Abstract: Total social electricity consumption is a vital indicator measuring the electric energy consumption in various industrial and domestic processes of society. It accurately reflects the pace of societal development over a period. Power grid companies and government departments demand high accuracy in future predictions of total social electricity consumption, as higher precision enables better allocation of electric power resources to maintain system stability. This paper first introduces two classic neural network prediction models and collects data like the 2022 regional GDP and total output value of the construction industry from the data published by the government of County A, filling missing values using linear trend and smoothing methods. Subsequently, using SPSS software, multilayer perceptron（MLP） and radial basis function（RBF） models were established, and the processed data were used for training and testing these models. After analyzing the residuals and the importance of independent variables, the accuracy of the prediction results was evaluated. The study concludes that the RBF model demonstrates higher accuracy and stability in predicting total social electricity consumption.