SAMSE 2018 IOP Conf. Series: Materials Science and Engineering 490 (2019) 062027
IOP Publishing doi:10.1088/1757-899X/490/6/062027
and the organics in the mixture are chemically converted into the gases like methane and carbon dioxide, as well as microbial bacterial plastids under the action of the microorganisms. Based on the analysis of production behavior and process mechanism of the UMIC device, with the combination of experts’ experience and knowledge as well as the sensitivity analysis of field data, 8 process variables that affect the COD of the treatment system were selected as the input variables of the model, and they are: influent COD/mg•L -1 , influent SS/mg•L -1 , influent pH, influent flow/m 3 , influent temperature/°C, circulating pool level/%, effluent pH and effluent temperature/°C, while the output variable of the model is effluent COD/mg•L -1 . Two sample data collection methods were adopted, one of which was that the mill’s distributed control system DCS was used to collect 8 process variables, and the other was that the on-site sampling laboratory obtained effluent COD through offline test (Sun Jun et al., 2017). After the collection of the mill’s field operation data from July 2016 to February 2017 was completed, the missing data was directly removed, then the abnormal data was identified and deleted, and finally the initial sample matrix set containing 175 sample individuals was obtained.
1-Bar screener; 2-Blending pond; 3- Preliminary clarifier; 4-Pump; 5- Regulating container; 6-Anaerobic reactor; 7-Cycling wstandpipe; 8- Anoxic pool; 9- The secondary clarifier; 10-Biogas storage tank; 11-Nutrient auto-count pipette Fig 1. Flow chart of anaerobic degradation for paper mill wastewater 2.2. OCS-PCA-PSO-LSSVM Soft-sensor Method 2.2.1. PCA Technology The independent variable matrix n p u X of the obtained initial sample data was recorded, where n is the number of sample individuals, i.e. the sample size, and p is the number of process variables. PCA technology (Jolliffe et al., 2002) was namely to project X from the Euclidean space to the latent vector space of the pivot element. T T ¦ d k k k=1 X=TQ +E= tq +E (1) Where, k t is the k th extracted pivot element, k q is the load vector used to extract the pivot element, and E is the final residual matrix. In essence, the construction of the PCA latent vector space is to represent most of the dynamic information in the initial process variables in the sample data by extracting d pivot elements ( d p d ) (Sun Jun et al., 2017), of which, the information contribution of the k th pivot element can be calculated according to Formula (2).
p
(2)
1 k K O O ¦ / k k k
2.2.2. Soft-sensor Optimization Model PSO-LSSVM The least square support vector machine (LSSVM) is an extension of the standard SVM method (Cristianini et al., 2000), and the main ideas of the algorithm are summarized as follows:
3
Made with FlippingBook Digital Publishing Software