An Optimized Speed Curve Generation Algorithm for Trains Based on CNN

With the continuous development of train control and communication technologies, trains are also developing towards automatic control and autonomous control. The train speed curve generation is one of the key research contents. Currently, some promising optimization algorithms of the train speed curve failure to meet the punctuality requirements in practical applications. For reducing the time of generating speed curve and improving the punctuality of the speed curve generation algorithm, an optimized speed curve generation algorithm for trains on the basis of Convolutional Neural Network(CNN) is proposed in this paper. This algorithm can learn the speed curves produced by different optimization algorithms or the speed curves of trains driven by excellent drivers, on the premise of ensuring energy-saving, punctuality, and other performance indexes of the curve. Based on the existing speed curve optimization algorithms, taking the Rolling Optimization Algorithm of Train Speed Curve(ROATSC) as an example without loss of generality, this paper uses the speed curve produced by this algorithm to train the CNN model, and uses the freight train as an example for simulation. Simulation results are achieved with different train weights and lengths, the CNN model speed curve fits the ROATSC algorithm speed curve with an approximation accuracy index 98%, which is higher than that of the Multilayer Perceptron(MLP) and Long Short-Term Memory network(LSTM) models. After testing, the speed curve produced by this method can not only decrease the generating time of speed curve, but also maintain the ROATSC algorithm performance. As seen in simulation results, the presented algorithm provides an additional choice to apply some algorithms of train speed curve generating in Automatic Train Control (ATO) system with the real-time requirement.