With the continuous expansion of power grid scale and the continuous implementation of Energy Internet construction, the long-distance and large-capacity electric energy exchange between regional power grids is increasingly frequent, which makes the stability problem to a wide range of attention. Therefore, this paper proposes a transient voltage control method based on physics information and reinforcement learning, which is called Physics-Informed Reinforcement Learning. This method combines the physical model and the data-driven model of power system, and takes the constraints in the physical model as the constraints of the data-driven model to accelerate the convergence rate of the model, so as to realize the rapid scheduling of transient voltage instability. Finally, an example of IEEE-9 bus system is given to verify the effectiveness and superiority of the proposed method.
This paper presents a novel deep reinforcement learning (DRL)-based method for the adaptive control of transient voltage in power systems. First, we propose a neural network-based nonlinear controller (TVCON) designed to modulate each voltage-source converter (VSC), such as photovoltaic systems or energy storage systems, that actively contributes to transient voltage control. Subsequently, all distributed TVCONs can collaborate to rapidly restore system voltage during fault transients by centrally optimizing their parameters (weight coefficients). Specifically, the optimization is conducted periodically using incremental DRL to efficiently update the TVCONs’ parameters in accordance with the practical operating conditions of the system and VSCs. Consequently, the provision of transient reactive current by VSCs, which have operational constraints related to current saturation, can be feasibly and adaptively controlled by the TVCONs while considering their steady-state active current outputs. Additionally, the inappropriate sacrifice of VSCs’ active current and the resulting adverse impacts can be effectively mitigated. Finally, simulations conducted on a modified IEEE 14-Bus system validate the proposed method.
With the increasing integration of renewable energy sources and power electronic converters, Grid-Forming (GFM) technologies such as Virtual Synchronous Generators (VSGs) have emerged as key enablers of future power systems. However, conventional VSG control strategies with fixed parameters often fail to maintain transient stability under dynamic grid conditions. This paper proposes a novel adaptive GFM control framework based on physics-informed reinforcement learning, targeting transient power angle stability in systems with high renewable penetration. An adaptive controller, termed the 3N-D controller, is developed to periodically update the virtual inertia and damping coefficients of VSGs based on real-time system observations, enabling anticipatory adjustments to evolving operating conditions. The controller leverages a reinforcement learning architecture embedded with physical priors, which captures the high-order differential relationships between rotor angle dynamics and control variables. This approach enhances generalization, reduces data dependency, and mitigates the risk of local optima. Comprehensive simulations on the IEEE-39 bus system with varying VSG penetration levels validate the proposed method’s effectiveness in improving system stability and control flexibility. The results demonstrate that the physics-embedded GFM strategy can significantly enhance the transient stability and adaptability of future power grids.
The increasing integration of renewable energy sources and the deep fusion with power electronics have introduced new challenges to the stability of power systems, making traditional prediction methods inadequate for real‐time, high‐precision control. This paper presents a power system‐specific methodology designed to solve high‐dimensional, nonlinear dynamic prediction problems, which is demonstrated through an application to rotor‐angle trajectory prediction in power systems. Our proposed method, which combines model‐based approaches with data‐driven techniques, leverages the strengths of both paradigms. At its core, the method employs the Transformer deep learning architecture to capture the temporal dynamics of system transients. Additionally, physics‐informed neural networks are integrated to ensure that key physical principles are embedded within the model, enhancing both accuracy and convergence speed. While the proposed methodology has the potential for broader applications in dynamic modeling, this paper focuses on its application to rotor‐angle trajectory prediction, which is a critical task in transient stability assessment. This scenario was chosen as a representative use case to validate the effectiveness and feasibility of the approach under realistic power system conditions. The method was tested on the modified IEEE‐39 bus system, with results showing that it delivers fast, accurate, and physically consistent predictions without a large amount of labelled data. This demonstrates the method's potential for supporting online control and maintaining system stability in future power grids.
12 February 2026 in Electric Power Systems Research