Algal blooms have various effects on drinking water supply systems; thus, proper monitoring is essential. Traditional visual identification using a microscope is a time-consuming method and requires extensive labor. Recently, advanced machine learning algorithms have been increasingly applied for the development of object detection models. The You-Only-Look-Once (YOLO) model is a novel machine learning algorithm used for object detection; it has been continuously improved in newer versions, and a tiny version of each standard model presented. The tiny versions applied a less complicated architecture using a smaller number of convolutional layers to enable faster object detection than the standard version. This study compared the applicability of the YOLO models for algal image detection from a practical aspect in terms of classification accuracy and inference time. Therefore, automated algal cell detection models were developed using YOLO v3 and YOLO v4, in which a tiny version of each model was also applied. The cell images of 30 algal genera were used for training and testing the models. The model performances were compared using the mean average precision (mAP). The mAP values of the four models were 40.9, 88.8, 84.4, and 89.8 for YOLO v3, YOLO v3-tiny, YOLO v4, and YOLO v4-tiny, respectively, demonstrating that YOLO v4 is more precise than YOLO v3. The tiny version models presented noticeably higher model accuracy than the standard models, allowing up to ten times faster object detection time. These results demonstrate the practical advantage of tiny version models for the application of object detection with a limited number of object classes.

Environmental sensors are utilized to collect real-time data that can be viewed and interpreted using a visual format supported by a server. Machine learning (ML) methods, on the other hand, are excellent in statistically evaluating complicated nonlinear systems to assist in modeling and prediction. Moreover, it is important to implement precise online monitoring of complex nonlinear wastewater treatment plants to increase stability. Thus, in this study, a novel modeling approach based on ML methods is suggested that can predict the effluent concentration of total nitrogen (TN_eff) a few hours ahead. The method consists of different ML algorithms in the training stage, and the best selected models are concatenated in the prediction stage. Recursive feature elimination is utilized to reduce overfitting and the curse of dimensionality by finding and eliminating irrelevant features and identifying the optimal subset of features. Performance indicators suggested that the multi-attention-based recurrent neural network and partial least squares had the highest accurate prediction performance, representing a 41% improvement over other ML methods. Then, the proposed method was assessed to predict the effluent concentration with multistep prediction horizons. It predicted 1-h ahead TN_eff with a 98.1% accuracy rate, whereas 3-h ahead effluent TN was predicted with a 96.3% accuracy rate.

In order to optimize the process operation against fluctuating influent water quality, it is essential to apply process control and simulate the process for deriving the optimal operation method. To simulate the process, the ASM2d model and ADM model of the IWA have been widely used for the simulation of the nitrogen and phosphorus removal process and biogas production process, which consist of ordinary differential equations. In order to simulate a sewage treatment plant, it is essential to simulate the steady state of a process under a given set of disturbances and operating conditions. However, the disadvantage is that the calculation time is long when analyzing the ordinary differential equations. In order to shorten the computation time, we propose an improved Newton-Raphson method. As a result, the ASM2d and the ADM1 were able to simulate the processes 32.3 times and 8 times faster than ordinary differential equation analysis, respectively.