Application of deep learning algorithms driven by multi-source building environment data in green building design

The increasing demand for precise prediction of indoor comfort and energy efficiency in the design of green buildings with the help of multi-source environmental data. Traditional techniques are most often founded on small amounts of data and linear formulas that preclude them the ability to depict dynamic interactions of complex dynamics. To overcome these deficiencies, a deep learning (DL)-based model, A Greylag Goose-tuned Bidirectional Attention-based Convolutional Network (GGO-BA-CNN) is proposed and it integrates Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory networks (Bi-LSTM), and attention mechanism, the hyperparameters of which are optimized using Greylag Goose Optimization. The model is trained with 2400 samples of the indoor conditions, outdoor weather, energy consumption, and occupancy factors, where preprocessing is performed at removing noise, filling missing values, and normalizing, and the Fast Fourier Transform (FFT) is used to extract features and correlation analysis. The experimental findings have better performance with Coefficient of Determination (R 2 ), Mean Absolute Error (MAE), and Mean Squared Error (MSE) of 0.98, 0.008, and 0.049, respectively, and a classification accuracy of 88.77% in comfort categorization. Such findings indicate that the proposed framework is appropriate to model the compound spatial-temporal interactions and form adequate predictions of building energy usage and indoor comfort. The study also reveals the prospects of integrating multi-source data with DL to smart, dynamic, and sustainable green building design.