The Evolving Role of Building Automation and Management Systems (BAMSs)
In theory, building automation and management systems (BAMSs) can provide all the components and functionalities required for analyzing and operating buildings. However, in reality, these systems can only ensure the control of heating, ventilation, and air conditioning (HVAC) systems. Many other tasks are left to the operator, such as evaluating buildings’ performance, detecting abnormal energy consumption, identifying changes needed to improve efficiency, ensuring the security and privacy of end-users, and more.
To address these limitations, there has been a movement toward developing artificial intelligence (AI) and big data analytic tools, as they offer various new and tailor-made solutions that are incredibly appropriate for practical buildings’ management. Typically, they can help the operator in (i) analyzing the tons of connected equipment data; and (ii) making intelligent, efficient, and on-time decisions to improve the buildings’ performance.
This paper presents a comprehensive systematic survey on using AI-big data analytics in BAMSs. It covers various AI-based tasks, such as load forecasting, water management, indoor environmental quality (IEQ) monitoring, occupancy detection, and more. The first part of this paper adopts a well-designed taxonomy to overview existing frameworks. A comprehensive review is conducted about different aspects, including the learning process, building environment, computing platforms, and application scenario.
The second part aims at providing the reader with insights into the real-world application of AI-big data analytics. Thus, three case studies that demonstrate the use of AI-big data analytics in BAMSs are presented, focusing on energy anomaly detection in residential and office buildings and energy and performance optimization in sports facilities. Lastly, future directions and valuable recommendations are identified to improve the performance and reliability of BAMSs in intelligent buildings.
The Growing Importance of AI-Big Data Analytics in BAMSs
Building automation and management systems (BAMSs) are intelligent systems of both hardware and software, connecting HVAC systems, lighting, security, and other systems to communicate on a single platform. BAMSs deliver crucial information to operators and/or users on the operational performance of buildings, which aim at promoting energy efficiency and optimizing water consumption, enhancing the safety and comfort of the occupants, reducing maintenance costs, extending the life cycle of the utilities, and more.
This is possible by networking a plethora of sensors and components responsible for the monitoring and operation of mechanical, security, fire, lighting, HVAC, and humidity control and ventilation systems. With the broad utilization of information and communication technologies (ICTs), sensing and measurement technologies along with cloud computing, big data storage, and data analytics, conventional BAMSs are being revolutionized.
Vast quantities of building automation and management data are produced, gathered, and saved. This has offered an excellent opportunity for implementing big data mining and analysis in BAMSs. In this context, as the quantity of data collected in BAMSs is enormous, the “big data” phenomena is surfacing this field and revolutionizing the way we manage data by using AI-big data analytics tools.
Accordingly, with advanced sensing and metering technologies in BAMSs, data split into multiple modalities and many variables can create a comprehensive source of information to analyze. This allows for more targeted analysis, but also means that more powerful, intelligent, and sophisticated tools are needed to identify the most enormous patterns/variables.
As a consequence, the big data analytics market in the building energy sector is expected to grow at a compound annual growth rate (CAGR) of 11.28%, during the forecast period, 2021–2026. Data collection in the building industry is becoming all-embracing. This wealth of big data allows informed data-driven decision-making by designers, facilities managers, and owners during building design, operation, and retrofit.
On the other hand, for existing or outdated buildings to make full use of the services offered by the flourishing data analytics market, the necessary enhancement to the existing system for deploying the new technology must be addressed and sorted out. The main challenges of gathering and analyzing data of old buildings are the outdated technologies and the conventional error-prone data collection means.
Nevertheless, data analytics can assist in designing and implementing the new system adaptation and existing system renovation. Besides, it is of utmost importance to know the current state of AI-based building automation before presenting the actual study concerning user input, demand, response, energy-saving, and automation.
In this respect, it is obvious that AI adds new dimensions to building automation environments by enabling autonomous data analysis for operation optimization. Therefore, many AI-based contributions have recently emerged as key solutions for (i) predicting building occupancy, (ii) forecasting thermal comfort, (iii) boosting energy saving, and (iv) enabling demand-side response.
Additionally, as mentioned in previous studies, people can spend up to 90 percent of their lives in buildings; this highlights the importance of user input, behavioral data, and behavioral analytics for optimizing and automating building operations. To that end, a significant research effort is ongoing to develop AI-based behavioral change technologies to promote energy saving in residential and office buildings, understand consumers’ demand patterns for successful demand response development, optimize occupants’ thermal comfort, transform water management, improve fault detection and diagnosis, and more.
Moreover, AI-based big data analytics are contributing to building automation by making BAMSs self-learning, self-configuring, self-diagnosing, and self-commissioning. Additionally, using AI-based analytics can adapt existing building systems to promote the deployment of BAMSs with fewer investments from building owners.
From another hand, as AI models are very competent to learn common human error patterns, their use in big data analytics is significant. They can (i) detect and resolve possible flaws in datasets, (ii) learn by watching how the operators and users interact with the analytics programs, and identify anomalies and surface unexpected insights from large-scale datasets fast.
In this context, AI models assist operators and users of BAMSs to perform the different tasks related to the big data cycle, among them the operations of collecting, pre-processing, aggregating, storing, analyzing, and extracting various kinds of features. The integration of AI-big data analytics can (i) optimize energy and operational efficiency, (ii) automate monitoring and control through wireless platforms, (iii) provide quick and better decision making, (iv) smartly control the facility and reduce risk failures, (v) lower life cycle costs, and (vi) increase safety and security measures with ease.
A Taxonomy of AI-Big Data Analytics Frameworks in BAMSs
Due to the importance of using AI-big data analytics in BAMSs, a plethora of works have been proposed to (i) address different challenges, (ii) improve and automate building operation, and (iii) optimize building user experience. In addition, different reviews have been introduced to discuss the advances made in this research topic.
However, most of them have only focused on addressing one task at a time, e.g., energy management, rather than covering multiple BAMS tasks together (e.g., water management, occupancy detection, comfort optimization, fault diagnosis and anomaly detection (FDAD), etc.). To that end, we present in this paper a comprehensive systematic survey reflecting the latest developments in the field of AI-big data analytics and their utilization in BAMSs from different perspectives.
Figure 1 portrays a structured analysis framework that helps in overviewing existing techniques and shedding light on the organization of the presented framework. The first step in any AI process is system learning. This can take four primary forms: supervised learning, unsupervised learning, semi-supervised, and reinforcement learning.
Unsupervised Learning
Unsupervised learning learns from raw data without prior knowledge and mainly deals with unlabeled datasets. Although it does not need to annotate data as supervised learning, the learning phase can be more computational as all the possibilities are checked. The accuracy is lower since there are no corresponding outputs (labels).
Unsupervised learning is a category of ML algorithms used for separating data (e.g., energy consumption observations, ambient conditions, etc.) into different classes or clusters following a specific goal. Clustering algorithms usually pertain to one of the following groups: hybrid, fuzzy-based, model-based, and density-based approaches.
Using the clustering process facilitates the classification tasks when dealing with various problems, such as anomaly detection of energy consumption, IEQ monitoring and detection of pollutants, detection of abnormal water consumption, and more. K-means, C-means, and fuzzy C-means (FCM) were among the most investigated clustering approaches.
Dimensionality reduction techniques can also be employed in diverse ML tasks to classify data while promoting low computational costs as they first remove irrelevant observations. Accordingly, a plethora of frameworks have been proposed in the literature to explore the applicability of dimensionality reduction schemes in BAMSs, including PCA, LDA, QDA, MDA, Isomap, KPCA, t-SNE, MDS, and TSVD.
Supervised Learning
Supervised learning is applied for the case of labeled energy datasets. Despite its high performance, the necessity of labeled data causes some difficulties in real-world applications. It refers to conventional ML models that attempt to derive some conclusions from the input data given in the training process, and hence aim at predicting the class labels/categories for a new set of data.
Classification models have widely been deployed in existing BAMS-based big data analytics frameworks to perform different tasks, e.g., energy forecasting, energy balancing, IAQ monitoring, energy optimization, fault and anomaly detection. Typically, SVM, KNN, DT, ANN, MLP, ELM, and LR are among the famous classification models deployed in BAMSs.
Regression models have also gained popularity in smart buildings and smart energy systems because most of them are easy to implement and interpret, and efficient to train. Diverse regression models have been proposed to analyze BAMSs’ data, e.g., SVR, LR, AR models, RT, and RFT.
Deep Learning (DL)
Deep neural networks (DNNs) aim at simulating the behavior of the human brain “albeit far from matching its ability”, which allows them to learn from large-scale datasets. In addition to the capability of NNs with a single layer for making approximate predictions, DNNs have further benefits via (i) optimizing and refining the classification accuracy when additional hidden layers are considered, and (ii) identifying the most informative features of the data.
DNNs have become the state-of-the-art methods in various ML-based domains, similarly in BAMSs, they are attracting greater attention. They have widely used for energy forecasting, IEQ monitoring, occupancy detection, and more. Various hybrid models have also been built by combining the aforementioned models with other DL architectures, such as CNN-LSTM, CNN-BiLSTM, CNN-GRU, DFNN-LSTM, RBFNN-CNN, and more.
Statistical Models
They refer to mathematical models embodying an ensemble of statistical rules used to generate data samples, predict the relationships between one or diverse random/non-random variables, or classify them. Widely used statistical models include BN, NB, GAM, BBN, RBM, CRBM, and FCRBM. In BAMSs, they have been used for different tasks, such as selecting the most energy-efficient primary HVAC systems, building energy and water retrofitting, energy forecasting, assessing energy efficiency, NILM, gas usage prediction, IEQ monitoring, and more.
Semi-Supervised Learning (SSL)
SSL refers to the process of training ML models using a small portion of labeled data along with a large number of unlabeled observations. Then, the ML models should be able to learn and make predictions on new data. It falls between unsupervised learning and supervised learning, which is also considered as a special instance of weak supervision.
SSL has been widely used for fault and anomaly detection in BAMSs, such as fault detection and diagnosis in air handling units (AHUs), attack detection in HVAC systems, chiller fault diagnosis, and load monitoring.
Reinforcement Learning (RL)
Reinforcement learning is a field in artificial intelligence that involves an agent that develops the knowledge of the best strategy to follow to accomplish a defined objective by trial and error given the interaction with its environment. It can be categorized as traditional RL (TRL) methods and deep RL (DRL), which represents the evolution of the traditional methods where DL models are used to approximate the state and/or action value.
TRL-based BAMS applications include occupancy prediction, HVAC control, and controlling the HVAC system and windows for mechanical and natural ventilation. Recently, DRL is becoming a significant focus of scientists in BAMSs and many other research fields, such as indoor and domestic hot water temperature control, office HVAC systems control, energy optimization and thermal comfort control, and more.
Ensemble Methods
Ensemble methods are a class of ML that deploy different aggregation strategies for combining multiple learning models and then achieving better predictive performance compared to the use of a unique learning algorithm. It implies the gradual development of an ensemble learning using a set of ML models, where every new model occurrence is trained for emphasizing the training occurrences that previous models misclassified.
Ensemble models have been used in BAMSs for energy forecasting, heating and ventilation load prediction, HVAC optimization, load disaggregation and monitoring, water monitoring, IEQ monitoring, and occupancy detection.
Building Environments and Their Unique Characteristics
Buildings range in size, function, construction, design, and other attributes. Additionally, they present varying levels of potential hazards and risks to the occupants and the surrounding environment. However, buildings are primarily classified based on the utilization purpose that governs occupancy profile, sophistication level, and building design requirements.
Residential Buildings
Residential buildings are mainly for private occupancy, designed and built for individuals or groups, providing the necessary facilities and utilities to satisfy living requirements. Spaces in residential buildings involve several activities, including sleeping, sitting, conveniences, cooking, dining, and others. They exist in various sizes and have different occupancy rates, with a low occupancy density generally characterizing them.
Office Buildings
Office buildings are where people perform routine tasks, execute assignments and jobs for their employers, or provide passive or active, free of charge or remunerated services to the public. Familiar workplaces are office buildings such as law and corporate firms, commercial companies, post offices, banks, courtrooms, and similar places where people are involved in lengthy desk jobs or light-weight activities.
Healthcare Centers
The indoor environment in healthcare centers is critical for the health, well-being, safety, and comfort of patients, visitors, and the staff, as well as for the medical utilities and services. It has to comply with specific standards related to temperature, infection, and odor control. Healthcare centers require a clean and sterile environment, and air ventilation and infection control are essential to control the potential contaminants and other suspended microorganisms.
Sports Facilities
Sports facilities involve areas where individuals or groups engage in physical exercise, participate in athletic competitions, or attend sporting events. They encompass large and various spaces involving different types of activities. Sports facilities have distinct requirements for air conditioning and ventilation, thermal comfort, and lighting with unique usage and occupancy patterns.
Commercial Buildings
Commercial buildings have at least 50% of their floor spaces for commercial activities, such as malls, retail, and food services. They demand maintaining a clean and well-conditioned environment. For example, in restaurants and coffee shops, compliance with proper food storage and preparation standards is required to reduce the risk of spoiling food and eliminate the risk of incidents jeopardizing the well-being of the users as well as the reputation of the restaurants.
Industrial Buildings
Industrial buildings include buildings used for the generation and distribution of power, manufacturing products, the processing of raw materials, and many others. They have minimal and relatively low user flow for security purposes, such that they are only accessible to individuals with privileges. However, they involve energy-intensive and delicate machinery.
Academic Buildings
Academic buildings are used to conduct teaching activities such as schools, academies, universities, colleges, technical institutes, etc. A convenient and safe environment in academic facilities is an essential requirement for the education process. It affects the well-being and comfort of students, faculty members, and other staff, hence their productivity and working efficiency.
Computing Architectures for AI-Big Data Analytics in BAMSs
The advancement of cloud computing platforms has opened new opportunities for BAMSs to take control of operations on a large scale. Significant efforts have been devoted to developing cloud-based big data analytics solutions in BAMSs.
However, some drawbacks are still causing issues to users and operators, among them (i) the increased cost and communication overheads, (ii) the privacy and security concerns, especially when private data is transmitted to a centralized server for processing.
Edge computing refers to performing data pre-processing, data fusion for different sources, and AI-big data analytics at the edge of the network, i.e., sensor nodes. It enables optimizing cloud computing platforms due to its capability to use the processing power of IoT devices for filtering, pre-processing, aggregating, and storing IoT sensor data.
Fog computing represents a decentralized computing strategy where data storage, data processing, and computing resources are located in the middle layer situated between edge devices and cloud. BAMSs can benefit from streaming data over a layer of fog devices (or nodes) to become more connected, where data can be analyzed to detect abnormalities and autonomously react.
Hybrid computing refers to the case when the aforementioned computing architectures, i.e., edge computing, fog computing, and cloud computing, are used together to process and analyze data. Based on the application scenario and computation requirement
