Rising spatiality of transient frequency dynamics in grid-connected renewable energy systems is necessitating complex cooperative inertia allocation and frequency control. To this end, a physics-informed deep reinforcement learning (PI-DRL) control strategy is proposed. First, a nodal rate of change of frequency constrained virtual inertia allocation is proposed upon improved frequency divider and synchronizing power coefficient. It prompts adaptability of the entire scheme in varying unit commitment. Then, upon a recent ASF model, a learning-augmented spatial average system frequency (LA-SASF) model is devised to reduce order of frequency dynamics. It deconstructs the center of inertia frequency into nodal frequencies, such that tractable physics of spatial frequency can be analyzed to train PI-DRL. Thereafter, an alternating training architecture is tailored to concomitantly evolve the PI-DRL and the LA-SASF model. The training scheme is finally developed on the environment with ongoing stochastic disturbances to help learn spatial patterns of frequency responses. Case studies on IEEE 39-bus system manifest that, our approach beats conventional control strategy regarding the effective-ness of nodal frequency security, with 36.7% lower nodal frequency deviation, and the necessity of cooperative inertia allocation and frequency control is verified through ablation tests.

Existing ac/dc power flow computations necessitate sequential convergence-oriented trial-and-error under various dc control modes, rising computational burden. This article thus proposes a physics-guided multiagent graph learning (PG-MAGL) method toward real-time power flow analysis with dc control mode adaptation. The tailored graph structure with built-in dc control modes and state variables is first advanced to ensure topology adaptability. Then, MAGL is proposed to enable adaptive jump over dc control modes. The trick is to organize multiagents to parameterize power flow solutions under various dc control modes and set aside trigger signals according to the operational violations of converters for the agent switching to the follow-up agent. To clarify the trigger signals, an augmented Lagrangian method-based PG-MAGL method is finally designed. It relaxes the control boundaries into the violation minimizers and enforces other constraints, such that dc control switching can be identified by the only violation signal. Utilizing inductive biases to rectify experiential biases in pure data-driven models, PG-MAGL enables precise inference of dc control mode feasibility. Case study shows that, relative to the other seven data-driven rivals, only the proposed method matches the performance of the model-based baseline, also beats it in efficiency beyond ten times.

In this article, a novel data-driven robust voltage regulation method employing the multiagent soft actor–critic algorithm for photovoltaic-rich distribution grids considering storage lifetime and topology flexibility is proposed. In the proposed scheme, the active and reactive power from distributed energy storage system (ESS) are coordinated to deliver effective voltage support. To account for the long-term influence of ESS behavior on its lifetime, the life costs associated with the energy throughput are firstly formulated into the reward function of the Markov game-based voltage regulation model. Then, the topology status is represented by continuous variables transformed via Gumbel-softmax and embedded into the local observation of ESS agents for being aware of topology variations due to operational reconfiguration. In addition, to enhance the robustness of the voltage regulation method against imperfect measurements, the designed state space incorporates solely partially observed information from the entire distribution networks. Numerical simulations on IEEE 69-bus and IEEE 141-bus test systems confirm the outperforming of the proposed method over the previously implemented voltage regulation approaches.

Electricity prices behave more irregular patterns due to uncertainties and effects of mixed-temporal primary energy markets. Thus, it is challenging to precisely forecast them. To conquer this barrier, an interpretable interval prediction method that seamlessly unifies cross-energy and electricity markets is proposed. At the outset, to clarity the feature rising the aberrant electricity prices, several exogenous, and multitemporal features from other primary energy markets, such as natural gas and coal markets, are unified to settle our database. Then, a Gaussian mixture model (GMM)-lightweight gradient boosting machine hybrid detector is presented to isolate and foresee the outlier sequence of electricity prices. A hybrid LSTNet-kernel density estimation (LSTNet-KDE) method is further proposed to enable outlier-adaptive interpretable interval prediction. Specifically, the LSTNet contributes to amalgamating multitemporality across markets and predicting the principal trends, and the KDE serves to encapsulate the uncertainty for the GMM-foreseen outliers. The method further merges with the Shapley additive explanations technique, such that exogenous latent features that induce electricity prices outliers can be finally comprehended. The numerical study on the real-world Danish electricity market verifies that, our proposed method beats other rivals in terms of precision, especially notable in forecasting outliers of electricity prices.

Larger-scale stochastic power systems urge the development of real-time alternating current optimal power flow, artificial intelligence (AI) thus becomes an alternative. However, traditional AI only imitates experiences, and cannot follow in-depth physics. This may cause an undesired nongeneralizability and topology intractability. To address this issue, a physics-guided graph neutral network (PG-GNN) is proposed. The PG-GNN firstly capture the physical constraints by a dual Lagrangian. Besides, the branch features of power grids are fully exploited to allow the PG-GNN to master tremendous topological patterns. To further manage the out-of-distribution topology, stability property of the PG-GNN is proved, then upon this evidence, an online transfer learning is proposed to allow the PG-GNN to fast master the unexpected topology. Numerical tests on benchmarks show that, the proposed method holds well topology-transferability, enables near or even better solutions than conventional optimizer, but merits much more than 100 times efficiency.