The accuracy of city gas daily load forecasting is of great significance to the optimal design, rational scheduling and stable operation of gas supply system.This paper comprehensively considers many factors of daily load forecasting and establishes a combined forecasting model PCA-GM-BPNN, which is based on Principal Component Analysis (PCA), Grey Model (GM) and BP Neural Network (BPNN).The model first uses the grey optimization model to predict the prediction value that is equal to the number of training samples of BP neural network as the correction sequence of the input layer neurons,then through PCA to reduce the dimension of the impact factors including the correct sequence,and then several principal components with more than 85% of the cumulative contribution rate are trained as input neural networks for neural network training.The analysis of practical application results shows that the daily load MAPE value of Nanle county in this combined model is 5.44%, which is less than the other four load forecasting models, and it is a rather effective method for daily load forecasting of city gas.