Based On The Improved Neural Network Short-term Traffic Flow Prediction Research

Short-term traffic flow forecasting is an important prerequisite for traffic controland guidance, and its forecasting performance is the key for the traffic flow guidancesystem to effectively realize. This thesis analyzes the basic characteristics, the mainparameters and data acquisition techniques of traffic flow, and selects traffic volume asforecast object. Traffic flow typically exhibits the characteristics of nonlinearity,randomness, uncertainty and ambiguity, and neural network just has better nonlinearmapping ability, learning ability and adaptive ability. Therefore, this paper focuses onthe research of short-term traffic flow forecasting based on neural network. As the corepart of feedforward neural networks, BP neural network is widely used, but it still hassome shortcomings, such as sensitive to the initial value, easy to fall into the localminimum value and that the network structure is usually determined by the empiricalequation. Due to these shortcomings, BP neural network is improved, and the improvedalgorithm is used for short-term traffic flow forecasting.Firstly, in allusion to the shortcomings of sensitive to the initial value and easy tofall into the local minimum value, this paper proposes an improved fusion algorithmbased on the analysis of the existing fusion algorithms of ant colony algorithm and BPneural network with known neural network structure, the main work includes,(1) trafficflow data preprocessing and determines the structure of BP neural network according tothe data pretreatment.(2) Establishes the basic fusion model of ant colony algorithm andBP neural network, and analyzes the impact of ant colony algorithm parameters.(3)Divides the search space evenly and uses the local search method of BP algorithm toimprove the basic fusion algorithm, so as to improve the quality of solutions which antshave found.Secondly, in allusion to the determination of neural network structure byexperience, this paper proposes a fusion algorithm which makes use of ant colonyalgorithm to optimiz the neural network structure, the weights and the thresholdssimultaneously. By using the global optimization ability of ant colony algorithm, thealgorithm can find an optimal network structure and meanwhile give a set of initial weights and thresholds, then the algorithm is applied to short-term traffic flowforecasting.Finally, this article simulates the above-mentioned algorithms using MATLAB7.6,when BP neural network has known neural network structure, the simulation resultsshow that the improved fusion algorithm is better than the basic fusion algorithm, whichcan improve the solution quality that ants have found. Introducing traffic flowpredictive error index, the simulation results show that the second algorithm can find astable structure of BP neural network, and has a higher prediction accuracy.