SAMSE 2018 IOP Conf. Series: Materials Science and Engineering 490 (2019) 062027
IOP Publishing doi:10.1088/1757-899X/490/6/062027
accounting for 13.0% of the total industrial wastewater discharge. The chemical oxygen demand (COD) in the discharged wastewater is 335,000 tons, accounting for 13.1% of the total industrial COD emission. In recent years, with the increasing shortage of water resources, production water has become a problem that restricts the development of paper-making enterprises. At present, in order to solve the environmental pollution due to paper-making wastewater and realize resource utilization, biogas production through anaerobic digestion has become a main method. The anaerobic digestion process under the action of microorganisms is featured as multi-factor influence, dynamic variability, complex nonlinearity (Yang Hao et al., 2016), etc. and the mechanism model thereof is difficult to construct, so the real-time operation control and optimization and calibration that affect safe production and effluent water production conditions cannot be realized. The production effectiveness of the industrialization process of anaerobic digestion for paper-making wastewater is often measured by the effluent COD. However, the current COD testing of enterprises is mostly realized by timed manual sampling and laboratory analysis. The test results cannot be obtained till several hours later, so the real-time performance is poor (Xu Lisha et al., 2012). In case that a COD on-line analyzer is installed on site, failure often occurs, resulting in loss of data. And also, the maintenance is difficult and the instrument is expensive(Langergraber et al., 2004; Bourgeois et al., 2010). With the improvement of enterprise automation as well as the deep integration of informationization and industrialization, the methods like pivot element regression, partial least squares regression, neural network, support vector machine and fuzzy logic have been used for the data modeling and operational control of the performance indicators including COD concentration, volatile fatty acid (VFA), dissolved oxygen, suspended solids (SS) concentration and gas production in the process of paper-making wastewater treatment (Bourgeois et al., 2010; Haimi et al., 2013) Choi et al., 2001; Wan et al., 2011; Huang et al., 2015 Dürenmatt et al., 2012; Zhou Hongbiao et al., 2017; Liu Lin et al., 2017; Tang Wei et al., 2017). With respect of the method selection, Wan et al. (2011) designed an adaptive fuzzy inference system integrating fuzzy subtractive clustering and PCA technologies, of which the fuzzy subtractive clustering is used to identify the model structure, and PCA is used to reduce the complex collinearity between variables as well as the dimensionality. The model accuracy with this integrated method is higher than that with the BP neural network method in the performance test about the COD and SS concentration prediction of paper-making wastewater. Wang Yao et al. (2017) chose the LSSVM method to predict the COD and SS concentrations. The results show that the soft-sensor model created by optimizing the LSSVM method parameters via the PSO algorithm has a higher prediction accuracy. The LSSVM method based on minimum structural risk is widely used in soft-sensor modeling because of its features of low dependence on sample data, less parameters to be estimated, and strong generalization ability (Souza et al., 2016; Wang et al., 2015; Fortuna et al., 2007; Liu Bo et al., 2015; Zheng Rongjian et al., 2017). However, the prediction accuracy of the soft-sensor model based on the offline sample data architecture, will gradually decline in the face of dynamic changes in continuous production processes. In order to solve the above problem, this paper proposes an OCS-PCA-PSO-LSSVM soft-sensor method integrating data analysis technology and regression modeling, which can eliminate the complex collinearity between variables and achieve dimensionality reduction via PCA technology; then, implement the LSSVM method to establish the nonlinear relationship between input and output variables, and realize the optimization of LSSVM model parameters by means of PSO; and finally, initiate the online calibration strategy (OCS) in case the prediction deviation of the new sample individual exceeds the set error limit, iteratively updating the soft-sensor model in an adaptive manner. 2. Materials and Methods 2.1. Process and Data Collection With the wastewater anaerobic treatment system of a papermaking mill as the test object of application, the production process is shown in Fig. 1, in which the ascending multistage internal circulation anaerobic reactor UMIC is the main device. The UMIC reactor works based on the principle of granular sludge (Ruggeri et al., 2015; Zhang Yi et al., 2014), namely, the papermaking wastewater is thoroughly mixed with anaerobic microbial sludge after being pumped into the reactor by a lift pump,
2
Made with FlippingBook Digital Publishing Software