基于改进粒子群优化长短时记忆神经网络的脱硫系统SO_(2)预测模型

Prediction model of SO_(2) concentration in desulfurization system based on improved particle swarm optimization LSTM

针对燃煤电厂脱硫系统出口SO_(2)质量浓度难以稳定控制的问题,提出了一种基于改进粒子群算法(IPSO)优化长短时记忆(LSTM)神经网络的IPSO-LSTM预测模型。首先利用主成分分析(PCA)计算各个变量的贡献率从而筛选出模型的辅助变量,实现辅助变量的降维。其次,利用改进粒子群算法确定LSTM神经网络的神经元数量、学习率和迭代次数。最后,将选定的辅助变量作为IPSO-LSTM预测模型的输入,预测出口SO_(2)质量浓度,采用国内某电厂2×600 MW机组脱硫数据进行仿真,并与相关11种模型进行对比。仿真结果表明,本文模型预测误差最小,其均方根误差为0.98 mg/m^(3),平均相对误差为1.81%;与传统LSTM、LSSVM模型相比,预测精度分别可提高72%和81%;与其他相关模型相比,改进的PSO可以提高PSO的全局寻优能力和收敛速度,当LSTM神经网络具有2层隐含层时,IPSO-LSTM模型预测精确度最高。

Aiming at solving the problem that it is difficult to control the SO_(2)content at outlet of the desulfurization system in coal-fired power plant stably,a prediction model IPSO-LSTM based on long short-term memory(LSTM)neural network optimized by improved particle swarm optimization(IPSO)is proposed.Firstly,principal component analysis(PCA)is used to calculate the contribution rate of each variable to filter out the auxiliary variables of the model,so as to achieve the dimensionality reduction of the auxiliary variables.Secondly,the improved particle swarm algorithm is used to determine the number of neurons,the learning rate and the number of iterations of the LSTM neural network.Finally,the selected auxiliary variables are used as the input of the IPSOLSTM model to predict the SO_(2)content at the outlet.The desulfurization data of a domestic 2×600 MW power plant is used for simulation and the result was compared with 11 related models.The simulation results show that,compared with other conventional models,when the LSTM neural network has two hidden layers,the IPSO-LSTM model has the highest prediction accuracy,and the improved PSO can improve the global optimization capability and convergence speed of the PSO.