The energy performance in buildings is influenced by many factors, such as ambient weather conditions, building structure and characteristics, occupants and their behaviors, operation of sublevel components like heating, ventilation and air-conditioning systems. These complex properties make the prediction, analysis or fault detection/diagnosis of building energy consumption very difficult to perform accurately.

This book focuses on up-to-date data mining and machine-learning methods to solve these problems. This book first presents a review of recently developed models for solving prediction, analysis or fault detection/diagnosis of building energy consumption, including detailed and simplified engineering methods, statistical methods and artificial intelligence methods. Then, the methodology to simulate energy consumption profiles for single and multiple buildings is presented. Based on these datasets, support vector machine (SVM) models are trained and tested to do the prediction. The results from extensive experiments demonstrate high-prediction accuracy and robustness of these models. A recursive deterministic perceptron (RDP) neural network model is then used to detect and diagnose faulty building energy consumption. In the experiment, the RDP model shows a very high-detection ability. A new approach, based on the evaluation of RDP models, is also proposed here to diagnose faults. Since the selection of subsets of features significantly influences the performance of the model, the optimal features are selected based on the feasibility of obtaining them and on the scores they provide under the evaluation of two filter methods. Experimental results confirm the validity of the selected subsets and show that the proposed feature selection method guarantees the model accuracy and reduces the computational time. One challenge of predicting building energy consumption is to accelerate the model training when amounts of data are very large. To address this issue, this book proposes an efficient parallel implementation of SVMs based on the decomposition method. The parallelization is performed on the most time-consuming part of the training. The underlying parallelism is conducted with a shared memory Map-Reduce paradigm, making the system particularly suitable for multicore and multiprocessor systems. Experimental results show that our original implementation offers a high speedup compared to Libsvm, and it is superior to the state-of-the-art message passing interface (MPI) implementation of Pisvm in both computational time and storage requirement.

The aim of this book is to explain and illustrate, using concrete examples, the recent techniques of data mining and machine learning for solving prediction, analysis or fault detection/diagnosis of building energy consumption. We follow a didactic approach and gradually introduce mathematical and computing concepts where appropriate, and whenever the need arises to enhance understanding. This book is intended primarily for Master students in all fields of engineering concerned with building energy analysis. It may also interest any engineer working in the fields of data mining and knowledge discovery for energy management and energy efficiency. Portions of this book have been used, for a number of years by the authors, in seminars, keynotes and lectures on artificial intelligence at Wuhan University of Science and Technology (China), University Paris Saclay (France), Conservatoire National des Arts et Métiers (France), École Centrale des Arts et Manufactures (France), École Supérieure des Sciences et Technologies de l’Ingénieur de Nancy (France), University Duisburg-Essen (Germany), Chuo University (Japan), Doshisha University (Japan), Keio University (Japan) and University of Electro Communications (Japan).

Frédéric MAGOULÈS

Hai-Xiang ZHAO

November 2015