Data Generation using Latin Hypercube Sampling Approach for Power System Dynamic Security Assessment

Data driven dynamic security assessment using Machine Learning (ML) models based on advanced artificial intelligence techniques plays a very important role in the field of power systems. Database robustness (redundant and highly informative) to manage security issues under abnormal situations and system operating limits affects ML model performance. Given current data availability of unusual circumstances and the need for more thorough techniques to producing such datasets which contains all dynamically critical locations in the power system operating space, this paper introduces a novel approach based on smart and exceptional Latin hypercube sampling techniques to generate the highly quality datasets. Convex relaxation techniques are employed to explore the highly informative regions by using certificate of infeasibility for a specified security threshold. It can efficiently build a database of all dynamically critical power system operating stages (offline) and train ML models for real-time security assessment. The suggested method is computationally efficient (having compact size dataset with minimum computational efforts, and time) and tractable for huge and complex power systems. The suggested method creates datasets for two typical IEEE systems (14 and 57bus) to train the Machine Learning model to assess the system's operating state (insecure or secure) including the stability margin.