短期负荷预测相空间重构法参数优选的数值测试与分析

采用混沌方法中相空间重构法的局部线性法进行短期负荷预测时,需要优选3个参数,即负荷记录序列的延时时间、嵌入相空间的维数以及选择邻近点时使用的距离。数值测试表明,按混沌理论优选的延时时间和嵌入相空间的维数一般不是负荷预测的最适合参数。这2个参数的取值和搭配对预测误差的影响最大,其次才是选择邻近点时使用的距离。这是由于负荷记录不是严格混沌的,而是以双周期为主。对测试结果的分析表明,优选的延时时间,在离线预测时可以选择使负荷记录中的双周期成分延时相轨迹出现最小重叠的延时时间;在线预测时是使负荷取样序列具有最小方差。此外,还确认采用负荷记录的“平衡点 混沌”拆分通常可以降低预测误差。

IMPROVING PRECISION OF SHORT TERM LOAD FORECASTING BY NUMERICAL TESTING IN LOCAL LINEARIZATION METHOD OF PHASE SPACE RECONSTRUCTION

Three parameters, which are the delay time of load record series, embedded space dimension and distance in choosing the neighboring points, need to be determined in forecasting short term loads by the local linearization method of phase space reconstruction. Numerical simulation test shows that the delay time and embedded space dimension given by chaos theory usually are not the best ones for load forecasting. Their combination is the most important factor to reduce the load forecasting errors, and the distance is the secondary factor. The reason is that the load records are not rigorously chaotic but chiefly double periodical. Analyses show that, (1) in off-line forecasting the best delay time corresponds to that makes the minimal superposition in phase space trajectories of the double periodical component, (2) in on-line forecasting the best delay time corresponds to that makes the minimal standard deviation of the new series sampled from the load records. Furthermore, it is shown that splitting the load records series into "equilibrium point + chaos" can usually reduce the forecasting errors.