Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Computationally Efficient Electromagnetic Transient Power System Studies using Bayesian Optimization

Marius Kuhn, Evelyn Heylen, Willem Leterme

TL;DR

The paper tackles the high cost of EMT-type power-system simulations by introducing a Bayesian Optimization (BO) framework capable of handling optimization, active learning-based classification, and active discovery. By using Gaussian Process surrogates and acquisition functions (e.g., Expected Improvement), the approach achieves near-global optima with far fewer simulations, demonstrated on transformer energization problems where brute-force Monte Carlo or derivative-free optimizers require orders of magnitude more evaluations. Key findings show BO reduces simulations by factors up to 10–100 and that initialization and surrogate choice critically influence outcomes, especially in higher-dimensional input spaces. The work offers a versatile, data-efficient automation tool for complex, nonlinear power-system studies with potential broad applicability beyond optimization to classification and discovery tasks, enabling more efficient planning and risk assessment in EMT contexts.

Abstract

The power system of the future will be governed by complex interactions and non-linear phenomena at small time-scales, that should be studied more and more through computationally expensive software simulations. To solve the abovementioned problems, power system engineers face problems with following characteristics: (i) a computationally expensive simulator, (ii) non-linear functions to optimize and (iii) lack of abundance of data. Existing optimization settings involving EMT-type simulations have been developed, but mainly use a deterministic model and optimizer, which may be computationally inefficient and do not guarantee finding a global optimum. Furthermore, the main focus has been on optimization routines, and less attention has been paid to other tasks such as classification. In this paper, an automation framework based on Bayesian Optimization is introduced, and applied to two case studies involving optimization and classification. It is found that the framework has the potential to reduce computational effort, outperform deterministic optimizers and is applicable to a multitude of problems. Nevertheless, it was found that the output of the Bayesian Optimization depends on the number of samples used for initialization, and in addition, careful selection of surrogate models, which should be subject to future investigation.

Computationally Efficient Electromagnetic Transient Power System Studies using Bayesian Optimization

TL;DR

The paper tackles the high cost of EMT-type power-system simulations by introducing a Bayesian Optimization (BO) framework capable of handling optimization, active learning-based classification, and active discovery. By using Gaussian Process surrogates and acquisition functions (e.g., Expected Improvement), the approach achieves near-global optima with far fewer simulations, demonstrated on transformer energization problems where brute-force Monte Carlo or derivative-free optimizers require orders of magnitude more evaluations. Key findings show BO reduces simulations by factors up to 10–100 and that initialization and surrogate choice critically influence outcomes, especially in higher-dimensional input spaces. The work offers a versatile, data-efficient automation tool for complex, nonlinear power-system studies with potential broad applicability beyond optimization to classification and discovery tasks, enabling more efficient planning and risk assessment in EMT contexts.

Abstract

The power system of the future will be governed by complex interactions and non-linear phenomena at small time-scales, that should be studied more and more through computationally expensive software simulations. To solve the abovementioned problems, power system engineers face problems with following characteristics: (i) a computationally expensive simulator, (ii) non-linear functions to optimize and (iii) lack of abundance of data. Existing optimization settings involving EMT-type simulations have been developed, but mainly use a deterministic model and optimizer, which may be computationally inefficient and do not guarantee finding a global optimum. Furthermore, the main focus has been on optimization routines, and less attention has been paid to other tasks such as classification. In this paper, an automation framework based on Bayesian Optimization is introduced, and applied to two case studies involving optimization and classification. It is found that the framework has the potential to reduce computational effort, outperform deterministic optimizers and is applicable to a multitude of problems. Nevertheless, it was found that the output of the Bayesian Optimization depends on the number of samples used for initialization, and in addition, careful selection of surrogate models, which should be subject to future investigation.
Paper Structure (17 sections, 5 equations, 9 figures, 1 table, 1 algorithm)

This paper contains 17 sections, 5 equations, 9 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Automation Settings
  3. Optimization
  4. Classification
  5. Active Search and Discovery
  6. Automation Framework Elements
  7. Function $f(\boldsymbol{x})$
  8. Probabilistic surrogate model
  9. Acquisition Function
  10. Implementation
  11. Case Study
  12. Transformer Energization
  13. Brute Force Benchmark Solution
  14. Maximum Overvoltage (One Dimensional Input)
  15. Maximum Overvoltage (Three-Dimensional Input)
  16. ...and 2 more sections

Figures (9)

  • Figure 1: Workflow for Bayesian Optimization (a), Active Learning (b), Active Search (c) and Active Discovery (d).
  • Figure 2: Electrical circuit for transformer energization analysis.
  • Figure 3: Brute Force Benchmark Solution
  • Figure 4: Optimal value against function evaluations (1D Input)
  • Figure 5: Surrogate model and benchmark for BO
  • ...and 4 more figures