression of chaotic phase-space reconstruction in short term load forecasting of power systems are optimized. The optimal parameters are gained through numerical experiments. Firstly, the delay time is optimized by the smoothness and singularity of the sampled series, especially avoiding the singularity in matrix calculation. Secondly, the dimensions of the embedding phase space are selected according to the length of the embedded window, 24 h. Lastly, the numbers of neighboring vectors are selected according to three to five times of the embedding dimensions, instead of the distance formerly.