Research On Road Network Overtaking Prediction Based On Grey Theory And Deep Learning

With the development of urban transportation in China,the scale of China's urban road network has grown,and the traffic safety problems in cities have become more serious.Traffic accidents have emerged one after another.Among them,a wide range of overtaking is an important factor in generating traffic safety problems.The past overtaking data is difficult to obtain and predict.In this paper,we use the identification of electronic police and bayonet license plates,and it can be more accurate through upstream and downstream license plate recognition and time comparison,it can obtain the overtaking relationship between the vehicles on the road section,and on this basis,predict and analyze the overtaking characteristics of the urban roads.This paper proposes a grey Bernoulli model based on intelligent optimization algorithm optimization and a deep recurrent neural network for short-term prediction and long-term prediction of overtaking problems.The sequence accumulating method of the gray system can effectively eliminate the disturbance of the external irrelevant information,and can accurately predict the overtaking situation in a short time.The associative memory function of the recurrent neural network can achieve the preservation of high quality information and the elimination of noise,through long-term prediction and short-term forecasting can make a reasonable decision on the road network overtaking situation.Specifically,the main contributions of this article are as follows:(1)In this paper proposes a short-term overtaking prediction method based on grey theory and gives an optimization method.The volatility of road overtake data is large.In short-term forecasting,classical statistical models tend to have large errors.The idea based on grey theory can weaken its randomness.In order to obtain the optimal model hyperparameters,the differential evolution algorithm and quantum genetic algorithm are used to optimize the cumulative order and background value construction coefficients of the grey model.(2)In terms of long-term prediction,in this paper we firstly comb the classical statistical methods,and secondly propose the GRU neural network for road network over-vehicle prediction,which has strong associative memory function,but it has less computation than LSTM and can effectively predict the long-term situation of road network overtaking problem.(3)To verify the validity of the proposed model.The actual data of Xinghu Street-Modern Road Section of Suzhou Industrial Park,Suzhou City,Jiangsu Province was selected for testing.The results show that the grey model proposed for short-term overtaking prediction has higher precision than other grey models.The GRU for long-term prediction is more accurate,strong,robust and generalized than the other two long-term prediction models.