Recently, large language models (LLMs) have achieved significant progress in automated code generation. Despite their strong instruction-following capabilities, these models frequently struggled to align with user intent in the coding scenario. In particular, they were hampered by datasets that lacked diversity and failed to address specialized tasks or edge cases. Furthermore, challenges in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) led to failures in generating precise, human-intent-aligned code. To tackle these challenges and improve the code generation performance for automated programming systems, we propose Feedback-driven Adaptive Long/short-term memory reinforced Coding OptimizatioN (i.e., FALCON). FALCON leverages long-term memory to retain and apply learned knowledge, short-term memory to incorporate immediate feedback, and meta-reinforcement learning with feedback rewards to address global-local bi-level optimization and enhance adaptability across diverse code generation tasks. Extensive experiments show that FALCON achieves state-of-the-art performance, outperforming other reinforcement learning methods by over 4.5% on MBPP and 6.1% on Humaneval, with the code publicly available. https://anonymous.4open.science/r/FALCON-3B64/README.md.

Modern text-to-image generation systems have enabled the creation of remarkably realistic and high-quality visuals, yet they often falter when handling the inherent ambiguities in user prompts. In this work, we present Twin-Co, a framework that leverages synchronized, co-adaptive dialogue to progressively refine image generation. Instead of a static generation process, Twin-Co employs a dynamic, iterative workflow where an intelligent dialogue agent continuously interacts with the user. Initially, a base image is generated from the user's prompt. Then, through a series of synchronized dialogue exchanges, the system adapts and optimizes the image according to evolving user feedback. The co-adaptive process allows the system to progressively narrow down ambiguities and better align with user intent. Experiments demonstrate that Twin-Co not only enhances user experience by reducing trial-and-error iterations but also improves the quality of the generated images, streamlining creative process across various applications.

Understanding the composition of measured demand at the substation level is essential for understanding and modeling the highly complicated behaviors of numerous loads, which is very challenging due to their dynamic nature. This article presents a flexible, real-time decomposition framework that separates individual load sources from aggregate substation measurements while providing interpretability through load-type classification. Our approach employs an enhanced dual-path neural network to adaptively decompose load signals, supported by an encoder–decoder-based source-counting mechanism that estimates the number of load sources dynamically. To accommodate varying load counts, we include a multilayer convolutional module for flexible separation mask application. Extensive case studies with aggregate measurements containing different numbers of load sources demonstrate our model’s adaptability, achieving a mean-squared error reduction of up to 15% and a signal-to-distortion ratio improvement of 10% over baseline methods, with computational times suitable for real-time implementation. Comparative studies further show that our solution achieves consistent load-type classification accuracy of over 95%, enabling reliable insights for operational decisions in grid management.

This article proposes a transient stability-constrained unit commitment (TSC-UC) model using input convex neural networks (ICNNs). An ICNN is trained to learn the transient function that maps prefault operation conditions (e.g., generator power output) to the transient stability index (TSI), which can be further utilized to identify the transient status (e.g., stable or unstable). Transient stability evaluation is conducted using the learned ICNN, without discretizing differential-algebraic equations (DAEs) or interacting with the time-domain simulation tools. Based on the convexity of ICNNs, the trained ICNN is exactly encoded as a linear programming (LP) model and integrated into conventional UC models to form a TSC-UC model. To impose transient stability constraints and expedite the solution process, the proposed TSC-UC model is decomposed into a UC master problem and two subproblems (i.e., network feasibility check subproblems and transient stability check subproblems). The decomposed problem is then iteratively solved using the Benders decomposition. Simulation tests are conducted in the New England 39-bus test system and IEEE 118-bus test system to verify the validity of the proposed approach.

Grid emergency voltage control (GEVC) is paramount in electric power systems to improve voltage stability and prevent cascading outages and blackouts in case of contingencies. While most deep reinforcement learning (DRL)-based paradigms perform single agents in a static environment, real-world agents for GEVC are expected to cooperate in a dynamically shifting grid. Moreover, due to high uncertainties from combinatory natures of various contingencies and load consumption, along with the complexity of dynamic grid operation, the data efficiency and control performance of the existing DRL-based methods are challenged. To address these limitations, we propose a multi-agent graph-attention (GATT)-based DRL algorithm for GEVC in multi-area power systems. We develop graph convolutional network (GCN)-based agents for feature representation of the graph-structured voltages to improve the decision accuracy in a data-efficient manner. Furthermore, a cutting-edge attention mechanism concentrates on effective information sharing among multiple agents, synergizing different-sized subnetworks in the grid for cooperative learning. We address several key challenges in the existing DRL-based GEVC approaches, including low scalability and poor stability against high uncertainties. Test results in the IEEE benchmark system verify the advantages of the proposed method over several recent multi-agent DRL-based algorithms.

Power system stability and control have become more challenging due to the increasing uncertainty associated with renewable generation. The performance of conventional control is highly driven by the physics-based offline-developed dynamic models that can deviate from the actual system characteristics under different operating conditions and/or configurations. Data-driven approaches based on online measurements can be a better solution to addressing these issues by capturing real-time operation conditions. This article describes a novel fully data-driven probabilistic framework to derive a linear representation of postcontingency grid dynamics and online prescribe control based on the derived model to enhance transient stability. The complex nonlinear power system dynamics is approximated by a linear model by using multiple neural network modules that infer distributions of the observations and introducing a Koopman layer to sample possible Koopman linear models from the inferred distributions. The trained model features linearity that can be easily incorporated into the existing linear control design paradigm and ease the controller design process. The effectiveness of Koopman-based control designs is validated through comparative case studies, which demonstrate increased prediction accuracy and control performance when applied to a power system with heterogeneous generator dynamics.

Distribution network reconfiguration is an effective method to face the problem of power fluctuation in the power system. Previous studies have focused on mathematical optimization techniques with complex modeling processes and heuristic algorithms with time-consuming solving processes to obtain the optimal reconfiguration strategy. In this article, a hybrid data-driven and model-based distribution network reconfiguration (HDNR) framework is proposed, where the model-based module includes model reduction and goal-oriented clustering to cluster the identical reconfiguration strategies. Here, the data-driven module is implemented through a long short-term memory network to learn the mapping mechanism between load distribution and optimal reconfiguration strategies. The model-driven module and the data-driven module are coupled through the proposed hierarchical network recovery process, which presents the reconfiguration results layer by layer. Finally, the numerical case study on the IEEE 33-bus, IEEE 119-bus, and IEEE 123-bus network shows the validity of the proposed HDNR framework. It is shown that the solution space is reduced, which contributes to reducing computation time and resources. Moreover, the obtained accuracy of the reconfiguration strategy is higher than most existing research even with limited data samples.

Distributed energy resources bring about challenges related to the participation of an increasing number of prosumers with strong social attributes in peer-to-peer (P2P) energy sharing markets, resulting in the increased complexity of socio-technical systems. Previous research has focused on energy sharing analysis based on rational games without considering the social attributes of prosumers, which are not typically used in real scenarios. In this article, an interdisciplinary P2P energy sharing framework that considers both technical and sociological aspects is proposed. It is based on prospect theory (PT) and stochastic game theory, in which the prosumers work as followers with subjective load strategies, while an energy sharing provider (ESP) serves as the leader with a dynamic pricing scheme. A subjective utility model with risk utility (RU) determined by PT is designed for prosumers, and a profit model for dynamic prices is suggested for ESP. Moreover, a solution algorithm that consists of interpolation and curve fitting to obtain the RU function, the aggregation of prosumers to a Markov decision process, and a differential evolution algorithm to solve the game are proposed to solve the problems of the “curse of dimensionality” and discreteness arising from the social attributes of prosumers. Numerical analysis reveals the results of the Stackelberg equilibrium and demonstrates the effectiveness of this method in terms of the social behavior of prosumers, i.e., radicalness when losing and conservatism when gaining.

The uncertainty of power resources introduces significant challenges for classic load modeling approaches. Moreover, load parameter identification techniques are affected by various load components with highly nonlinear time-varying behaviors and dependencies. This article presents a new deep generative architecture (DGA) based on the long short-term memory (LSTM) network for probabilistic time-varying parameter identification (PTVPI). In contrast to previous methods that merely compute point estimations of load parameters, our objective is to learn the continuous probability density function (PDF) of load parameters for composite load modeling (CLM) with ZIP load and induction motor. The proposed DGA learns complex temporal patterns from the time-varying parameters/measurements to estimate load parameters in a probabilistic fashion. Leveraging the LSTM network, our DGA computes deep temporal states and state transitions of load parameters. An encoding neural network extracts useful latent variables from the captured temporal states that are further mapped by a decoding neural network into the observed load parameters; hence, learning the underlying PDF of these parameters. Numerical results on the 68-bus New England and New York Interconnect System with four CLMs show accurate results for PTVPI in terms of various probabilistic estimation metrics, including reliability, sharpness, and continuous ranked probability score.

Battery swapping stations (BSSs) are ideal candidates for fast frequency regulation services (FFRS) due to their large battery stock capacity. In addition, BSSs can precharge batteries for customers and the batteries that are not in charging can provide a stable regulation capacity to the market. However, uncertainties, such as ACE signals and the EV per-hour visit counts, introduce stochastic nonlinear dynamics into the operation of a BSS-based FFRS. Currently, there is no quantification method to ensure its optimal economical operation. To close this gap, in this article, we propose a novel deep Q-learning-based FFRS capacity dynamic scheduling strategy. This method can autonomously schedule the hourly regulation capacity in real time to maximize the BSS's revenue for providing FFRS. Case studies using real-world data verify the efficacy of the proposed work.

In recent years, with the rapid growth of rooftop photovoltaic (PV) generation in distribution networks, power system operators call for accurate forecasts of the behind-the-meter (BTM) load and PV generation. However, the existing forecasting methodologies are incapable of quantifying such BTM measurements as the smart meters can merely measure the net load time series. Motivated by this challenge, this article presents the spatiotemporal BTM load and PV forecasting (ST-BTMLPVF) problem. The objective is to disaggregate the historical net loads of neighboring residential units into their BTM load and PV generation and forecast the future values of these unobservable time series. To solve ST-BTMLPVF, we model the units as a spatiotemporal graph (ST-graph) where the nodes represent the net load measurements of units and edges reflect the mutual correlation between the units. An ST-graph autoencoder (ST-GAE) is devised to capture the spatiotemporal manifold of the ST-graph, and a novel spatiotemporal graph dictionary learning (STGDL) optimization is proposed to utilize the latent features of the ST-GAE to find the most significant spatiotemporal features of the net load. STGDL utilizes the captured features to estimate the historical BTM load and PV measurements, which are further used by a deep recurrent structure to forecast the future values of BTM load and PV generation at each unit. Numerical experiments on a real-world load and PV data set show the state-of-the-art performance of the proposed model, both for the BTM disaggregation and forecasting tasks.

Accurate estimation of customer baseline load (CBL) is a key factor in the successful implementation of demand response (DR). CBL technologies implemented at utilities currently are primarily designed for large industrial and commercial customers. The U.S. Federal Energy Regulatory Commission (FERC) order 745 states that DR owners, including residential customers, can sell their load reduction in the wholesale market. However, since residential load is random and un-schedulable, this tends to inherently degrade the effectiveness of existing CBL technologies. In this paper, a novel SAE based CBL method for residential customers that uses the data reconstruction capability of a stacked autoencoder (SAE) is described. In the model, two SAEs are synchronously trained-one SAE generates a pseudo-load pool and the second one is used to select a pseudo-load to reconstruct a residential CBL. A support vector machine (SVM) classifier is self-trained to conduct the pseudo-load selection. The proposed strategy is validated using a real data set consisting of 328 residential customers' smart meter readings. Benchmarks from other machine learning techniques and existing CBL methods are compared with the proposed method. Test results show that the accuracy of the residential CBL reconstruction significantly improves when compared with existing methods, such as HighXofY and exponential moving average.

Grid modernization calls for increasing requirements of electric grid operation with enhanced sustainability and resilience [1]. In particular, distribution grids serve as a critical venue to bridge bulk upstream transmission and generation systems and a large number of downstream end users on the customer side, playing a significant role in modern electric grids for multiple purposes (e.g., renewable energy integration, power flow distribution, and end-user power quality enhancement) [2]. Under a normal grid operation condition, the increasing penetration level of renewable energy sources imposes new challenges on conventional distribution grid infrastructure (e.g., protection malfunction [3] and voltage violation [4]); on the other hand, in an extreme grid operation scenario, it is urgently needed to restore grid services after severe power outages, such as those caused by natural disasters [5]. In particular, for critical infrastructures, an efficient grid service restoration strategy should be implemented to avoid further damage over an extended period of time.

Peer-to-peer energy sharing in the distribution networks (DN) is an emerging issue with the large-scale development of photovoltaic (PV) prosumers. The DN can be classified into energy-shared regions (ESR) to enable the zonal energy trading. A Stackelberg-game-based energy-sharing framework is recommended for DN with multi-ESR, where the energy-sharing provider (ESP) works as a leader with dynamic pricing for multi-ESR, whereas PV prosumers serve as followers with the demand response's (DR) ability to choose an ESR to link and modify their flexible loads. A profit maximization model, along with multi-ESR pricing and a network usage fee, is designed for the ESP operation in this article. This involves a utility model with DR strategies, including ESR selection and load adjustment, which is proposed for the prosumers. Moreover, the presence and uniqueness of the Stackelberg equilibrium are being provided. Finally, through the use of a real system, the simulation results show that the ESP profit and prosumers can be increased whereas the impact of PV uncertainty and variability on the utility grid is reduced.

This paper presents a novel nonlinear dictionary learning (DL) model to address the energy disaggregation (ED) problem, i.e., decomposing the electricity signal of a home to its operating devices. First, ED is modeled as a new temporal DL problem where a set of dictionary atoms is learned to capture the most representative temporal features of electricity signals. The sparse codes corresponding to these atoms show the contribution of each device in the total electricity consumption. To learn powerful atoms, a novel deep temporal DL (DTDL) model is proposed that computes complex nonlinear dictionaries in the latent space of a long short-term memory autoencoder (LSTM-AE). While the LSTM-AE captures the deep temporal manifold of electricity signals, the DTDL model finds the most representative atoms inside this manifold. To simultaneously optimize the dictionary and the deep temporal manifold, a new optimization algorithm is proposed that alternates between finding the optimal LSTM-AE and the optimal dictionary. To the best of authors' knowledge, DTDL is the only DL model that understands the deep temporal structures of the data. Experiments on the Reference ED Data Set show an outstanding performance compared with the recent state-of-the-art algorithms in terms of precision, recall, accuracy, and F-score.

The power grid is rapidly transforming, and while recent grid innovations increased the utilization of advanced control methods, the next-generation grid demands technologies that enable the integration of distributed energy resources (DERs)- and consumers that both seamlessly buy and sell electricity. This paper develops an optimization model and blockchainbased architecture to manage the operation of crowdsourced energy systems (CESs), with peer-to-peer (P2P) energy trading transactions (ETTs). An operational model of CESs in distribution networks is presented considering various types of ETT and crowdsourcees. Then, a two-phase operation algorithm is presented: Phase I focuses on the day-ahead scheduling of generation and controllable DERs, whereas Phase II is developed for hour-ahead or real-time operation of distribution networks. The developed approach supports seamless P2P energy trading between individual prosumers and/or the utility. The presented operational model can also be used to operate islanded microgrids. The CES framework and the operation algorithm are then prototyped through an efficient blockchain implementation, namely, the IBM Hyperledger Fabric. This implementation allows the system operator to manage the network users to seamlessly trade energy. Case studies and prototype illustration are provided.

Dynamic microgrids (MG)have shown great potential in distribution system restoration as they could be frequently and actively reconfigured as per request from system operator. MGs usually have smaller inertia compared to conventional power systems thus require more delicate controller for operation safety. In addition, most distribution systems have unbalanced configurations which requires dedicated regulations on system operation states in both positive and negative sequence. This paper proposes a distributed secondary control strategy for unbalanced dynamic MGs operation. System frequency and voltage regulation are constantly enabled, along with voltage unbalance (VU)mitigation on critical load buses. The proposed control strategy provides automatic management of DGs within the same MG and seamless system transition as reconfiguration is requested. Detailed controller designs for both positive and negative sequence regulations are presented. Small-signal stability analysis of the proposed controller is derived and its dynamic performances are validated using MATLAB/Simulink.

We present a perfect formulation for a single generator in the unit commitment problem, inspired by the dynamic programming approach taken by Frangioni and Gentile. This generator can have characteristics such as ramp-up/ramp-down constraints, time-dependent start-up costs, and start-up/shut-down limits. To develop this perfect formulation, we extend the result of Balas on unions of polyhedra to present a framework allowing for flexible combinations of polyhedra using indicator variables. We use this perfect formulation to create a cut-generating linear program, similar in spirit to lift-and-project cuts, and demonstrate computational efficacy of these cuts in a utility-scale unit commitment problem. The online supplement is available at https://doi.org/10.1287/ijoc.2017.0802 .

Cybersecurity and resiliency of wide-area monitoring, protection, and control (WAMPAC) applications is critically important to ensure secure, reliable, and economical operation of the bulk power system. WAMPAC relies heavily on the security of measurements and control commands transmitted over wide-area communication networks for real-time operational, protection, and control functions. The current “N-1” security criterion for grid operation is inadequate to address malicious cyber events; therefore, it is important to fundamentally redesign WAMPAC and to enhance energy management system applications to make them attack resilient. In this paper, we present three key contributions to enhance the cybersecurity and resiliency of WAMPAC. First, we describe an end-to-end attack-resilient cyber-physical security framework for WAMPAC applications encompassing the entire security life cycle including risk assessment, attack prevention, attack detection, attack mitigation, and attack resilience. Second, we describe a defense-in-depth architecture that incorporates attack resilience at both the infrastructure layer and the application layer by leveraging domain-specific security approaches at the WAMPAC application layer in addition to traditional cybersecurity measures at the information technology infrastructure layer. Third, we discuss several attack-resilient algorithms for WAMPAC that leverage measurement design and cyber-physical system model-based anomaly detection and mitigation along with illustrative case studies. We believe that the research issues and solutions identified in this paper will open up several avenues for research in this area. In particular, the proposed framework, architectural concepts, and attack-resilient algorithms would serve as essential building blocks to transform the “fault-resilient” grid of today into an “attack-resilient” grid of the future.

Recent severe power outages caused by extreme weather hazards have highlighted the importance and urgency of improving the resilience of the electric power grid. As the distribution grids still remain vulnerable to natural disasters, the power industry has focused on methods of restoring distribution systems after disasters in an effective and quick manner. The current distribution system restoration practice for utilities is mainly based on predetermined priorities and tends to be inefficient and suboptimal, and the lack of situational awareness after the hazard significantly delays the restoration process. As a result, customers may experience an extended blackout, which causes large economic loss. On the other hand, the emerging advanced devices and technologies enabled through grid modernization efforts have the potential to improve the distribution system restoration strategy. However, utilizing these resources to aid the utilities in better distribution system restoration decision making in response to extreme weather events is a challenging task. Therefore, this paper proposes an integrated solution: a distribution system restoration decision support tool designed by leveraging resources developed for grid modernization. First, we review the current distribution restoration practice and discuss why it is inadequate in response to extreme weather events. Then, we describe how the grid modernization efforts could benefit distribution system restoration, and we propose an integrated solution in the form of a decision support tool to achieve the goal. The advantages of the solution include improving situational awareness of the system damage status and facilitating survivability for customers. The paper provides a comprehensive review of how the existing methodologies in the literature could be leveraged to achieve the key advantages. The benefits of the developed system restoration decision support tool include the optimal and efficient allocation of repair crews and resources, the expediting of the restoration process, and the reduction of outage durations for customers, in response to severe blackouts due to extreme weather hazards.

In this article, we perform a comprehensive survey of the technical aspects related to the implementation of demand response and smart buildings. Specifically, we discuss various smart loads such as heating, ventilating, and air-conditioning (HVAC) systems and plug-in electric vehicles (PEVs); the power architecture with multibus characteristics; different control algorithms such as the hybrid centralized and decentralized control and the distributed coordination among buildings; the communication technologies and network architectures; and the potential cyber-physical security issues and possible mechanisms for enhancing the system security at both cyber and physical layers. The current status of the demand response in United States, Europe, Japan, and China is reviewed, and the benefits, costs, and challenges of implementing and operating demand response and smart buildings are also discussed.

Plug-in hybrid electric vehicles (PHEVs) and plug-in electric vehicles (PEVs) have received increasing attention because of their low pollution emissions, petroleum independence, and high fuel economy. The large market penetration of these vehicles is dramatically changing the view of the power distribution system. Unlike other power loads, these vehicles can be connected to power grids anywhere and anytime, which brings more spatial and temporal diversity and uncertainty. There is an urgent need to investigate the impact of PHEV/PEV charging on the power distribution system considering multidisciplinary complexities (e.g., driving behavior, route and departure time choice, charging station location, engineering, policy, economic, environment, technology, and social impact). This paper consolidates the modeling and simulation of power distribution system and transportation network in order to assess the emerging electric vehicle technologies. Moreover, this paper proposes a comprehensive co-modeling/simulation framework for investigating the impact of the electrification of transportation in the real world.