Towards Probabilistic Dynamic Security Assessment and Enhancement of Large Power Systems

TL;DR

The paper tackles probabilistic dynamic security assessment for large power systems under load/generation variability and uncertain cascade propagation. It introduces a three-step PDSA framework: building a database of credible states from weather-driven Monte Carlo years and market actions, sampling with per-contingency stopping criteria and optional screening, and handling uncertain protection behavior during fast cascading outages, followed by security enhancement via interpretable ML. Key contributions include rigorous statistical accuracy indicators for per-contingency sampling ($SE_i \le \epsilon R$), stability-based screening to reduce compute, a protection-uncertainty handling mechanism, and ML-driven root-cause analysis to guide mitigations. Demonstrated on the RTS-GMLC system, the method identifies critical contingencies and scalable computation, with scalability discussions suggesting feasibility for grids with thousands of buses. The approach offers practical pathways for operators to quantify risk and prioritize security investments.

Abstract

This paper proposes a novel methodology for probabilistic dynamic security assessment and enhancement of power systems that considers load and generation variability, N-2 contingencies, and uncertain cascade propagation caused by uncertain protection system behaviour. In this methodology, a database of likely operating conditions is generated via weather data, a market model and a model of operators' preventive actions. System states are sampled from this database and contingencies are applied to them to perform the security assessment. Rigorous statistical indicators are proposed to decide how many biased and unbiased samples to simulate to reach a target accuracy on the statistical error on the estimated risk from individual contingencies. Optionally, a screening of contingencies can be performed to limit the computational burden of the analysis. Finally, interpretable machine learning techniques are used to identify the root causes of the risk from critical contingencies, to ease the interpretation of the results, and to help with security enhancement. The method is demonstrated on the 73-bus reliability test system, and the scalability to large power systems (with thousands of buses) is also discussed.

Figures (9)

• Figure 1: Flowchart of the proposed PDSA methodology
• Figure 2: Prediction of the relevance of protection-related uncertainties using the indicator from HandlingProtections. (a) Slow sequence (reference). (b) Fast sequence for which the system is unlikely to be affected by protection-related uncertainties. (c) Fast sequence likely to be affected: $3_c$ occurs before $2_a$. (d) Fast sequence likely to be affected: a new event ($4_d$) occurs.
• Figure 3: Network layout of the RTS-GMLC from RTS-GMLC. New interconnections are represented in black.
• Figure 4: Market dispatch for a typical day in January (left) and July (right)
• Figure 5: Number of sampled operating conditions and statistical accuracy for all delayed-clearing N-1 contingencies sorted in decreasing order of likelihood