Machine - Learning - Based Prediction Models for Power Systems: Techno - economic Analysis of Voltage Stability and Cost Optimization

This paper introduces a novel framework that integrates advanced machine learning methods with classical power system analysis to address voltage stability and cost optimization challenges in modern grids. In an era where power systems incorporate distributed renewable sources and operate near their physical limits, ensuring voltage stability and economic efficiency is essential. An infinite bus model coupled with a continuation power flow approach is employed to simulate voltage dynamics and delineate the evolution of active and reactive power under maximum power transfer conditions. A meticulously designed long short-term memory (LSTM) network, featuring multiple recurrent layers and nonlinear activation functions, is implemented to accurately forecast voltage stability metrics. Complementary evaluations using convolutional neural networks (CNN) further enhance predictive accuracy by extracting intrinsic spatial relationships from tabular data and mitigating overfitting. Experimental results, obtained in a Python-based Jupyter Notebook environment, demonstrate near-perfect accuracy for the LSTM model and robust generalization for the CNN model. Statistical measures such as accuracy, precision, recall, and the F1 score substantiate the efficacy of the classification framework in distinguishing stable from unstable operating conditions. This integration of data-driven predictive analytics with traditional power system modeling significantly advances both technical performance and economic decision-making, thereby contributing to the development of resilient, adaptive, and cost-effective energy infrastructures.