Abstract:

The sewage treatment process has the characteristics of hysteresis，time-varying，strong coupling and nonlinearity. It is a complex system with multi-dimension input and multi-dimension output. According to the real-time influent quality，it is possible to judge whether some parameters of the future effluent quality exceed the standard，so as to adjust the process parameters in advance，which is of great significance to the optimization of the sewage treatment system. In order to solve the problem of long lag in the detection of total nitrogen（TN） in wastewater，a soft sensor model，based on Gaussian function nonlinear decreasing step firefly algorithm，was proposed to optimize support vector regression parameters（GNFA-SVR）. The improved firefly algorithm（FA） was used to optimize the penalty factor and kernel parameters of SVR. The characteristic components of influent quality were taken as the input of the prediction model to establish the GNFA-SVR total nitrogen of effluent prediction model. Simulation results showed that compared with LFA（linear decreasing step firefly algorithm）-SVR and FA-SVR prediction model，the GNFA-SVR model had fast convergence speed and high prediction accuracy，and the prediction error was within ±0.5 mg/L，which could achieve accurate prediction of total nitrogen of effluent.

Key words: firefly algorithm, support vector regression, parameter optimization, total nitrogen of effluent, prediction model, Gaussian function

摘要：

污水处理流程具有滞后性、时变性、强耦合、非线性等特点，是一个多维输入、多维输出的复杂系统。根据实时进水水质判断未来出水水质的一些参数是否存在超标可能，以便提前对工艺参数进行调整，对污水处理系统的优化具有重要意义。针对污水处理出水总氮（TN）检测大滞后的问题，提出一种基于高斯函数非线性递减步长的萤火虫算法优化支持向量回归参数（GNFA-SVR）的软测量模型。通过改进萤火虫算法（FA）对支持向量回归机（SVR）的惩罚因子和核参数寻优，将污水进水水质的特征组分作为预测模型输入，建立GNFA-SVR出水TN预测模型。仿真结果表明，与LFA（线性递减步长的萤火虫算法）-SVR和FA-SVR预测模型相比，GNFA-SVR组合模型参数寻优收敛速度快、预测精度高，且预测误差在±0.5 mg/L以内，可实现出水TN的精准预测。

关键词: 萤火虫算法, 支持向量回归机, 参数优化, 出水总氮, 预测模型, 高斯函数

CLC Number: 

• TP18

• X703