Research And Application Of Audio Signal Based On Deep Learning In Metro Passenger Flow Estimation

With the deepening of urbanization and vehicle population surging, the daily life of residents have been seriously affected by growing traffic congestion in many cities. Urban rail transportation as a large capacity, high efficiency of public transportation has unparalleled great advantage in improving the efficiency of the residents travel to ease the pressure on urban traffic aspects. The railway passenger estimation is undoubtedly the most important work. How to improve the accuracy of the estimated rail transit will be a key issue associated with rail transportation planning, construction and operation phase.In this article, firstly, the background and significance of the traffic estimation, the current situation of domestic and foreign research and the development and application of deep learning are studied; Analyzes the construction of the subway, the subway traffic characteristics, the subway traffic impact factors and the long-term and short-term traffic estimation method in detail; Related theory of acoustic detection and estimation system, the depth study of the theory, method and restricted boltzmann machine(RBM) model were studied. Depth model overcomes the automatic encoder directly on the efficiency of multilayer network training problem. Then analyzes the experimental environment, data preprocessing, experimental data set, analyze the audio data processing by the depth of the characteristics of unsupervised feature learning DBN model. And the experimental results and the application of the subway traffic estimation are studied and analyzed.Simulation results show that the number of iterations for 50 cases, set different classification number simulation, when the classification number is 7 refactoring minimum error(9.50), Then put the data into the trained 4096-1000-500-250-2 model, the depth of the RBM model statistical color to red, black, green, shallow blue, pink, yellow, dark blue clustering number, then 0 to 50, 51-100 tag samples in the sample concentration training, comparing the result of the match number and the number of different clustering was consistent. It is concluded that each color clustering points corresponding to the subway traffic passenger information, and then realize the estimation of the subway traffic passenger.