Optimization of water reuse and modelling by saline composition with nanoparticles based on machine learning architectures

Disposal of solid wastes by landfill is a more widely utilized disposal technique compared to others due to its economic advantages as well as simplicity (Rajput et al. 2022). Leachate develops in landfill areas as a result of water's impact on waste composition, precipitation, and physical, chemical, and biological processes. Wastewater with a high level of contaminants, including monovalent and multivalent ions, bacteria, and heavy metals, is referred to as leachate. Chemical composition of leachate is influenced by the type of trash in the landfill, waste content, landfill age, and environmental conditions (Zhong et al. 2022). Chemical oxygen demand (COD), total phosphorus (TP), and organic nitrogen concentrations in normal leachate vary between 140 and 152,000 mg/L, 0.1 and 30 mg/L, and 14 and 2,500 mg/L. In landfill regions, liners with low hydraulic conductivity (k ≤ 10 − 7 cm/s) are employed to stop leachate from getting into groundwater. Numerous variables, including void ratio, medium porosity, pore size, soil type, liquid type, and contaminant concentration, all affect hydraulic conductivity (Fu et al. 2021). The structure and characteristics of the clay are altered throughout the leachate's passage through the clayey soil, altering its hydraulic conductivity. The leachate's increased ion concentration, ionic valence, and organic matter concentration all help to boost hydraulic conductivity, while its increased microbial population and concentration of suspended solids help to lower it. Some of the factors that contribute to improvement in hydraulic conductivity as a result of leachate-liner interaction include thinned diffuse double layer (DDL) thickness of the clay, agglomeration, flocculation of clay particles, formation of cracks in soil structure, and a rise in void ratio. Growing global population increases water usage and daily pollution production (Alshehri et al. 2021). Approximately 2.4 billion people worldwide do not have access to reliable sanitation services, both in urban and rural regions. Within 20 years, 2 billion more people are expected to reside in towns and cities, mostly in developing nations, increasing the demand for sanitation. Sewage is released untreated and damaging lakes, rivers, and coastal areas in more than 90% of developing nations. Traditional sanitation practises, which rely on water-wasting toilets, are not an appropriate answer for industrialized and developing nations. The sanitation systems were created and planned under the assumption that human excreta was a type of waste that could be disposed of in the environment and be assimilated by it. EcoSan is a cycle of a sustainable closed system used in sanitation systems that can bridge the gap between agriculture and sanitation. The EcoSan strategy emphasizes resource management and offers a comprehensive approach to environmentally and financially sound sanitation. The major goal is to promote sustainable development by reducing nutrient and water cycles while using less resources and energy. The EcoSan system is a different method for avoiding the drawbacks of traditional wastewater systems (Melesse et al. 2020).

Smart technology and newly developed artificial intelligence (AI) and machine learning (ML) are filling a void in water applications that were previously left unfilled by conventional methods and ways of thinking. Smart technology and newly developed AI and ML are filling a void in water applications that were previously left unfilled by conventional methods as well as ways of thinking. According to some projections, the water industry will spend up to 10% of over $90 billion in investments that are anticipated to mature by 2030 on AI. Because of its universality, resilience, and relative ease of design, AI, ML, and smart methods are projected to methods as well as solve complex and challenging problems in water applications, resulting in cost savings and process optimization. Water as well as wastewater treatment, monitoring of natural methods, and precision/water-based agriculture are a few water-related applications that have seen substantial ML usage. These industry studies rely on a variety of ML methods, with the most popular ones being artificial neural network (ANN), recurrent neural network (RNN), radio frequency (RF), support vector machine (SVM), and adaptive neuro-fuzzy inference system (ANFIS), with sporadic use of AI techniques like fuzzy inference systems (FISs). ANN-RF and SVM-RF are two examples of hybrid algorithms that combine two ML systems in applications. Studies have shown success using both AI and ML to optimize modelling processes in water-based settings (Chang et al. 2022).

Contribution of this research is as follows:

• 1.

  To propose a novel technique in the optimization of water reuse with nanoparticle analysis based on ML architecture.

• 2.

  Optimization of water reuse is carried out based on nanoparticle solar cells for water treatment.

• 3.

  Saline composition has been analyzed using a gradient discriminant random field based on saline water treatment.

Intelligent scissors, distance regularized level set evolution, and graph cuts are some recently developed advanced segmentation algorithms that have enhanced performance, which reduces high-throughput advantage of computerized data analysis. The concentration of dissolved oxygen (DO) in wastewater can be significantly reduced by a high temperature, which causes aquatic species to perish (Aghilesh et al. 2021). A drop in DO will cause nitrogen to be transformed into nitrate molecules, which will also produce unpleasant odours. The wastewater's odour and health may be harmed if the hue gradually turns from grey to blackish brown (Wang et al. 2021). Wastewater must be treated before being released freely into a body of water in order to protect public health (Jiang et al. 2021). Water is necessary for human living, but wastewater needs our attention because it is the primary source of most diseases. Additionally, the main source of sickness is water (Wei et al. 2022). 80% of disease outbreaks in 2002 were attributed to water contamination by microbes, according to WHO data (Wang et al. 2020). Because losses are a result of the uncertainty introduced by risks, risk management is essential for lowering losses (Li et al. 2021). Effective management is required to reduce these risks. As part of this management, procedures must be carried out effectively in order to accomplish an objective. Pre-analysis, prediction, and control are all effective ways to lower possible risks (Nam et al. 2022). Therefore, risk management is required to lower the likelihood of failure and offer convenience for handling prospective risks. The methodical process of identifying environmental threats, reviewing probable outcomes, and controlling the amount of environmental risk is one specific illustration of environmental risk management (Astray et al. 2021). The four steps of risk management for the environment are issue formulation, risk analysis, risk characterization, and risk management (Zhao et al. 2020). Large datasets can be efficiently classified using ML techniques in a variety of domains, including biology and text mining (Alqahtani et al. 2022). Particle shape categorization is now accomplished using deep learning techniques like CNN algorithms and ANN (Bagheri et al. 2019). This method produces good results for particular particle forms, but in order to produce training datasets, it requires prior knowledge about the latter. An unsupervised ML technique, which does not need prior input or data training, is preferable to accomplish generalization in data analysis (Hoa et al. 2019). Unsupervised ML techniques may effectively evaluate the size as well as shape information of nanoparticles, according to recently published studies; however, they are only applicable to nanoparticles with distinct shapes, significant visual contrast, and good dispersions, such as gold nanorods (isolated particles). Due to a variety of nanoparticle compositions, dispersions, and imaging settings, it is actually challenging to ensure a sharp contrast as well as a homogeneous background in energy management images (Zeng et al. 2018; Chen et al. 2021).

Wastewater processing plants can benefit from adaptive ML models by minimizing downtime and increasing profits through more efficient management. The use of an ensemble technique increased prediction accuracy by 5% when compared to using only the individual base models (Zaghloul & Achari 2022). Controlling chlorination using AI approaches has been shown to be successful while predicting disinfection by-products (DBP) concentrations and key parameters for adsorption and membrane-filtration processes using ML models has shown similar promise. By establishing the connection between input and output among system variables, we may improve AI and ML application efficiency that suffers from differences in the quality of data used for testing and training (Lowe et al. 2022). Successful real-time surveillance, optimization, uncertainty prediction, and fault identification for intricate environmental systems are all possible due to ML algorithms, which proved useful in predicting the uncertain behaviours of treatment procedures. Hyperparameter tweaking is used during the machine learning algorithm (MLA) selection process to discover the optimal solution in the least amount of time while using the fewest available computing resources (Sundui et al. 2021).

From the related study findings, it was found that there is a gap in the application of nanoparticles in solar panels to improve water reuse and to determine the saline composition. Also, the application of ML techniques in water reuse is very limited. Incorporating these approaches in the present study makes it novel and distinct with the adoption of datasets from bright-field tunnelling electron microscope (BF-TEM) and annular dark-field scanning TEM (ADF-STEM).

Pre-processing begins with separating the image's background from the particles. Setting a global brightness threshold to remove the backdrop from the image is one approach that is frequently used to do this. This method takes advantage of contrast in brightness between particles as well as background in TEM image. The inhomogeneity of the background presents a problem non-depending on this method to create accurate background-removed photos. This inhomogeneity might be caused by differences in the sample backdrop matrix's thickness, surface unevenness, or by uneven electron microscope illumination. When a global filtering threshold is used, inhomogeneous background causes regional background to be both over- as well as under-filtered in the removal method. Because of this, the technique either fails to effectively isolate particles or introduces artefacts, which distort shapes. Such insufficient background removal can compromise the precision of particle shape data (Wen et al. 2021).

In constructing the PV, there are two layers of silicon present: a negative N layer and a positive P layer that is doped with boron. The tempered glass-coated PV module absorbs solar energy when it is exposed to the sun. After a certain amount of time, the energy absorbed exceeds the band gap energy level, which causes electrons to travel from conduction band to valence band via that band. Due to this, electrons in the conduction band can flow freely and form electron–hole pairs. Since the flow of electric current is caused by the movement of electrons, the electricity produced during the process is used to power the load. In order to generate adequate electricity, an array configuration is insufficient because these systems experience numerous losses (Al-Ezzi & Ansari 2022). The ideal method for maximizing each string's effectiveness is MPPT (maximum power point tracking). The most power possible from PV modules can be obtained by using such control approaches.

Gradient discriminant random field-based saline water composition analysis:

The equation shows that VAR{h_t+1 − h_t}, 0, indicating that gradient updates do not contribute uniformly. It's intriguing to note that ∇ψ_w(d_t) 2 and ∇ψ_w(d_t), the deciding variables in this equation, depend on d_t, indicating a relationship between h_t+1h_t and d_t.

The anticipated consensus error V(k) representing the extra disturbances brought on by the variations in solutions. Two extra terms are probably insignificant in the long run, though, assuming V(k) decays quickly enough relative to U(k), and we would infer that convergence rate of U(k) is similar to R(k) for SGD. for ⁠. Plugging this into Equation (21) leads to inequality ⁠. Hence, when ⁠, we have that

A challenging combinatorial optimization issue is referred to as the minimization of the equation in (19).

The machine used for the experiment has the following hardware components: an Intel Core i5 7200U processor, 8 GB of RAM, a 1 TB hard drive, and NVIDIA GTX 760MX graphics. Additionally, Python 3.5 environments were utilized to simulate how the suggested strategy might be put into practise. In order to establish results of offered technique, we carried out a statistical analysis by evaluating expected performance.

Nanoparticle description: Drop-casting nanoparticle dispersions onto Cu TEM grids with the holey carbon film coating was used to create the TEM samples of nanoparticles. ADF-STEM or BF-TEM modes of the JEOL F200 (Akishima, Tokyo, Japan) were used to photograph the nanoparticle TEM samples.

Table 1 gives analysis for TEM dataset. Here, the dataset analyzed are BF-TEM and ADF-STEM in terms of specificity, computational cost, kappa coefficient, training accuracy, and mean average precision.

The following parameters must be understood in order to perceive and comprehend the categorization model that has been put into use: True Positives (Tp) – If the actual class labels and expected class labels match, this calculates to true. True Negatives (Tn) – If actual class labels match projected class labels, this evaluates to false. False Positives (Fp) and False Negatives (Fn) specifications indicate whether actual class labels correspond to predicted class by the model or the opposite (Patel et al. 2022; Gerevini et al. 2023).

Figure 5(a)–5(e) gives analysis for ADF-STEM dataset. Proposed technique attained specificity of 79%, computational cost of 55%, kappa coefficient of 49%, training accuracy of 85%, and mean average precision of 66%; existing ANN attained specificity of 72%, computational cost of 51%, kappa coefficient of 45%, training accuracy of 79%, and mean average precision of 60%; CNN attained specificity of 76%, computational cost of 53%, kappa coefficient of 48%, training accuracy of 83%, and mean average precision of 63%. These metrics scores show that by categorizing salt particles, our deep learning classification model makes precise classifications and predictions. The implemented method was compared with comparable salinity-based prediction methods in order to demonstrate that our classification approach is reliable and produces cutting-edge results (Khan et al. 2022). We sought to compare it with best ML algorithms used to evaluate salinity because there has not been much research done on identification of seawater salinity concentration, making it impossible to compare it with other methods used in this field of research.

This research proposes a novel technique in the optimization of water reuse with nanoparticle analysis based on ML architecture. The saline composition has been analyzed using gradient discriminant random field based on saline water treatment. The produced solution functions as seawater and utilized to identify concentration of salt particles in seawater because concentration of salt particles in sea or seawater ranges to about 35 ppt. This could be helpful for desalinated water treatment plants. It was noted that our adopted method had shown to be effective and precise in terms of precision. Additionally, a real-world dataset was used to target salinity of seawater utilizing a combination of ML methods, allowing performance as well as resilience of proposed method to be evaluated. Proposed technique attained specificity of 79%, computational cost of 55%, kappa coefficient of 49%, training accuracy of 85%, and mean average precision of 66%.

The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University Abha 61421, Asir, Kingdom of Saudi Arabia for funding this work through the Large Groups Project under grant number RGP.2/142/44.

Deanship of Scientific Research at King Khalid University Abha 61421, Asir, Kingdom of Saudi Arabia through the Large Groups Project under grant number RGP.2/142/44.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.