基于自适应粒子群优化BP神经网络的氨法烟气脱硫效率预测

Efficiency Prediction of Ammonia Flue Gas Desulfurization Based on Adaptive PSO-BP Model

针对氨法烟气脱硫效率的预测问题,建立了以脱硫系统运行中8个主要参数作为输入变量的BP神经网络模型,采用粒子群优化算法(PSO)对建立的BP神经网络模型的权值进行优化,提出基于粒子群优化算法的BP神经网络(PSO-BP)预测新模型,并利用某电厂脱硫系统20组运行数据对该模型进行了验证.结果表明:采用PSO算法对BP神经网络的权值和阈值进行寻优,避免了网络局部极小值的出现,提高了网络的泛化能力,采用PSO-BP预测模型可以对氨法烟气脱硫效率进行较高精度的预测.

To improve the prediction accuracy of ammonia flue gas desulfurization（FGD） efficiency, a BP neural network model was established by taking 8 main operation parameters as the input variables, fol lowing which a new PSO-BP model was set up based on particle swarm optimization（PSO） algorithm via optimization on the weight of the previously-established BP network model. The new PSO-BP model was verified with 20 sets of operation data from a desulfurization system of power plant. Results show that net work local minimum can be avoided by using PSO algorithm to optimize the weight and threshold of BP neural network, thus the generalization capability of network can be improved, and subsequently high accuracy can be obtained in predicting ammonia FGD efficiency by using PSO-BP model.