Regional centralized control of hydropower operation requires continuous condition monitoring and anomaly analysis of multi-power station units and their auxiliary machines. Focusing on the hydroelectric generator set, speed regulation system, main transformer and auxiliary equipment, this paper constructs a data mining and analysis framework, and maps vibration, swing, temperature, head, flow, active power, pressure pulsation and event quantity into a four-dimensional working condition portrait of “physics-environment-health-business”. In terms of method, statistical features, short-time Fourier transform features and wavelet packet process features were fused, and support vector machine, random forest and long short-term memory network were combined to complete steady-state identification, transient tracking and joint discrimination, and index deviation analysis, abnormal pattern extraction, early warning classification and feedback write-back were realized. Experimental results show that the Accuracy and F1 of the test set reach 95.6%and 95.1%respectively, and the average Accuracy and F1 of the field pilot are 95.3%and 94.7%respectively, which can support fine-grained state discrimination, anomaly interpretation analysis and operation and maintenance decision support under regional centralized control conditions.
