基于人工鱼群-径向基神经网络的NO_(x)预测模型

NO_(x) Prediction Model Based on Artificial Fish Swarm-Radial Basis Function Neural Network

针对燃煤电厂脱硝系统入口NO_(x)质量浓度迟延较大、难以测量的问题,提出了人工鱼群算法(AFSA)优化径向基神经网络(RBFNN)预测模型。利用互信息确定模型输入变量,运用K-近邻互信息算法预估迟延时间;采用具有强泛化能力的RBFNN建立相空间重构的辅助变量和主导变量的预测模型,并运用AFSA确定RBFNN的最优参数组合,克服输入规律不明和参数随机性的影响。最后将AFSA-RBFNN预测模型与RBFNN、PSO-RBFNN预测模型进行对比验证。结果表明:AFSA-RBFNN预测模型的均方根误差、平均绝对百分比误差最小,运行时间最短,表明该模型的泛化能力、预测精度明显优于其他模型,并能够解决粒子群算法的局部收敛和运行时间长的问题。

Aiming at the problem that NO_(x) concentration at the inlet of SCR system for coal-fired power plant was delayed and difficult to measure,a prediction model based on artificial fish swarm algorithm(AFSA)optimized radial basis function neural network(RBFNN)was proposed.The input variables of the model were determined by mutual information while the delay time was estimated by K-nearest neighbor mutual information.Moreover,a prediction model of the auxiliary variables and dominant variables of phase space reconstruction was established by using RBFNN with strong generalization ability,and the optimal parameter combination of RBFNN was determined by AFSA,which can overcome the influence of unknown input rules and random parameter.Finally,the AFSA-RBFNN prediction model was compared with RBFNN and PSO-RBFNN prediction models.Results show that the root mean square error,the average absolute percentage error and the running time of the AFSA-RBFNN prediction model are minimum,indicating that its generalization ability and prediction accuracy are obviously better than those of other models,and the model can solve the problems of local convergence and long running time of the particle swarm optimization algorithm.