Building Energy Prediction Research Articles

Residential building energy consumption estimation: A novel ensemble and hybrid machine learning approach

In recent decades, there has been a substantial rise in both worldwide energy consumption and the accompanying increase in Carbon Dioxide (CO2) emissions, primarily propelled by population growth and the escalating demand for personal comfort. Operational energy consumption in buildings constitutes about 30% of the world's total final energy usage, underscoring the significance of predicting building energy usage for effective energy planning, management, and conservation. This study has enhanced the prediction of heating load (HL) and cooling loads (CL) for residential buildings using novel and dependable machine learning (ML) techniques. The study utilized two base models: Adaptive Boosting (ADA) and Extreme Gradient Boosting (XGBoost). An ensemble consisting of ADA and XGB was constructed to improve the model's performance, aligning with the principles of the Dempster-Shafer theory. To optimize the efficiency of ADA and XGB, five innovative optimizers, namely Victoria Amazonica Optimization (VAO), Giant Trevally Optimizer (GTO), Covariance Matrix Adaptation Evolution Strategy (CMAES), Coyote Optimization Algorithm (COA), and Mountain Gazelle Optimizer (MGO), were integrated. Statistical analysis has been employed to evaluate the performance of the proposed models. The results highlight the effectiveness of CMAES in optimizing the XGB and ADA models for predicting HL and CL. The most accurate result was achieved by the XGCM hybrid model, as evidenced by the impressive total R2 value of 0.9934 in HL and 0.9911 in CL prediction. The experimental findings illustrate that the suggested approach exhibits superior predictive performance across various scenarios of building energy consumption.

Application and evaluation of the evolutionary algorithms combined with conventional neural network to determine the building energy consumption of the residential sector

Residential uses a significant amount of energy; hence, encouraging sustainability and lessening environmental effects requires minimizing energy consumption in this sector. This study focuses on applying and evaluating evolutionary algorithms combined with conventional neural networks to predict building energy consumption in the residential sector. The primary objectives were to assess the performance of three evolutionary algorithms – Heap-Based Optimizer (HBO), Multiverse Optimizer (MVO), and Whale Optimization Algorithm (WOA) – in comparison to each other and to determine their effectiveness in predicting energy consumption. Each algorithm was integrated into the neural network framework to optimize the prediction model. Training and testing datasets were employed to evaluate the performance of the models. Two key statistical indices, Root Mean Square Error (RMSE) and R-squared (R2), were utilized to assess the accuracy of the predictions. The results of the evaluation demonstrated varying performances among the three evolutionary algorithms. MVO achieved the highest scores for both RMSE (48.55082 in training and 68.44517 in testing) and R2 (0.99184 in training and 0.98236 in testing) on both training and testing datasets, indicating superior predictive accuracy compared to HBO and WOA. These findings underscore the importance of algorithm selection in optimizing predictive models for energy consumption forecasting. Further research may explore hybrid approaches or parameter tuning to enhance the performance of evolutionary algorithms in this domain. Overall, this study contributes to advancing energy forecasting techniques, with potential implications for energy management and conservation efforts in the residential sector.

Innovative hybrid prediction method integrating wavelet threshold decomposition and entropy-based model selection strategy for building energy consumption prediction

Building energy prediction plays a pivotal role in the building energy management system, and an efficient prediction method is crucial to ensure the effective operation of building systems. Existing research enhances prediction accuracy by combining decomposition algorithms with prediction methods to separate non-stationary information from the original sequence. However, the decomposition process and the subsequent prediction of multiple decomposed subsequences result in significant time overhead, making it unsuitable for scenarios with high temporal requirements, such as short-term energy consumption forecasting. Additionally, a single prediction model struggles to adapt to the diverse features of decomposed subsequences, creating prediction accuracy and efficiency bottlenecks. Addressing these challenges, this study proposes a hybrid prediction method that combines wavelet threshold decomposition (WTD) with an entropy-based model selection strategy. The method initially utilizes the efficient WTD algorithm to decompose the original energy consumption data into three fixed subsequences. Subsequently, the permutation entropy for each subsequence is computed, and based on the comprehensive performance of four prediction models (ANN, LSTM, CNN-LSTM, and TCN) on sequences with different entropy values, the optimal prediction model is selected. Finally, the prediction results of the subsequences are aggregated to obtain the ultimate prediction outcome. Experimental results demonstrate that the proposed method significantly outperforms baseline methods in both prediction speed and accuracy across five real building energy consumption datasets with distinct operational modes.

Enhancing multi-scenario data-driven energy consumption prediction in campus buildings by selecting appropriate inputs and improving algorithms with attention mechanisms

Effective building energy prediction is vital for sustainable development, especially with an increasing focus on flexibility and elasticity in building energy usage. However, challenges persist in input and algorithm selection including suitable input selection and the balance of calculation time consumption and prediction precision. This study proposed a method for suitable input selection and introduced a CNN-LSTM-SE model for enhancing prediction precision. The CNN-LSTM-SE model integrates a convolutional neural network (CNN) process and a squeeze-and-excitation (SE) block attention mechanism into a long short-term memory (LSTM) neural network. Experiments across seven buildings demonstrate that selecting inputs improves the prediction performance with a 13.94 % decrease in MAPE (mean absolute percent error), a 3.36 % increase in R2 (coefficient of determination), and an 11.20 % decrease in CV-RMSE (coefficient-of-variation of root mean square error) on average compared to not selecting inputs. Moreover, implementing CNN-LSTM-SE yields a 21.44 % MAPE decrease, a 6.25 % R2 increase, and a 21.82 % CV-RMSE decrease compared to alternative algorithms. Simultaneously employing suitable inputs and CNN-LSTM-SE balances time consumption and prediction precision optimally.

BO-STA-LSTM: Building energy prediction based on a Bayesian optimized spatial-temporal attention enhanced LSTM method

In predicting building energy (affected by seasons), there are issues like inefficient hyperparameter optimization and inaccurate predictions, it is unclear whether spatial and temporal attention improves performance. This study proposes a method based on Bayesian Optimization (BO), Spatial-Temporal Attention (STA), and Long Short-Term Memory (LSTM). Seven improved LSTM models (BO-LSTM, SA-LSTM, TA-LSTM, STA-LSTM, BO-SA-LSTM, BO-TA-LSTM, and BO-STA-LSTM) are compared with LSTM and the impacts of seasonal variations on BO-STA-LSTM are analysed using different sample types and time domain analysis. To further demonstrate the efficiency of the proposed method, comparisons with convolutional neural network (CNN) and temporal convolutional network (TCN) are performed, and followed by validation with new datasets. The findings indicate that adding STA and BO to LSTM enhances the average R2 of prediction performance by 0.0885. BO alone contributes 0.0717, while adding attention improves 0.0560. BO-STA-LSTM achieves higher prediction accuracy for similar test and training samples or a test sample size of 14,016, effectively capturing seasonal, trend, and peak energy patterns. Additionally, BO-STA-LSTM outperforms CNN and TCN, demonstrating superior prediction accuracy.

Deep learning for prediction of energy consumption: an applied use case in an office building

Abstract Non-residential buildings are responsible for more than a third of global energy consumption. Estimating building energy consumption is the first step towards identifying inefficiencies and optimizing energy management policies. This paper presents a study of Deep Learning techniques for time series analysis applied to building energy prediction with real environments. We collected multisource sensor data from an actual office building under normal operating conditions, pre-processed them, and performed a comprehensive evaluation of the accuracy of feed-forward and recurrent neural networks to predict energy consumption. The results show that memory-based architectures (LSTMs) perform better than stateless ones (MLPs) even without data aggregation (CNNs), although the lack of ample usable data in this type of problem avoids making the most of recent techniques such as sequence-to-sequence (Seq2Seq). Graphical abstract

Development of a rapid assessment tool for integrating thermal comfort in early design stage of energy-efficient office buildings

Comprehending and predicting energy performance of existing buildings and new constructions is crucial towards decarbonization. Instead of utilizing simulation software, multiple regression models are utilized to predict building energy consumption while ensuring indoor thermal comfort to speed up this process. However, previous predictive models prioritize reducing energy demand, with limited focus on thermal comfort. This study aims to support decision-making during retrofitting and new construction planning though developing a prediction model. An air-conditioned office building served as a reference building for simulation. 21 design parameters were analyzed, including aspects of weather, building envelope, internal loads, ventilation, and temperature settings. Stepwise regression results unveiled the crucial variables in the final model, with 8, 9, 13, and 6 variables remaining for peak cooling load, annual cooling load, overheating hours (WE), and Environmental Quality Index for thermal comfort (EQITC) in the perimeter zones, and 5, 6, 8, and 5 variables for the core zones, respectively. Furthermore, insights into the important variables regarding cooling load and thermal comfort were respectively provided. Weather- and envelope-related variables, such as cooling degree-days, global solar radiation, solar heat gain coefficient (SHGC), and U-value, have the highest impacts on cooling load. For thermal comfort, variables including temperature setpoint, occupant activity level, and factors related to window sunlight transmission performance, such as SHGC, window area ratio, and overhang projection ratio, proved to be influential. Overall, this study provided accurate models for assessing optimal strategies for energy efficiency and thermal comfort during the early design phases, advancing building performance practices.

A novel hybrid model based on modal decomposition and error correction for building energy consumption prediction

Accurate energy consumption prediction is pivotal in supporting efficient grid scheduling and enhancing renewable energy penetration. This paper proposes a novel hybrid prediction model called CEEMDAN-SE-EC-BiLSTM. The proposed model utilizes Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to decompose the original energy consumption data into Intrinsic Mode Functions (IMFs). Afterwards, Sample Entropy (SE) is employed to evaluate the complexity of the IMFs. The complexed IMFs are then transformed into simpler IMFs using a second decomposition. All IMFs are predicted using Bidirectional Long Short-Term Memory (BiLSTM) to attain the initial predicted sequence. To further refine the prediction, an Error Correction (EC) technique is introduced. This involves subtracting the initial predicted sequence from the original sequence to obtain the error sequence, which is then decomposed using CEEMDAN. Subsequently, the decomposed error sequence is predicted using BiLSTM to generate the error prediction sequence. Finally, the predicted results and the error prediction are combined to obtain the final energy forecast sequence. To verify the superior performance of the proposed method, the model is tested on five selected buildings and benchmarked against seven prediction models. The experimental results indicate that the proposed hybrid prediction model achieves the highest accuracy in building energy prediction.

Predicting building energy consumption in urban neighborhoods using machine learning algorithms

Assessing building energy consumption in urban neighborhoods at the early stages of urban planning assists decision-makers in developing detailed urban renewal plans and sustainable development strategies. At the city-level, the use of physical simulation-based urban building energy modeling (UBEM) is too costly, and data-driven approaches often are hampered by a lack of available building energy monitoring data. This paper combines a simulation-based approach with a data-driven approach, using UBEM to provide a dataset for machine learning and deploying the trained model for large-scale urban building energy consumption prediction. Firstly, we collected 18,789 neighborhoods containing 248,938 buildings in the Shanghai central area, of which 2,702 neighborhoods were used for UBEM. Simultaneously, building functions were defined by POI data and land use data. We used 14 impact factors related to land use and building morphology to define each neighborhood. Next, we compared the performance of six ensemble learning methods modeling impact factors with building energy consumption and used SHAP to explain the best model; we also filtered out the features that contributed the most to the model output to reduce the model complexity. Finally, the balanced regressor that had the best prediction accuracy with the minimum number of features was used to predict the remaining urban neighborhoods in the Shanghai central area. The results show that XGBoost achieves the best performance. The balanced regressor, constructed with the 9 most contributing features, predicted the building rooftop photovoltaics potential, total load, cooling load, and heating load with test set accuracies of 0.956, 0.674, 0.608, and 0.762, respectively. Our method offers an 85.5%-time advantage over traditional methods, with only a maximum of 22.75% of error.

Study on the Impact of Building Energy Predictions Considering Weather Errors of Neighboring Weather Stations.

Weather data errors affect energy management by influencing the accuracy of building energy predictions. This study presents a long short-term memory (LSTM) prediction model based on the "Energy Detective" dataset (Shanghai, China) and neighboring weather station data. The study analyzes the errors of different weather data sources (Detective and A) at the same latitude and longitude. Subsequently, it discusses the effects of weather errors from neighboring weather stations (Detective, A, B, C, and D) on energy forecasts for the next hour and day including the selection process for neighboring weather stations. Furthermore, it compares the forecast results for summer and autumn. The findings indicate a correlation between weather errors from neighboring weather stations and energy consumption. The median R-Square for predicting the next hour reached 0.95. The model's predictions for the next day exhibit a higher Prediction Interval Mean Width (139.0 in summer and 146.1 in autumn), indicating a greater uncertainty.

A data-driven framework for fast building energy demand estimation across future climate conditions

Abstract The rapid and precise forecasting of building energy requirements plays a crucial role in decision-making processes for architects during the early design phase. This study introduces a data-driven framework capable of projecting energy demands in the context of evolving climate conditions. We evaluated four widely-used machine learning algorithms. Our results indicated that a hybrid approach, integrating Catboost and Bayesian optimization, excelled in both accuracy and efficiency for predicting building energy demand under climate change conditions. The framework proposed herein has potential applications in fostering sustainability in early-stage architectural design.

Performance assessment of cross office building energy prediction in the same region using the domain adversarial transfer learning strategy

Precise and reliable building energy consumption prediction is of great significance for building energy management and future energy planning. To mitigate the adverse impact of data limitation on the performance of building energy consumption prediction model, a transfer learning strategy was introduced. However, the operation modes of different building energy systems are different, resulting in a large difference in the distribution of building energy consumption related data. In this paper, the relevant data of 25 buildings from the same region of the United States were selected. On this basis, two similarity calculation strategies are compared, and three similarity measurement methods were used to analyze the impact of different source buildings on the performance improvement of transfer learning models. The results show that choosing the appropriate source building can improve the transfer learning performance by up to 82.56% when compared to the no-transfer model (the baseline model). When higher similarity source buildings are selected, the corresponding average PIR is about 24.7% higher than the total average, which can make the transfer model performance improvement more obvious. This paper provides insights into the selection of source domains for transfer learning strategies under the same region.

Statistical and Artificial Intelligence-Based Tools for Building Energy Prediction: A Systematic Literature Review

Statistical and Artificial Intelligence-Based Tools for Building Energy Prediction: A Systematic Literature Review

Effect of different generated weather climate data file on building energy consumption in different climates of Iran

Accurate estimation of building energy consumption is critical, and it can usually be calculated using simulation software like Energy Plus. The input weather data for software is typically a typical or reference year as a standard weather file, such as TMY2. To best estimate building energy consumption, TMY2 data must be generated intelligently. In the present study, first, several TMY2 data files have been generated from different periods (5–21 years) of available measured weather data (9 separate files). Available measured weather data is for 1998 to 2019. Also, a TMY2 file has been generated from the long-term average for the total period (1998–2019), and two TMY2 files have been generated using the CCWeatherGen software for 2020 and 2050. In the next step, the calculated building energy consumption from those files for an educational building will be compared using Energy Plus software. The primary purpose of this study is to investigate the impact of different generated TMY2 (from different weather data periods) on energy consumption and determine the best period to generate TMY2 files from the available period of 5 to 21 years for the capital of Iran (Tehran), which has a semi-arid climate, and Tabriz, which has a cold and mountainous climate. These cities have different climates. Results show that for Tehran, building heating energy consumption changes by about 10% from other generated TMY2 files (one file generated from 2009 to 2019 and another generated from the 2007 to 2019 period). The result shows that generating TMY2 files from different weather data periods affects building energy consumption. So, a suitable strategy for generating TMY2 files is necessary. As a new strategy, the present study proposes that TMY2 files should be generated based on building cooling and heating energy from recent years. Also, because climate change is an important parameter, a TMY2 file should be generated from recent years’ weather data to accurately predict building energy consumption.

Sequential attention deep learning architecture with unsupervised pre-training for interpretable and accurate building energy prediction with limited data

ABSTRACT Predicting building energy consumption is important for energy efficiency and reducing carbon emissions. However, deep learning (DL) models for energy consumption forecasting often have limited predictive generalization and lack explainability. To address these challenges, this research adopts a sequential attention deep learning architecture (SADLA) that uses attention mechanisms to learn the importance of different features for predicting cooling energy consumption, which helps building occupants and managers to make informed decisions about energy optimization. The model was trained and tested on comprehensive and scarce datasets in terms of recording periods and predictor numbers. The data were derived from a three-month field experiment in a commercial office building in Seoul, Korea. Furthermore, models were constructed with other prevalent algorithms like Long Short-Term Memory, deep neural networks, and Extreme Gradient Boosting for comparative assessment. The results from the one- and two-week datasets and reduced features suggested that the SADLA-based models surpassed others in model generalization (R2 = 0.961, 0.967, and 0.976 respectively). However, all algorithms demonstrated comparable performance with the three-month dataset achieving an average R2 of 0.970. These findings underscore the potential of the adopted model in addressing data scarcity in existing and new buildings for accurate cooling energy prediction.

Transferability and robustness of a data-driven model built on a large number of buildings

Data-driven energy prediction models have shown a great importance in building energy management. However, these models require sufficient operational data to ensure prediction accuracy, posing great challenges for buildings with scarce data. Transfer learning has emerged as a key strategy to overcome this issue, enabling cross-building prediction even with limited data availability. While existing studies have mainly focused on a single or a few specific buildings to train models, this study aims to explore the transferability of building energy prediction models across a large number of buildings. Initially, energy consumption data retrieved from 327 buildings were selected as source domain to create a pre-trained model. Different volumes of data in the target domain were then utilized to fine-tune the pre-trained model, and the resulting accuracy distribution and accuracy improvement achieved by transfer learning were examined. The study also evaluated model robustness by conducting transfer procedures at 20 different time nodes. The occurrence of negative transfer was also monitored. The results show that transfer learning can significantly improve prediction accuracy when compared with the baseline model, with a median MAPE value improving from 18.31 % to 7.76 % when using only 7 days data. Meanwhile, transfer learning using 7 days data outperformed direct prediction using 180 days of data. However, negative transfer may occur, although at a low rate, and is not related to data volume. In addition, there is a possibility that a model with high general accuracy yield biased results at certain time nodes. This work provides valuable insights into the advantages and limitations of transfer learning in building energy prediction model, which facilitates the exploitation of existing building data source for advanced data analytics.

At the roots of the energy performance gap: Analysis of monitored indoor air before and after building retrofits

The growing concern about climate change and energy security has fostered energy efficiency measures to reduce building consumption in many European countries. The policies and incentives behind these improvements typically rely on pre-calculated expected energy savings. However, evidence shows that the actual energy performance after such interventions often falls short of the expected targets. People's behavior is one of the causes of the energy performance gap between measured and predicted building energy performance. This study contributes to this discussion by analyzing data on air temperature, relative humidity, and Volatile Organic Compounds monitored in eleven apartments in Milan (Italy) before and after building renovation. These data were then used to simulate two representative flats, thus obtaining their energy demand for space heating. The analysis of the measured data shows that users adapt differently to building retrofits. Under the assumption of constant moisture generation during the periods monitored, some occupants appear to increase air change rates and reduce indoor air temperatures, while others show the opposite behaviour. These trends could be related to the fact that some users prioritize air quality over thermal comfort and vice versa. Energy simulations suggest that the former behaviour leads to a larger energy performance gap.

Filling time-series gaps using image techniques: Multidimensional context autoencoder approach for building energy data imputation

Building energy prediction and management has become increasingly important in recent decades, driven by the growth of Internet of Things (IoT) devices and the availability of more energy data. However, energy data is often collected from multiple sources and can be incomplete or inconsistent, which can hinder accurate predictions and management of energy systems and limit the usefulness of the data for decision-making and research. To address this issue, past studies have focused on imputing missing gaps in energy data, including random and continuous gaps. One of the main challenges in this area is the lack of validation on a benchmark dataset with various building and meter types, making it difficult to accurately evaluate the performance of different imputation methods. Another challenge is the lack of application of state-of-the-art imputation methods for missing gaps in energy data. Contemporary image-inpainting methods, such as Partial Convolution (PConv), have been widely used in the computer vision domain and have demonstrated their effectiveness in dealing with complex missing patterns. Given that energy data often exhibits regular daily or weekly patterns, such methods could be leveraged to exploit the regularity of the data to learn underlying patterns and generate more accurate predictions for missing values. To study whether energy data imputation can benefit from the image-based deep learning method, this study compared PConv, Convolutional neural networks (CNNs), and weekly persistence method using one of the biggest publicly available whole building energy datasets, consisting of 1479 power meters worldwide, as the benchmark. The results show that, compared to the CNN with the raw time series (1D-CNN) and the weekly persistence method, neural network models with reshaped energy data with two dimensions reduced the Mean Squared Error (MSE) by 10% to 30%. The advanced deep learning method, Partial convolution (PConv), has further reduced the MSE by 20%–30% than 2D-CNN and stands out among all models. Based on these results, this study demonstrates the potential applicability of time-series imaging in imputing energy data. The proposed imputation model has also been tested on a benchmark dataset with a range of meter types and sources, demonstrating its generalizability to include additional accessible energy datasets. This offers a scalable and effective solution for filling in missing energy data in both academic and industrial contexts.

Comparative Analysis of Data-Driven Algorithms for Building Energy Planning via Federated Learning

Building energy planning is a challenging task in the current mounting climate change scenario because the sector accounts for a reasonable percentage of global end-use energy consumption, with a one-fifth share of global carbon emissions. Energy planners rely on physical model-based prediction tools to conserve energy and make decisions towards decreasing energy consumption. For precise forecasting, such a model requires the collection of an enormous number of input variables, which is time-consuming because not all the parameters are easily available. Utilities are reluctant to share retrievable consumer information because of growing concerns regarding data leakage and competitive energy markets. Federated learning (FL) provides an effective solution by providing privacy preserving distributed training to relieve the computational burden and security concerns associated with centralized vanilla learning. Therefore, we aimed to comparatively analyze the effectiveness of several data-driven prediction algorithms for learning patterns from data-efficient buildings to predict the hourly consumption of the building sector in centralized and FL setups. The results provided comparable insights for predicting building energy consumption in a distributed setup and for generalizing to diverse clients. Moreover, such research can benefit energy designers by allowing them to use appropriate algorithms via transfer learning on data of similar features and to learn personalized models in meta-learning approaches.

Reducing carbon emissions in the architectural design process via transformer with cross-attention mechanism

IntroductionThe construction industry is one of the world’s largest carbon emitters, accounting for around 40% of total emissions. Therefore, reducing carbon emissions from the construction sector is critical to global climate change mitigation. However, traditional architectural design methods have some limitations, such as difficulty in considering complex interaction relationships and a large amount of architectural data, so machine learning can assist architectural design in improving design efficiency and reducing carbon emissions.MethodsThis study aims to reduce carbon emissions in the architectural design by using a Transformer with a cross-attention mechanism model. We aim to use machine learning methods to generate optimized building designs that reduce carbon emissions during their use and construction. We train the model on the building design dataset and its associated carbon emissions dataset and use a cross-attention mechanism to let the model focus on different aspects of the building design to achieve the desired outcome. We also use predictive modelling to predict energy consumption and carbon emissions to help architects make more sustainable decisions.Results and discussionExperimental results demonstrate that our model can generate optimized building designs to reduce carbon emissions during their use and construction. Our model can also predict a building’s energy consumption and carbon emissions, helping architects make more sustainable decisions. Using Transformers with cross-attention mechanism models to reduce carbon emissions in the building design process can contribute to climate change mitigation. This approach could help architects better account for carbon emissions and energy consumption and produce more sustainable building designs. In addition, the method can also guide future building design and decision-making by predicting building energy consumption and carbon emissions.