Temperature Prediction On Blast Furnace Based On Interval Wavelet Neural Network

Iron and steel industry is the pillar industry of our country and plays an extremely important role in the rapid growth of national economy. Blast furnace is the core equipment of iron making in blast furnace smelting process, and the furnace temperature control which is the basis of the blast furnace process control is the most important. Furnace temperature is the most directly reflect of the operation situation of the blast furnace, and the silicon content of hot metal and furnace temperature is direct ratio relation. Take the shortcomings of traditional intelligent algorithms into consideration, which is the badly anti-interference ability and higher sensitive to the accuracy of the data, this thesis studies the interval neural network algorithm to make up for these deficiencies and forecast the silicon content of hot metal to achieve the purpose of predicting temperature of a blast furnace.This thesis, taking BaYuJuan 1# blast furnace in Anshan Iron & Steel Company as the background, studies the blast furnace temperature prediction method based on interval wavelet neural network. The specific work is as follows:According to the analysis of the interval neural network, the research about interval neural network algorithm is mainly focused on the situation that the weight and threshold values are interval value, and the input value is point value. The inverse modification method of the weight is analyzed in detail and the method is validated by numerical experiment. According to the characteristics of the blast furnace temperature prediction model, wavelet is introduced into the interval neural network to solve the shortcomings, which are the slowly learning rate and easily fallen into local minimum. Two types of interval wavelet neural network, relaxing and tight, are constructed in this thesis. The manners of the parameter modification method of the weights, scale factor "a" and shift factor "b" are provided too. The simulation analysis verifies that the width of the prediction interval is decreased significantly.The interval neural network and interval wavelet neural network models are established to suitable for the prediction of blast furnace temperature. Firstly, the dimension of the multidimensional input variables is reduced based on PCA method and new data is received which is used to train network.Finally, this thesis predict the BaYuJuan 1# blast furnace temperature using the interval wavelet neural network model, and calculate the shooting and the relative error of these three methods to verify the effect and feasibility. According to the comparison of these three methods, the effect of the tight interval wavelet neural network is better than the other two. Thus this method can be used in the blast furnace temperature prediction.