Samuel Chevalier

Samuel Chevalier, Petr Vorobev, Konstantin Turitsyn

Locating the sources of forced low-frequency oscillations in power systems is an important problem. A number of proposed methods demonstrate their practical usefulness, but many of them rely on strong modeling assumptions and provide poor performance in certain cases for reasons still not well understood. This paper proposes a systematic method for locating the source of a forced oscillation by considering a generator's response to fluctuations of its terminal voltages and currents. It is shown that a generator can be represented as an effective admittance matrix with respect to low-frequency oscillations, and an explicit form for this matrix, for various generator models, is derived. Furthermore, it is shown that a source generator, in addition to its effective admittance, is characterized by the presence of an effective current source thus giving a natural qualitative distinction between source and nonsource generators. Detailed descriptions are given of a source detection procedure based on this developed representation, and the method's effectiveness is confirmed by simulations on the recommended testbeds (eg. WECC 179-bus system). This method is free of strong modeling assumptions and is also shown to be robust in the presence of measurement noise and generator parameter uncertainty.

Samuel Chevalier, Petr Vorobev, Konstantin Turitsyn

Since forced oscillations are exogenous to dynamic power system models, the models by themselves cannot predict when or where a forced oscillation will occur. Locating the sources of these oscillations, therefore, is a challenging problem which requires analytical methods capable of using real time power system data to trace an observed oscillation back to its source. The difficulty of this problem is exacerbated by the fact that the parameters associated with a given power system model can range from slightly uncertain to entirely unknown. In this paper, a Bayesian framework, via a two-stage Maximum A Posteriori optimization routine, is employed in order to locate the most probable source of a forced oscillation given an uncertain prior model. The approach leverages an equivalent circuit representation of the system in the frequency domain and employs a numerical procedure which makes the problem suitable for real time application. The derived framework lends itself to successful performance in the presence of PMU measurement noise, high generator parameter uncertainty, and multiple forced oscillations occurring simultaneously. The approach is tested on a 4-bus system with a single forced oscillation source and on the WECC 179-bus system with multiple oscillation sources.

Samuel Chevalier, Paul D. H. Hines

With the continued deployment of synchronized Phasor Measurement Units (PMUs), high sample rate data are rapidly increasing the real time observability of power systems. Prior research has shown that the statistics of these data can provide useful information regarding network stability, but it is not yet known how this statistical information can be actionably used to improve power system stability. To address this issue, this paper presents a method that gauges and improves the voltage stability of a system using the statistics present in PMU data streams. Leveraging an analytical solver to determine a range of "critical" bus voltage variances, the presented methods monitor raw statistical data in an observable load pocket to determine when control actions are needed to mitigate the risk of voltage collapse. A simple reactive power controller is then implemented, which acts dynamically to maintain an acceptable voltage stability margin within the system. Time domain simulations on 3-bus and 39-bus test cases demonstrate that the resulting statistical controller can out-perform more conventional feedback control systems by maintaining voltage stability margins while loads simultaneously increase and fluctuate.

Jochen Stiasny, Samuel Chevalier, Rahul Nellikkath et al.

Deep decarbonization of the energy sector will require massive penetration of stochastic renewable energy resources and an enormous amount of grid asset coordination; this represents a challenging paradigm for the power system operators who are tasked with maintaining grid stability and security in the face of such changes. With its ability to learn from complex datasets and provide predictive solutions on fast timescales, machine learning (ML) is well-posed to help overcome these challenges as power systems transform in the coming decades. In this work, we outline five key challenges (dataset generation, data pre-processing, model training, model assessment, and model embedding) associated with building trustworthy ML models which learn from physics-based simulation data. We then demonstrate how linking together individual modules, each of which overcomes a respective challenge, at sequential stages in the machine learning pipeline can help enhance the overall performance of the training process. In particular, we implement methods that connect different elements of the learning pipeline through feedback, thus "closing the loop" between model training, performance assessments, and re-training. We demonstrate the effectiveness of this framework, its constituent modules, and its feedback connections by learning the N-1 small-signal stability margin associated with a detailed model of a proposed North Sea Wind Power Hub system.

Samuel Chevalier, Spyros Chatzivasileiadis

Machine learning, which can generate extremely fast and highly accurate black-box surrogate models, is increasingly being applied to a variety of AC power flow problems. Rigorously verifying the accuracy of the resulting black-box models, however, is computationally challenging. This paper develops a tractable neural network verification procedure which incorporates the ground truth of the non-linear AC power flow equations to determine worst-case neural network prediction error. Our approach, termed Sequential Targeted Tightening (STT), leverages a loosely convexified reformulation of the original verification problem, which is an intractable mixed integer quadratic program (MIQP). Using the sequential addition of targeted cuts, we iteratively tighten our formulation until either the solution is sufficiently tight or a satisfactory performance guarantee has been generated. After learning neural network models of the 14, 57, 118, and 200-bus PGLib test cases, we compare the performance guarantees generated by our STT procedure with ones generated by a state-of-the-art MIQP solver, Gurobi 11.0. We show that STT often generates performance guarantees which are far tighter than the MIQP upper bound.

Samuel Chevalier, Petr Vorobev, Konstantin Turitsyn et al.

Despite wide-scale deployment of phasor measurement unit technology, locating the sources of low frequency forced oscillations in power systems is still an open research topic. The dissipating energy flow method is one source location technique which has performed remarkably well in both simulation and real time application at ISO New England. The method has several deficiencies, though, which are still poorly understood. This paper borrows the concepts of passivity and positive realness from the controls literature in order to interpret the dissipating energy flow method, pinpoint the reasons for its deficiencies, and set up a framework for improving the method. The theorems presented in this paper are then tested via simulation on a simple infinite bus power system model.

Samuel Chevalier, Ilgiz Murzakhanov, Spyros Chatzivasileiadis

Computational tools for rigorously verifying the performance of large-scale machine learning (ML) models have progressed significantly in recent years. The most successful solvers employ highly specialized, GPU-accelerated branch and bound routines. Such tools are crucial for the successful deployment of machine learning applications in safety-critical systems, such as power systems. Despite their successes, however, barriers prevent out-of-the-box application of these routines to power system problems. This paper addresses this issue in two key ways. First, for the first time to our knowledge, we enable the simultaneous verification of multiple verification problems (e.g., checking for the violation of all line flow constraints simultaneously and not by solving individual verification problems). For that, we introduce an exact transformation that converts the "worst-case" violation across a set of potential violations to a series of ReLU-based layers that augment the original neural network. This allows verifiers to interpret them directly. Second, power system ML models often must be verified to satisfy power flow constraints. We propose a dualization procedure which encodes linear equality and inequality constraints directly into the verification problem; and in a manner which is mathematically consistent with the specialized verification tools. To demonstrate these innovations, we verify problems associated with data-driven security constrained DC-OPF solvers. We build and test our first set of innovations using the $α,β$-CROWN solver, and we benchmark against Gurobi 10.0. Our contributions achieve a speedup that can exceed 100x and allow higher degrees of verification flexibility.

Samuel Chevalier, Duncan Starkenburg, Krishnamurthy Dvijotham

In the field of formal verification, Neural Networks (NNs) are typically reformulated into equivalent mathematical programs which are optimized over. To overcome the inherent non-convexity of these reformulations, convex relaxations of nonlinear activation functions are typically utilized. Common relaxations (i.e., static linear cuts) of "S-shaped" activation functions, however, can be overly loose, slowing down the overall verification process. In this paper, we derive tuneable hyperplanes which upper and lower bound the sigmoid activation function. When tuned in the dual space, these affine bounds smoothly rotate around the nonlinear manifold of the sigmoid activation function. This approach, termed $α$-sig, allows us to tractably incorporate the tightest possible, element-wise convex relaxation of the sigmoid activation function into a formal verification framework. We embed these relaxations inside of large verification tasks and compare their performance to LiRPA and $α$-CROWN, a state-of-the-art verification duo.

Vladimir Dvorkin, Samuel Chevalier, Spyros Chatzivasileiadis

Gas network planning optimization under emission constraints prioritizes gas supply with the least CO$_2$ intensity. As this problem includes complex physical laws of gas flow, standard optimization solvers cannot guarantee convergence to a feasible solution. To address this issue, we develop an input-convex neural network (ICNN) aided optimization routine which incorporates a set of trained ICNNs approximating the gas flow equations with high precision. Numerical tests on the Belgium gas network demonstrate that the ICNN-aided optimization dominates non-convex and relaxation-based solvers, with larger optimality gains pertaining to stricter emission targets. Moreover, whenever the non-convex solver fails, the ICNN-aided optimization provides a feasible solution to network planning.

Eren Tekeler, Xiangru Zhong, Huan Zhang et al.

Security-Constrained DC Optimal Power Flow (SC DCOPF) is an important tool for transmission system operators, enabling economically efficient and physically secure dispatch decisions. Although CPU-based commercial solvers (e.g., Gurobi) can efficiently solve SC-DCOPF problems with a reasonable number of security constraints, their performance degrades rapidly as both system size and the number of contingencies grow into thousands. In this paper, we design a computational graph representation of the SC-DCOPF-based market-clearing problem, inspired by the third ARPA-E Grid Optimization Competition. Using a tool from the neural network verification community known as Interval Bound Propagation (IBP), we quickly compute bounds on the optimal objective across the full set of N-1 contingencies. Our results demonstrate that IBP can compute certified bounds with mean optimal solution gaps below 3.98% on small cases, and it can efficiently scale up to 8,316 bus systems with thousands of contingencies.

Omid Mokhtari, Samuel Chevalier, Mads Almassalkhi

This paper presents a novel structure-preserving, Kron-based reduction framework for unbalanced distribution feeders. The method aggregates electrically similar nodes within a mixed-integer optimization (MIP) problem to produce reduced networks that optimally reproduce the voltage profiles of the original full network. To overcome computational bottlenecks of MIP formulations, we propose an exhaustive-search formulation to identify optimal aggregation decisions while enforcing voltage margin limits. The proposed exhaustive network reduction algorithm is parallelizable on GPUs, which enables scalable network reduction. The resulting reduced networks approximate the full system's voltage profiles with low errors and are suitable for steady-state analysis and optimal power flow studies. The framework is validated on two real utility distribution feeders with 5,991 and 8,381 nodes. The reduced models achieve up to 90% and 80% network reduction, respectively, while the maximum voltage-magnitude error remains below 0.003 p.u. Furthermore, on a 1000-node version of the network, the GPU-accelerated reduction algorithm runs up to 15x faster than its CPU-based counterpart.

Saba Rafiei, Samuel Chevalier

By exploiting the observed tightness of dual rotated second-order cone (RSOC) constraints, this paper transforms the dual of a conic ACOPF relaxation into an equivalent, non-conic problem where dual constraints are implicitly enforced through eliminated dual RSOC variables. To accomplish this, we apply the RSOC-based Jabr relaxation of ACOPF, pose its dual, and then show that all dual RSOC constraints must be tight (i.e., active) at optimality. We then construct a reduced dual maximization problem with only non-negativity constraints, avoiding the explicit RSOC inequality constraints. Numerical experiments confirm that the tight formulation recovers the same dual objective values as a mature conic solver (e.g., MOSEK via PowerModels) on various PGLib benchmark test systems (ranging from 3- to 1354-buses). The proposed formulation has useful performance benefits, compared with its conic counterpart, and it allows us to define a bounding function which provides a guaranteed lower bound on system cost. While this paper focuses on demonstrating the correctness and validity of the proposed structural simplification, it lays the groundwork for future GPU-accelerated first-order optimization methods which can exploit the unconstrained nature of the proposed formulation.

Oleksii Molodchyk, Omid Mokhtari, Samuel Chevalier et al.

Real-time control of distribution networks requires accurate information about the system state. In practice, however, such information is difficult to obtain because real-time measurements are available only at a limited number of locations. This paper proposes a novel data-driven power flow (DDPF) framework for balanced radial distribution networks. The proposed algorithm combines the behavioral approach with the DistFlow model and leverages offline historical data to solve power flow problems using only a limited set of real-time measurements. To design DDPF under sparse measurement conditions, we develop a sensor placement problem based on optimal network reductions. This allows us to determine sensor locations subject to a predefined sensor budget and to explicitly account for the radial nature of distribution networks. Unlike approaches that rely on full observability, the proposed framework is designed for practical distribution grids with sparse measurement availability. This enables data-driven power flow for real-time operation while reducing the number of required sensors. On several test cases, the proposed DDPF algorithm could demonstrate accurate voltage magnitude predictions, with a maximum error less than 0.001 p.u., with as little as 25% of total locations equipped with sensors.

Alyssa Kody, Samuel Chevalier, Spyros Chatzivasileiadis et al.

Nonlinear power flow constraints render a variety of power system optimization problems computationally intractable. Emerging research shows, however, that the nonlinear AC power flow equations can be successfully modeled using Neural Networks (NNs). These NNs can be exactly transformed into Mixed Integer Linear Programs (MILPs) and embedded inside challenging optimization problems, thus replacing nonlinearities that are intractable for many applications with tractable piecewise linear approximations. Such approaches, though, suffer from an explosion of the number of binary variables needed to represent the NN. Accordingly, this paper develops a technique for training an "optimally compact" NN, i.e., one that can represent the power flow equations with a sufficiently high degree of accuracy while still maintaining a tractable number of binary variables. We show that the resulting NN model is more expressive than both the DC and linearized power flow approximations when embedded inside of a challenging optimization problem (i.e., the AC unit commitment problem).

Samuel Chevalier, Jochen Stiasny, Spyros Chatzivasileiadis

Recent advances in deep learning have allowed neural networks (NNs) to successfully replace traditional numerical solvers in many applications, thus enabling impressive computing gains. One such application is time domain simulation, which is indispensable for the design, analysis and operation of many engineering systems. Simulating dynamical systems with implicit Newton-based solvers is a computationally heavy task, as it requires the solution of a parameterized system of differential and algebraic equations at each time step. A variety of NN-based methodologies have been shown to successfully approximate the trajectories computed by numerical solvers at a fraction of the time. However, few previous works have used NNs to model the numerical solver itself. For the express purpose of accelerating time domain simulation speeds, this paper proposes and explores two complementary alternatives for modeling numerical solvers. First, we use a NN to mimic the linear transformation provided by the inverse Jacobian in a single Newton step. Using this procedure, we evaluate and project the exact, physics-based residual error onto the NN mapping, thus leaving physics ``in the loop''. The resulting tool, termed the Physics-pRojected Neural-Newton Solver (PRoNNS), is able to achieve an extremely high degree of numerical accuracy at speeds which were observed to be up to 31% faster than a Newton-based solver. In the second approach, we model the Newton solver at the heart of an implicit Runge-Kutta integrator as a contracting map iteratively seeking a fixed point on a time domain trajectory. The associated recurrent NN simulation tool, termed the Contracting Neural-Newton Solver (CoNNS), is embedded with training constraints (via CVXPY Layers) which guarantee the mapping provided by the NN satisfies the Banach fixed-point theorem.