Enhanced Learning Model of LSTM-GAN Hybrid Network in the Dynamic Prediction of Building Energy Consumption

Building energy consumption plays a crucial role in energy management and sustainable urban development, and integrating smart meters and Internet of Things (IoT) devices offers a wealth of data on energy usage. However, the nonlinear, nonstationary, and time-dependent elements of building energy use pose challenges for LSTM algorithms to capture effectively, especially in dynamic contexts with abrupt variations. The objective of this research is to develop an effective Intelligent Long Short-Term Memory-Generative Adversarial Network (ILSTM-GAN) learning model for predicting the consumption of energy of dynamic buildings. To improve the capacity to forecast intricate and variable patterns of energy usage, the research combines LSTM and GAN. The dataset comprises hourly energy usage, weather variables, and building metadata. In preprocessing, missing values were filled, and categorical variables were encoded using one-hot encoding. For feature extraction, the Wavelet Transform (WT) was applied to isolate patterns and trends from noise, thereby improving temporal resolution and feature quality. The proposed ILSTM-GAN model captures long-term temporal dependencies in sequential data and incorporates adversarial training to generate realistic and robust predictions, thereby enhancing the modeling of nonlinear building energy usage. The ILSTM-GAN model achieves performance metrics such as accuracy, precision, recall, and F1-scores above 95–99%. The ILSTM-GAN model shows strong adaptability to temporal fluctuations, facilitating effective energy management and supporting energy efficiency initiatives.