A Procedure for Automating Energy Analyses in the BIM Context Exploiting Artificial Neural Networks and Transfer Learning Technique

Abstract

One of the main benefits of Building Information Modelling is the capability of improving the decision-making process thanks performing what-if tests on digital twins of the building to be realized. Pairing BIM models to Building Energy Models allows designers to determine in advance the energy consumption of the building, improving sustainability of the construction. The challenge is to consider as many elements involved in the energy balance as possible and shuffling their parameters within a certain range. In this work, the automatic creation of a relevant set of design options to be analyzed for searching the optimum has been carried out. Firstly, the usual workflow that would be applied manually has been automatically followed by running scripts and codes, depending just on the initial setup given by the user. Although the procedure is very resource consuming, the main advancement relies in the reduction of the manual intervention and the possibility of creating large datasets of design options, avoiding gross errors. Secondly, Artificial Neural Networks and Transfer Learning techniques are applied to speed up the process of dataset creation. With such approach, the same dataset has been created, with about 30% of initial data and without significant loss of accuracy.

Transfer Learning for Day-Ahead Load Forecasting: A Case Study on European National Electricity Demand Time Series

“…Additionally, depending on the nature of the use case [69,70], MLPs can be preferred over a recurrent neural network, including LSTMs. Considering these findings, alongside the numerous applications of MLP-based TL in the energy sector [57,[71][72][73], MLPs clearly demonstrate their usefulness as a base model within the contemporary TL landscape.…”

Relationship between digital twin and building information modeling: a systematic review and future directions

“…Thus, Bastos Porsani et al (2021) proposed a semi-automated workflow from BIM to BEM that could improve the design process. In addition, Demianenko and De Gaetani (2021) proposed a system for simulating renovation scenarios that incorporated BIM and artificial neural networks-based models to predict the total energy consumption, life cycle cost and LCA.…”

“…In addition, Demianenko and De Gaetani (2021) proposed a system for simulating renovation scenarios that incorporated BIM and artificial neural networks (ANN)-based models to predict the total energy consumption, life cycle cost, and life cycle assessment.…”

Online transfer learning strategy for enhancing the scalability and deployment of deep reinforcement learning control in smart buildings

“…However, reusing previous knowledge from various sources could be beneficial in the context of smart buildings. Therefore, in recent years TL was implemented in smart buildings in the context of load prediction [39,40,41], occupancy detection and activity recognition [42,43], building dynamics [44,45,46] and system control [47,48].…”