Maria Vintan

Real-time environmental monitoring is increasingly supported by Internet of Things (IoT)-based sensing infrastructures and machine learning techniques capable of processing high-frequency urban environmental data. However, many existing solutions remain focused primarily on data acquisition, visualization, or isolated prediction tasks, with limited integration of predictive analytics and human expertise within a unified decision-support framework. This limitation highlights the need for intelligent monitoring approaches that combine continuous sensing, short-term prediction, and Human-in-the-Loop interpretation. Hybrid Human–AI Collaborative Networks (HCNs) are increasingly relevant for supporting environmental monitoring and decision-making in smart cities. This paper proposes an Artificial Intelligence of Things (AIoT)-enabled collaborative framework that integrates distributed uRADMonitor sensors, cloud-based data management, machine learning models, and human stakeholders into a unified monitoring ecosystem. A distributed uRADMonitor sensing network was deployed across multiple locations in Sibiu for continuous environmental monitoring, while the machine learning experiments presented in this study were conducted independently using three sensing-node datasets collected from different neighbourhoods in Sibiu. The same overall modelling and validation methodology was applied to each dataset, while the predictor set reflected the environmental variables available at each sensing node. Several machine learning algorithms were evaluated for Air Quality Index (AQI) prediction, including Linear Regression, Random Forest, and Gradient Boosting. Under a random train–test split, Random Forest was the best-performing contemporaneous model across all three sensing nodes, although predictive performance varied substantially between locations. In contrast, the short-term forecasting experiment provided the operationally relevant predictive component, with Ridge Regression achieving positive skill relative to persistence across forecasting horizons from 5 to 120 min.

With the continued expansion of the power systems, the accuracy of fault diagnosis on power transmission lines is crucial to ensure their safe and stable operation. A single-phase-to-ground fault can occur anywhere along an overhead transmission line. The amplitude of the fault current and its variation over time depend on the distance between the transformer substation and the fault location. This paper introduces a MATLAB/Simulink programmatic modelling framework for a faster simulation of fault scenarios in a given power system. It also presents the single-phase-to-ground fault current signal features for various fault locations. Both the amplitude of the current and the voltage decreases when the distance from the busbars of the station increases, with the current declining steeper than the voltage.

Energy communities represent an important step towards clean energy; however, their management is a complex task due to various factors such as fluctuating demand and energy prices, variable renewable generation, and external factors such as power outages. This paper investigates the effectiveness of a Reinforcement Learning agent, based on the Proximal Policy Optimisation (PPO) algorithm, for energy management across three different energy community configurations. The performance of the PPO agent is compared against a Rule-Based Controller (RBC) and a baseline scenario using solar generation but no active management. Simulations were run in the CityLearn framework to simulate real world data. Across the three evaluated community configurations, the PPO agent achieved its greatest improvement over a single run in the scenario where all participants were prosumers (Schema 3), with a reduction of 9.2% in annual costs and carbon emissions. The main contribution of this work is demonstrating the viability of Reinforcement Learning agents in energy optimization problems, providing an alternative to traditional RBCs for energy communities.

Energy communities represent an important step towards clean energy, however their management is a complex task due to various factors such as fluctuating demand and energy prices, variable renewable generation, and external factors such as power outages. This paper investigates the effectiveness of a Reinforcement Learning agent, based on the Proximal Policy Optimisation (PPO) algorithm, for energy management across three different energy community configurations. The performance of the PPO agent is compared against a Rule-Based Controller (RBC) and a baseline scenario using solar generation but with no active management. Simulations were run in the CityLearn framework to simulate real world data, and results show that the PPO agent was up to 9.2% more effective in reducing annual costs and carbon emissions than the RBC, its effectiveness increasing in scenarios which allowed control over resources such as photovoltaic generation and battery storage. The main contribution of this work is demonstrating the viability of Reinforcement Learning agents in energy optimization problems, providing an alternative to traditional RBC controllers for energy communities.

The increasing complexity of electrical power installations requires more and more sophisticated mathematical models for the analysis of their operating regimes in relation to transient regimes. This requirement can be solved by using numerical mathematical models implemented in professional programming environments. MATLAB/Simulink is such a programming environment that allows for the analysis of transient regimes caused by faults occurring in electrical power installations. In this paper, the transient regime caused by the phase-to-ground fault is analyzed using a numerical model of a 20 kV nominal voltage power network, implemented in the MATLAB/Simulink programming environment. The numerical model was validated by comparing the obtained results with those experimentally determined in a real 20 kV network. Using the numerical model, we can analyze how the zero-sequence voltage of the 20 kV bus bars and the fault current in the 20 kV network, which are neutrally treated with a Petersen coil, are influenced by the following parameters: the initial phase of the voltage at the fault location; the regime in which the medium voltage network operates (resonance, under-compensated, or over-compensated); the insulation state (value of the electrical resistance of the insulation); and the value of the resistance at the fault location. The differences between the experimentally obtained results and those obtained using the numerical model are as follows: for the fault current, 6.67% if R_t = 8 Ω and 4.94% if R_t = 268 Ω; for the zero-sequence voltage, 3.21% if R_t = 8 Ω and 6.19% if R_t = 268 Ω.

The electrical energy supply of industrial equipment is provided by electrical power stations with high- (HT), medium- (MV) and low-voltage (LV) busbars. Consumers are connected to either MV or LV busbars. In this paper, a real power station was considered, through which the gasoline extraction from the well gas installation is powered. Electric consumers (electric motors) supplied by a MV busbar have active power of 13.54 MW, and those fed by a LV busbar have active power of 6.4 MW. Since electrical consumers operate in explosive environments, the design and operating conditions are more severe than in the case of electrical installations operating in non-explosive environments. The case of a single phase-to-ground fault occurring on the HV transmission lines feeding the power station has been analysed. First, the mathematical models for the calculation of the phase voltages, the dissymmetry and asymmetry coefficients, the reduction coefficient of the plus sequence component, and the effective values of the phase voltages were established. The influence of the source impedance (the equivalent impedance of the HV transmission lines) and of the neutral point configuration of the HV/MV medium-voltage transformer on the calculated quantities was analysed. Then, the results obtained using the established mathematical models were compared with those obtained experimentally by provoking a single-phase-to-ground fault near the HV busbars of the real power station. This study has shown that the de-symmetrisation of the phase voltages of the MV and LV busbars is lower when using the Y/Δ connection for the HV/MV transformer. As a result, it is recommended the Y/Δ connection be used for this transformer, instead of the Y₀/Δ connection.

The paper analyses how the characteristics of the medium-voltage network with insulated neutral and the conditions under which the single line-to-ground fault occurs (insulation condition and value of the fault resistance) influence the zero-sequence components of the voltage of the transformer station medium-voltage bus bar, the fault current, the currents of the faulted line and the currents of the other non-faulted power lines connected to the transformer station bus bars. Assuming the waveform of the medium voltage bus voltages in the transformer station is known, the influence of the fault resistance on the total distortion coefficient of the fault current is analysed. This establishes the conditions under which a single line-to-ground fault can be detected by controlling the total distortion coefficient of the fault current and the currents of the fault-free lines. It also examines the conditions under which a single line-to-ground fault can be selectively detected by checking the effective value and direction of flow of the zero-sequence currents of the medium-voltage lines during the fault. The results obtained by the calculation were verified experimentally in the medium voltage network considered in the study. The acceptable differences, considering the degree of accuracy with which the parameters of a medium voltage network are known, between the results obtained by calculation and those obtained experimentally show that the simplifying assumptions accepted for the mathematical models used in the calculation are correct.

One important problem in the operation of medium voltage networks is the detection of a single-line-to-ground fault in its incipient state, when the fault resistance values are very high. In a medium voltage (MV) distribution network with a neutral grounding resistor (NGR), one of the methods employed to discriminate a single line-to-ground fault is the use of an overcurrent relay with an operating characteristic adjusted according to the effective value of the current flowing through the limiting resistor. In case of a single line-to-ground fault with a high fault resistance value, the correct tripping settings of the protective relay require the precise computation of this current. In comparison to the assumptions made by the models from the literature—the three-phase voltage system of the medium voltage busbars is symmetrical and there are no active power losses in the network insulation—the model proposed in this paper considers the pre-fault zero-sequence voltage of the medium voltage busbars and the active power losses in the network insulation, which is necessary in certain fault conditions where the use of the former leads to unacceptable errors.

Settings of protection are essential to ensure the sensitivity and selectivity needed to detect defects. Making the correct settings requires the calculation of the fault currents with as little error as possible. Fault currents are influenced by the parameters of the electrical networks, including the state of the insulation and the Petersen coil, which changes during their operation electrical networks. This paper analyzes how the insulation parameters of medium voltage power lines, the parameters of the Petersen coil used to treat the neutral of the medium voltage electrical network and the value of the resistance at the fault location influence the fault current in the case of a single-phase fault. The large number of single-phase faults that occur in medium voltage electrical networks justifies this analysis. The symmetrical components method was used to calculate the fault current. The results obtained by calculation were verified experimentally by causing a single-phase-to-ground fault in a real medium voltage network. The paper presents the situations in which the analytical calculation of the single-phase-to-ground fault current can lead to inadmissibly large errors, even over 50%, but also the situations in which the errors fall below 3%.

In this paper we will present a new approach regarding the documents representation in order to be used in classification and/or clustering algorithms. In our new representation we will start from the classical "bag-of-words" representation but we will augment each word with its correspondent part-of-speech. Thus we will introduce a new concept called hyper-vectors where each document is represented in a hyper-space where each dimension is a different part-of-speech component. For each dimension the document is represented using the Vector Space Model (VSM). In this work we will use only five different parts of speech: noun, verb, adverb, adjective and others. In the hyper-space each dimension has a different weight. To compute the similarity between two documents we have developed a new hyper-cosine formula. Some interesting classification experiments are presented as validation cases.