Intelligent Train Operation Models Based On Data Mining

Due to the advantages of high efficiency, economy, energy saving and environmental protection, urban rail transit has irreplaceable role in solving the problem of traffic congestion, energy consumption and environmental pollution. Automatic Train Operation (ATO) system, as a significant equipment of replacing drivers, could control the speed of train real-time which has a direct impact on the energy consumption of train operation, comfort of passengers, the running time and stopping error. In pursuit of green economy and low-carbon life, researching more superior operation models have important practical significance for reducing the operating costs and improving transport efficiency of urban rail transit.According to the observation of subway drivers’driving strategies and analysis of driving data, we find that experienced drivers can operate the train to the specified location on time with a few times changing of handle, and idle running time is long, energy consumption is low, riding comfort is high. So, combined with data mining methods and expert experience, a new control strategy based on data of manual driving is first presented to achieve the multi-objective control of train operation. The main research work is as follows:Firstly, according to the acquisition of field data in Yizhuang Line, combined with running time error, stopping error, switching times of controller’s output, energy consumption and comfort these five performance index, the outstanding data was sorted out and sieved out to establish the standard database.Secondly, the data mining algorithms used in this paper are expressed. Mine the data of continuous operation handle train and discrete operation handle train using two regression algorithms (B-CART and L-CART) and three classification algorithms (KNN, B-CART and A-CART) respectively. Dig out the outstanding driver strategy, and describes the expert experience and heuristic stopping algorithm. Build the Simulink simulation model of train operation control system and Graphical User Interface (GUI) in MATLAB software, and we call it ITO model simulation platform.Finally, this model is simulated in the platform with actual line data of Beijing Yizhuang subway. Simulation results show that the train is kept at a steady state which completely corresponds to drivers’ driving law. Compared with Proportion Integral Derivative (PID) control, this model is better with higher comfort, less energy consumption, and it meets the requirements of running time and stopping precision. Compared with the average level of manual driving, also this model has a better performance. For the continuous operation handle train, L-CART algorithm is better than B-CART algorithm; for the discrete operation handle train, A-CART algorithm is best in three classification algorithms. Furthermore, results of simulations with different parameters of train model verify the robustness of ITO model and analyze the relativity of ITO model, and there are significant correlations between running time error, stopping error and the model parameters. Complex cases of speed limits and steep gradient verify the adaptability of ITO model.