为了实现非侵入式负荷监测的功能并提高负荷辨识准确率，文中提出一种基于机器学习的负荷辨识方法。在家用电器电流波形及各次谐波特征的数据中，采用核主成分分析方法（Kernel Principal Components Analysis，KPCA），解决非线性特征提取与降维的问题，最大限度抽取特征信息。再利用一维卷积核提取时序特征并压缩后输入到XGBoost模型，得到负荷辨识结果。通过在实验室中采集数据进行算法验证，文中提出算法在各类用电器的识别中均具有较高的准确率。
英文摘要:
In order to realize the function of non-intrusive load monitoring and improve the accuracy rate of load identification, a load identification method based on machine learning is proposed in this paper. In the data of current waveform and harmonic characteristics of household appliances, Kernel principal components analysis (KPCA) is used to solve the problem of nonlinear feature extraction and dimension reduction, and extract feature information to the maximum extent. One dimensional convolution kernel is used to extract time series features and compress them into the XGBoost model to obtain load identification results. The algorithm is verified by the data collected in the laboratory. The proposed algorithm has high accuracy rate in the identification of all kinds of electrical appliances.