Water-turbine generator set bearing abnormal state prediction method based on deep belief network

A deep belief network-based water-turbine generator set bearing abnormal state prediction method belongs to the technical field of electric power big data, and adopts a grey correlation theory to analyze a coupling relationship among related characteristic quantities of a generator bearing, extracts key characteristics of a bearing operation state according to the degree of correlation, and reduces the data dimension; determining a bearing key characteristic dynamic threshold value, comprehensively analyzing the bearing operation condition and the bearing health state by adopting a nonlinear state evaluation method, and dynamically updating the bearing key characteristic threshold value; a bearing key feature data set and a dynamic threshold value are input into a prediction model based on PSO-DBN, a PSO learning factor and an inertia weight are dynamically updated by using an adaptive method, the optimization performance of PSO on DBN parameters is improved, prediction model errors are reduced, the capabilit