Exploring Time -Series Machine Learning for Voltage Prediction in Air Conditioning Systems

Abstract

This paper presents a comprehensive evaluation of machine learning algorithms for voltage timeseries prediction in power systems. We compare six models: Linear Regression, Random Forest, XGBoost, Support Vector Regression (SVR), k-Nearest Neighbors (KNN), and Long Short-Term Memory (LSTM) networks using multiple metrics (MAE, MSE, RMSE, R2). Our results demonstrate that LSTM networks achieve superior performance with MAE of 7019.20V and R2 of 0.915, effectively capturing temporal dependencies in voltage fluctuations. Surprisingly, Linear Regression performs competitively (R2=0.974), suggesting significant linear components in voltage dynamics. The study provides practical insights for power system operators, recommending LSTM for accurate voltage stability assessment and Linear Regression for computationally efficient monitoring. The findings highlight the importance of algorithm selection based on specific grid monitoring requirements.