AI-Driven Building Energy Prediction and Multi-Objective Scheduling Using LSTM-GRU-MPC Integration

Abstract

With the increasing energy consumption of buildings, traditional energy management methods are challenging to cope with complex real-time demand and supply changes. Artificial intelligence-based prediction and optimal scheduling methods are urgently needed to improve energy efficiency and reduce energy consumption. The primary goal is to enhance the energy efficiency of buildings through intelligent prediction and scheduling techniques. To achieve this, we propose the use of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, which are well-suited for capturing the temporal dependencies inherent in energy consumption data. Secondly, in the aspect of energy optimal scheduling, a multi-objective optimal scheduling method is designed by combining model predictive control (MPC) and artificial intelligence technology, and the strategy is adjusted and optimized according to dynamic constraints. This paper integrates big data and Internet of Things technology into the building energy management system to further improve management efficiency. It uses the collaborative optimization of edge computing and cloud computing to achieve the combination of real-time scheduling and long-term planning. The temperature, humidity, and energy consumption data show a certain regularity in this building energy consumption prediction analysis. For example, the average energy consumption within the building is 23.7 kWh, and the maximum energy consumption value is 91.2 kWh, showing an apparent fluctuation in the building's energy demand. LSTM achieved a Mean Absolute Error (MAE) of 12.3 kWh, outperforming linear regression (MAE = 25.6 kWh) and GRU (MAE = 13.8 kWh) in long-term temporal dependency capture. The MPC-based scheduling reduced peak energy consumption by 13.5% (from 91.2 kWh to 78.9 kWh) and total energy cost by 18%, while maintaining an 85% thermal comfort satisfaction rate.