【作者】 周非；

【导师】 张永相；

【作者基本信息】 西华大学 ， 机械电子工程， 2006， 硕士

【摘要】 降低汽车排放和油耗是涉及环境保护和能源的国际性问题,从发动机控制角度而言,精确控制汽油机瞬态空燃比是解决这一问题的重要措施,也是一个具有挑战性的研究课题。对此,作者开发了采用RBF神经网络控制和逆控制空燃比的方法,并基于发动机平均值模型进行了这两种控制策略的MATLAB离线仿真和SIMULINK在线仿真。 首先,介绍了研究汽车发动机空燃比(特别是瞬态空燃比)控制的必要性和重要性,以及国内外研究现状。同时,根据对国内外汽车发动机模型的综述分析,得到在不同控制策略下选择发动机模型的方法。 接着,在分析国内外各种控制策略的基础上,采用神经网络和逆控制对汽车发动机空燃比进行控制的策略。通过对已有的神经网络控制方案进行分析比较,得出采用RBF神经网络的结论。逆控制控制空燃比的控制策略是作者提出来的,作者在中外刊物上没有找到用此方法控制汽油机空燃比的文献。文中对神经网络和逆控制的相关知识做了必要的介绍。 文中采用的发动机模型是由丹麦技术大学Elbert Hendricks教授和他的同事们提出来的,此模型具有较高的精度,它对不同于建模的三种发动机在整个运行域上的误差标准差仅为2～3%,对同一发动机采用不同进气歧管和喷油系统时也具有相同精度水平,被研究者们广泛采用。 在发动机模型和控制方案确定后,作者编写了RBF神经网络和逆控制的MATLAB离线仿真程序,并借鉴课题前期成果的方法,把自适应补偿算法加到控制程序中,从而构成自适应RBF神经网络控制和自适应逆控制策略。分别对简单油膜方程和复杂油膜方程作了MATLAB仿真分析比较,两种控制策略在瞬态过程都得到了很高的空燃比控制精度。随后还与课题前期的CMAC更多还原

【Abstract】 Decreasing emission and fuel consumption of automobile is an international issue about environmental protection and source of energy.To control the transient AFR(air-fuel ratio) of gasoline engine in high precision is an important measure to solve this issue, but it is a difficult and challenging subject which is well known of the world. Aiming to this subject, the author develops RBF(radial basis function) neural network and adverse model and applies it to control AFR of gasoline engine in this paper. The author then emulates the two controllers off-line under MATLAB software platform and emulates it on-line under SIMULINK software platform based on mean engine model.The author begins with the necessity and importance of AFR control(espe-cially the transient AFR) in automobile engine and researches on this issue that have been done both domestically and abroad. Then, a method which chooses a model under the different strategy of control is gotten based on the synthetically analyzing of engine models that have existed till now.Subsequently, the author adopts neural network and adverse model to be the controller of gasoline engine as the result of analyzing existed strategy. Furthermore, the author analyzes these neural networks that have been used in control area respectively and get the result that RBF neural network is a very good choice for AFR control of gasoline engine. The strategy that is adopted to control AFR is brought up by the author and nothing is found in all publications about this area. Knowledge which is related to neural network and adverse control is introducednecessarily in this paper.The gasoline engine model which was brought up by professor Elbert Hend-ricks and his colleague who works in Denmark technology university is adopted in this paper. This engine model has very high precision and its standard error is only 2-3% in whole working range for three different engine models. It has the similarly precision level to the same engine adopting different manifold and fuel injection system. So it is adopted widely.After engine model and its control strategy have been established , the author compiled MATLAB programs about RBF neural network and adverse model and emulated them off-line. In the meanwhile, a kind of adaptive compensation algorithm was added in the programs in order to constitute adaptive RBF neural network and adaptive adverse control system. Both simple fuel film of engine and complicated one have been emulated and these two kinds of control strategy all get very good result. Subsequently, analysis and comparison are accomplished between this result and the result which is gotten by CMAC neural network.Ultimately, the author uses S_function to express controllers of these two control strategy under SIMULINK software platform, so the SIMULINK emulation on-line model based on S_function is accomplished. Two controllers’ robustness and adaptability are tested by changing load^ parameter of fuel film> replacing function of fuel film and adding random noise in ignition angle in the SIMULINK emulation model. It is showed that both error is inside ± 2% between ideal AFR and AFR which is gotten using these two controllers, and both two controllers have good robustness. In the end , all kinds of results are analyzed synthetically, compared with the result which is gotten by CMAC neural network controller.There are 67 references, 80 figures and 9 tables in this paper.更多还原