The building structure and household behavior of end-users are different. The heating datasets of end-users have features of large amount, strong nonlinearity, long response time, etc. In the original data space, it is hard to implement anomaly detection by the clustering analysis. The problem is the serious data crossing that greatly reduces the accuracy. In this paper, the high dimensional Gaussian mixture clustering (HGMM) is proposed to map datasets in original space to high-dimensional space for clustering. Kernel function mapping, inner product, decomposition of high-dimensional feature space are used to improve clustering accuracy and avoid dimensional disaster. Industrial big data ingestion and analysis platform ( IBDP) is established. The clustering and anomaly detection accuracy of K-Means, Gaussian mixture model (GMM), constant false alarm rate, and HGMM are compared. The proposed method could improve the clustering accuracy to 90. 72% and reduce the detection error rate to 5. 92% . Four types of abnormal heating patterns are identified and analyzed. The proposed HGMM could be used to effectively analyze the residential heating characteristics, detect the abnormal datasets, help reduce the heating energy consumption, and realize the building energy saving.