Energy Efficiency Prediction in Commercial High-Rise Buildings Using CLPSO-Optimized Multi-Output BiLSTM Neural Network

Abstract

Rapid growth of commercial high-rise construction in the Delhi/NCR composite climate has intensified concerns about operational energy consumption, particularly for heating, ventilation, and air-conditioning (HVAC) systems. In such buildings, envelope-related parameters---including geometry, orientation, glazing, and roof/wall characteristics---strongly influence heating and cooling loads, directly affecting electricity demand and carbon emissions. This study proposes a data-driven framework that combines a Comprehensive Learning Particle Swarm Optimizer (CLPSO) with a multi-output Bidirectional Long Short-Term Memory (BiLSTM) neural network to accurately predict heating and cooling loads from building envelope features.The framework follows a complete pipeline: data cleaning, exploratory analysis, feature scaling and selection, outlier detection, CLPSO-based hyperparameter optimization, BiLSTM training, and comparison with established machine-learning algorithms. Using the public "Energy Efficiency" dataset (768 samples, eight input features, and two targets), the proposed CLPSO-BiLSTM model achieves a mean squared error (MSE) of 3.83 and coefficient of determination (R²) of 0.9633 for heating load, and an MSE of 6.99 with R² of 0.9245 for cooling load[1]. These results substantially outperform linear regression, decision tree, random forest, ridge, and lasso regression baselines evaluated under identical conditions.The findings demonstrate that CLPSO-optimized deep recurrent architectures provide robust generalization on multimodal, nonlinear response surfaces typical of building-energy problems, and they support early-stage design decision-making for energy-efficient commercial high-rise buildings in composite climates such as Delhi/NCR.