Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Data-Driven Multi-Stage OPF

Oleksii Molodchyk, Philipp Schmitz, Alexander Engelmann, Karl Worthmann, Timm Faulwasser

TL;DR

The paper addresses multi-stage OPF in large-scale power systems by replacing explicit topology identification with a data-driven predictive control approach for a linear descriptor model that captures DC power flow with battery storage. It leverages Willems' fundamental lemma to build a non-parametric input–output representation from offline data and solves a constrained OCP over a horizon $L$ with a short initial trajectory length, avoiding explicit network parameter estimation. On a modified 118-bus test system, the data-driven approach achieves a closed-loop performance within about 0.5% of a classical topology-based controller while maintaining feasible line flows and practical compute times, despite not requiring generator/demand locations. The work demonstrates the viability of data-driven control for power-system operation and outlines directions for incorporating measurement noise and voltage data in future work.

Abstract

The operation of large-scale power systems is usually scheduled ahead via numerical optimization. However, this requires models of grid topology, line parameters, and bus specifications. Classic approaches first identify the network topology, i.e., the graph of interconnections and the associated impedances. The power generation schedules are then computed by solving a multi-stage optimal power flow (OPF) problem built around the model. In this paper, we explore the prospect of data-driven approaches to multi-stage optimal power flow. Specifically, we leverage recent findings from systems and control to bypass the identification step and to construct the optimization problem directly from data. We illustrate the performance of our method on a 118-bus system and compare it with the classical identification-based approach.

Towards Data-Driven Multi-Stage OPF

TL;DR

The paper addresses multi-stage OPF in large-scale power systems by replacing explicit topology identification with a data-driven predictive control approach for a linear descriptor model that captures DC power flow with battery storage. It leverages Willems' fundamental lemma to build a non-parametric input–output representation from offline data and solves a constrained OCP over a horizon with a short initial trajectory length, avoiding explicit network parameter estimation. On a modified 118-bus test system, the data-driven approach achieves a closed-loop performance within about 0.5% of a classical topology-based controller while maintaining feasible line flows and practical compute times, despite not requiring generator/demand locations. The work demonstrates the viability of data-driven control for power-system operation and outlines directions for incorporating measurement noise and voltage data in future work.

Abstract

The operation of large-scale power systems is usually scheduled ahead via numerical optimization. However, this requires models of grid topology, line parameters, and bus specifications. Classic approaches first identify the network topology, i.e., the graph of interconnections and the associated impedances. The power generation schedules are then computed by solving a multi-stage optimal power flow (OPF) problem built around the model. In this paper, we explore the prospect of data-driven approaches to multi-stage optimal power flow. Specifically, we leverage recent findings from systems and control to bypass the identification step and to construct the optimization problem directly from data. We illustrate the performance of our method on a 118-bus system and compare it with the classical identification-based approach.

Paper Structure

This paper contains 8 sections, 1 theorem, 24 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries and Problem Statement
  3. Linear descriptor systems
  4. Data-driven Predictive Control
  5. Data-Driven Multi-Stage OPF
  6. Connection to topology identification
  7. Numerical Experiments
  8. Conclusion and Outlook

Key Result

Lemma 1

Assume that eqn:lds_tr is R-controllable. Let $(\mathbf{u}^{\mathrm d}, \mathbf{w}^{\mathrm d}, \mathbf{y}^{\mathrm d})_{[1:T]}$ be an offline-recorded length-$T$ input-output trajectory of system eqn:lds_tr such that the joint input signal $\left(\mathbf{u}^{\mathrm d}, \mathbf{w}^{\mathrm d}\right

Figures (1)

  • Figure 1: Maximum power flow over all lines during closed-loop control for the three considered controllers.

Theorems & Definitions (3)

  • Definition 1: R-controllability and R-observability dai.1989
  • Lemma 1: Schmitz2022a
  • Remark 1: Simplifying the non-parametric system representation \ref{['eqn:dd_sys']}