Ensemble Learning Model-Based Test Workbench for the Optimization of Building Energy Performance and Occupant Comfort

Abstract

Buildings consume tremendous energy for the improvement of living and working conditions. Control of daylight-artificial light has the potential to improve energy performance and occupant comfort in buildings. This research proposes an intelligent generalized ensemble learning technique to develop a novel control strategy for Venetian-blind positioning (up-down movement with static slat angle of 45 • ) of different window orientations. The proposed model helps to maintain occupant comfort and energy saving in a commercial building. The performance of the ensemble learning approach compared against Gaussian process regression, support vector regression and artificial neural network using conventional statistical indicators. Finally, the proposed data-driven model implemented in a real-time Labview-myRIO platform for the experimental validation. The data-driven model is compared with the baseline model and with the uncontrolled blind condition in terms of daylight glare, and energy consumption of lighting and air-conditioning system in the building. The data-driven model is derived using two years of data collected from a fuzzy-based daylight-artificial light integrated scheme. The blind position providing reduced energy consumption and daylight glare along with setpoint illuminance and temperature are validated. A high dynamic range image with EVALGLARE software used to verify the visual comfort based on daylight glare probability. While evaluating the overall energy savings, the ensemble learning model consumes 17% less power than the uncontrolled system and 15% less power than the baseline system. Here, though we are not controlling the air-conditioning system, the experimental validation confirmed that the air-conditioning system significantly reduces its energy consumption.INDEX TERMS Window blind control, data-driven models, ensemble learning, bayesian optimization, daylight glare, labview, myRIO, energy comparison, lighting control, air-conditioning.

A state-of-the-art artificial intelligent techniques in daylighting controller: models and performance

“…This section tries to provide the answer regarding research queries RQ2 and RQ3. Model-based adaptive predictivebased control systems for shading devices using real-time data and robust control methods are becoming increasingly popular to improve occupant visual/thermal comfort and energy economy (Gunay et al, 2017;Huchuk et al, 2016;Sanjeev Kumar et al, 2020b). Xiong and Tzempelikos (2016) and Shen and Tzempelikos (2017) proposed a model-based predictive control for shading devices with real-time data and demonstrated robust control schemes with the potential to improve the indoor environment while reducing lighting energy consumption.…”

“…Ayoub (2020) present a detailed review of the literature on machine-learning approaches to predict daylighting performance inside buildings, with an emphasis on prediction, various algorithms, data representation, and evaluation metrics. Regression and classification-based predictive models prominently used in the literature are based on algorithms like ANN, Support Vector Machine (SVM), Decision Tree (DT), Ensemble Modest electricity energy reductions of 0.9 kWh/m 2 only with adaptive lighting controls and 1.2 kWh/m 2 with both adaptive lighting and blind controls Sanjeev Kumar et al (2020b) Real-time data driven predictive models for thermal and visual comfort. Luminaire dimming and ensemble learning based window blind position prediction; Daylight glare assessment models using SVM, decision tree, ensemble tree, and ANN The model consumes 17% less power than the uncontrolled system and 15% less power than the baseline system…”