Application Of Optimized BP Neural Network In Coal Thermal Transition

Based on the characteristics of energy consumption structure in China and to meet the requirements of energy security and environmental protection, efficient clean conversion of coal and its by-products utilization has become an inevitable trend. But coal pyrolysis, gasification and other thermal conversion approaches often require harsh conditions such as high temperature and pressure. At the same time, the coal structure and the reaction mechanism are so complicated that the law of the conditions and product properties between the various reactions is difficult to summarize. In artificial neural network, BP neural network is a strong non-linear empirical model. However, the traditional BP neural network has slow convergence and easy to fall into local minima and other defects. Therefore, we use a variety of methods to optimize BP neural network, and to establish the liquefaction residue catalytic gasification model and the coal pyrolysis reaction model which based on BP neural network. That will provide an important reference and basis for process design and optimization of coal and its by-product conversion process.In the liquefaction residue catalytic gasification model, we select catalyst relative molecular mass, the melting point and addition of the catalyst, the final gasification temperature as network inputs, gasification rate, the maximum gasification rate and the reaction index as network outputs. First, we establish multi-output prediction model, includes BP multi-output prediction model, improved BP multi-output prediction model and variable excitation function BP multi-output prediction model. The average relative errors of these models' prediction samples are all less than 6%, which have a good prediction effect. Compering three models we found that, improved BP multi-output prediction model and variable excitation function BP multi-output prediction model are better than BP multi-output prediction model weather in convergence speed or in prediction accuracy. Then, we establish BP single-output prediction model and analyze the model. We can see the type of catalyst is greatest influence factor, followed by the final gasification temperature, the addition of catalyst has least effect.For coal pyrolysis system, we establish helium atmosphere coal pyrolysis model, helium mixing hydrogen atmosphere coal pyrolysis model and different atmosphere coal pyrolysis model. After algorithm comparison, we select PSO-GA-BP algorithm to establish model. We make seven kinds of coal's weightlessness data as the training sample to predict the residual mass percentage of another kind of coal at each temperature. Thereby we obtain the weightlessness curve. The average relative errors of three models'prediction samples are 1.38%?0.75%?1.18%, and the predicted weightlessness curve is consistent with actual weightlessness curve, which has a high prediction precision. It also suggests that PSO-GA-BP model can effectively overcome the slow convergence and easy to fall into local minima and other defects of BP network, which has a good applicability to the complex systems containing large amounts of data. Through the analysis of the model, we know that temperature, ash and the C content of coal have more significant influence on the process of weightlessness.