Data driven model improved by multi-objective optimisation for prediction of building energy loads
Introduction
There have been several approaches proposed to enhance the energy efficiency of buildings in many countries in recent decades. For instance in Europe, it was estimated in 2010 that 60 billion Euros could be saved annually by improving EU buildings' energy performance by 20% [1].
Every attempt to optimise the energy performance of buildings involves a series of calculations to estimate the energy consumption and create an index, such as an ‘energy performance indicator’ or ‘use intensity’ from the measured data [2,3]. Most prevailing optimisation methods are simulation-based where the energy-related objectives (i.e. energy consumption or gas emissions) are calculated by a Building Performance Simulation (BPS) tool such as EnergyPlus, TRNSYS and ESP-r. This approach restricts the computing complexity of the algorithms to BPSs' calculation time. As such, when a vast range of solutions are defined, the calculation and optimisation process may become extremely costly and cumbersome [4]. For this reason, most of the studies which focused on decision making for energy performance improvement of buildings either investigated basic and simple optimisation models or targeted retrofitting only one or two parts of envelopes to pare-down total calculation time and cost. It should also be noted that the majority of studies targeted residential buildings, and there are only a few examples of research related to tertiary buildings. A key component of achieving global development and meeting climate change mitigation targets is the optimisation of the entire building stock. This process requires significant testing and planning to deliver.
With the tremendous growth in the amount of valid and attainable datasets of buildings and collection of Big Data from smart buildings, there is an increasing interest in the employment of Artificial Intelligent (AI) methods specifically Machine Learning (ML) techniques for analysing, modelling, and predicting building data [4,5].
The precision and suitability of the data and the relationships inferred from it become a critical fact in the successful application of ML models. As ML methods build s model over a historical dataset, the main and most important step for having accurate predictions is the extraction of relevant features. Depending on the nature of predictions (the energy indicator and forecasting period), this variable could include simple basic weather indices (e.g. temperature and humidity) or complex building characteristics and climate parameters. Previous research has demonstrated that rather than feature extraction the process of tuning a model itself not only increases the predictive accuracy but also reduces model complexity, ease of use, and consistency of predictions [6]. It has been argued that considering occupancy in retrofit decision-making, particularly in populated real estate properties such as higher education buildings, could leverage energy efficiency [7].
ML techniques have been widely used for modelling building energy loads and performance. Traditionally, the default values for hyper-parameters have been used in this field. However, in recent years researchers have started to tune the ML models to have more accurate predictions of energy metrics [[8], [9], [10], [11]]. Tuning ML model hyper-parameters using a grid search can be time-consuming when a complex method is chosen such as Artificial Neural Networks (ANN) [13] or models based on decision trees.
When MLs are utilised for forecasting multiple measures such as heating and cooling loads, models need to be optimised for both the targets [12,13]. This procedure, in turn, increases the time required for processing and improves the usability of MLs.
In the proposed method, evolutionary-based multi-objective optimisation (MOO) algorithm was employed to smartly explore the ML model's configuration parameters space and suggest a set of packages for maximising ML accuracy for both heating and cooling load predictions. This study applied a Random-Forest (RF) model because a python implementation is capable of providing the multivariate forecasting.
Section 2 provides an overview of the preceding studies with regards to tuning ML models with the purpose of building energy indicators forecasting. Afterwards, the RF method and the studied dataset are described in Section 3. Section 4 presents the results of the proposed ML optimisation approach. The final section provides detailed discussions and recommendations for future work.
Section snippets
Background and motivations
Machine learning algorithms are categorised into two groups: supervised learning, in which the data is labelled, and unsupervised learning, where there is no target for the records in the dataset. Supervised learning is a regression analysis or a set of classifications linking inputs factors (X) to single or multiple “output” variables (Y). Whereas, in unsupervised learning, data is organised into clusters by pulling out similarities between various samples within the dataset. As such,
Methodology
ML models work as black boxes, meaning that the detailed relations of energy performance and building characteristics and weather data are not provided. As mentioned earlier, the initial phase of data-driven modelling is the extraction of a feature set for representing the energy system. Surrogate methods model a system with fewer features than engineering approaches. However, formulating a logical set of variables for these models is both essential and laborious, particularly when modelling
Results
This study used Python programming language and packages for implementing the proposed algorithms. The study used a PC with Intel Core i7–6700 3.4GHz CPU, 32GB RAM (with no utilisation of GPU processing) for running the experiments.
Using conventional Grid search method requires further investigation to decide the topmost hyper-parameters for the ML model. Besides, the existing solutions are not developed to calculate the accuracy of predicting multiple targets. Hence, a custom function is
Conclusion
This research addresses the issues regarding inaccurate modelling of building energy loads using ML techniques. As mentioned in the reviewed literature, most research studies used MLs without model optimisations, and they proposed to model each energy metric, such as heating and cooling loads separately. The latest attempt to enhance the performance of those data-driven models included exhaustive exploration of variable parameters to choose the best performing model. This paper has proposed a
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
The research presented in this paper was co-funded by The Data Lab (Edinburgh, UK) and arbnco Ltd. (Glasgow, UK), through DataLab SFC Earmarked Grant Agreement: PO DL 00033. This work would also not be feasible without the generous PhD funding for the first author, which was cofunded by the Engineering The Future scheme from University of University of Strathclyde and the Industry Funded Studentship Agreement with arbnco Ltd. (Studentship Agreement Number: S170392-101).
References (57)
- et al.
A review on sustainable construction management strategies for monitoring, diagnosing, and retrofitting the building’s dynamic energy performance: focused on the operation and maintenance phase
Appl. Energy
(2015) - et al.
Analysis of an energy efficient building design through data mining approach
Autom. Constr.
(2011) - et al.
Tuning machine learning models for prediction of building energy loads
Sustain. Cities Soc.
(2019) - et al.
Trees vs neurons: comparison between random forest and ANN for high-resolution prediction of building energy consumption
Energy and Buildings
(2017) - et al.
Forecasting energy consumption of multi-family residential buildings using support vector regression: investigating the impact of temporal and spatial monitoring granularity on performance accuracy
Appl. Energy
(2014) - et al.
Short-term load forecasting in a non-residential building contrasting models and attributes
Energy and Buildings
(2015) - et al.
Artificial neural networks for the prediction of the energy consumption of a passive solar building
Energy
(2000) - et al.
Prediction of energy demands using neural network with model identification by global optimization
Energy Convers. Manag.
(2009) - et al.
Pseudo dynamic transitional modeling of building heating energy demand using artificial neural network
Energy and Buildings
(2014) - et al.
Forecasting diurnal cooling energy load for institutional buildings using Artificial Neural Networks
Energy and Buildings
(2016)
A prediction model based on neural networks for the energy consumption of a bioclimatic building
Energy and Buildings
Hourly prediction of a building’s electricity consumption using case-based reasoning, artificial neural networks and principal component analysis
Energy & Buildings
Building’s electricity consumption prediction using optimized artificial neural networks and principal component analysis
Energy and Buildings
Comparison between detailed model simulation and artificial neural network for forecasting building energy consumption
Energy and Buildings
The prediction of heating energy consumption in a model house by using artificial neural networks in Denizli-Turkey
Adv. Eng. Softw.
Modeling of the energy demand of the residential sector in the United States using regression models and artificial neural networks
Appl. Energy
The gap between predicted and measured energy performance of buildings: a framework for investigation
Autom. Constr.
Artificial neural networks for energy analysis of office buildings with daylighting
Appl. Energy
Artificial neural networks to predict energy performance and retrofit scenarios for any member of a building category: a novel approach
Energy
Applying support vector machines to predict building energy consumption in tropical region
Energy and Buildings
Applying support vector machine to predict hourly cooling load in the building
Appl. Energy
Predicting hourly cooling load in the building: a comparison of support vector machine and different artificial neural networks
Energy Convers. Manag.
Prediction of building energy consumption using an improved real coded genetic algorithm based least squares support vector machine approach
Short-term prediction of electric demand in building sector via hybrid support vector regression
Appl. Energy
Accurate quantitative estimation of energy performance of residential buildings using statistical machine learning tools
Energy and Buildings
Predictive modeling for US commercial building energy use: a comparison of existing statistical and machine learning algorithms using CBECS microdata
Energy and Buildings
Random Forest based hourly building energy prediction
Energy and Buildings
Methods for benchmarking building energy consumption against its past or intended performance: an overview
Appl. Energy
Cited by (59)
Review of the building energy performance gap from simulation and building lifecycle perspectives: Magnitude, causes and solutions
2024, Developments in the Built EnvironmentBuilding energy loads prediction using bayesian-based metaheuristic optimized-explainable tree-based model
2023, Case Studies in Construction MaterialsData-driven prediction and optimization toward net-zero and positive-energy buildings: A systematic review
2023, Building and EnvironmentApplication of the hybrid neural network model for energy consumption prediction of office buildings
2023, Journal of Building Engineering