Water is one of the most essential natural resources for the basic necessities of life. It is a scarce resource because water shortages already exist in many regions. The challenge presented by water scarcity will be exacerbated since climate change is aggravating the contradiction between world development and water demand . In order to address this problem, researchers have been developing technologies to enhance water supplies and optimize water resource management so that the efficiency of the usage of water can be improved [1, 2].
Wastewater reclamation and reuse are promising solutions to cope with water scarcity. They can expand water supplies, provide environmental benefits and increase economical sustainability. The successful wastewater reclamation and reuse depend on the development of wastewater treatment technologies. Usually, wastewater treatment includes primary, secondary, and tertiary treatment. Primary treatment clarifies wastewater by the separation of larger solids. Secondary treatment removes the dissolved and suspended organic matter in the effluent from primary treatment. Tertiary treatment further purifies wastewater so that it can be used for industrial, agricultural, and municipal supplies .
The paper-making industry generates a great amount of wastewater, which is a major pollution source around the world. The treatment of wastewater in pulp and paper mills has attracted attention from researchers for decades [4, 5]. As wastepaper recycling continues to increase because of economic and environmental initiatives, deinking becomes one of the most important steps that determine the performance of wastewater treatment . Deinking is to remove the printing ink from newsprint, printed paper, and so on, when these papers are defibrated for wastepaper recycling . Flotation deinking is the most widely used deinking technology, and the other methods used for deinking are bleaching deinking, enzymatic deinking, and washing deinking, etc. [8, 9]. The wastewater during paper deinking processes is of a very complex composition, including a large number of ink particles, fine fibers, fillers, paper additives, and other substances . The characteristics of the wastewater after deinking processes are different depending on the deinking technologies, which leads to the various wastewater treatment methods adopted. For example, the deinking wastewater after enzymatic deinking has low biological oxygen demand (BOD) and chemical oxygen demand (COD), and it is hard to be biodegraded . Like other wastewater, the deinking wastewater treatment mainly includes physical, chemical, and biological treatment processes [12, 13]. Physical treatment can remove suspended solids through filtration, sedimentation, flotation, and so on, which is used in primary and tertiary treatment . Chemical treatment methods include coagulation and flocculation, adsorption, oxidation, etc., and they also can be used in primary and tertiary treatment . The deinking wastewater is commonly pretreated using coagulation and flocculation to reduce the cost and improve the performance of wastewater treatment . Biological processes such as fungal treatment, aerobic, and anaerobic digestion, are used for secondary treatment [5, 16]. In order to improve the performance of deinking wastewater treatment, an integrated system with multiple treatment methods is usually implemented [15, 17]. Simstich et al. used a thermophilic aerobic membrane bioreactor, which was a combination of biological treatment and ultrafiltration, to treat deinking wastewater . Yu et al. combined coagulation and flocculation, activated sludge process, and a Fenton reaction system for primary, secondary, and tertiary treatment .
Coagulation and flocculation have long been utilized to process water and wastewater. In addition, they can be applied to various industries to remove miscellaneous contaminants because of their effectiveness and convenience. Coagulants and flocculants are chemical agents that can be used to complete the two-phase process, coagulation and flocculation, to remove contaminants . Coagulants added to wastewater destabilize the suspended particles and allow for the aggregation and sedimentation of suspended particles, while flocculants are bridging compounds and they facilitate the formation of floc. Floc is the networks of clustering micro-floc and macro-floc of aggregated fine particles, which can be efficiently removed by physical methods .
The primary condition of forming floc in wastewater is the contact and collision among the various particles. There are three main ways for the particles to come into contact with each other in wastewater: the Brownian movement of the particles, collisions caused by differences in the particles’ settling velocities, and the hydraulic effect of the flowing water especially the shear conditions . The contact and collision caused by Brownian motion are significant if the particles are small enough. As the networks of floc increase, the effect due to Brownian movement is negligible . The second way of collision and aggregation can be due to the different settling speeds among particles. It does play a role in floc formation and sedimentation. However, compared with the strong disturbance of water flow in the flocculation jar during mixing processes, the effect of relatively low settling speeds among particles is limited . Especially at the initial stage of flocculation, the particles are small, their settling speeds are low, and the difference among the settling speeds is even smaller. Thus, the contact and collision of particles mainly depend on the hydraulic effect of the flowing water in the flocculation jar . Rapid mixing is crucial because it facilitates the growth of floc by expediting the dispersion of coagulants and improving the collision efficiency of suspended containments. Though, high intensity and long duration of rapid mixing cause micro-floc breakage and reduce floc re-growth potential [23, 24]. Slow stirring takes longer, and usually, the floc size is constant indicating a dynamic balance between floc growth and break . The increase of slow stirring speed results in the decrease of steady-state floc size. The hydraulic conditions significantly affect coagulation and flocculation processes, which in turn determine the performance of deinking wastewater treatment . However, modeling coagulation and flocculation processes is complex, and there are contradictory recommendations of hydraulic conditions in the literature . Therefore, it is imperative to better understand the relationship between the hydraulic conditions and the efficiency of flocculation in order to address complex wastewater treatment issues. The goal of this paper is to evaluate the optimal hydraulic conditions that enable efficient flocculation and result in a substantial reduction of the turbidity of deinking wastewater.
Machine learning is a family of algorithms that enable computers to accomplish tasks through learning from usually limited datasets presented to them, where information about the task is not available thoroughly. Machine learning techniques have an extensive variety of applications in all aspects of modern life, including classification, recognition, optimization, prediction, and so on. As awareness of environmental issues rises, machine learning algorithms have seen a wide range of applications in wastewater treatment [25,19,27]. In this paper, machine learning methods were utilized to examine the hydraulic conditions for efficient flocculation during the treatment process of deinking wastewater. Support vector machines (SVM), Gaussian process regression (GPR), and genetic algorithms (GA) are frequently used machine learning algorithms. Researchers have been inspired to take advantage of these algorithms to explore new applications [28,22,30]
Support vector machines are supervised learning algorithms that can be utilized for classification and regression analysis. SVMs determine the decision boundary by constructing the maximum margin hyperplane which has the maximum distances with the nearest data points of all separated classes . SVMs can choose different kernel functions appropriately to perform not only linear classification but also nonlinear classification in a higher-dimensional space . In contrast to neural networks, SVMs do not need a large number of observation data for training, which is suitable for tasks with limited prior data. SVMs are relatively easy to implement, and their training speed is fast. SVMs are also one of the most robust prediction algorithms. Therefore, SVMs have been extensively applied to solve real-world problems.
Support vector regression (SVR) utilizes the same principles as SVMs to implement regression analysis. SVR can interpret the datasets presented to it and identify the underlying relationship between the input variables and the output response in order to predict a decision outcome [33, 34]. The essential goal of SVR is to find the most appropriate function over the training datasets, which is the hyperplane that has the maximum number of data points.
Gaussian process regression is another machine learning method to solve nonlinear regression problems. It is a nonparametric Bayesian approach using probability distribution to estimate and predict uncertainties . Since GPR does not focus on individual data samples, it can work well with smaller datasets . Moreover, GPR can use kernel functions to integrate prior knowledge to capture the inherent properties in the sample datasets. Appropriate selections of kernel functions can optimize the modeling of observed datasets .
Genetic algorithms belong to the family of evolutionary algorithms (EAs) which are inspired by Darwin’s theory of evolution. Individuals in a population undergo descent with modification, and the mechanism of natural selection leads to the survival of the fittest individuals. GAs are adaptive heuristic search algorithms that have been extensively utilized in global optimization . In order to find the optimal solution of a fitness function, individuals of a population generated randomly have to go through the evolution of generations by selection, crossover, migration, and mutation . The iterative processes terminate when the population has converged, i.e., the change of the population can be ignored or the maximum number of generations has been reached.
In this study, the performance of flocculation in the deinking wastewater treatment, which was induced by different hydraulic conditions in the flocculation jar, was investigated. The performance of flocculation was indicated by the measured turbidities after each treatment. The hydraulic conditions included the fast stirring speed and time, low stirring speed and time, and temperature. The boundary values for hydraulic conditions are infinite since any speed, time and temperature could be chosen to conduct the coagulation and flocculation processes. In order to simplify the task, each condition was divided into four levels which are frequently adopted. The effect on the performance of wastewater treatment was examined under different combinations of these levels.
Based on the orthogonal array test, 16 different combinations of hydraulic conditions were implemented, and the corresponding turbidity was measured after each run. Machine learning methods were utilized to analyze the experimental results and determine the optimal hydraulic conditions. The machine learning methods used in this study were provided by MATLAB toolboxes. Both SVR and GPR were adopted to model the relationship between the hydraulic conditions and the performance of deinking wastewater treatment. The performance of the two methods was compared to determine which model represented the experimental data better. SVR was chosen to develop the model relating the hydraulic conditions to the performance of wastewater treatment. After the model was set up, GA was applied to evaluate the optimal combination of the hydraulic conditions resulting in the best performance of wastewater treatment, i.e., the lowest turbidity. GA determined the optimal conditions based on the fitness function, the constraints, and the upper and lower bounds. Taking into account the real-life conditions of flocculation, a set of hydraulic conditions was adopted.