Thesis Title: Scenario-Based Auto Data Generation Framework Using Hardware-in-the-Loop Testbed for Deep Learning-Based Application

​

Summary:​

        This thesis introduces the architecture and deployment of a scenario-driven autonomous data generation framework purpose-built for deep learning (DL) applications within Distributed Energy Resource (DER)-integrated distribution networks. With the accelerated integration of DERs, modern power systems have become increasingly complex, facing challenges such as stochastic power injection, dynamic load conditions, and evolving cyber-physical vulnerabilities. ​To mitigate the inherent limitations of conventional dataset acquisition methods, the proposed framework facilitates the automated synthesis of high-resolution, feature-rich datasets encompassing a diverse set of operating conditions. These include transient line faults, steady-state anomalies, and cyberattack-induced perturbations, all emulated in real-time via a Hardware-in-the-Loop (HIL) Cyber-Physical Systems (CPS) testbed powered by OPAL-RT.​​

     The system exposes a Python-based configuration interface, enabling dynamic control over parameters such as voltage magnitude, current phase, fault injection types, attack vectors, and time domain characteristics. An integrated API orchestrates the full data pipeline—from event triggering and waveform acquisition to structured dataset generation—ensuring reproducibility, traceability, and scalability for ML model development. ​The framework generated a corpus of labeled datasets tailored for anomaly detection tasks. These were validated using a Long Short-Term Memory (LSTM) neural network optimized for temporal pattern recognition. The model attained 99.01% classification accuracy, affirming the quality and utility of the synthetic datasets for critical grid intelligence applications.​​

       This research not only addresses the data scarcity problem in smart grid ML workflows but also lays the groundwork for next-generation grid analytics, including real-time diagnostic systems, federated learning architectures, and edge AI deployment in operational substations.

 

Tech Stack

• Languages & Libraries: Python, NumPy, Pandas, Scikit-learn, TensorFlow/Keras

• Simulation & Systems: OPAL-RT, RT-Lab, MATLAB

• Data Engineering: Custom Feature Engineering Scripts, Real-Time Signal Logging, Data Normalization Pipelines

• Modeling: LSTM (Time Series), Hyperparameter Tuning, Performance Evaluation (Precision, Recall, F1)

• Automation: Python Scripting, API-driven Test Configuration, Dataset Orchestration Framework

​

What the Framework Offers

• Plug-and-play simulation control: Automate grid events like faults or attacks using parameterized Python scripts. 

• Custom dataset generation: Create large-scale, annotated datasets tailored for ML/DL workflows.

• Realistic test scenarios: Use Hardware-in-the-Loop for high-fidelity signal simulation.

• Rapid prototyping: Evaluate ML models quickly using freshly generated, labeled datasets.

• High-performance validation: Achieved 99.01% anomaly detection accuracy using LSTM on generated data.

• Reproducible & extensible: Easily integrate into other AI pipelines or smart grid applications.