Research On Interpretable Methods Of Intelligent Algorithms For Urban Traffic Networks

In recent years,urban development is growing faster and faster,and the traffic demand is also growing rapidly.However,the frequent traffic accidents in the urban transportation network and the increasingly serious road congestion in the morning and evening rush hours have affected people’s daily travel and caused widespread concern from all walks of life.The main trend of future urban transportation development is to build an intelligent and efficient urban transportation network.Intelligent algorithms in urban traffic networks can perform functions such as urban traffic flow prediction,traffic signal intelligent road control and traffic event detection.However,these functions are based on a data-driven approach to obtain valid information from complex nonlinear traffic data,from which the reasons for the obtained decision results cannot be verified.Therefore,it is difficult to understand and explain the decision-making process of intelligent algorithms during expert consultation,and it cannot be a reliable basis for expert decision-making,which leads to a difficult communication gap between intelligent algorithms and expert decision-making and becomes a key obstacle to the further application of intelligent algorithms in urban transportation networks.In order to make the intelligent algorithm transparent and interpretable,an interpretable model of intelligent algorithm based on random forest-neural network accompaniment is constructed to realize the interpretability of intelligent algorithm in urban traffic network.In order to make the intelligent algorithm transparent and interpretable,we construct an interpretable model of intelligent algorithm based on random forest-neural network accompaniment to achieve the interpretability of intelligent algorithm in urban transportation network.The main work accomplished is as follows.(1)Construct an interpretable concomitant model for intelligent algorithms in urban traffic networks.To address the problems of opaque information and unintelligible decision results in the decision-making process of intelligent algorithms in urban traffic networks,a random forest-neural network concomitant model is established,and two submodels are trained with the same data set to establish the relationship of parallel composition.The mapping of uninterpretable neural network neuron activation vectors in the decision space is completed by using the interpretable rules of random forest to realize the interactive mapping of neural network neuron activation vectors and interpretable rules of random forest,and its feasibility is verified by traffic flow prediction example analysis to lay the foundation for subsequent research.(2)Construct an interpretable rule base for intelligent algorithms of urban traffic networks.For the trained neural network sub-model and random forest sub-model,given the same input,the neuronal activation vector of the neural network can be directly obtained;then the interpretable random forest rule extraction method is introduced to extract the optimal rules of the random forest sub-model,and the two form a one-to-one correspondence,which is stored in the interpretable rule base item by item to form the interpretable rule base dataset,and finally the interpretable rule base dataset is formed by The final analysis of the traffic flow prediction example of the interpretable rule base verifies the feasibility and rationality of the rule extraction and rule base construction methods for the traffic flow dataset,and provides data sample support for the subsequent interpretable mapping network.(3)Construct an interpretable mapping network for urban traffic network intelligence algorithm.For the interpretable rule base dataset,an interpretable mapping network is constructed to learn the mapping relationships in the dataset and further verify the rationality and effectiveness of the interpretable model of urban traffic network intelligence algorithm.The interpretable mapping network is trained with a convolutional neural network as the encoder,a long and short-term memory network as the decoder,and an interpretable rule dataset as the basis.The trained network model is used to analyze the morning and evening peak traffic congestion at the south intersection of the Ankang Hanjiang First Bridge and to give the decision basis of the network model,and to analyze the causes of traffic congestion and propose road reconstruction plans based on the decision basis of the network model.After simulation and field modification,it is verified that the explainable mapping network trained by the explainable rule dataset is capable of decision making and can provide the network model decision basis.The results show that the interpretable mapping model of urban traffic network intelligent algorithm is correct and feasible,and can provide the decision model and realize the interpretability of the intelligent algorithm.