Short-term traffic flow prediction system based on SARIMA-GA-Elman combined model

Abstract

The invention discloses a short-time traffic flow prediction system based on an SARIMA-GA-Elman combined model, and belongs to the technical field of traffic flow prediction of an intelligent traffic system. The method is realized through the five steps of modeling of an SARIMA model, Elman-RNN prediction, SARIMA-GA-Elman combined prediction, linear prediction and nonlinear prediction. The invention provides a novel optimization method to solve the problem that training of the neural network often consumes a large amount of time cost. The method comprises the following steps: firstly, training an Elman-RNN weight and a threshold value based on GA to obtain an interval near value of an optimal solution, and then continuously training by using an Elman-RNN gradient descent algorithm to obtain a final weight and a final threshold value; the optimization mode is very similar to transfer learning, and the training efficiency of the model can be improved. According to the method disclosed by the invention, the Elman-RNN nonlinear model is constructed by utilizing a prediction error subjected to SARIMA prediction. The prediction error excludes the interference of a linear factor and a daily cycle factor, and comprises the nonlinear characteristics of the traffic flow. The two models fully consider periodic, linear and nonlinear characteristics of traffic flow.

Description

technical field [0001] The invention belongs to the technical field of vehicle flow forecasting in intelligent traffic systems, and in particular relates to a short-term traffic flow forecasting system based on a SARIMA-GA-Elman combined model. Background technique [0002] The linear statistical model is a traffic flow prediction model established from the perspective of mathematical statistics to study the law of similarity between future data and historical data. Linear statistical models can only capture the linear characteristics of traffic flow, but cannot make predictions for complex nonlinear characteristics. Currently, they are usually combined with neural networks for prediction. [0003] Nonlinear theoretical models mainly include wavelet theory, fractal theory, catastrophe theory, and chaos theory. At present, the nonlinear theoretical model is not used as a single model to predict traffic flow, but more to be combined with other models to predict traffic flow. ...

AI Technical Summary

Problems solved by technology

Hou Q[37] proposed an adaptive short-term traffic flow forecast hybrid model, which uses linear autoregressive integrated moving average (ARIMA) method and nonlinear wavelet neural network (WNN) method to predict traffic flow, but the wavelet neural network The network cannot effectively avoid overfitting during training

Images