Study On Neural Networks Machine And Its Application In Control

This dissertation focuses on research of neural networks algorithm and applications of neural networks theory in the control field.Neural networks can be thought as a kind of mathematic tool which is independent on model.Neural networks have strongly adaptive and learning capability and are suitable for those objects which has uncertainty and non-linearity.Neural networks theory is an important branch field of artificial intelligence. This dissertation gives a brief summary on neural networks theory and researches the improved algorithm of a neural networks and applications of neural networks in the control,pattern recognition,fault diagnose.The main contributions of the dissertation are as following:1.We introduce neural networks' development history and review its basic content and learning theory.We also analyse and expatiate the research at present in the world mainly from two aspects of algorithm and application, respectively.2.On the base of detailedly analysing the fourier neural networks, we find this neural networks have the characteristic which can transform the nonlinear mapping into linear mapping. So,we improve the original learning algorithm based on nonlinear optimization and propose a novel learning algorithm based on linear optimization(This dissertation adopts the least squares method).The novel learning algorithm highly improve convergence speed and avoid local minima problem.Because of adopting the least squares method,when the training output samples were affected by white noise, this algorithm have good denoising function. 3.The modeling method based on mechanism analysis and identification method always exits unmodeled high-order part and the modeling method based on neural networks usually has not good enough generalization capability.We fuse above two kinds of modeling method and put forward a hybrid modeling method based on mechanism analysis, identification and RBF neural networks. This paper proposed a hybrid modeling method based on mechanism analysis, identification and RBF neural networks.First,Get a object's low-order model by the mechanism analysis and identification method.Second,adopt RBF neural networks modeling method to compensate unmodeled high-order model.The sum of the low-order model and high-ordermodel is the hybrid model.This kind of hybrid model has more accuracy than a model based on mechanism analysis and identification and has more generalization capability than a model based on neural networks.4.To a kind of nonlinear time-varying system whose state can not be measured, proposes a kind of state observer based on BP neural networks inverse model and the state observer can real-time observe the system's state.S.We studied how neural networks are applied to fault diagnosis and concretely study how neural networks are applied to freeway traffic incident detection and proposes two novel detection algorithms.The first is a freeway traffic incident detection novel algorithm based on RBF Neural Networks and SOM neural networks.Use a RBF neural networks to get the freeway traffic flow model and use the model to obtain the output prediction. The residual signals will be gotten from the comparison between the actual and prediction states.Use a SOM neural networks as a classifier to classify characteristics contained in the residuals.So,we can detect whether the traffic incident has happened.This algorithm can detect not only whether the traffic incident has happened but also the level of accidental congestion caused by the traffic incident The second is another freeway traffic incident detection novel algorithm based on ART2 neural networks.This algorithm uses the freeway traffic flow model and ART2 neural networks as observer and classifier,respectively.The residual signals will be gotten from the comparison between the actual and estimated value of observer.Use the ART2 neural networks to classify characteristics contained in the residuals.So.we can detect whether the traffic incident has happened.This algorithm can recognize new pattern at the same ti...