Although India's solar power generation capacity is substantial, accurate forecasting of this capacity is challenging because to the country's extreme weather variability. Projections of future solar power generation are now less precise than they once were because of these fluctuations caused by things like seasons and erratic weather predictions. In this paper we have focused on three objectives, first is to identify the prominent locations in India where the abundant solar power is available and can be harnessed to maintain the system's sustainability. The second is to hybridize the two well-known machine learning algorithms to develop an efficient and effective hybrid ML Algorithm to achieve more accurate forecasting results; furthermore, to improve the ML model's efficiency and reduce the iteration time, the optimization of the hyperparameters has been performed with the genetic algorithm technique. The other and last objective is to train the developed model with the historical data of the proposed region on the developed hybrid ML model to testify our results and forecast the solar power with higher accuracy. According to the proposed strategy, the Jodhpur location has been chosen with the available historical meteorological data. Based on the collected meteorological and solar energy data, three separate non-overlapping groups are formed: training data (the upper 70
Addressing a restoration challenge within a sustainable radial electrical distribution system involves resolving issues to re-energize loads downstream of sectors impacted by a permanent fault. This research article presents a Non-dominated sorting Binary Particle Swarm Optimization method for solving a sustainable electrical radial distribution system's multi-objective in-service restoration problems. The proposed hybridized methodology is faster in computation than other methods, such as the conventional genetic algorithm, non-dominated sorting genetic algorithm, reactive tabu search, and conventional particle swarm optimization. Single- and multiple-fault case studies have been taken to analyse full-service restoration. The simulation study and comparative analysis have been performed on standard IEEE 10-bus (System I), 13-bus (System II), 33-bus (System III), and a practical 173-bus (System IV). The system I, NSBPSO, with and without PFC for the single-fault case, shows a difference in execution time of 25.89 and 4.27 s, respectively. Similarly, for the multi-fault case, the difference in execution time is 10.3 and 5.36 s, respectively. The power losses found from load flow results after applying M1 are 104.50 kW, while M5 is 106.86 kW. The proposed method, M1, executed in less time compared to other methods to proof a reliable and promising solutions in electrical distribution system.
Due to the growing number of electric vehicles (EVs), public charging stations are faced with the need for intelligent scheduling frameworks. The current methods of slot allocation cause high wait times, high power losses and poor user experience. In this manuscript, a hybrid framework is proposed to schedule EV charging sessions in an optimal manner while maintaining a quality-of-service (QoS) guarantee. The uncertainty-aware predictions of the charging session duration for each EV arrival is generated by the Bayesian constitutive artificial neural network (BCANN) component, which allows the scheduler to reserve a time slot for high-uncertainty charging sessions. The Kirchhoff’s law algorithm (KLA) subsequently optimizes charging schedules, using an objective that combines total charging cost with the loss cost. The proposed BCANN-KLA framework is simulated and compared with three benchmark methods such as many-objective stochastic competition optimization algorithm, Chaotic Harris Hawks Optimization algorithm and the artificial neural network based scheduler. The results of statistical evaluation over 30 independent runs have shown that the scheduling efficiency of BCANN-KLA is 98.74
The deterministic day-ahead (DA) market does not accommodate the uncertainty from variable renewable energy sources (VRES), often leading to increased balancing costs. Alternatives based on stochastic programming (SP) and adaptive robust optimisation (ARO) have been proposed in the literature but have yet to be implemented in functioning DA markets. This paper compares the deterministic, SP, and ARO market-clearing approaches for the co-optimization of energy and reserves, highlighting their advantages and drawbacks. Three case studies inspired by the power systems of the Netherlands, France, and Germany are examined using in-sample and out-of-sample analyses. Sensitivity analyses explore the impacts of sample size and conservativeness parameters on model performance. The findings reveal that SP and ARO models enhance socio-economic welfare (SEW) and significantly reduce the need for balancing services after the DA compared to the deterministic approach. Notably, the deterministic model yields the lowest SEW on days with high forecasted VRES output, whereas uncertainty-based models demonstrate the greatest benefits under high VRES penetration. Paving the way towards real implementation, we show that the trade-offs between economic performance, integration of VRES, and balancing needs after the DA depend on both system characteristics and the uncertainty model applied.
Grid-connected photovoltaic (PV) systems require fast and robust maximum power point tracking (MPPT) to maintain high energy yield and grid power quality under rapidly varying irradiance, temperature fluctuations, and partial shading conditions. Conventional MPPT methods such as Perturb and Observe (P O), Incremental Conductance, and shallow neural networks often suffer from slow convergence, steady-state oscillations, and mis-tracking under dynamic operating conditions. To address these limitations, this paper proposes a hybrid deep-learning-based MPPT strategy combining Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Transformer attention (C-BiLSTM-TrA) for grid-connected PV systems. The CNN extracts spatial correlations among PV voltage, current, irradiance, and temperature, the BiLSTM captures bidirectional temporal dependencies, and the Transformer attention mechanism models long-range contextual relationships, enabling accurate and predictive MPPT control. The predicted optimal operating point is used to control a Landsman DC–DC converter, while a PLL-based three-phase inverter ensures grid synchronization and power quality. The proposed controller was implemented in a HIL environment to verify real-time feasibility, and the performance results presented in this paper are based on both simulation and HIL validation. Simulation and Hardware-in-the-Loop (HIL) validation outcomes validate that the proposed method achieves a tracking efficiency of 99.2
