Predicting Pulverized Coal Plasma Ignition Performance by BP Neural Network

Abstract

To make sure the major factors and their influence for pulverized coal plasma ignition(PCPI), the way predicting the PCPI was investigated in this paper. The back propagation(BP) neural network was used to established a prediction model which can study by itself for PCPI. Then the sample database was set up by simulating the PCPI in kinds of conditions. After that, the prediction model was trained by sample database to improve the prediction level. At last, the prediction model was used to predict the PCPI in new conditions and the prediction error is under 0.004. The research show that the BP neural network can predict the PCPI correctly. In this paper, the BP neural network was applied to predict the PCPI innovatively, and the prediction efficiency increase highly and the prediction accurancy does not deline. Introduction During the operation PCPI, plasma torch injects into the pulverized coal stream to form a stable flame, which is poured into the furnace of boiler. PCPI technology has attracted attentions worldwide because it can be used in the pulverized coal fired boilers to realize the startup or stable combustion in part load operation. It has been a promising way to reduce oil consumption in coal fired power plants. Up to now, PCPI systems have been used in about 550 boilers in China, which have a total capability of near 230 GW. The PCPI processes have been investigated. Among these, Masaya Sugimoto et al.[1] investigated the ignition processes with different coal in a drop tube, discussed the power demand to realize the success of ignition under differential excess air ratio. E.I. Karpenko et al. [2], focused on a 200 MW boiler, studied the plasma ignition processes with Reynolds Average Navier Stocks (RANS) simulation and experiments in this boiler. Zhang xiaoyong et al. [3], studied how to design the multistep plasma ignition burner with RANS and experiments. With the development of PCPI both in practical applications and theory investigations, how to predict the main parameters of the PCPI becomes a challengeing problem. Traditional simulation methods, such as RANS and LES, can not meet command in practice because of prediction efficiendy. The back propagation(BP) neural network is an proper approach to slove this problem. It has been used in both laminar and turbulent reactive flows, as an alternative to the conventional kinetics evaluation, which can reduce the CPU cost largely [4~7]. But the applicability of BP neural network in two phase flows, especially in the pulverized coal combustion processes, still remains to be demonstrated. How to predict the PCPI by BP neural network was investigated in this paper. Establish Prediction Model The back propagation(BP) neural network is the most popular network architecture now. The transfer function of the neurons in BP neural network is Sigmoid differentiable function, so it can deal with the nonlinear mapping problems. The momentum—adaptive learning rate method is used to improve the performance of algorithm. This improved BP algorithm is to add a proportion which is proportional to the variable quantity of last weight to each weight adjustment quantity and adjust the learning rate automatic in the learning iteration process. After one cycle of the learning sample, the learning rate will change according the variation of errors. Because of momentum coefficient, the network can avoid the trap of local International Conference on Modeling, Analysis, Simulation Technologies and Applications (MASTA 2019) Copyright © 2019, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Intelligent Systems Research, volume 168