In order to make better use of the user's electricity consumption information in the smart grid and improve the forecasting accuracy, a classified short-term electricity consumption forecasting and correcting method for residential users considering time-of-use electricity price is proposed. Firstly, the fuzzy clustering is used to classify the users according to their electricity usage behavior, and the electricity price, electricity consumption and weighted weather and date influence factors are used as the predictive model input. Secondly, according to the simulation comparison, the BP, Elman and LSTM neural network algorithms are chosen for corresponding users to construct the prediction model. Finally, the modified algorithm is used to correct the peak-to-valley value with large error, and the corrected classification prediction results are added up to obtain the overall predicted value. Taking a certain community in Yunfu City, Guangdong Province as an example, the method is simulated and compared with random forest and CART algorithms. Experimental results show that the proposed method has higher prediction accuracy. This work is supported by Science and Technology Project of China Southern Power Grid Company (No. 035300KK 52150007 and No. 031800KK52170073).