Abstract
In order to improve the accuracy of short-term traffic flow prediction, a combined model composed of artificial neural network optimized by using Genetic Algorithm (GA) and Exponential Smoothing (ES) has been proposed. By using the metaheuristic optimal search ability of GA, the connection weight and threshold of the feedforward neural network trained by a backpropagation algorithm are optimized to avoid the feedforward neural network falling into local optimum, and the prediction model of Genetic Artificial Neural Network (GANN) is established. An ES prediction model is presented then. In order to take the advantages of the two models, the combined model is composed of a weighted average, while the weight of the combined model is determined according to the prediction mean square error of the single model. The road traffic flow data of Xuancheng, Anhui Province with an observation interval of 5 min are used for experimental verification. Additionally, the feedforward neural network model, GANN model, ES model and combined model are compared and analysed, respectively. The results show that the prediction accuracy of the optimized feedforward neural network is much higher than that before the optimization. The prediction accuracy of the combined model is higher than that of the two single models, which verifies the feasibility and effectiveness of the combined model.
Highlights
In recent years, with the development of economy and rural urbanization, the urban population has increased significantly, and the traffic demand has risen sharply
In order to combine the non-linear fitting advantage of neural network with the linear fitting advantage of Exponential Smoothing (ES) model, we propose a combined model of feedforward neural network optimized by the Genetic Algorithm (GA) and ES
The prediction accuracy of the feedforward neural network optimized by GA is much higher than that of the feedforward neural network trained by a traditional backpropagation algorithm, and the value of MSE is reduced by 16.9%
Summary
With the development of economy and rural urbanization, the urban population has increased significantly, and the traffic demand has risen sharply. Linear model is simple in structure, weak in adaptability to various random disturbance factors, and poor in adaptability when applied to short-term prediction field, making it difficult to meet the accuracy requirements of traffic flow prediction. Chang and Yoon [30] applied the nearest neighbour non-parametric regression method in short-term traffic flow prediction. This method requires sufficient historical data to predict more traffic flows with changing trends when there is enough data. According to previous research results, feedforward neural network can realize any non-linear mapping, and it has prominent advantages in shortterm prediction, but it is prone to fall into local minimum value and slow learning. In order to combine the non-linear fitting advantage of neural network with the linear fitting advantage of ES model, we propose a combined model of feedforward neural network optimized by the Genetic Algorithm (GA) and ES
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
