Abstract: The load forecasting of power system is of great significance to the safe and stable operation of power system. In order to solve the problem of gradient descent and disappearance in the traditional RNN algorithm, the Long Short-Term Memory idea is introduced into the load forecasting process, and the Bidirectional Long Short-Term Memory neural network is used to capture the features of the data series from both forward and backward actions and deal with the data dependency better. Further, EMD and PSO are used to optimize the model, and the optimal model of each IMF (Intrinsic Mode Function) is obtained. The corresponding PSO-BiLSTM prediction model is constructed for IMF sequences, and finally the predicted values of the whole load are obtained by combining multiple sets of predicted values. Taking the actual load as an example, it is proved that the proposed algorithm can effectively improve the prediction accuracy.