Research on highway traffic flow prediction based on a hybrid model of ARIMA–GWO–LSTM

As an important part of the transportation industry, highway transportation, compared with other modes of transportation, has the advantages of high flexibility, wide adaptability and wide coverage. Traditional time series-based traffic forecasting is difficult to solve the problem of nonlinear and non-stationary characteristics. To solve these problems, a highway traffic forecasting method based on ARIMA–GWO–LSTM combination model is proposed. Firstly, the preliminary prediction of the UK M25 highway traffic using the Autoregressive Integrated Moving Average Model (ARIMA) model obtains linear prediction results. Then, a highway traffic prediction model based on the Long Short-Term Memory Network (LSTM) is established, and automatic optimization of the LSTM hyperparameters is achieved through the introduction of the Gray Wolf Optimizer (GWO) into the traditional LSTM model, and the residuals are corrected using the LSTM model to obtain the nonlinear prediction results; the model finally combines the linear and nonlinear prediction results to obtain the predicted value of the traffic flow of the UK highway. Finally, with the help of the prediction error indexes, such as MAE, RMES, MAPE, and R², the hybrid model is compared and analyzed with the LSTM, ARIMA, SVR, ARIMA-SVR, and ARIMA-LSTM models. The results show that the ARIMA–GWO–LSTM model reduces the MAE by 77.66%, the RMSE by 75.57%, the MAPE by 75.06%, and the R² is closer to 1 than the single model. Verifying that the hybrid model cannot only extract long-term dependencies of data, but also effectively capture linear and nonlinear features of data, thus achieving more accurate prediction.