基于灰色投影改进随机森林算法的电力系统短期负荷预测

发布时间：2018-10-20 09:23

【摘要】：针对短期负荷预测领域传统的机器学习算法(如人工神经网络、支持向量机等)存在的诸如泛化性能不强、参数和模型结构确定困难等问题,将随机森林回归算法引入短期负荷预测领域。同时应用投影原理改进了传统的灰色关联相似日选取算法,提出了一种基于灰色投影改进随机森林算法的电力系统短期负荷预测组合方法。基于灰色投影的相似日选取方法,采用灰色关联度判断矩阵表征历史样本与待预测日影响因素间的关联关系,并用熵权法确立影响因素的权重对判断矩阵加权,最后利用各个样本关联度投影值排序得到相似日集合。采用随机森林算法建立预测模型,利用灰色投影筛选出的相似日样本集合训练模型,最后输入预测日特征向量(天气预报数值、日类型等)完成预测。以浙江电网某县级市的负荷数据作为实际算例,并将上述方法与支持向量机方法以及未作灰色投影改进的随机森林算法进行对比。实验结果表明,新方法具有较高的预测精度和鲁棒性。
[Abstract]:The traditional machine learning algorithms in the field of short-term load forecasting (such as artificial neural networks, support vector machines, etc.) have some problems, such as poor generalization performance, difficulty in determining parameters and model structure, etc. The stochastic forest regression algorithm is introduced into the field of short-term load forecasting. At the same time, the traditional grey correlation similarity day selection algorithm is improved by applying the projection principle, and a combination method of short-term load forecasting for power system based on grey projection improved stochastic forest algorithm is proposed. Based on the method of similar day selection based on grey projection, the grey correlation degree judgment matrix is used to characterize the correlation between historical samples and the influence factors to be predicted, and the weight of the influencing factors is established by entropy weight method to weight the judgment matrix. Finally, the similar day set is obtained by using the projection values of each sample correlation degree. The prediction model is established by using stochastic forest algorithm, and the training model of similar daily sample set selected by grey projection is used. Finally, the predicted daily eigenvector (weather forecast value, day type, etc.) is inputted to complete the prediction. Taking the load data of a county city in Zhejiang power grid as an example, the above method is compared with the support vector machine method and the stochastic forest algorithm without grey projection. Experimental results show that the new method has high prediction accuracy and robustness.
【作者单位】：北京交通大学国家能源主动配电网技术研发中心;
【基金】：国家自然科学基金资助项目(51277009)~~
【分类号】：TM715

【参考文献】