Research On Data Processing Based On Gaussian Process

According to the given parameter model, computer can get the accurate output from input. Since the condition of the parameter model changes, the parameters cannot be updated immediately, the output will be inaccurate. Supervised learning has the ability to get the corresponding relation from the sample data. Corresponding relation will be change when conditions change. It will reduce the deviation of the output. Especially for the complex function relationship which parameter model is hard to define, supervised learning can also get the prediction results.In the actual train-operation, there are many objective factors changes such as track and train condition, weather and terrain that would not allow parameters model change simultaneously, delay and error accrued. In order to control the train better and improve the efficiency and safety, this thesis illustrated by the case of the train arriving, collected the data and used the gaussian process regression model on the simulation; Compared with the parameters model, it analyzed their advantages and disadvantages. It has a realistic significance for the practical application of the gaussian processes model. The thesis emphasized on several aspects as follows:(1)The gaussian process regression model was set up based on its mathematical characteristics and the main factors that influenced the predictive distribution of the model were analyzed.(2)The control strategy and the speed monitoring process of the train were analyzed. According to the foundation of train brake dynamics, the parameters model of train brake and the parameters model of the train target distance control were established by using the forward and reversed iteration. It was concluded that when the velocity change or the distance change was enough little, the distance of common brake and emergency brake which two methods calculated were basically the same and could satisfy the provision of the practical application.(3)The thought of using the nonparametric model on the simulation of train brake was proposed. According to different data, the train brake models based on multiple data and online data were established by using the gaussian processes regression model of supervised learning. It was concluded that the nonparametric model was closer to the actual braking process and the parameter model was more suitable for the train speed limit; through the simulation of different amount of training data, it came to the requirement of training data about the model based on the online data.The result indicates that the gaussian processes regression model in the nonparametric model is of feasibility and adaptability for the research of the train brake process.