基于FRBPSO-RBF神经网络的污水BOD5软测量方法

Soft Sensing Method of BOD5 in Sewage based on FRBPSO-RBF Neural Network

污水处理过程中污水BOD5难以实时准确测量,故软测量方法逐渐被用于污水BOD5的预测,其中RBF神经网络软测量方法应用广泛,但存在训练过程易陷入局部极值等问题。为提高RBF神经网络的预测精度,提出了基于适应度排名的粒子群算法(fitness ranking based particle swarm optimization,FRBPSO),根据适应度排名与迭代次数确定惯性权重的大小,并根据粒子个体历史最优值的排名与迭代次数确定自我学习因子与社会学习因子的大小,并将FRBPSO算法引入RBF神经网络的参数训练中。基于13个基准测试函数与其他3个粒子群优化算法对比,实验结果显示FRBPSO算法的寻优能力相对较强。再将基于FRBPSO算法的RBF神经网络用于构建污水BOD5软测量模型,仿真结果表明,在测试数据中,FRBPSO-RBF软测量模型的平均绝对误差比PSO-RBF软测量模型、DAIW-RBF软测量模型、SCVPSO-RBF软测量模型分别降低了0.7178、0.2402、0.5851,平均绝对百分比误差分别降低了0.47%、0.15%、0.33%,均方根误差分别降低了0.0034、0.0015、0.0039。与其他3个基于PSO算法的BOD5软测量模型相比,FRBPSO-RBF模型具有较高的BOD5预测精度。

During the wastewater treatment process,it is difficult to accurately measure the BOD5 of wastewater in real time.Therefore,soft measurement methods are gradually being used to predict BOD5 in wastewater.Among them,the RBF neural network soft measurement method is widely used,but there are problems such as the training process being easily trapped in local extremes.In order to improve the prediction accuracy of the RBF neural network,a fitness ranking based particle swarm optimization algorithm(FRBPSO)is proposed.The size of the inertia weight is determined accord-ing to the fitness ranking and the number of iterations,and the size of the self-learning factor and social learning factor are determined according to the ranking of the individual's historical optimal value and the number of iterations.Then,the FRBPSO algorithm is introduced into the parameter training of the RBF neural network.Based on 13 benchmark test functions and comparison with other 3 particle swarm optimization algorithms,experimental results show that the FRBPSO algorithm has relatively strong optimization ability.Then,the RBF neural network based on FRBPSO algorithm is used to construct a soft measurement model for wastewater BOD5.Simulation results show that in test data,compared with PSO-RBF soft measurement model,DAIW-RBF soft measurement model and SCVPSO-RBF soft measurement model,the aver-age absolute error of FRBPSO-RBF soft measurement model decreased by 0.7178,0.2402 and 0.5851 respectively;The average absolute percentage error decreased by 0.47%,0.15%,and 0.33%respectively;The root mean square error decreased by 0.0034,0.0015 and 0.0039 respectively.Compared with other three BOD5 soft measurement models based on PSO algorithm,FRBPSO-RBF model has higher BOD5 prediction accuracy.