Abstract: The significance of load forecasting for coal-fired power plants lies in the fact that it is possible to know in advance the demand for electricity in the future period of time, so as to rationally arrange the operation and downtime of power generation equipment for maintenance, avoid energy waste, and improve the efficiency of power generation;moreover, in the context of coal-fired power plants participating in in-depth peaking and coal blending, it is necessary to predict in advance the future period of time in order to ensure that the coal blending heat generation capacity is adapted to the demand for loads and to improve the efficiency of combustion. In this paper, a load forecasting method for coal-fired power plants based on the fast Non-dominated Sorting Genetic Algorithm Ⅱ（NSGA-Ⅱ） optimized Vector Autoregression（VAR） model is proposed. The method takes the historical superheated steam time series, the historical reheated steam time series, and the historical power generation series together as the input VARiables of the VAR model to predict the power generation load in the next 8 hours, and at the same time uses the NSGA-Ⅱ algorithm to optimize the order and intercept of the VAR model, thus improving the accuracy of the prediction model.In the testing stage, the data sample interval from October 25, 2022 to October 30, 2022 for a unit in Shanghai is selected to establish the initialized prediction model; the model effect is tested on the sample interval from 8: 00 on October 31, 2022 to 16:00 on November 1, 2022, and the VAR model is optimized according to the test results using the NSGA-Ⅱ algorithm; and the VAR model is optimized on the sample interval from 8:00 on October 2, 2022 to 16:00on November 2, 2022 using the NSGA-Ⅱ algorithm; the VAR model is optimized according to the test results in the test stage. The prediction accuracy of the optimized model is further tested on the sample interval from 8:00 on November 2, 2022 to 16: 00 on November 3, 2022, using the NSGA-Ⅱ algorithm to optimize the VAR model based on the test results. The results show that the root-mean-square error of the prediction is 15.341 MW, and the average absolute error is 7.839 MW, which is improved compared with other time series prediction models. Therefore, the model can be practically applied to the load forecasting of similar coal power units, thus providing a reference for subsequent operation decisions.