Research On Future Vehicle Speed Prediction Model On Roads Based On Deep Learning

With the development of China’s economy,the progress of society,the improvement of living standards,and the acceleration of urbanization,urban traffic has gradually been improved.Road has become not only the main link connecting cities,but also the lifeline of the city.However,due to the increase in the number of private cars and the disruption of urban construction,urban road congestion has become increasingly serious.The congested roads greatly increase the travel costs of residents,and will cause local pollution of the urban environment.Accurately predicting the future vehicle speed of urban roads can help solving the urban traffic congestion problems.The speed is affected by many factors,such as weather,holidays,regional locations and so on.The main research and contents of this thesis are as follows:the average vehicle speed of the road is estimated using taxi trajectory data.For taxis with high real-time requirements,large data volume and complex road environment,the matching algorithm based on road geometry information is used to realize the spatial division of GPS data.As an algorithm for neural network parameter optimization,the traditional selection method in genetic algorithm only saves the best individual in one population,but does not consider the damage of population diversity of other individuals.The method is easy to fall into local optimum problems.In order to solve this problem,a BP neural network model is designed based on improved genetic optimization algorithm.By improving the quality genetic algorithm selection operation to improve the performance of genetic algorithm,BP neural network model can be better optimized.And this method is applied to road speed prediction modeling.The future vehicle speed of the road has a great relationship with the speed of the previous period,and the predicted features is time-relevant.Therefore,an end-to-end neural network model FSPOL is designed.The LSTM network is used to process the road speed information with spatio-temporal characteristics.the fully connected layers in the neural network are combined to fuse multiple factors that affect vehicle speed.The comparison experiment results show that the prediction accuracy of the vehicle speed prediction model FSPOL is better than the commonly used methods.