Research On The OD Estimation Method Based On Optimal Artificial Neural Network

Traffic demand analysis is an important part of traffic planning and traffic impact analysis,and the current situation OD traffic volume is often the basic basis of the link.With the development of China’s transportation industry,the scale of road network has become very large,relying on traditional methods to obtain OD has become very difficult,and using road section traffic for OD inversion has become an important means to obtain the status quo OD.Based on this,the feasibility of neuronal network algorithm in OD inversion and how to improve its effectiveness were studied in this thesis,and a specific method was formed.First,the limitations of existing OD backpropagation methods were analyzed and the influencing factors of OD backpropagation were identified.On this basis,the feasibility analysis of BP neural network inverse OD was conducted,and the basic idea and methodological framework of the OD inverse method based on BP neural network were established.Secondly,by analyzing the principle of BP neural network algorithm and taking the road network in Alashan region of Inner Mongolia as an example,the method of OD inverse inference model construction based on BP neural network was introduced,which included the method of determining the indexes of model input and output layers,the method of determining the number of implied layers,the method of determining the number of nodes of implied layers and the method of determining the parameters of BP neural network.The input layer of this OD backpropagation model was 74 roadway flow data,and the output layer was 110 OD data.For the limitations of BP neural network and the basic characteristics of genetic algorithm,the optimization method of model parameters based on genetic algorithm was studied.The thesis also investigated the method of obtaining virtual samples and the method of analyzing the accuracy of model backpropagation.The optimization and training of the case model were completed using Matlab,and the OD calculation of the case and the analysis of the usage effect were carried out.Finally,the same case was used to compare and analyze with other two models.The results showed that the average relative error of the research method in this thesis was 7.47%,which was significantly lower than the average relative errors of 10.18%and 11.09% of the Trans CAD backpropagation model and the unoptimized BP neural network model.It showed that the backpropagation accuracy of this method was higher than that of other models.