Power data prediction model optimization method and system

The invention provides an electric power data prediction model optimization method and system, the electric power data prediction model optimization method is applied to object type data processing, and the method comprises the steps: collecting and obtaining electric power structured information, and generating electric power state sample data according to the electric power structured information; classifying and identifying the power state sample data to obtain sample label information, and obtaining training model parameters and prediction target data from a preset training model; processing the training model parameters and the prediction target data to obtain sample loss parameters; processing the sample label information according to a sample loss parameter so as to obtain adaptive sample mining data; and carrying out convergence processing on the preset training model by using the adaptive sample mining data. According to the method, the online difficult sample mining loss function is designed, the we