Study Of Wavelet Neural Network And Its Application To Predict The Slab Temperature In Reheating Furnace

Neural network has very strong nonlinear approximation ability, self-study ability, self-adaptive ability and stronger fault tolerance, thus it can trail and catch object's change and change tendency very well because of the influence of various uncertain factor, as well as it can reflect well the internal mechanism in complex industrial process. So it is a kind of method that has been studied and used in many fields in recent years. But when it is applied in practice, the training time of network is long, convergence speed is slow and local minimum value can't avoid, so that its application has been restricted in the control of real time. In order to make neural network get better application, wavelet transform is led into neural network to form a new type wavelet neural network. Wavelet neural network combines the local property of wavelet transform with the advantages of neural network, so it has overcome theblindness of structural design of former neural network and it has better approximation ability, fast convergence speed and can avoid the local minimum value efficiently. Wavelet neural network chosen in this paper is replacing the conventional sigmoid function in single hidden layer neural network with wavelet function. At the same time, the weights of input layer to implicit layer and the valves of implicit layer are substituted by the dilation parameter and the translation parameter of wavelet function, and output layer is linear neuron, which overlay linearly the wavelet dilation coefficient of hidden layer and outputs it. According to some existent problems in wavelet neural network algorithm, this paper makes plenty of research work on it and has put forward some measures to improve network. The main contents are (1) a kind of method to select the wavelet neural network initial values; (2) how to definite the number of hidden layer nodes of the network; (3) the self-regulating learning algorithm of wavelet neural network.Slab reheating furnace is an important equipment used to reheat the slabs before rolling in hot steel rolling industry. In reheating furnace, slab temperature and its distribution are important targets to appraise the slab heating quality as well as realize reheating furnace optimization control. But the uncertainty and complexity existing in reheating furnaceproduction course makes it difficult to use traditional method to establish precise prediction temperature model. At present, it has become an important research direction to study a kind of slab temperature prediction model that has simple structure and is fit for on-line operation.So the wavelet neural network algorithm put forward in this paper is applied to predict the slab temperature of reheating furnace, and it has been demonstrated that the predicting precision of wavelet neural network is higher, the adaptability is stronger and the convergence speed is also faster than simply neural network obviously by simulation results.