Magnesium Reduction Degree Prediction Based On BP Neural Network Optimized By Genetic Algorithm And Temperature Field Simulation In Reduction Pot

In Pidgeon Process the establishment of magnesium reduction degree predictionmodel, getting the functional relationship between process parameters and magnesiumreduction degree is beneficial to improve the automation level of Pidgeon Process, thusguides the optimal choice of process parameters in order to increase magnesiumreduction degree, reduce production cost and achieve energy conservation andemissions reduction. Magnesium production process is a very complicated nonlinearprocess, in which the process parameters that influence the magnesium reduction degreeare much and complicated. Achieving the nonlinear relation of the process parametersand magnesium reduction degree with the traditional mathematical model is not only alarge amount of circulation, and it is difficult to guarantee the accurac y of the modelwhich can predict the magnesium reduction degree relatively precisely. Using theartificial neural network technology which has the characterisitics of self-learning,self-organization and self-adaption calculates automatically input samples in computerscan obtain the function relation and variation law of the process parameters andmagnesium reduction degree and establish an accurate prediction model. The model canbe used as technical basis for magnesium smelting process.The principle and characteristics of BP neural network and genetic algorithm areintroduced. With research background of magnesium smelting process in an enterpriseof north of Shaanxi, Pidgeon process is narrated briefly and the influence of variousprocess parameters on magnesium reduction degree is analysed. The magnesium reduction degree prediction model based on BP neural network optimized by geneticalgorithm with the input including calcined dolomite activity, silicon ratio, pelletizingpressure, reduction time, reduction temperature, vacuum degree and the outputmagnesium reduction degree is established to study the relationship between the processparameters and magnesium reduction degree on the basis of these above. The model isrehearsed and tested by the actual production data. The results show that the predictionmodel could relatively precisely predict the magnesium reduction degree, the hit rate ofthe model with△ηMg±1.0%is about96%, the maximum error is less than1.3%.RMSC said the root-mean-square error of the actual and estimated magnesiumreduction degree, the RMSC of GA-BP prediction model is0.4210. Compared withstandard BP neural network prediction model GA-BP prediction model has higherprediction accuracy.Temperature field variation in reduction pots in Pidgeon process is simulated bythe finite element analysis software ANSYS, the temperature distribution and variationin reduction pots in down-draft kiln and regenerative reduction furnace is analysed. Theresults show that the regenerative reduction pots are heated relatively uniformly andhave high transfer efficiency, the central pellets reach the reaction temperature within9hours, a saving of about an hour by comparison to the down-draft kiln.