基于聚类加权随机森林的非侵入式负荷识别

非侵入式负荷识别是实现用能管理的重要监测手段，而随机森林因其良好的泛化能力和鲁棒性应用于负荷识别领域。针对传统随机森林算法忽略决策树分类能力的差异、投票不公平的问题，提出了一种基于层次聚类的加权随机森林算法。首先，提取各类负荷开关状态下负荷特征量，建立特征数据库用于训练原始随机森林模型。然后，利用有功功率差检测总线信号中的开关事件，并提取负荷特征量作为验证集和测试集；验证集采用层次聚类选择法获得每个聚类中分类精度最高的决策树，测试集采用加权投票策略实现负荷识别。通过实验验证，说明相比于传统的机器学习算法，该算法可以实现更高的识别精度，准确率可达96.2%。

Non-invasive Load Identification Based on Clustering Weighted Random Forest

Non-invasive load identification is an important monitoring method for energy management， and random forest is applied to non-invasive load identification because of its good generalization ability and robustness. A weighted random forest algorithm based on hierarchical clustering is proposed to solve the problem of ignoring the difference of decision tree classification ability and unfair voting. Firstly， various load characteristics under load switching states are extracted， and the feature database is established to train original random forest model. Then， active power difference is used to detect switching events from bus signals， and load characteristics is extracted as verification set and test set. Its verification set adopts hierarchical clustering selection method to obtain decision tree with the highest classification accuracy in each cluster， and its test set adopts weighted voting strategy to achieve load identification. Experimental results show that the proposed algorithm can achieve higher recognition accuracy with an accuracy of 96.2% compared with traditional machine learning algorithms.