Background: With the development of smart grids, accurate electric load forecasting has become increasingly important as it can help power companies in better load scheduling and reduce excessive electricity production. However, developing and selecting accurate time series models is a challenging task as this requires training several different models for selecting the best amongst them along with substantial feature engineering to derive informative features and finding optimal time lags, a commonly used input features for time series models. Methods: Our approach uses machine learning and a long short-term memory (LSTM)-based neural network with various configurations to construct forecasting models for short to medium term aggregate load forecasting. The research solves above mentioned problems by training several linear and non-linear machine learning algorithms and picking the best as baseline, choosing best features using wrapper and embedded feature selection methods and finally using genetic algorithm (GA) to find optimal time lags and number of layers for LSTM model predictive performance optimization. Results: Using France metropolitan’s electricity consumption data as a case study, obtained results show that LSTM based model has shown high accuracy then machine learning model that is optimized with hyperparameter tuning. Using the best features, optimal lags, layers and training various LSTM configurations further improved forecasting accuracy. Conclusions: A LSTM model using only optimally selected time lagged features captured all the characteristics of complex time series and showed decreased Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for medium to long range forecasting for a wider metropolitan area.

Optimal Deep Learning LSTM Model for Electric Load Forecasting using Feature Selection and Genetic Algorithm: Comparison with Machine Learning Approaches †

Energy consumption and efficiency in buildings: current status and future trends

Predicting electricity consumption for commercial and residential buildings using deep recurrent neural networks

Modeling and forecasting building energy consumption: A review of data-driven techniques

Deep learning for estimating building energy consumption

In the European Union, the building sector is one of the largest energy consumer with about 40% of the final energy consumption. Reducing consumption is also a sociological, technological and scientific matter. New methods have to be devised in order to support building professionals in their effort to optimize designs and to enhance energy performances. Indeed, the research field related to building modelling and energy performances prediction is very productive, involving various scientific domains. 
Among them, one can distinguish physics-related fields, focusing on the resolution of equations simulating building thermal behaviour and mathematics-related ones, consisting in the implementation of prediction model thanks to machine learning techniques. This paper proposes a detailed review and discussion of these works. First, the approaches based on physical (‘‘white box’’) models are reviewed according three-category classification. Then, we present the main machine learning (‘‘black box’’) tools used for prediction of energy consumption, heating/cooling demand, indoor temperature. Eventually, a third approach called hybrid (‘‘grey box’’) method is introduced, which uses both physical and statistical techniques. The paper covers a wide range of research works, giving the base principles of each technique and numerous illustrative examples

/pdf/state-of-the-art-in-building-modelling-and-energy-lwuxvh5pjq.pdf

State of the art in building modelling and energy performances prediction: A review

Automatic reconstruction of 3D building models from scanned 2D floor plans

3D digital modeling, Building Information Modeling (BIM) and numerical simulation are widely recognized as essential components of building design support tools, but require a significant amount of digital data to truly achieve their potential. Currently, they are mostly applied in the design and construction of new buildings but rarely in renovation projects, since few digital data are available for the majority of existing buildings. It is therefore urgent to devise reliable and effective approaches to the generation of 3D digital (BIM) models of existing buildings. This recognition is widely shared and has resulted in a substantial amount of research work and significant innovations in various fields: 3D laser scanning, images processing, etc. With the aim of bringing some significant contribution to this state-of-the-art, this paper provides a critical review of the methods and tools for generating 3D building models from 2D drawings, developing along two complementary lines: a wide-spectrum assessment of 3D generation techniques, and a more focused, in-depth review of 2D drawings-based approaches (from image processing to BIM creation and validation). The review follows a well-defined methodology and builds on the work of more than 100 relevant references. It includes substantial discussions to highlight the strengths, weaknesses and preferential applications of the reviewed research works, and provides a research agenda. The study particularly highlights that the state-of-the-art is fragmented: most research works focus on specific, limited steps of the 3D models generation process, but no solution has yet been able to tackle the whole generation chain. An additional conclusion is that the selection of the most effective approach largely depends on the intended application, and on project-specific constraints. Also, the study highlights that significant benefits could be drawn from combining existing approaches.

Review: reconstruction of 3D building information models from 2D scanned plans

BIM (Building Information Model) applications cover many areas such as quality control, monitoring, structural analysis and space management. BIM is also widely used as a basis for energy simulation as it contains most of the data needed for energy analysis. However, buildings feature multi-physics phenomena in which energy is only one aspect; therefore it is often needed to jointly perform different kinds of simulations and, to properly connect BIM to the related simulation environments and formats. In line with this concern, our research work focuses on the issue of heterogeneous multi-models interoperability for simulation. The aim is to devise an approach and to develop tools that can support and enhance connections between BIM authoring tools and simulation tools, covering various physical dimensions. 
To implement this approach, we propose to rely on some well-proven and focused standards from the Building Smart International consortium, namely the IDM (Information Delivery Manual) and the Model Views Definition (MVD). In this respect, this paper brings two contributions. The first is an in-depth review and discussion of the interoperability issues of the Architecture, Engineering and Construction (AEC) area, in particular since the advent of BIM technologies. The second is an account of the first results of our research work, focusing on the definition and implementation of a well-structured and standards-compliant approach to the generation of energy simulation models from BIM IFC (Industry Foundation Classes) models. This work entailed defining a mapping between the building elements as defined in the targeted simulation engine (the simulation engine of the French thermal regulation), and the corresponding IFC elements and attributes. As the IFC do not cover all the required elements for energy simulation, further enrichment of the IFC – either through ad hoc extension or through connection to complementary data sources – was defined and implemented.

A standardized approach to BIM and energy simulation connection

In order to popularize the use of building simulation during the design process, a possible solution is to identify and integrate the different components of a simulation system - such as a geometry converter and a calculation engine in a generalist Building Information Model (BIM) software platform. With this in mind, this paper presents a research work aiming at integrating a Building Energy Performance Simulation (BEPS) service in a versatile serviceoriented BIM platform. It particularly focuses on the methodology used to obtain a complete Building Simulation Model (BSM) ready for calculation from an IFC (Industry Foundation Classes)-based BIM.

Sylvain Robert

Papers

State of the art in building modelling and energy performances prediction: A review

Automatic reconstruction of 3D building models from scanned 2D floor plans

Review: reconstruction of 3D building information models from 2D scanned plans

A standardized approach to BIM and energy simulation connection

Towards a bim-based service oriented platform: application to building energy performance simulation