Smart Sampling for Reduced and Representative Power System Scenario Selection

Renke Huang,Xinya Li,Xinda Ke,Sohom Datta,Xueqing Sun,Qiuhua Huang,Z. Jason Hou

doi:10.1109/oajpe.2021.3093278

Abstract

With increasing penetration of renewable energy and active market participation, power system operation scenarios and patterns have increased exponentially. This has led to challenges in identifying a good subset of scenarios for routine planning, operation, and emerging machine learning applications. To address these challenges, we develop an approach integrating comprehensive exploratory data analyses and smart sampling techniques to identify and select a small subset of representative power system scenarios that maintain the coverage of system scenarios and operation envelope, therefore, leading to very efficient, yet representative studies and analysis. We propose a hierarchical Latin Hypercube Sampling (LHS) technique for smart sampling, which allows free-form distributions of system load and considers generator commitment status along with generation levels. A set of performance metrics are also defined for systematic evaluation of the adequacy and efficiency of the sampled cases. The developed approach and metrics are demonstrated using the Texas 2000 bus system in this paper and will be extended to the more complex real world systems such as Western Interconnect System.

Highlights

For a large-scale modern power system, grid operation conditions and contingency scenarios are subject to increased uncertainties that need to be quantified based on existing information and considered for further power system security and reliability analyses
A wide range of grid operation conditions are often required in power system control analyses, such as emergency control by deep reinforcement learning (DRL)[1], [2] or solving AC power flow by deep supervised learning framework [3]
This paper proposes a well-designed modification of Latin Hypercube Sampling (LHS) to deal with such data, incorporating hierarchical sampling and LHS with dependence (LHSD)

Summary

INTRODUCTION

For a large-scale modern power system, grid operation conditions and contingency scenarios are subject to increased uncertainties that need to be quantified based on existing information and considered for further power system security and reliability analyses. [5], [6] to address uncertainties in analyses for energy systems Such techniques can be classified into three categories: sampling-based method, forecasting-based method, and optimization-based method[6], to be used for generating scenarios for single, or multiple parallel variables, such as wind power[7], or load[8]. Our hierarchical sampling is flexible and adaptive to the real-world problems, dealing with high-dimensional data with hierarchical structures and mixture of data types It significantly increases the sampling efficiency, while maintaining the probabilistic properties and grid characteristics from the perspective of operational patterns and generation planning.

DATA DESCRIPTION

METHODOLOGY

Distribution Fitting

Summary of Down-selected Power Flow Base Cases And Contingencies

Validation

Findings

CONCLUSION