Study On Automatic Transmission Method Of Construction Vehicle Based On Support Vector Machine

Automatic transmission technology is the landmark of intelligent construction vehicles development in which the shift schedule, called shift-decided also, is the core problem for automatic transmission. The shift schedule is about gear's changing rule concerning the control parameters. Traditional shift schedule is built based on the precise object's precise mathematic model, while for the construction vehicles like loader, bulldozer or grade, it is rather difficulty to build a precise model under the complicated working conditions because of the greatly changed load. So a variety of the intelligent control technology is introduced into the area n recent years, like fuzzy control and neural network.In the fuzzy shift-decided system, both the input and output variables are expressed by fuzzy variables. But its membership functions are determined in advanced according to the survey results and the experts'opinions. If it is determined, the membership function will be fixed so it loses the ability of self-learning. The shift-decided method based on neural network makes empirical risk minimization as the summarized principle, so only when the training sample number tends to endless, its performances can reach the theoretical optimal. But in the shift-decided problem, the training sample is usually limited, so the previous shift-decided method is not ideal. A new developed machine learning method in recent years-support vector machine provides a more effective mean for us to resolve this problem.Our research is based on the project (No.59705005) named"engineering vehicle electronic control for saving energy"supported by National Natural science foundation of China and doctoral fund project (20020183003)"Research on four-parameter automatic transmission technology". This article with the aid of support vector machine method gets good classification performance on limited samples, nonlinear and high dimensional feature space. Against the character of automatic transmission of hydraulic driveling system, and in order to improve the working performances of hydrodynamic driving system, playing maximum dynamic and economic performance, we propose to use support vector machine to make the automatic transmission vehicle's shift decision. So we have made deep research on support vector machine method and its use in shift-decided.In chapter one, based on automatic transmission technical information, the development trends of vehicle's automatic transmission and process of automatic transmission shift schedule are outlined. We analyze the inadequate of the traditional control methods and smart stall control algorithm in solving the problem of decision-making deficiencies, and use the support vector machine to make the stall decision. At last the urgent issues in the process of the support vector machine method implementation are discussed.In chapter two, at first we introduce the components and characteristics of project vehicle power-train system, use the test data of hydraulic drive system, and set up the mathematical model through theoretical analysis and system modeling approach. At the same time, the characteristics of engine and torque converter are analyzed, and the performance evaluation criteria of the engine working with hydraulic torque converter are obtained. Through the analysis of the vehicle dynamics to set up an engineering vehicle dynamics model, we give the vehicles power equation. Finally, the principle of power split of construction vehicle power-train system is introduced.In chapter three, BP network is one of the most applied artificial neural net-work. It is based on traditional statistics. The content of traditional statistics is the progressive theory when the sample is infinity, that is, the statistic nature when the sample data tends to infinity, but the sample data is usually limited in the practical issues. So it is difficult to get the ideal applied effects by the algorithm supposing that the sample data is infinity. But support vector machine is based on the statistics theory, it can overcome the problem that neural network can't avoid.In chapter four, facing this practice problems like shift-decided, first we need transform it to the mathematic model that can be solved by support vector machine that is model selection. The model of support vector machine contain: feature selection, support vector machine's selection, support vector machine's kernel function as well as parameter selection. After determination of kernel function, the related parameters'selection is the key to choose the model. The generalization performance of the SVM classifier can be influenced by kernel parameters and errors penalty parameter C directly. By choosing RBF kernel function parameter( C ,γ) for example, this article analyses the character of bilinear Search method and grid search method that is a method to choose the parameter and the step of the algorithm's realization, and proposes bilinear grid search. Experiment results show that the proposed bi-linear grid search has the advantage of good performance and high predict accuracy compared to grid search, it can improve the study performance and enhance the study precision.In chapter five, traditional support vector machine is designed for two classified questions. The shift-decided of construction vehicle is a typical multi-classification problem. This paper analysis the questions and shortcomings of the existing multi-classification method, and improve the existing support vector machine multi-classification method, which is based on bintree-coding. With the character of construction vehicle shift-decided has higher request of decided speed, it proposes a support vector machine multi-classification method based on bintree-coding. This method distributes the classifier at each node and thus constitutes a multi-classification support vector machine. For N-type questions, it only needs construction N-1 SVM classifier, and this method does not exist the domain that can not be identified, and it is not necessary to traverse all the SVM classifier when classification is made. The SVM classifiers can significantly improve the speed of training and classification. This paper combines with operating characteristics of the vehicle to build a bintree structure and give the solving process. Through training, get the model parameters, and by the ability test, accuracy rate is above 97%. Several simulations demonstrate:in the case of the classification accuracy is almost not affected, this method can significantly reduce the classification time, and improve the speed of training and decision.In chapter six, the goal of the automatic transmission system of the construction vehicle is to make shift-decided with time according to the change of the controlled plant, this is a key link in the progress of the industrialization of the automatic transmission technology, and also is a difficult problem of applying the support vector machine theoretical achievement to project practice. In order to solve the reliability and real-time problems of shift decision, combined with the features of automatic transmission in construction vehicle hydraulic system, this article introduces the development process of ZL50 wheel loader's automatic transmission electronic control unit. We elaborate the design and development model of electronic control systems, complete the design of the system. Based on introduction of the main performance and the internal resources of automotive electronics ASIC MC9S12DP128, we highlight the design of signal acquisition, output control, man-machine interface, communication and power supply circuit, and combined with the engineering characteristics of the system, add the anti-jamming measures to meet the multi-mission, reliability, real-time demand of automatic transmission electronic control system. ECU software applies the modular design to ensure the system hardware to achieve control mission reliably.In chapter seven, an experimental study for automatic transmission system is made. We make the test of automatic transmission control system on test stand of electric control system, and verify electronic control unit and automatic transmission control algorithm based on support vector machine. According to the sample data to train the ECU, we get support vector machine model parameters, and determine the shift schedule. ECU makes different shift decision in different working conditions in accordance with the shift strategy through training, and output the test results and analysis the conclusions. The results of experiment show that the control system of automatic transmission can make training according to the supplied sample data, get the support vector machine model parameters, and obtain energy-saving shift schedule that is suitable to improve system efficiency and power of construction vehicle. According to environmental conditions, we apply these shift strategies to make shift decision of the test condition, and optimize shift transmission to make torque converter to work at high-performance areas, aim to achieve energy-saving control.Chapter eight is the summary of the whole dissertation, and the main research achievements and innovative discovers. Suggestion and prospect of automatic transmission research are given.