论文部分内容阅读
以某企业造纸废水厌氧处理系统为对象,基于装置运行及过程机理分析,结合专家经验知识,选择8个过程变量为输入变量,并借助现场传感器采集这些变量的工业运行数据,进而构建出水水质指标化学需氧量(COD)输出变量的PCA-PSO-LSSVM软测量模型:首先,选用主成分分析法(PCA)执行数据样本输入变量的预处理,以消除变量间的相关性,完成输入变量的降维和主成分提取;然后,实施主成分与出水COD间的最小二乘支持向量机(LSSVM)数据建模;考虑到LSSVM模型中核函数宽度和惩罚因子对模型性能有较大影响,再通过粒子群优化算法(PSO)完成上述两个参数的全局寻优;最后,将所建成的PCA-PSO-LSSVM软测量模型应用于未知样本数据的预测,得其均方根误差2.17%,极大误差4.19%。结果表明,本文所构建的软测量模型预测精度高,泛化性能与稳定性好,可为造纸废水厌氧出水COD在线预测及该处理系统的优化控制提供指导。
Taking a papermaking wastewater anaerobic treatment system of an enterprise as an example, eight process variables are selected as input variables based on device operation and process mechanism analysis, combined with expert experience and knowledge, and the industrial operation data of these variables are collected by field sensors so as to construct a water quality PCA-PSO-LSSVM Soft-Sensing Model for Indicator Chemical Oxygen Demand (COD) Output Variables: First, principal component analysis (PCA) was used to perform preprocessing of data sample input variables to eliminate the correlation between variables and to complete the input variables Then, the least square support vector machine (LSSVM) data modeling between principal component and effluent COD was carried out. Considering that the kernel function width and penalty factor in LSSVM model had a great influence on the performance of the model, Particle swarm optimization algorithm (PSO) accomplishes the global optimization of the above two parameters. Finally, the PCA-PSO-LSSVM soft-sensing model is applied to the prediction of unknown sample data, with a root mean square error of 2.17% Error 4.19%. The results show that the soft measurement model constructed in this paper has high prediction accuracy and good generalization and stability, which can provide a guideline for predicting the COD of anaerobic effluent from papermaking wastewater and the optimal control of the treatment system.