The invention discloses a building energy consumption prediction method based on a recurrent neural network and a multi-task learning model, relates to the technical field of comprehensive energy management, and solves the technical problems of parallel prediction of multiple types of energy consumption, guarantee of prediction precision and shortening of model training time. The method comprisesthe following steps: acquiring a building energy consumption data sample; The method comprises the following steps of: carrying out missing data processing by utilizing a plurality of similar time point data averaging, constructing a plurality of learning tasks according to an energy consumption type, a time step length and initial time, then normalizing a data set of each learning task, and measuring the similarity among a plurality of task training sets by using a Pearson correlation coefficient. After the similarity among multiple tasks is ensured, a neural network model is created and trained, and finally, a multi-task CIFG-LSTM neural network model is used for predicting composite energy consumption. According to the energy consumption prediction method, the multivariate energy consumption can be predicted at the same time, close connection and interaction between the energy consumption are fully utilized, and the prediction precision and speed are improved.