Study On Neural Network For The Recognition Of Blast Furnace Top Gas Temperature Distribution

Gas flow distribution is correlative with the temperature distribution in blast furnace, the shape of the cohesive zone, the smooth state of blast furnace and using status of gas flow , finally influence the smelting index of blast furnace. The target of blast furnace operation mainly is to achieve proper and optimum gas flow distribution. At the same time, the gas flow distribution also is the important foundation for blast furnace operators to estimate the state of blast furnace.The blast furnace is a closed system, and the change of gas flow distribution is invisible, so the only method of estimate the status of blast furnace is to utilize records of all kinds of sensors. The conventional ways of identifying gas flow distribution is mathematic model based absolute mechanism. But this method is too complex , be subjected to on line operation and is difficult to on line analyzing and control , so can not to feedback information in time and to adjust burden rule. The artificial neural network model has strong fault - tolerant performance, learning performance, self-adaptive performance and non - linearity map ability, and it is adaptive to solve some problems like non - determinacy inference of complex causal relation , judgment , recognition , classification and so on. The artificial neural network model is defined by network topological structure, neuron characteristic function and learning method. The production process of blast furnace is a large - scale distribution parameter system. We can obtain all kinds of detection value in altitude - direction and radius - direction by many sensors. After that, to recognize their character distribution patterns by artificial neural network, and using these to diagnose and control the condition of blast furnace.In fact, gas flow has not immovable distribution patterns, so this paper recognize gas flow distribution pattern by a self-organization neural network. The self organization neural network is a kind of unsupervised learning network, it can learns automatically from surrounding, can self -organize, self - stabilize and large-scale parallel - process random multi and complex two - dimensional patterns, and find some significative rules from inputted data in the condition of unsupervised data. We arranged 25 gas flow distribution patterns (5X5) from multiple production data of the Ne 1 blast furnace of Bao Steel, the 25 patterns are mapped on a two - dimensional net diagram , but also the coordinates of those similar patterns is also closer in the diagram, namely, the method has classified function. Using this model, the operators can recognize expediently gas flow'distribution. The model describes expediently actual gas flow distribution, and establishes expediently the relation between temperature data, burden mode, charge shape, the ore - to - coke ratio, and so on. The model is a sub - model of distribution model of Ne 1 blast furnace of Bao Steel, and is working on line.