Due to the slow convergence speed of the model with many hidden layers in the LSTM (long short-term memory) recurrent neural network, the updating of its weights and thresholds depends on the gradient descent algorithm, which may lead to the local extremum phenomenon in the weight correction of the network nodes, resulting in the reduction of the generalization ability of the LSTM neural network model. Based on this, this paper proposes an optimized LSTM neural network model based on APSO (accelerated particle swarm optimization) algorithm (APSO-LSTM). In this model, root mean square error is designed as an appropriate value function, and APSO algorithm is used to build an optimization system to optimize the weights of each neuron node globally, so as to improve the prediction performance of the model. The experimental results on the classic DataMarket and UCI datasets show that the prediction accuracy of APSO-LSTM model is significantly improved compared with the traditional LSTM model, which verifies the effectiveness of APSO-LSTM model.