Research On Gas Production Prediction Of Texaco Coal Water Slurry Gasifier Based On Improved BP Neural Network

Coal gasification is an important technology to use coal cleanly and efficiently.As the core equipment of coal gasification technology,gasifier plays an important part in producing gas safely,economically and environmentally friendly.Texaco coal water slurry gasifier occupies an important commercial position in the domestic market.Therefore,this paper predicts the gas production of this type of gasifier,which is of great significance for guiding and adjusting the production plan and synthesizing downstream chemical products.Different from the traditional chemical simulation model of gasifier syngas,this paper proposes an artificial intelligent model based on principal component analysis and improved adaptive genetic algorithm to optimize BP neural network from the perspective of data analysis.15 variables that affect the gas production are collected from three aspects:coal quality,coal water slurry and operation process.After eliminating outliers by the criteria of 3σ,there are 603 data,503 data are used as training sets to build simulation models,while the remain 100 data are used to predict the gas production.In order to eliminate the multicollinearity between variables and reduce the data dimension,the first five principal components with a contribution of 84.148%are extracted by principal component analysis as the input variables of BP neural network.This paper systematically introduces the basic theory of neural network,genetic algorithm and its adaptive improved algorithm.After using the maximum and minimum values to normalize the data,comparing the errors and convergence steps under different BP algorithms,LM algorithm is selected.By comparing the errors of different hidden layer nodes,the optimal number of hidden layer nodes is 10.The advantages of BP neural network are high nonlinear fitting capability and needless of the mathematical information,while the shortages are long training time and low global search ability.In this paper,genetic algorithm is introduced to optimize the initial connection weights and thresholds of BP neural networks,which reduces the prediction errors.In order to further reduce the prediction error,an adaptive improvement is proposed for the crossover and mutation with fixed probability in genetic algorithm,and compared with the results of particle swarm optimization.Finally,the results of PCA-BPNN,PCA-GA-BPNN,PCA-PSO-BPNN are compared with those of the PCA-improved AGA-BPNN model proposed in this paper.Experimental results:(1)The number of iterations in which the optimal fitness tends to be stable is reduced from 35 to 8.Hence,the number of model training is reduced and the prediction efficiency is improved.(2)The mean absolute percentage error of the model prediction results decreased from 0.0116 to 0.0103,the mean absolute error decreased from 3.5273×10~3 to 3.1581×10~3,the mean square percentage error decreased from0.0176 to 0.0140,and the root mean square error decreased from 5.2362×10~3 to 4.2403×10~3.The results show that the model proposed in this paper is better than the other three in the prediction of gas production.