Nowadays, increasing energy efficiency is used to face the great world challenges, such as energy security, air pollution, climate change and economic crises, among others. Energy efficiency alternatives have the power to optimize energy consumption and reduce greenhouse gas emissions, thus contributing positively to the preservation of natural ecosystems and human health. Additionally, energy efficiency alternatives help mitigate the economic effects in the workplace [1
]. Energy efficiency has aroused great interest in research worldwide, because energy consumption has increased in recent years, especially in the residential sector. For this reason, organizations and governments worldwide are proposing actions for energy conservation with the purpose of reducing energy-related inconveniences. The residential sector is attributed a high energy consumption; however, home automation systems (HAS), combined with IoT, are alternatives that promise to contribute to greater energy efficiency [2
Likewise, advances in energy conversion, communication, and information technologies have paved the way for a new generation of homes—smart homes—thus allowing people to improve aspects of their houses, such as comfort, convenience, safety, and entertainment, while simultaneously helping them to cut energy waste. Additionally, Home Energy Management Systems (HEMS) are important in achieving the goals of smart energy homes in many countries. Likewise, the smart home market is growing rapidly. In particular, it is improving in areas such as fire detection, lighting, entertainment, and energy efficiency systems, among others [3
]. In addition, in recent years, energy conservation action plans for the residential sector have gained prominence, since they contemplate indoor comfort and energy efficiency. In this sense, it is worth mentioning that user behavior patterns in terms of energy consumption change according to user needs and lifestyles, yet there is always an inclination to maintain indoor comfort at the expense of power saving. Therefore, to design and implement effective energy saving action plans, it is important to know not only the link between indoor comfort and energy consumption, but also home characteristics and home resident needs [4
On the other hand, in recent years, the Internet has impacted people’s daily lives through a new paradigm called IoT, which is present in smart homes, retail, education, government services, smart grids, agriculture, communication and business, among others. The IoT is the combination of various technologies belonging to application domains that interconnect objects or things through the Internet; by doing so, IoT-based things or objects acquire detection, monitoring, and remote-control capabilities. Some of the most common IoT technologies include cloud computing, Wireless Sensor Networks (WSN), Radio Frequency Identification (RFID), networks and communication, machine-to-machine (M2M) interaction, Real-Time Systems (RTS) and mobility support, among others [5
]. The IoT can collect, distribute, and analyze data to convert it into knowledge and information [7
]. Therefore, it is important to further research and develop energy optimization mechanisms across different IoT application domains. Moreover, appropriate energy saving and collection proposals and programming algorithms should continue to be sought, since renewable energy sources are becoming more important every day. Additionally, researchers estimate that the adoption of the IoT is based on the success of these energy optimization proposals [8
IoT devices for smart homes have restricted capabilities; hence, it is important to incorporate more data handling options to successfully collect, manage, and analyze large volumes of data. Some of these alternatives include machine learning and big data technologies. To collect and analyze large volumes of information, big data analytics technologies are used [9
]. Moreover, they allow large volumes of sensor data to be effectively analyzed and used [10
]. On the other hand, machine learning is part of artificial intelligence, because it is responsible for studying algorithms and statistical models based on patterns and inferences that systems use to meet their goals [11
]. Likewise, machine learning is broadly used in real-time applications due to its viability and robustness. In addition, machine learning provides solution alternatives to learning-based problems and identifies the background and characteristics of such problems to learn from them and thus increase system functioning. Finally, machine learning executes actions requiring previously obtained knowledge, which is classified as reinforcement learning, unsupervised, and supervised [12
Current challenges in smart homes are areas of opportunity for the IoT paradigm, machine learning, and big data technologies. In this work, we propose HEMS-IoT, a big data and machine learning-based smart home energy management system for home comfort, safety, and energy saving. Machine learning techniques and big data technologies are important in this work because they help our system analyze and classify energy consumption efficiency, identify user behavior patterns, and offer them increased comfort at home. HEMS-IoT relies on the machine learning algorithm J48 and Weka API to know energy consumption patterns and user behavior patterns. Likewise, we relied on RuleML and Apache Mahout to generate energy-saving recommendations based on user preferences to preserve smart home comfort and safety. Finally, to validate HEMS-IoT, we introduce a case study in which we monitor a smart home to ensure comfort and safety and reduce energy consumption. HEMS-IoT is an extension of the works presented in [13
] and [14
The remainder of this paper is structured as follows. Section 2
discusses works on IoT, big data technologies, energy efficiency strategies for smart homes, intelligent agents, and machine learning. Next, in Section 3
, we introduce the architecture of HEMS-IoT and discuss a case study in which we monitor a smart home to ensure comfort and safety and reduce energy consumption. In Section 4
, we present the results from the case study discussed in the before section. Finally, in Section 5
, we present the conclusions and the future work.
2. Related Work
The IoT is modifying citizens’ living environments by moving from a traditional home to a smart home [15
]. In smart homes, people can control, monitor, and manage energy consumption according to their lifestyle [3
]. In this section, we present a review of related works with IoT initiatives for energy efficiency in smart home. We pay close attention to those initiatives using machine learning and big data. For instance, in their work, Kang et al. [16
] proposed an IoT-based system that uses a three-level context creation model for environment-sensitive services in the domotic space. IMS was designed with open source software and hardware in order to extend in the IoT context. The system was tested as IMS-based smart home services in two scenarios: a smart home health care service and a disaster management service. Also, Adiono et al. [17
] introduced an optimization protocol for WSN through an architecture for a smart home. This architecture is divided into two environments, exterior and interior, which communicate through an access point. The user´s home can be controlled from anywhere and anytime by a smart phone. SQLite database system was used for the implementation and different tests for the validation. Lee et al. [18
] proposed a web services architecture for home service environments. Three layers shape the architecture: (1) information layer, (2) management layer, and (3) presentation layer. The service overlay network was used in this work to generate new service composition in the IoT context. By contrast, Montesdeoca-Contreras et al. [19
] implemented an IoT application for controlling and monitoring smart homes. Namely, the application allows users to monitor and control domotic devices through tactile functions or voice commands, including a safety net. The application was developed with App Inventor and Android Studio.
Chilipirea et al. [20
] proposed a method for creating models for IoT applications that facilitates the generation of robust, energy-efficient systems in a home security system. The model used the overlap between device characteristics to preserve energy and temporarily disable part of them. Elkhorchani and Grayaa [21
] proposed a shedding algorithm for home energy usage and an architecture of a smart home energy management system. This work was based on domestic renewable energy sources, wireless communication among domotic devices, a control system and a home management system, and on grid management. Likewise, Salman et al. [22
] proposed a smart system for energy efficient IoT-based homes with a cooling system was demonstrated the heat distribution in the kitchen area and virtual model of flow. The system remotely controls heating/cooling and lighting when an occupant leaves or enters the kitchen. In their work, Al-Ali et al. [9
] introduced an Energy Management System (EMS) for smart homes. The EMS relies on MQTT (Message Queue Telemetry Transport), is empowered with Business Intelligence (BI) and analytics and uses big data. The system was validated using HVAC (Heating, Ventilation and Air Conditioning) to simulate the small residential area systems.
Baker et al. [23
] introduced and tested the algorithm E2C2, an energy-aware multi-cloud IoT service composition which can generate energy efficient composition proposals by adding services from service providers that are scattered globally. By contrast, Matsui [24
] proposed an data provision system for both maintaining indoor comfort and decreasing electricity consumption with data provision to modify home resident behaviors. Fensel et al. [25
] presented the OpenFridge platform and approach for energy saving in electrical appliances, particularly refrigerators. The approach demonstrated the feasibility of users eventually using it up for data economy and interacting with semantic energy information. Additionally, Hossain et al. [26
] proposed an energy-efficient cyber–physical smart home system. Using cloud big data and computing, the system monitors the elderly to assist them in maintaining energy efficiency at home. In addition, in this work was proposed a smart multimedia-enabled middleware assistant to receive notifications relating to the status of domotic devices, visualize energy-efficient processes, share multimedia messages and control smart domotic devices through gestures.
Golam et al. [27
] proposed an architecture that uses web objects for offering energy efficient comfortable living services for smart home IoT services. In addition, a conceptual semantic ontology model was designed using the tool Protégé for use in the smart home scenario. Additionally, Chauhan and Babar [28
] presented a Web-of-Things (WoT) system to manage appliances. The system tries to meet some of the standard business needs and the quality of smart homes. The use of a Reference Architecture (RA)-centered approach for the evolution of the WoT and IoT systems was proposed in this work. In addition, Lanfur and Pérez [29
] implemented a real-time video streaming and transmission security system in a residential scale model. The system allows lights to be turned off and on, and doors to be closed and opened as occupants enter/leave a room. Also, the system relies on motion sensors to obtain information that can be visualized through a web interface. Iqbal et al. [30
] presented architecture that uses the ZigBee technologies to minimize unnecessary electrical energy usage in smart homes, based in the context of IoT. The architecture is powered by GRAPHX, Bit Data, SPARK, and Hadoop for data analysis.
Marinakis et al. [31
] proposed an architecture of a big data platform that can create, develop, maintain, and exploit smart energy services using cross-domain data. A web-based Decision Support System (DSS) according to the proposed architecture was developed to use multi-sourced information to generate management activity strategies in the domotic space. Jo and Ik-Yoon [32
] presented three intelligent models as IoT platform application services for a smart home: (1) intelligence energy efficiency as a service (IE2S) to perform the role of a server and process the information collected by IAT using the Mobius platform and an artificial TensorFlow engine for data learning, (2) intelligence service TAS (IST) to manage, and provide control the service stage, and (3) intelligence awareness target as a service (IAT) to manage the “things” stage. Filho et al. [33
] proposed STORm, a decision-making solution for residential environments that combines computational intelligence and fog computing. STORm retrieves, processes, disseminates, detects, and controls the information sent by sensors installed in a residential scenario to apply the decision-making process. Additionally, Tao et al. [34
] developed a multilayered architecture based on the cloud and a home automation anthology supported by the IoT. The anthology was used to address application heterogeneity, data representation, and knowledge.
Iqbal et al. [35
] developed an interoperable IoT-based platform for domestic environments using web-based objects and the cloud. The platform facilitates the control of domotic devices from different locations, provides important household information to analyze applications and various services, and tries to improve resource utilization. Yassine et al. [36
] proposed a platform for smart homes that combines big data analytics technology and IoT with cloud computing and fog. This platform is a fast and efficient solution that supports large volumes of smart home information. In addition, Matsui [37
] presented a HEMS to obtain information from a smart home, with the purpose of maintaining interior comfort and reducing energy consumption according to resident comfort preferences, which were previously provided and set through a web page. On the other hand, Terroso-Saenz et al. [38
] presented IoTEO, an IoT energy platform that attempts to be the first holistic solution for the management of IoT energy information. IoTEO relies on FIWARE to deal with energy quality and support data analytics. Likewise, Park et al. [39
] proposed a thermal comfort-based controller (TCC-V1) to decrease the energy consumed by cooling residential buildings. To energy optimal control, the controller uses the predicted mean vote (PMV).
Bouaziz et al. [40
] proposed EMA-RPL, a new energy efficient and mobility aware routing protocol which is based on Lossy Networks (the RPL standard) and the Routing Protocol for Low power. EMA-RPL enables the better sustaining of connectivity of conserving energy and mobile nodes. By contrast, Mancini et al. [41
] presented the characterization of three control systems based on user energy consumptions habits. The authors used a demand/response (DR) program and procedure to evaluate energy consumption and economic savings. Similarly, they conducted a financial analysis of the investment needed to implement the program. Sun et al. [42
] proposed an energy-efficient mechanism that optimizes IoT service compositions to support concurrent requests. The mechanism reduces energy consumption in the network and improves the exchange of IoT services among concurrent requests. In addition, Meena et al. [43
] presented a framework for optimal planning of hybrid energy conversion systems (battery energy storage system, photovoltaic cells, and wind turbine) in smart homes. The model aims at generating less costly and more reliable alternatives for smart homes for middle-class families. Alarif and Tolba [12
] proposed AQL (Adaptive Q-Learning), a reinforcement-based learning technique to increase network performance with reduced energy–overhead tradeoff in a smart device (sensor) cloud-assisted internet of things (CIoT).
Matsui et al. [4
] presented a study on energy usage patterns to then propose an energy conservation action plan. To this end, the researchers gathered real-time data on energy consumption and indoor temperature in a Tokyo residential area using IoT devices and sensors. Then, the data were analyzed using three types of method: a clustering algorithm, a correlation analysis, and a classification of indoor temperature for detached houses with different building structure and ages, and condominium apartments. From a similar perspective, Le and Benjapolakul [44
] conducted a study to test the energy yield of the rooftop photovoltaics (PV) systems based on machine learning techniques (i.e., multiple linear regression and bootstrap). The authors identified an association between technical configuration details of PV (number of inverters, number of panels, rated solar panel power, and rated inverter power) and the energy yield. In addition, Castro-Antonio et al. [45
] presented a Robotics Operation System (ROS) that integrates various types of services in a single smart home service system. The ROS provides services to smart homes without fully recognizing them through a collection of sensors, cameras, and by incorporating autonomous service robots that can move inside the home and interact with the residents.
Huh et al. [46
] introduced the design of a Smart Metering Control System and a series of tests based on power line communication (PLC) using smart agents (i.e., program in charge of collecting data or performing certain actions without user intervention). For the tests, the authors used a smartphone and an Android mobile application. On the other hand, the protocols in the system were developed with languages C++ and C. In addition, the system was developed and integrated in Java with the purpose of being a basic element for the Smart Grid. Likewise, Jung and Huh [47
] proposed a model to predict atypical data of a linear transmission point using the A-Deep Q-Learning algorithm in combination with the altered K-means algorithm. The goals were to identify atypical information of the same linear transmission point in big data, and to know the objective of its elements. Additionally, the model makes it easy to automatically select values from cluster k, which is based on unlabeled sensors and big data. In addition, Yassine et al. [48
] proposed a model to identify patterns of human activity to support the health care of people from their homes. The model makes use of big data to analyze the activities of the inhabitants at home. Additionally, pattern mining is proposed to analyze energy consumption variation in domotic devices, which depends on the behavior of the inhabitants. Also, in the model, FP-growth was used to identify patterns, whereas the k-means algorithm was used to know the relationship among devices influencing energy consumption.
Zhao et al. [49
] proposed a neutral blockchain-based data trading protocol within the big data market to increase the availability, fairness, and privacy of data trading. The blockchain infrastructure has the main advantage of debugging failures in any of the big data market points. Additionally, similarity learning was used to increase the quality and availability of information from data providers, and an extension of the double authentication preventing signature (DAPS) was carried out. Additionally, Risteska Stojkoska and Trivodaliev [50
] proposed a framework that integrates different components from IoT architectures/frameworks proposed in other works in order to efficiently integrate smart home objects in a cloud-centric IoT solution. Likewise, the authors identified a smart home management model for their architecture, along with the tasks to be performed at each architectural level. Finally, the authors discussed current challenges in smart home design, emphasizing such aspects as interoperability, information processing, and communication protocols. Rathore et al. [51
] proposed an IoT-based system that relies on different types of sensors installed in a home to contribute to smart city development and urban planning through big data. The system uses a four-tier architecture: (1) bottom tier-1, (2) intermediate tier-1, (3) intermediate tier-2, and (4) top tier. In addition, the system is implemented using Hadoop with Spark, voltDB, Storm or S4 for real-time information processing and data collection.
As the previous paragraphs show, a great range of applications and IoT tools seek to contribute to energy-saving efforts in smart houses. Likewise, note that many of the analyzed works established models, communication protocols, technologies, and security paradigms to guarantee the interoperability and integrity of domotic systems. However, it seems that only a few initiatives can automatically, and without much user intervention, handle and operate decisions of smart home devices connected to a domotic control system. From this perspective, HEMS-IoT is a solution that integrates and provides access to information from different IoT providers, domotic devices, and sensors. Likewise, our system analyzes the information collected from big data technologies and machine learning algorithms and offers knowledge on efficient energy consumption and home comfort through recommendations, rules, and alerts. Overall, HEMS-IoT monitors domotic devices and sensors in real time, provides energy-saving recommendations, and facilitates communication and interaction between devices and with users. The following section describes the architecture and functionality of HEMS-IoT and discusses our case study.
Energy efficiency has become a key research area, because energy consumption exponentially increases as years go by, particularly in the residential sector. If combined with the IoT paradigm, home automation systems are promising energy saving alternatives. The IoT can successfully collect, distribute, and analyze data to convert it to knowledge and information; however, IoT devices for smart homes have limited resources. To overcome this limitation, it is important to consider other data handling alternatives—such as machine learning and big data—to collect, manage, and analyze large volumes of data. The big data analytics technologies are used to obtain and analyze large amounts of data, whereas machine learning algorithms and statistical models based on patterns and inferences are needed by the system to meet its goals. Additionally, machine learning provides alternatives to learning-based problems and identifies the background and characteristics of such problems in order to learn from them and increase system functioning.
Current challenges in the residential sector related to energy consumption are areas of opportunity for the IoT paradigm, machine learning, and big data technologies. In this work, we proposed HEMS-IoT, a big data and machine learning-based smart home energy management system for home comfort, safety, and energy saving. Machine learning techniques and big data technologies are important in our work, because they are used by the system to analyze and classify energy consumption efficiency, identify user behavior patterns, and ensure home comfort. We used the machine learning algorithm J48 and Weka API to learn energy consumption patterns and user behavioral patterns. Also, RuleML and Apache Mahout were used to create energy-saving recommendations based on user preferences to preserve smart home comfort and safety. We presented a case study to validate HEMS-IoT, where we monitored a smart home to ensure home comfort and safety and reduce energy consumption. In conclusion, three factors were fundamental in achieving energy consumption reduction in this case study: (1) the commitment of smart home inhabitants to change their energy consumption habits, (2) follow-ups on the system’s energy-saving recommendations, and (3) the fact that the system allows users to modify the operating parameters of domotic devices. From this perspective, we trust that the results obtained in this work will be a motivation for more users to rely on HEMS-IoT when seeking a smart home management alternative that makes it possible to optimize energy consumption and thus make savings.
Our proposal has five main constraints. First, the HEMS-IoT mobile application only works on the Android operating system, even though we know that the tablet market is dominated by iOs. Second, our system is only compatible with some types of home automation sensor. Third, we only use big data technologies and the J48 machine learning algorithm. Fourth, the system does not generate customized energy-saving recommendations due to some limitations having arisen during the research process. Five, we did not collect data on energy consumption from domotic devices before the implementation of HEMS-IoT.
As future work, we will seek to implement HEMS-IoT on a larger scale and among smart homes using a greater number of domotic devices. Additionally, we will intend to incorporate location-based functionalities in HEMS-IoT by relying on the GPS from mobile devices. In this sense, we expect that HEMS-IoT will be able to estimate residents’ arrival time at home to increase comfort by performing actions such as playing music or turning the air conditioner in advance based on the mood of the user. Finally, we will seek to include solar panels—in order to minimize electrical energy consumption—and implement more advanced security strategies, such as blockchain and cyber security.