Building Energy Information: Demand and Consumption Prediction with Machine Learning Models for Sustainable and Smart Cities

. The building energy consumption plays an important role in the urban sustainability. The prediction of the energy demand is also of particular importance for developing smart cities and urban planning. Machine learning has recently contributed in the advancement of meth-ods and technologies to predict demand and consumption for building energy systems. This paper presents a state of the art of machine learning models and evaluates the performance of these models. Through a systematic review and a comprehensive taxonomy the advances of machine learning are carefully investigated and promising models are introduced


Introduction
The energy is one of the essential aspects of smart cities [1].The sustainability factor of urban development is a direct function of energy production and consumption of every city [2].The energy consumption of buildings is responsible for a great amount of energy used in an urban settlements [3].From this perspective the prediction of demand and consumption is essential in development of smart cities of the future [4].Machine learning [5][6][7][8][9] has recently well contributed in advancing the accurate and reliable prediction models [4][5][6][7][8][9].Figure 1 shows the exponential increase in using machine learning models in this realm within the past decade.The contribution of this paper is to investigate the application of novel machine learning models in shaping the future of smart and sustainable cities in terms of energy.
Fig. 1. rapid increase of using ML models in various scientist domains

Building energy demand prediction
Prediction of demand in building energy sector is essential for planning and managing energy systems.Table 1 presents top six studied developed by ML methods in building energy demand prediction.[10] To employ machine learning for Quantifying the effect of landscape composition and configuration on urban land surface temperatures

Generalized boosted regression (GBR)
-Land surface temperature -Machine learning [11] To present a comprehensive review about the application of machine learning as a solution in smart buildings ANN, DL, SVM, GA and SVR -Internet of Things -Machine learning [12] To compare different forecasting models for estimating the natural gas demand empirical models, RNN and LR -Machine learning -Deep learning [13] To employ different machine learning methods for Building Performance Simulation ANN, LSTM -Deep learning -Machine learning [14] To present a comprehensive state of the art of machine leaning methods for the prediction of building energy demand ANN and SVM based machine learning methods -Machine leaning -Building energy demand [15] To present a hybrid ensemble method to increase the accuracy of load demand estimation of PV for building energy sector Single and hybrid machine learning methods -Machine learning -Ensemble method Osborne and Alvares-Sanches [10] developed an innovative approach in the presence of machine learning technique (GBR) for Quantifying the effect of landscape composition and configuration on urban land surface temperatures.Based on findings of the study, GBR could successfully predict land surface temperatures with a high correlation coefficient (0.956) using 102,935 data.
Djenouri et al. [11] presented a comprehensive state of the art about the application of ML methods in buildings.ML has been employed as solution for occupants and energy or devices.ML can be used as a multi-disciplinary solution for building purposes but in general the type and the size of the building are main factors for considering the effectiveness of ML method.But the use of ML methods can be more successful in energy demand purposes in buildings by developing innovative approaches.Hribar et al. [12] developed a study for evaluating the forecasting capabilities of different methods including empirical, deep learning and LR models for the prediction of natural gas demand in the presence of daily and hourly datasets.All the methods have been employed in their single form.Evaluations have been performed by the use of MAE and MAPE factors.Based on results deep learning method have the best performance (with MAE 1.06 and 18.3 for hourly and daily datasets, respectively) compared with that of the other techniques.
Singaravel et al. [13] employed ANN while comparing with single, two and three layer LSTM method for the estimation of building sustainability.Machine learning methods can play an important role in reducing the processing time and increasing the sustainability by increasing the model accuracy.Comparing the accuracy values of models for the prediction have been performed by employing determination coefficient values.Based on results all of methods could successfully cope with the prediction task but two layered LSTM method have the best performance compared with others.Ahmad et al. [14] provided a comprehensive state of the art of the machine learning based prediction models for the estimation of building energy demand sector.In general methods have been divided into two main categories including ANN and SVM based machine learning methods as the most frequently used methods in this field of science.This paper also indicated the importance of machine learning methods in the sustainability of buildings energy demand.SVM based methods provided a higher accuracy compared with that of the ANN based methods.
Reza et al. [15] developed a novel hybrid ensemble method including neural ensemble, Bayesian model and wavelet transform method for the prediction of PV performance in the building energy demand sector.This method has been developed by comparing different single and hybrid machine learning techniques in the term of the normalized root mean square error.This study wants to emphasize on the importance of the hybrid methods over the single methods.The proposed hybrid method could successfully estimate the demand forecasting factors and increased the accuracy of the model significantly.
Table 2 present a brief comparison about the accuracy, reliability and sustainability of methods developed for forecasting the energy demand in building sector.Accuracy factor has been generated from the performance factors related to the training step and reliability has been generated from the performance factors related to the testing step.But, sustainability was a little difference and has been generated by comparing reliability, accuracy, processing time and other factors which have been considered by results of the reviewed articles.

Building energy consumption perdition
Building energy consumption is important as much as the importance of building energy demand.Prediction of energy consumption in building energy sector can be one of the main steps for reaching the sustainable buildings and is essential for planning and managing of energy systems.Table 3 presents top six studied developed by ML methods in building energy consumption prediction.

Reference
Contribution ML method keywords [16] To present a robust artificial neural network to explore complex building energy consumption data which have been generated from the simulation-Based Multi-Objective Optimization model ANN -Energy consumption -Machine learning [17] To develop an accurate machine learning method for energy prediction in buildings using data generated from internet of things technology MLP, LR,RF, SVM and GBM -Internet of things -Machine learning [18] To develop a long short term memory (LSTM) network to predict the energy consumers' behaviour based on their recent energy consumptions LSTM -Machine learning; -Smart grid [19] To develop an innovative hybrid deep learning method for the prediction of energy consumption in buildings Hybrid LSTM-GA -Deep learning -Machine learning [20] To develop a comprehensive survey about different machine learning methods developed for the prediction of energy consumption in buildings Single, hybrid and ensemble machine learning methods -Data mining -Machine learning [21] In order to develop machine learning methods for the prediction of energy load in building sectors.

SVM and NARX-RNN -Machine learning -Deep learning
Sharif and Hammad [16] developed a robust ANN method to explore complex building energy consumption data which have been generated from the simulation-Based Multi-Objective Optimization model.In fact, this study focuses on developing an accurate prediction method for energy consumption of buildings.Evaluating of results indicated that the developed ANN method benefits less time consuming as well as a high accuracy which increases the sustainability of the developed method.
Chammas et al. [17] developed a study for the prediction of the energy consumption in buildings using data generated from the IOT technology embedded in buildings.The proposed method is a prediction model based on MLP while comparing with LR, SVM, GBM and random forest.Methods have been compared in terms of determination coefficient, MAPE and RMSE.Dataset for training process was separated into three categories (no light, no date and weather only) for finding the effective variables on modelling process.Based on results, eliminating lights data have an importance effect on increasing the accuracy of the target model.The developed MLP model have a higher determination coefficient and lower RMSE and MAPE compared with that for other methods.
Fenza et al. [18] developed a LSTM method for the prediction of consumers' behaviours in the term of energy consumption.Time series data have been employed in order to develop the target network.Results have been evaluated using RMSE factor.Based on results, the proposed method has successfully cop with the task as well as providing the required sustainability for prediction phase.Almalaq and Zhang [19] developed an innovative prediction model for the estimation of energy consumption in buildings using LSTM and optimizing its parameters by GA methodology to take an evolutionary DL method.The evaluation phase for this study has been performed by the use of datasets related to residential and commercial buildings.Results indicated that the hybrid methods which take an evolutionary DL method presents an accurate and sustainable method for the prediction of energy consumption in buildings over the common DL methods.
Chou and Tran [20] developed a comprehensive survey for studying different machine learning techniques developed for the prediction of energy consumption in building sectors.Methods have been categorized into three main categories including single, hybrid and ensemble machine learning methods.Methods have been compared in terms of performance factors and sustainability index.Results indicated that in case of using single and ensemble methods, ANN based methods have the best prediction performance but in case of using hybrid methods, SVM based methods could present the best performance.In general hybrid methods are the proposed method from the view point of accuracy and sustainability.
Koschwitz et al. [21] developed predictive models in order to estimate the building energy load.The target models include the RBF based SVM and Nonlinear Autoregressive Exogenous Recurrent Neural Networks which have been developed by the historical data from residential buildings in Germany.Based on results, NARX-RNN provided a higher performance and sustainability in comparison with those for the SVM method.
Table 4 present a brief comparison about the accuracy, reliability and sustainability of methods developed for forecasting the energy consumption in building sector.

Discussions
Please here summaries what ML from above methods have been used most.And which ones are popular.And what is the trend.

Conclusion
This paper concludes that usage of machine learning in building energy information applications will be growing with even a higher rate that we've seen during the last decade.The ensemble and hybrid models have emerged and continue to advance for higher accuracy and better performance.Deep learning models also will bring tremendous amount of intelligence for better prediction models.

Table 4 .
the comparison results of methods for energy consumption in building sector