基于分位回归鲁棒极限学习机的短时负荷预测方法

Quantile Regression Based Robust Extreme Learning Machine for Short-term Load Forecasting

针对短期负荷预测对电力系统运行管理和优化调度的影响,提出一种基于分位回归鲁棒极限学习机的短时负荷预测方法,即先对所收集的历史负荷数据进行归一化处理,然后利用自相关分析提取最相关的历史负荷数据作为模型的输入变量,再融合鲁棒极限学习机和分位回归建立负荷预测基本模型,最后利用某电力公司2016年采样频率为30min的数据进行实例分析,试验数据表明相比极限学习机(ELM)、分位回归(QR)和分位回归支持向量机(QR-SVM),所提模型预测精度更高,验证了所提模型和算法的可行性和有效性。

Aiming at the influence of short-term load forecasting on power system operation management and optimal scheduling,a short-term load forecasting method based quantile regression robust extreme learning machine was proposed.Firstly,the collected historical load data should be normalized,and autocorrelation analysis was used to extract the most relevant historical load data as the input variables of the model.Then robust extreme learning machine and quantile regression were combined together to establish the basic load forecasting model.Eventually the load data sampled in 30 min from a power company of year 2016 were utilized as case analysis.Experiment results show that compared with the extreme learning machine（ELM）,quantile regression（QR）and quantile support vector machine（QR-SVM）,the proposed model has a higher prediction accuracy,thus certifying the proposed model is feasible and effective.