The invention relates to a medium and long-term power consumption forecasting method and device. The method comprises obtaining the training data and the test data of a least square support vector machine model for medium and long-term power consumption forecasting according to the external factor data of a historical day and the corresponding power consumption data; optimizing the parameters of least squares support vector machine model for medium and long term power forecasting by using a bat algorithm; using the least square support vector machine (LS-SVM) model to forecast the medium and long-term electricity consumption after the parameters are optimized. According to the technical proposal provided by the invention, based on the algorithm principle of least square support vector machine, in the case of limited sample data, the normalization parameters and kernel function parameters of LS-SVM model are optimized by the bat algorithm, so that the model has better prediction accuracy and generalization ability, and can effectively solve the problem of incomplete medium-and long-term electricity forecasting method.