Milan Stojković, PhD
Team Lead – Senior Research Associate
Veljko Prodanović, PhD
Senior Research Associate
Water quality prediction systems based on artificial intelligence
Water quality is of key importance for the environment, biodiversity and human health. In order to improve the decision-making process and ecological status of rivers, an artificial intelligence (AI)-based framework for monitoring, predicting water quality and alarming is proposed.
This framework is based on the existing water quality measurement system (SEPA – The Environmental Protection Agency of the Republic of Serbia), which includes daily measurements of water quality at several measuring stations in Serbia. The proposal extends and improves the existing measurement system by providing real-time water quality data at a more precise temporal resolution. Smart floating buoys with multiple sensors improve water quality monitoring and support the state monitoring system (SEPA).
In addition, it is possible to assess water quality along the river course using equipped drones for water sampling. Drones collect information from different river sections, at specific GPS points between floating buoys, which observe the transformation process of water quality, thus improving the spatial resolution of water quality measurements. The added value of the project is reflected in the use of satellite images, the analysis of which provides additional information on the quality of water along the river course.
Development of machine learning methods for water quality classification
Assessment of the ecological status of river systems is always a current topic of interest for scientists in the field of ecology and environmental protection engineering. One of the basic indicators of the ecological status of rivers is the concentration of orthophosphate anions, which determines water quality and the rate of eutrophication. By assessing the presence of macrophytes, as official biological parameters for the assessment of the ecological status, an in situ assessment of the ecological status can be applied.
The suitability of using macrophytes in the assessment of ecological status is based on their eurivalence, that is, their ability to survive in parts of the river course with elevated concentrations of orthophosphate anions. So far, a small number of researches based on the assessment of ecological status using machine learning methods have been carried out, in which biological parameters are used as input variables. The results of modeling the ecological status of the Danube obtained by applying eight “state-of-the-art” machine learning models (Support vector machines, k-nearest neighbor; Decision trees; Random forest; Extra trees; Naïve Bayes; Linear discriminant analysis; Gaussian process classifier) showed that the best results of modeling the ecological status of the river system based on the presence of macrophytes as input variables and classes of ecological status as output variables are provided by the methods Support vector machines and Tree-based models.
The application of machine learning methods in the assessment of ecological status does not require financial costs and enables the preservation of the concept of sustainable development. The architecture of machine learning models created for the purposes of this research can be applied in future research based on the assessment of ecological status based on the presence of macrophytes and concentrations of selected chemical parameters.
Systems based on artificial intelligence for predicting flood events under climate change conditions
Climate change is leading to major weather-related natural disasters, particularly affecting countries at a low and medium level of development that lack sophisticated flood defense systems. In order to reduce the negative impacts of floods, it is necessary to develop a reliable flood forecasting system that also represents the basis for investment decisions in order to adapt to climate change.
More recently, scientific papers and Google’s flood prediction system have shown that AI-based solutions can provide timely flood warnings. Building on previous research, a flood prediction system is being developed under climate change conditions for smaller torrential streams where floods occur immediately after intense rainfall.
Assessment of solar potential in urban areas
Global warming is one of the main problems of today. The work to slow it down, or even better stop it, is intensifying all over the world. Many countries are moving towards replacing the use of traditional fossil fuels with alternative clean energy sources.
Solar energy is a sustainable alternative to fossil fuels and has a low impact on the environment. Recently, it has been used more and more and every country has the potential to produce it. In Serbia, rooftop solar power plants were a rarity until a few months ago, but with the adoption of appropriate regulations, citizens have been enabled to produce green energy for their own consumption. Today, solar expansion is taking place in Serbia, there are currently more than 370 registered producer-consumers with a total capacity of 5.7 MW, and another 100 MW is in the process of being connected to the grid.
In order to increase the use of solar energy in Serbia, a solar 3D urban model was created for part of the municipality in Belgrade, which was developed for the calculation and visualization of the solar energy potential of building roofs. LIDAR data was obtained from the city of Belgrade, on the basis of which a 3D model and calculations were made. Suitable roofs were selected in terms of surface area, slope and orientation, and it was obtained that this part of the municipality in Belgrade can produce almost 20,000 MWh per year.
Development of digital twins of water management systems
A significant increase in the available measured data and computer capacities attracts great interest in the formation of “digital twins” of water management systems. The term “digital twin” refers to a replica of a physical system that imitates the functioning of a real system. Such replicas use real-time measurement data to simulate the expected and critical behavior of the physical system, respectively, under regular conditions and under conditions outside the envelope of the designed system.
A digital replica of the altimeter water management system “Pirot” was developed, with the aim of assessing the dynamic resilience of the system during hazardous events. For the functioning of the digital replica, it is necessary to use cascade models that include a hydrological model, a water management model, as well as a stochastic model for the generation of hazardous events and risk assessment.
Based on the generated data, a model based on neural networks was developed for risk prediction, which can be used to manage accumulation in the event of hazardous events.
Assessment of water resources system resilience under hazardous events using system dynamic approach and artificial neural networks
The objective of this research is to propose a novel framework for assessing the consequences of hazardous events on a water resources system using dynamic resilience. Two types of hazardous events were considered: a severe flood event and an earthquake. Given that one or both hazards have occurred and considering the intensity of those events, the main characteristics of flood dynamic resilience were evaluated. The framework utilizes an artificial neural network (ANN) to estimate dynamic resilience. The ANN was trained using a large, generated dataset that included a wide range of situations, from relatively mild hazards to severe ones. A case study was performed on the Pirot water system (Serbia). Dynamic resilience was derived from the developed system dynamics model alongside the hazardous models implemented. The most extreme hazard combination results in the robustness of 0.04, indicating a combination of an earthquake with a significant magnitude and a flood hydrograph with a low frequency of occurrence. In the case of moderate hazards, the system robustness has a median value of 0.2 and a rapidity median value of 162 h. The ANN’s efficacy was quantified using the average relative error metric which equals 2.14% and 1.77% for robustness and rapidity, respectively.
Assessing water resources systems’ dynamic resilience under hazardous events via a genetic fuzzy rule-based system
In this paper, the use of a novel genetic fuzzy rule-based system (FRBS) is proposed for assessing the resilience of a water resources system to hazards. The proposed software framework generates a set of highly interpretable rules that transparently represent the causal relationships of hazardous events, their timings, and intensities that can lead to the system’s failure. This is achieved automatically through an evolutionary learning procedure that is applied to the data acquired from system dynamics (SD) and hazard simulations. The proposed framework for generating an explainable predictive model of water resources system resilience is applied to the Pirot water resources system in the Republic of Serbia. The results indicate that our approach extracted high-level knowledge from the large datasets derived from multi-model simulations. The rule-based knowledge structure facilitates its common-sense interpretation. The presented approach is suitable for identifying scenario components that lead to increased system vulnerability, which are very hard to detect from massive raw data. The fuzzy model also proves to be a satisfying fuzzy classifier, exhibiting precisions of 0.97 and 0.96 in the prediction of low resilience and high rapidity, respectively.
Post COVID-19 Thoughts: Controversies and Merits of the Technology Progress
Within a century, we are witnessing the technology’s development acceleration in such steps that just the changes in the last couple of decades made a difference in our lives. What was a “Science fiction” technology to the previous century generations, has become a common commodity for the 21st century generations. And for many born in the 20th century these changes are big. According to the universal laws that are in physics known in a form of “each action creates the reaction”, it is intuitively known that these changes are not to be accepted without the advent of some major concerns.
Machine Learning for Water Quality Assessment Based on Macrophyte Presence
The ecological state of the Danube River, as the world’s most international river basin, will always be the focus of scientists in the field of ecology and environmental engineering. The concentration of orthophosphate anions in the river is one of the main indicators of the ecological state, i.e., water quality and level of eutrophication. The sedentary nature and ability to survive in river sections, combined with the presence of high levels of orthophosphate anions, make macrophytes an appropriate biological parameter for in situ prediction of in-river monitoring processes. However, a preliminary literature review identified a lack of comprehensive analysis that can enable the prediction of the ecological state of rivers using biological parameters as the input to machine learning (ML) techniques. This work focuses on comparing eight state-of-the-art ML classification models developed for this task. The data were collected at 68 sampling sites on both river sides. The predictive models use macrophyte presence scores as input variables, and classes of the ecological state of the Danube River based on orthophosphate anions, converted into a binary scale, as outputs. The results of the predictive model comparisons show that support vector machines and tree-based models provided the best prediction capabilities. They are also a low-cost and sustainable solution to assess the ecological state of the rivers.
Application of machine learning in river water quality management: a review
Machine learning (ML), a branch of artificial intelligence (AI), has been increasingly used in environmental engineering due to the ability to analyze complex nonlinear problems (such as ones connected with water quality management) through a data-driven approach. This study provides an overview of different ML algorithms applied for monitoring and predicting river water quality. Different parameters could be monitored or predicted, such as dissolved oxygen (DO), biological and chemical oxygen demand (BOD and COD), turbidity levels, the concentration of different ions (such as Mg2+ and Ca2+), heavy metal or other pollutant’s concentration, pH, temperature, and many more. Although many algorithms have been investigated for the prediction of river water quality, there are several which are most commonly used in engineering practice. These models mostly include so-called supervised learning algorithms, such as artificial neural network (ANN), support vector machine (SVM), random forest (RF), decision tree (DT), and deep learning (DL). To further enhance prediction power, novel hybrid algorithms, could be used. However, the quality of prediction is not only dependent on the applied algorithm but also on the availability of previously mentioned water quality parameters, their selection, and the combination of input data used to train the ML model.
Failure Conditions Assessment of Complex Water Systems Using Fuzzy Logic
Climate change, energy transition, population growth and other natural and anthropogenic impacts, combined with outdated (unfashionable) infrastructure, can force Dam and Reservoir Systems (DRS) operation outside of the design envelope (adverse operating conditions). Since there is no easy way to redesign or upgrade the existing DRSs to mitigate against all the potential failure situations, Digital Twins (DT) of DRSs are required to assess system’s performance under various what-if scenarios. The current state of practice in failure modelling is that failures (system’s not performing at the expected level or not at all) are randomly created and implemented in simulation models. That approach helps in identifying the riskiest parts (subsystems) of the DRS (risk-based approach), but does not consider hazards leading to failures, their occurrence probabilities or subsystem failure exposure. To overcome these drawbacks, this paper presents a more realistic failure scenario generator based on a causal approach. Here, the novel failure simulation approach utilizes fuzzy logic reasoning to create DRS failures based on hazard severity and subsystems’ reliability. Combined with the system dynamics (SD) model this general failure simulation tool is designed to be used with any DRS. The potential of the proposed method is demonstrated using the Pirot DRS case study in Serbia over a 10-year simulation period. Results show that even occasional hazards (as for more than 97% of the simulation there were no hazards), combined with outdated infrastructure can reduce DRS performance by 50%, which can help in identifying possible “hidden” failure risks and support system maintenance prioritization.
Estimation of Large River Design Floods Using the Peaks-Over-Threshold (POT) Method
This research analyzes the peaks-over-threshold (POT) method for designed flood estimation needed to plan river levees, spillways and water facilities. In this study, a one-parameter exponential probability distribution has been modified by including the coefficient of λ, which represents an average number of floods and enables return period calculation within the specified period of time. The study also compares results using the Log-Pearson Type III distribution of maximum annual flows and a standard exponential distribution of the selected peaks over the threshold level. The aforementioned approach represents the standard mathematical tools for river flood design, while the proposed modification of the exponential distribution highlights the estimation of flood quantiles with longer return periods (e.g., 100, 1000 and 10,000 years). Moreover, the sensitivity analysis of the threshold selection is proposed to assist in the flood design flow estimation alongside the proposed modification of the exponential probability distribution. The study was carried out at the Danube River, and the Novi Sad hydrological station (Republic of Serbia) was used for the long-term recorded period from 1876 to 2015. The results suggest that the POT method derives more reliable estimates of design floods than the traditional statistical tools for flood estimation. The results suggest the theoretical values of the water level of the 10,000 years return period is equal to 867 cm, while the Log-Pearson Type III distribution of annual maximum flows underestimated this value for 14 cm.
Colored noise in river level oscillations as triggering factor for unstable dynamics in a landslide model with displacement delay
In the present paper we examine the effect of the noise in river level oscillation on the landslide dynamics. The analysis is conducted in several phases. In the first phase, we analyze the multi-annual level oscillation of the Kolubara and the Ibar river (Serbia). Based on the observed dataset, wesuggest adeterministic modelfor the river level oscillation with the additional contribution of the noise part, which we confirm to have the properties of colored noise. In the second phase of the research, we introduce the influence of the river-level oscillation, with the included effect of colored noise in the spring-block delay model of landslide dynamics. Results of the research indicate conditions under which the effect of river noise has both stabilizing and destabilizing effects on the landslide dynamics. The effect of noise intensity D and correlation time ε is systematically analyzed in interaction with delayed interaction, spring stiffness and friction parameters. It is determinedthatthelandslidedynamicsissensitivetothechangeofnoiseintensity and that the increase of noise intensity leads to onset of unstable landslide dynamics. On the other hand, results obtained indicate that the examined model of landslide dynamics is rather robust towards the change of correlation time ε. Interaction of this parameter and some of the friction parameters leads to stabilization of landslide dynamics, which confirms the importance of the influence of the noise color in river level oscillations on the landslide dynamics.
Digital Solution to Estimate Solar Power Potential of Rooftops in City of Belgrade
Global warming is one of the main issues of today. The work to slow it down or even better stop it is getting more and more intense all over the world. Many countries are moving towards replacing the use of traditional fossil fuels with clean energy alternatives. Solar energy is a sustainable alternative to fossil fuels and has a low impact on the environment. It has been increasingly used recently because the price of solar panels has decreased, and the cost of electricity has increased significantly. In Serbia, rooftop solar power plants were a rarity until 2022, but the adoption of an appropriate regulatory framework enabled citizens to produce green energy for self-consumption. In order to increase the use of solar energy in Serbia, a solar 3D urban model of the part of the municipality in Belgrade was developed for the calculation and visualization of the solar energy potential of buildings rooftops. We received LiDAR data from the city of Belgrade, on the basis of which we created a 3D model and calculations. We have used Digital Surface Model (DSM) and Building Footprints layer to create solar radiation layer which takes into account the position of the sun throughout the year and at different times of day. The model accounts for the obstacles, such as nearby trees and buildings that may block sunlight, as well as the slope and orientation of the rooftops. Suitable roofs in terms of surface area, slope and orientation were selected and it was obtained that this part of the municipality in Belgrade can produce almost 20000 MWh per year.
