Abstract:Quasi-static time series (QSTS) simulations have great potential for evaluating the grid's ability to accommodate the large-scale integration of distributed energy resources. However, as grids expand and operate closer to their limits, iterative power flow solvers, central to QSTS simulations, become computationally prohibitive and face increasing convergence issues. Neural power flow solvers provide a promising alternative, speeding up power flow computations by 3 to 4 orders of magnitude, though they are costly to train. In this paper, we envision how recently introduced grid foundation models could improve the economic viability of neural power flow solvers. Conceptually, these models amortize training costs by serving as a foundation for a range of grid operation and planning tasks beyond power flow solving, with only minimal fine-tuning required. We call for collaboration between the AI and power grid communities to develop and open-source these models, enabling all operators, even those with limited resources, to benefit from AI without building solutions from scratch.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are as follows: With the expansion of the power grid scale and the large - scale integration of Distributed Energy Resources (DERs), in Quasi - Static Time Series (QSTS) simulations, the computational cost of iterative power flow solvers is too high and the convergence problems are becoming increasingly serious. Specifically: 1. **High computational complexity**: As the number of power grid nodes increases, the computational complexity of traditional iterative power flow solvers grows quadratically, resulting in overly long computation time and difficulty in meeting the requirements of practical applications. 2. **Convergence problems**: When approaching the limit of power grid operation, the convergence of iterative solvers deteriorates. Especially when dealing with large - scale and complex power grids, non - convergence may occur. 3. **Lack of flexibility**: Traditional methods have difficulty coping with the uncertainties and dynamic changes brought by more DERs and electric vehicles (EVs) in future power grids. To solve these problems, the paper proposes to use Neural Power Flow Solvers based on Foundation Models (FMs). These models can significantly accelerate power flow calculations and can be applied to a variety of power grid operation and planning tasks with a small amount of fine - tuning, thereby improving economic feasibility and reducing the dependence on large - scale computing resources. ### Key point summary: - **Problem background**: QSTS simulations are used to evaluate the adaptability of power grids to large - scale DERs, but traditional methods face challenges in computational efficiency and convergence. - **Solution**: Introduce Neural Power Flow Solvers and use AI foundation models to reduce computational costs and improve convergence. - **Expected effect**: Through a one - time training process, the Neural Power Flow Solvers can significantly accelerate the calculation speed during operation and at the same time reduce the frequency of convergence problems. In terms of formulas, the main concepts involved in the paper include, but are not limited to: - Solving power flow equations: \(\mathbf{f}(\mathbf{x})=0\), where \(\mathbf{x}\) represents power grid state variables (such as voltage amplitude and phase angle), and \(\mathbf{f}\) is a nonlinear function. - Neural network mapping: Directly from input \(\mathbf{u}\) to output \(\mathbf{y}\), that is, \(\mathbf{y}=\mathbf{g}(\mathbf{u};\theta)\), where \(\theta\) is the model parameter. In this way, the paper aims to promote the cooperation between the AI and power system communities, jointly develop and open - source these foundation models, so that more power grid operators can benefit from the progress of AI technology.

Accelerating Quasi-Static Time Series Simulations with Foundation Models

Optimal Power Grid Operations with Foundation Models

A Perspective on Foundation Models for the Electric Power Grid

Fast Solutions in Power System Simulation through Coupling with Data-Driven Power Flow Models for Voltage Estimation

Convergence Enhancement for Neural Network Integrated Power System Time Domain Simulation

Applying Computational Grid Technology to Power System

Distributed Power Flow and Distributed Optimization -- Formulation, Solution, and Open Source Implementation

An Approach Inspired by Quantum Mechanics for the Modeling of Large Power Systems

Simulating the Smart Grid

Cyber-Constrained Optimal Power Flow Model for Smart Grid Resilience Enhancement

Early Exploration of a Flexible Framework for Efficient Quantum Linear Solvers in Power Systems

Artificial Intelligence and Mathematical Models of Power Grids Driven by Renewable Energy Sources: A Survey

Modeling and Simulation for On-Chip Power Grid Networks by Locally Dominant Krylov Subspace Method

Improving Long-Term Electricity Time Series Forecasting in Smart Grid with a Three-Stage Channel-Temporal Approach

From Co- Toward Multi-Simulation of Smart Grids based on HLA and FMI Standards

Framework for artificial intelligence analysis in large-scale power grids based on digital simulation

Solving AC Power Flow with Graph Neural Networks under Realistic Constraints

Probabilistic power flow based on physics-guided graph neural networks

Approximating Power Flow and Transmission Losses in Coordinated Capacity Expansion Problems

New Generation UHVAC/DC Power Grid Simulation Platform Architecture

Acceleration of Power System Dynamic Simulations using a Deep Equilibrium Layer and Neural ODE Surrogate