The formulation of expressway traffic policy needs to accurately predict the traffic volume. Based on this, Long short-term memory (LSTM) machine learning model was selected to study it. Aiming at the problem of parameter determination in LSTM model, Particle swarm optimization (PSO) algorithm was selected to improve it. At the same time for the PSO algorithm in the particle position update problem, the meaning of each parameter in the formula for the entry point for improvement, the PSO algorithm formula for the original static inertia weights and learning weights into the iteration number and the particle position will change with the change of the dynamic value, so as to achieve the purpose of searching for the purpose of improving accuracy. Based on this, the improved PSO-LSTM model is constructed. Finally, through the calculation and analysis of an example, the working days and rest days of expressway are predicted respectively. The results show that the root mean square error of the evaluation index is increased by 12.19% and 10.97%, the average absolute error is increased by 17.06% and 15.17%, and the square absolute percentage error is increased by 24.56, respectively. The algorithm shows that the improved PSO-LSTM model plays a significant role in traffic volume forecasting, and has strong anti-interference ability. It can provide a more reliable basis for the rational formulation of policies.