| In every department of national economy, plate and strip has extensive and important application. One of the quality indexes is shape(that is flatness). The precision of plate and strip acquired by department is becoming more and more high. As a result, shape control becomes the key technique and important develop direction for modern times high precise rolling mill. As far as the technique is concerned, theoretical basis and key scientific problem is the theory and mathematic model of shape control.Shape control is the key technique and international latest research discussion. Based on the neural network, one branch of artificial intelligence and ability theory, this paper selects INTELLIGENT CONTROL AND RECOGNITION OF FLATNESS FOR COLD ROLLING STRIP MILL as research task with theoretical and engineering sense, makes embedded theoretical research and industrial application for the plant such as the pattern recognition of HC rolling shape, the strategy of shape control and advanced enactment model.Flatness pattern recognition is the key to shape control. There are many problems to be solved in the following. For example, different topologic configuration network is needed to accomplish recognition assignment when the width is changing. The large learn assignment, slow convergence and local minimal in the network also make the problem more difficult. To get over the above problems, first the flatness pattern recognition with only 6-input and 3-output has been established based on Lerande orthodoxy polynomial, not only with simple structure but also physical significance definitude, it has higher discriminating precision. The GA-BP neural network optimal algorithm based on Euclid distance between normal patterns is exploited. The network uses the Euclid distance between the signal and standard stylebook as its input. The combination of adaptive genetic and BP algorithm is used to optimize the configuration of network and the weight. With the genetic algorithm's global optimization and BP algorithm's local extraction, the optimal neural network is produced. Using the optimal neural network, the optimal performance and-HI-the best configuration can be maintained in the training. Therefore the configuration of network is unchanged when the width is changing. In other words, it realizes the intelligence of the shape signal's pattern recognition and brings out basis for the establishment of shape control scheme.Improve on traditional LSM flatness pattern recognition, direct recognition method has been built, it's simple, applied and high precision. In conclusion, this paper not only puts forward two new methods of pattern recognition, but also solves the problem that traditional or fuzzy logic pattern recognition can not solve or solve well. It enlarges the range of pattern recognition and gets over the defect and deficiency in the traditional shape pattern recognition based on the least square.It is known that the search of the traditional genetic algorithm has some blindness such as the slow convergence and the unstable speciality. Particularly when the system is large. To advance the optimal effect, it needs much time. Besides, it is not fit with the precise adjustment and is difficult to conform the place.A new adaptive genetic algorithm with BP algorithm to optimize weight is backed up. The algorithm which combines the merits of the global convergence of genetic algorithm with fast local researching of BP algorithm not only intensifies the gradual convergence and evolution ability but also advance the speed of convergence, precision of training and generalization. To sump up, the new algorithm avoids the above problems and can be applied to training the network sufficiently.Hydraulic pressure bend roller is the basic segment of AFC system. It is known that the dynamic characteristic and stability of hydraulic pressure bend roller has key action on the whole AFC system (primary means quick regulation and transition interval). For the nonlinear, time varying and uncertainty existing in th...
|
PDF Full Text Request