The removal process of activated sludge in sewage treatment plants is very nonlinear, and removal performance has a complex causal relationship depending on environmental factors, influent load, and operating factors. In this study, how causal relationships are expressed in collected data was identified by structural equation modeling. First, path modeling was attempted as a preliminary step in structural equation model (SEM) construction and, as a result, the nutrient-removal mechanism could not be sufficiently represented as a direct causal relationship between measured variables. However, as a result of the deduced SEMs for effluent total nitrogen (T-N) and total phosphorus (T-P) concentrations, accompanied by exploratory factor analysis to extract latent variables, a causal network was formed that describes the direct or indirect effect of the latent factors and measured variables. Hereby, this study suggests that it is possible to construct an SEM explaining the nutrient-removal mechanism of the activated-sludge process with latent variables. Moreover, nonlinear features embedded in the mechanism could be represented by SEM, which is a model based on linearity, by including causal relations and variables that were not derived by path analysis. This attempt to model the direct and indirect causalities of the process could enhance the understanding of the process, and help decision making such as changing the driving conditions that would be required.The efficacy of the second approach, the attempt to understand via statistical analysis of the obtained data from the activated-sludge process, mainly depends on the quality and quantity of the obtained data. Unlike the mathematical model-based approach, this method makes it possible to use various measurement variables that cannot be utilized for mathematical models such as pH and oxidation reduction potential (ORP). Besides the sensor variables, operational factors such as the F/M ratio and airflow rate can be analyzed with influent/effluent water-quality variables within a dataset with the same utility. The most popular methodology applied to data analysis is principal component analysis (PCA), which has been adapted for sewage-treatment-process monitoring and identification of the operational state [8][9][10]. For the other approaches to enhance the understanding of the processing mechanism, there have been attempts to use signal processing and pattern recognition to detect operational abnormality or sensor faults. Wimberger et al. (2008) [11] applied a signal filter to detect the faults of the sewage-treatment process of sequencing batch reactor types. Baklouti et al. (2018) [12] tried to detect sensor faults using an improved particle filter. Chow et al. (2018) [13] developed a fault-detection and alarm algorithm based on the correlation analysis of signals of ultraviolet (UV) spectroscopy.A causal model can be considered to produce a more intuitive result for the purpose of supporting the judgment of the process operator and improving understanding the pr...

Phytoplankton blooms have a simple generation mechanism but myriad influencing factors. The compounding effects of water quality, weather conditions, and geological factors are site-specific to single river basin and can exhibit distinctive characteristics. To investigate the site-specific patterns of Chlorophyll-a outbreaks for each sub-part of the river, we statistically analyzed the results of water quality monitoring for the Nakdong River to classify the locations of the monitoring stations into upstream, midstream, and downstream sites. Then, based on the combination of the influencing factors, the rules explaining Chlorophyll-a concentration levels were extracted and interpreted using a decision tree algorithm, CHAID. The results revealed that for upstream, weather conditions were the primary factor influencing phytoplankton blooms because of the relative absence of pollutants compared with the midstream and downstream. For midstream, weather conditions and nutritional factors were influential in the generation of phytoplankton blooms. For downstream, the notable amount of pollutants originating from upstream not only reflected a high nutrient level for the phytoplankton but also caused the water quality factor to be the primary cause of phytoplankton blooms. The deduced tree, a tool for data-driven modeling, demonstrated its usefulness by extracting practical influencing factors and patterns of interest from the given data.

No abstract