A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Development of a Machine Learning Framework Based on Occupant-Related Parameters to Predict Residential Electricity Consumption in the Hot and Humid Climate
https://orcid.org/0000-0002-9016-9429
Graphical abstract Display Omitted
Highlights The RF regressor and classifier are developed using occupant-related inputs. The time of year is the most significant parameter, with an importance of 49.44%. After feature importance analysis, the ten least effective factors are removed. The RF regressor shows remarkable performance even with 16 inputs.
Abstract Occupant-related variables constitute one of the most significant groups of factors influencing residential building energy consumption. However, prediction methods often oversimplify these parameters, leading to substantial discrepancies between predicted and actual consumption. To address this issue, the present study aims to develop a machine learning framework for predicting electricity consumption in residential buildings based on occupant-related factors. The study incorporates twenty-six inputs, including occupant characteristics such as demographics, occupancy, behavior, and behavioral efficiency, two time-related factors, and three extra parameters related to equipment (refrigerator age, hot water source, and type of electricity meter) for training and testing the Random Forest (RF) algorithm in both regression and classification forms. The results indicate that the trained RF regressor exhibits well performance (R2 Train = 0.989, R2 Test = 0.916, MAETrain = 0.81, MAETest = 2.21, RMSETrain = 1.27, and RMSETest = 3.45). Furthermore, feature importance analysis reveals that the most significant parameter is the time of year, representing weather conditions, followed by the number of occupants, neighborhood, indoor set-point range, mean age of occupants, window opening, and cooling system mode. Even after removing the least impactful factors, the model maintains strong performance with the 16 most important variables (R2 Train = 0.986, R2 Test = 0.910, MAETrain = 0.83, MAETest = 2.25, RMSETrain = 1.31, and RMSETest = 3.49). Additionally, the RF classifier is designed for problems with 2, 4, 6, and 8 classes based on energy consumption ranges. The results of this model demonstrate that the 2-class model achieves the highest performance (AccuracyTest = 0.963, MAETest = 0.04, and RMSETest = 0.19). However, it lacks detailed categorization of homes based on electricity consumption. On the other hand, the 4-class and 6-class models strike a good balance between prediction performance and the level of detail. In conclusion, the proposed method can accurately predict residential electricity consumption and can serve as a valuable reference for researchers and utility managers when formulating energy reduction policies and comparing the effectiveness of different strategies.
Development of a Machine Learning Framework Based on Occupant-Related Parameters to Predict Residential Electricity Consumption in the Hot and Humid Climate
https://orcid.org/0000-0002-9016-9429
Graphical abstract Display Omitted
Highlights The RF regressor and classifier are developed using occupant-related inputs. The time of year is the most significant parameter, with an importance of 49.44%. After feature importance analysis, the ten least effective factors are removed. The RF regressor shows remarkable performance even with 16 inputs.
Abstract Occupant-related variables constitute one of the most significant groups of factors influencing residential building energy consumption. However, prediction methods often oversimplify these parameters, leading to substantial discrepancies between predicted and actual consumption. To address this issue, the present study aims to develop a machine learning framework for predicting electricity consumption in residential buildings based on occupant-related factors. The study incorporates twenty-six inputs, including occupant characteristics such as demographics, occupancy, behavior, and behavioral efficiency, two time-related factors, and three extra parameters related to equipment (refrigerator age, hot water source, and type of electricity meter) for training and testing the Random Forest (RF) algorithm in both regression and classification forms. The results indicate that the trained RF regressor exhibits well performance (R2 Train = 0.989, R2 Test = 0.916, MAETrain = 0.81, MAETest = 2.21, RMSETrain = 1.27, and RMSETest = 3.45). Furthermore, feature importance analysis reveals that the most significant parameter is the time of year, representing weather conditions, followed by the number of occupants, neighborhood, indoor set-point range, mean age of occupants, window opening, and cooling system mode. Even after removing the least impactful factors, the model maintains strong performance with the 16 most important variables (R2 Train = 0.986, R2 Test = 0.910, MAETrain = 0.83, MAETest = 2.25, RMSETrain = 1.31, and RMSETest = 3.49). Additionally, the RF classifier is designed for problems with 2, 4, 6, and 8 classes based on energy consumption ranges. The results of this model demonstrate that the 2-class model achieves the highest performance (AccuracyTest = 0.963, MAETest = 0.04, and RMSETest = 0.19). However, it lacks detailed categorization of homes based on electricity consumption. On the other hand, the 4-class and 6-class models strike a good balance between prediction performance and the level of detail. In conclusion, the proposed method can accurately predict residential electricity consumption and can serve as a valuable reference for researchers and utility managers when formulating energy reduction policies and comparing the effectiveness of different strategies.
Development of a Machine Learning Framework Based on Occupant-Related Parameters to Predict Residential Electricity Consumption in the Hot and Humid Climate
https://orcid.org/0000-0002-9016-9429
Graphical abstract Display Omitted
Highlights The RF regressor and classifier are developed using occupant-related inputs. The time of year is the most significant parameter, with an importance of 49.44%. After feature importance analysis, the ten least effective factors are removed. The RF regressor shows remarkable performance even with 16 inputs.
Abstract Occupant-related variables constitute one of the most significant groups of factors influencing residential building energy consumption. However, prediction methods often oversimplify these parameters, leading to substantial discrepancies between predicted and actual consumption. To address this issue, the present study aims to develop a machine learning framework for predicting electricity consumption in residential buildings based on occupant-related factors. The study incorporates twenty-six inputs, including occupant characteristics such as demographics, occupancy, behavior, and behavioral efficiency, two time-related factors, and three extra parameters related to equipment (refrigerator age, hot water source, and type of electricity meter) for training and testing the Random Forest (RF) algorithm in both regression and classification forms. The results indicate that the trained RF regressor exhibits well performance (R2 Train = 0.989, R2 Test = 0.916, MAETrain = 0.81, MAETest = 2.21, RMSETrain = 1.27, and RMSETest = 3.45). Furthermore, feature importance analysis reveals that the most significant parameter is the time of year, representing weather conditions, followed by the number of occupants, neighborhood, indoor set-point range, mean age of occupants, window opening, and cooling system mode. Even after removing the least impactful factors, the model maintains strong performance with the 16 most important variables (R2 Train = 0.986, R2 Test = 0.910, MAETrain = 0.83, MAETest = 2.25, RMSETrain = 1.31, and RMSETest = 3.49). Additionally, the RF classifier is designed for problems with 2, 4, 6, and 8 classes based on energy consumption ranges. The results of this model demonstrate that the 2-class model achieves the highest performance (AccuracyTest = 0.963, MAETest = 0.04, and RMSETest = 0.19). However, it lacks detailed categorization of homes based on electricity consumption. On the other hand, the 4-class and 6-class models strike a good balance between prediction performance and the level of detail. In conclusion, the proposed method can accurately predict residential electricity consumption and can serve as a valuable reference for researchers and utility managers when formulating energy reduction policies and comparing the effectiveness of different strategies.
Development of a Machine Learning Framework Based on Occupant-Related Parameters to Predict Residential Electricity Consumption in the Hot and Humid Climate
Qavidel Fard, Zahra (author) / Sadat Zomorodian, Zahra (author) / Tahsildoost, Mohammad (author)
Energy and Buildings ; 301
2023-10-24
Article (Journal)
Electronic Resource
English
Development of an Integrated Residential Humid Climate HVAC System
| British Library Conference Proceedings | 2008
Effective Features to Predict Residential Energy Consumption Using Machine Learning
| British Library Conference Proceedings | 2019