The invention provides a terminal access data abnormality detection method based on machine learning. The method comprises the following steps: processing received current terminal access data according to a predetermined multi-dimensional feature library to obtain a feature vector of each piece of access data, wherein at least one piece of access data is included in the current terminal access data; taking the obtained feature vectors as input information of a machine learning model, determining a detection result of the current terminal access data, wherein the machine learning model is a pre-trained abnormality access detection model passing a correctness test. The method utilizes the established abnormal access detection model to detect abnormal behaviors of terminal access data, and ensures the safe and reliable operation of an integrated energy metering collection system of electricity, water, heat and gas heat.