基于相似日选择与改进Stacking集成学习的短期负荷预测

Short-Term Load Prediction Based on Similar Day Selection and Improved Stacking Integration Learning

单一机器学习算法进行短期负荷预测存在着泛化能力受限的问题,本文将Stacking集成学习模型引入到短期负荷预测问题。对于在交叉验证下同一基学习器不同预测模型表现出预测准确度的差异性,根据预测精度对同一基学习器中不同预测结果进行赋权。考虑到不同环境下各影响因子对日负荷值影响程度不同,引入蚁狮算法(ALO)自适应的调整各个影响因子的权值,提高相似日选取方法的准确性。通过相似日选取方法筛选出的相似日集合样本训练改进的Stacking算法预测模型,利用中国北方某地区的负荷数据进行实际算例分析,实验结果表明,在面对负荷影响因素复杂且训练样本较多的情况下,本文所提的方法具有良好的鲁棒性、稳定性和预测精度。

The Stacking integrated learning model is introduced into the short-term load forecasting problem because of the limited generalization ability of a single machine learning algorithm.Different prediction models of the same base learner show different prediction accuracy under cross-validation,and different prediction results of the same base learner are weighted according to the prediction accuracy.Considering that the influence degree of each influence factor on daily load value is different under different environments,the ant lion algorithm(ALO)is introduced to adjust the weight of each influence factor adaptively to improve the accuracy of similar day selection method.Prediction model based on the improved Stacking algorithm is trained by the similar day selection method of similar day collection sample.The load data of a certain area in north China are used for analysis.The experimental results show that the influence factors in the face of load under the condition of more complex and the training sample,the proposed method has good robustness,stability and accuracy.