摘要
针对不同类型用户的短期负荷预测,目前应用较为广泛的为支持向量机与深度学习模型。针对最小二乘支持向量机(least squares support vector machine,LSSVM)模型中超参数难以确定、模型对数据质量要求较高等问题,而集成常规优化算法又会有寻优速度慢、易陷入局部最优等问题,提出了一种混合模型。首先使用层次聚类(hierarchical clustering,HC)对原始特征数据进行聚类进而为同一类预测日建立对应LSSVM模型,再通过改进的模拟退火算法(improved simulate anneal,ISA)对LSSVM中的超参数进行启发式搜索。最后通过对广东省佛山市某行业用户用电负荷进行负荷预测,与各种负荷预测模型性能进行对比,结果证明所提模型可有效提高负荷预测精度、缩短预测时间。
For short-term load forecasting of different types of users, support vector machine and deep learning model are widely used at present. A hybrid model is proposed to solve the problems of the least squares support vector machine model, such as the difficulty in determining the super parameters, the high data quality requirements of the model, and the slow optimization speed and easy to fall into the local optimization of the integrated conventional optimization algorithm. Firstly, the original feature data is clustered by hierarchical clustering and then the corresponding least squares support vector machine model is established for the same prediction day. Then, the super parameters in least squares support vector machine are heuristic searched by the improved simulated annealing algorithm. Finally, by comparing the performance of the load forecasting model with that of various load forecasting models, the results show that the proposed model can effectively improve the accuracy of load forecasting and shorten the forecasting time.
作者
郑乐
徐青山
冯小峰
ZHENG Le;XU Qingshan;FENG Xiaofeng(College of Cyberspace Security,Southeast University,Nanjing 210096,China;College of Electrical engineering,Southeast University,Nanjing 210096,China;Measurement Center,Guangdong Power Grid Co.,Ltd.,Guangzhou 518049,China)
出处
《电力需求侧管理》
2022年第5期51-57,共7页
Power Demand Side Management
基金
国家自然科学基金资助项目(51577028)。
关键词
短期负荷预测
层次聚类
LSSVM
改进模拟退火
short term load forecasting
hierarchical clustering
LSSVM
improved simulated annealing