The regulation of voltage in transmission networks is becoming increasingly complex due to the dynamic behavior of modern power systems and the growing penetration of renewable generation. This study presents a comparative analysis of three artificial intelligence approaches—Deep Q-Learning (DQL), Genetic Algorithms (GAs), and Particle Swarm Optimization (PSO)—for training agents capable of performing autonomous voltage control. A unified neural architecture was implemented and tested on the IEEE 30-bus system, where the agent was tasked with adjusting reactive power set points and transformer tap positions to maintain voltages within secure operating limits under a range of load conditions and contingencies. The experiments were carried out using the GridCal simulation environment, and performance was assessed through multiple indicators, including convergence rate, action efficiency, and cumulative reward. Quantitative results demonstrate that PSO achieved 3% higher cumulative rewards compared to GA and 5% higher than DQL, while requiring 8% fewer actions to stabilize the system. GA showed intermediate performance with 6% faster initial convergence than DQL but 4% more variable results than PSO. DQL demonstrated consistent learning progression throughout training, though it required approximately 12% more episodes to achieve similar performance levels. The quasi-dynamic validation confirmed PSO’s advantages over conventional AVR-based strategies, achieving voltage stabilization approximately 15% faster. These findings underscore the potential of neuroevolutionary algorithms as competitive alternatives for advanced voltage regulation in smart grids and point to promising research avenues such as topology optimization, hybrid metaheuristics, and federated learning for scalable deployment in distributed power systems.

Corrective congestion management in a transmission network with embedded HVDC requires coordinating continuous set-points and discrete topological actions under the full nonlinear AC power flow constraints, a mixed-integer nonlinear problem for which no polynomial-time solution is known. This paper formulates the task as a Markov decision process and solves it with a hybrid Soft Actor-Critic agent that handles both action types under one maximum-entropy objective: a squashed-Gaussian head for redispatch and the VSC-HVDC set-point, and two categorical heads trained through the Gumbel-Softmax Straight-Through estimator for generator and line disconnection. Each action subspace carries its own entropy temperature rather than a shared coefficient, and the behaviour of the three is measured. The state is 906-dimensional and covers branch loadings and reactive flows, bus voltages and angles, and the HVDC set-point. On the IEEE 118-bus system extended with a VSC-HVDClink,theagentclears98.6%ofa500-scenariointact-gridbatch and 97.4% of a 500-scenario batch that includes up to three simultaneous branch outages, in a mean of 1.5 and 1.7 corrective steps respectively, with an independent AC verification power flow confirming the resulting states. Inference requires a single forward pass, which makes the policy usable inside an operational decision cycle.

Rolling-horizon energy management systems (EMSs) and model predictive control (MPC) for microgrids in smart cities face a fundamental trade-off: finer temporal discretization improves operational performance but rapidly increases the size of the optimization problem and execution time, jeopardizing real-time feasibility. Furthermore, in short-horizon operation, only the first control actions are implemented, while long-horizon decisions primarily guide feasibility and constraints. This paper proposes a computation-aware temporal mesh design layer that jointly selects a variable granularity of blocks and an optimization horizon, explicitly bounded by market-aligned settlement steps and per-cycle computation budgets. Candidate configurations are represented as pairs ⟨B, H⟩, where B is a constant-step block programme, and H is the optimization horizon, and they are uniquely tracked through an auditable mesh signature. The method first evaluates a predefined, market-consistent set of solutions ⟨B, H⟩ to establish reproducible cost and execution-time benchmarks, then applies a greedy value-of-information (Greedy-VoI) search that generates valid neighbouring meshes through local refinement, thickening, and resolution reallocation without violating the basic requirements that every solution must meet. All candidates are evaluated using the same microgrid use case and the same comparative KPIs, enabling the systematic identification of near-optimal mesh–horizon designs for practical EMS implementation.

Studies of corrective congestion management, including solver benchmarking, operator training, security assessment and data-driven control, require a collection of operating points in which one or more branches are overloaded. Such cases are scarce in historical records, and although random perturbation of a base case does produce overloads, these concentrate on a small number of corridors and their severity cannot be prescribed. This paper presents a generator that produces reproducible corpora of physically consistent overloaded states, that is, converged AC solutions satisfying declared limits on the slack, on generator active power and on individual loads, but not a voltage band or generator reactive limits. The method combines a predictor–corrector sweep over a load-stress parameter with Newton–Raphson correction, a library of stress directions derived from distribution factors, and progressive tripping of overloaded branches to reach post-outage topologies. Each scenario is stored as a compact genotype that enables reconstruction of the complete operating point, so the corpus can be distributed as parameter vectors rather than solved cases. Applied to the IEEE 118-bus system extended with a VSC-HVDC link, the generator produced 253,378 overloaded states with maximum branch loadings from 101% to 199.82%, covering 60 distinct overloaded branches across intact-grid, single-outage and double-outage topologies. In the full-state reconstruction test the voltage magnitude discrepancy did not exceed $1.9\times {10}^{-7}$ pu.

Any study of corrective congestion management—benchmarking of optimisation solvers, operator training, security assessment, or data-driven control—requires a base of operating points in which one or more branches are actually overloaded. Such cases are scarce in historical records and are rarely obtained by random perturbation of a base case, which mostly yields non-convergent or operationally meaningless configurations. This paper presents a generator that builds reproducible corpora of physically consistent overloaded states from a full AC continuation power flow, with Newton–Raphson correctors, adaptive step control, a PTDF-based stress-direction library and Sherman–Morrison–Woodbury updates for the contingency sweep. Each scenario is stored as a compact genotype from which the complete operating point is reconstructed exactly, so a corpus can be distributed as parameter vectors instead of solved cases. A linearised variant based on Power Transfer and Line Outage Distribution Factors is described as a higher-throughput baseline for studies in which only thermal limits bind. Applied to the IEEE 118-bus system extended with a VSC-HVDC link, the AC generator produced 253,378 overloaded states covering 101% to 199.82% branch loading and 60 distinct overloaded branches, split evenly between intact-grid and single-outage topologies, with exact reconstruction verified for every stored scenario.

The integration of renewable energy sources into electrical grids introduces significant challenges, particularly in ensuring stability and reliability in dynamic, nonlinear environments. These sources create fluctuations, uncertainties, and voltage regulation issues that traditional control systems struggle to manage, compromising the grid’s ability to deliver a stable power supply. Addressing these challenges requires advanced, adaptive control solutions capable of responding to the variable nature of renewable energy. This article explores advanced techniques, focusing on reinforcement learning and Neuroevolution, to develop innovative control strategies for electrical systems. Neuroevolution, which combines neural networks with evolutionary algorithms, optimizes control without relying on gradient-based methods, making it suitable for complex, unpredictable scenarios. These approaches enhance grid stability, improve response times, and enable realtime anomaly detection and corrective actions, offering a resilient and efficient solution to the limitations of traditional control methods.