基于灰色关联度与LSSVM组合的月度负荷预测

Monthly Load Forecasting Based on Grey Relational Degree and Least Squares Support Vector Machine

由于月度负荷的二重趋势特性,其变化呈现出复杂的非线性组合特征,使预测精度一直不能达到令人满意的结果。针对月负荷的二重趋势特性和最小二乘支持向量机(least squares support vector machine,LSSVM)存在的数据输入维数大、训练时间长等缺点,提出一种基于灰色关联度与LSSVM组合的月度负荷预测方法。该方法通过计算灰色关联度来选择训练样本,选取LSSVM进行样本训练;将与待预测月高度相似的历史月负荷作为LSSVM的训练样本输入,剔除了冗余数据,减少了输入维数,提高了预测精度。通过实例验证和结果对比,证明了该方法可显著提高月负荷预测的精度。

Due to the dual trends of monthly load, i.e., the increasing and fluctuation, its variation is indicative of the feature of complex nonlinear combination, so its forecasting accuracy cannot be satisfactory. In allusion to the duplicity of monthly load and in view of such defects in least squares support vector machine （LSSVM） as high dimensionality and long training period, combining grey relational degree with LSSVM a method to forecast monthly load is proposed. In the proposed method, by means of calculating grey relational degree the training samples are determined and LSSVM is chosen to train the samples; historical monthly load data, that is highly similar to the to be forecasted monthly load, is taken as the input and by means of rejecting redundant data the dimensionality of the input is reduced, thus the forecasting accuracy can be improved. Utilizing LSSVM toolkit in Matlab 7.0, a simulation model of the proposed method is built; the RBF is chosen as the kernel function of LSSVM and through the modeling training by kernel function and paramters a final LSSVM model is achieved, then the to be trained samples chosen by grey relational degree is input into the final LSSVM model to perform the training; and then the maximum daily load of the forecasted month can be obtained. Comparing the forecasted results by the proposed method with forecasted results from only using LSSVM, it is proved that the accuracy of mothely load forecasting can be evidently improved by the proposed method.