Climate change as a result of global warming is gradually becoming an important issue of modern society. Its main cause is greenhouse gas emissions, which constitute the result of mainly anthropogenic activities involving the burning of fossil fuels, which currently supply more than 85% of the worldwide energy consumption, and their use is constantly increasing [1], despite the fact that they are finite resources [2]. Such applications include energy generation and energy related activities in domestic and tertiary buildings. Another important problem that theour society is facing is energy poverty, which currently remains a problem even for developed countries and is strongly linked to modern living standards, affecting the health and the emotional state of those facing it [3]. In order to address these issues, multiple options are examined towards the sustainability and accessibility of energy, including (a) the transition to clean renewable energy sources such as solar, wind, and hydroelectric energy [4], (b) the switch to more environmentally friendly fuels, such as shale gas and natural gas ^[5][6][5,6], even as transitory fuels towards full decarbonization by the year 2050 [7], and (c) the efficiency improvement of existing energy consumption units ^[8][9][8,9] such as industrial, commercial, residential, and public buildings. For the latter, in the European Green Deal, there are provisions that require Member States to renovate at least 3% of the total floor area of all public buildings annually and a multitude of other provisions for energy efficiency actions [10].

The emerging technologies of smart grids, artificial intelligence, Internet of Things, and smart homes provide the base of new solutions that can also lead to this direction. The aforementioned technologies provide a great opportunity to develop natural gas energy efficiency solutions, utilizing data from buildings in real-time and deploying control methodologies accordingly. The control methodologies developed for optimizing energy efficiency can also be used as the base for systems targeted at scheduling and shifting energy loads throughout the day, while taking into consideration user comfort. These solutions concern demand response (DR), which is the concept of increasing or decreasing an energy load in order for the supply to match the demand and, therefore, to help keep the grid stable. DR has been mainly implemented in electric loads during the last years, both in terms of research and of practical applications [11]. However, during the last years, the same concept was assessed for other energy carriers, including natural gas, and even for the combination of multiple energy sources, which defines integrated demand response [12].

Over the years, the evolution of the technical and commercial viability of new technologies such as IoT and artificial intelligence has created new opportunities to develop and deliver energy efficiency-oriented solutions. Indeed, multiple mid- and long-term plans have been established to drive towards a greener future, such as the European Union 2030 Climate and Energy Framework, which, among other key objectives, defined the targets of decreasing greenhouse gas (GHG) emissions by at least 40% and improving the overall EU energy efficiency by at least 32.5% until 2030 [13]. The EU also proposed a set of long-term targets, aiming to become completely climate-neutral by 2050 [14]. Based on the above, the EU updated its energy policy framework and published eight new energy rules aiming to make an impact in terms of the consumer perspective, the environment, and the economy [9]. These rules, most of which are defined by legislative initiatives, must be adopted by all EU countries and converted into national law, and they include directives towards building energy performance improvement and an energy efficiency increase [15].

The European Commission, through funding programs such as “Horizon 2020” and “Horizon Europe”, offers incentives to researchers to develop innovative solutions towards energy efficiency. Such an example is the “Secure, Clean and Efficient Energy” section of the “Horizon 2020” program, in which about six billion euros were invested for non-nuclear energy research purposes for the period 2014–2020, with the main priority being the energy efficiency sector [16]. In response to the energy efficiency targets set by the EU, the member states also implemented various measures to promote actions towards energy efficiency. For instance, Germany, which is one of the most energy efficient countries in the world at the moment, implemented a variety of policies and measures targeted both at residential and industrial buildings [17]. Regarding residential buildings, there are two energy efficiency incentives: (a) federal funding for the improvement of the efficiency of existing buildings and (b) tax reductions for the application of energy-related solutions.

Another, more global, approach was defined by the Paris Agreement [7], which was initiated in 2015, when it had effect on 55 countries, accounting for 55% of total global greenhouse gas emissions. More countries are continuously joining the agreement, accounting for a total of 189 parties that ratified until May 2020 [18]. Its main purpose is to limit the increase of the global average temperature to 1.5 °C compared to pre-industrial levels and to help developing countries make a transition to newer energy efficiency-oriented technologies.

In a similar fashion, the Clean Energy Ministerial was formed in 2010, which constitutes a global forum where major economies cooperate in order to share and promote the best practices and policies towards a global clean energy economy. Major global economies are represented among the participant countries, including China, which has also made commitments to minimize its carbon footprint by reducing their carbon dioxide emissions by 60–65% compared to 2005 levels by 2030.

3. Domestic Heating

Domestic heating is as necessary today as it has been for the entire human history so far, and its need is steadily increasing due to the respective increase of the world population. While for the largest part of the last century the fuel used to generate heat was oil or some of its byproducts, in the current century, other fuels and heating systems are gaining popularity [19], and amongst them is natural gas. Natural gas is a naturally born, non-renewable hydrocarbon gas mixture with multiple uses, including cooking, heating, mobility (in the form of compressed natural gas or CNG), and electricity generation in open-cycle and combined-cycle gas turbines (OCGTs and CCGTs, respectively). The reason for its extended use (natural gas dominates the European heating and energy supply [20]) is the fact that it offers a superior conversion efficiency compared to other fuels (coal, crude, oil products, etc.). It also emits considerably lower amounts of carbon dioxide when burned. This last feature aligns perfectly with the global initiative to reduce global warming, making natural gas the ideal fuel, both efficient and more environmentally friendly than other alternatives. This is the reason why natural gas has been designated by the European Commission as a transitory fuel towards the envisioned full decarbonization target in the year 2050. The technology of the boiler plays an important role in both the composition and amount of greenhouse gases emitted when gas is burned, but the key feature for any type of user, residential or industrial, is energy efficiency. Older technology boilers, also known as conventional boilers, are less efficient and environmentally friendly compared to condensing boilers. Condensing boilers consume less fuel and have a 23% lower environmental impact [21]. It is worth noting though that, according to a national study carried out in Italy, the national consumption for domestic heating has not increased substantially, and the average NO_x emissions decreased thanks to advancements in boiler technology, despite the increase in demand for natural gas during the period 1999–2011 [22]. In order for the fuel to reach the boiler, the house must be initially connected to the gas distribution network, analogously to the electrical grid. The gas enters the pipelines in the injection points (gas wells and/or storage facilities), and it flows through the transportation and distribution networks in order to reach the consumers’ houses to be deployed. This is the traditional approach, where combustion takes place locally to serve the needs of an individual house. In some cases, the expansion of modern-day cities and the rising number of buildings connected to the gas network call for a new, more centralized approach. A general model designed to achieve the coordinated development of centralized supply systems fueled by natural gas is opted for in some cities, by combusting gas in heating plants outside or nearby each city and distributing the heat energy through a district-heating system to the end-consumers within the city [23]. Considering all the characteristics of natural gas mentioned above, both (a) for the user’s economic benefit through the abundance of available customized solutions and (b) for reduced environmental impact, it becomes clear that natural gas will keep playing a major role in domestic heating in the foreseeable future.

4. Efficiency Optimization

Achieving higher energy efficiency and lower greenhouse gas emissions is a never-ending process leading to modifications in boilers and more efficient, environmentally friendly solutions. In this section, some well-known evolutions and breakthrough solutions are presented, categorized as follows:

• technological advancements in the construction phase;
• boiler operation manual improvement techniques;
• automated optimization during the boiler operational phase.

Each one of the above solutions contribute to the improvement of domestic boiler performance. The hereinafter described solutions belong to three main categories in chronological order, as summarized in Figure 1.

4.1. Boiler Technological Advancements in the Construction Phase

Boiler systems have been widely utilized for a long time as the driving force behind the industrial revolution, but initially they were not introduced in domestic environments until the middle of the 20th century in the form that are used today [24]. Traditional boilers were designed with only one mode of operation, i.e., on-off. This means that they could operate only at their full rated capacity. Thus, in many cases, the boiler would turn on, in order to satisfy the load, and then turn off again, multiple times, increasing the number of boiler cycles. This increase in boiler cycles leads to cycle losses, which makes the boiler less energy efficient and adds to the wear of the equipment. An important innovation that reduced the amount of boiler cycles is the boiler’s ability to modulate its output in a continuous manner, namely, to operate within a range between the minimum and the maximum modulation level [25]. Manufacturers started offering units with multiple firing rates, which allow the system to adapt its response according to the load and not operate in full mode. As a result, most modern boilers are modulated. Another major upgrade for domestic heating was the introduction of condensing boilers. In a traditional, non-condensing boiler, some heat is wasted in the form of hot gases released from the flue, whereas condensing boilers capture that heat and transfer it to water returning from the central heating system. This process results in a lower temperature of combustion products, recycling of the exhaust gas through the condensing heat exchanger, and reduced CO₂ emissions. All these characteristics make condensing boilers safer and more environmentally friendly, with the added benefit of being more efficient, reaching an efficiency level up to 99%. This high efficiency level is achieved by using waste heat in flue gas to preheat the cold water entering the boiler [26]. Inquiries have also been made in the direction of materials used for various components of the boiler. In a paper written by Liu et al. on emissions and thermal efficiency in premixed condensing gas boilers, two different types of burners were examined, metal fiber and stainless steel, in different heat loads and air rates to define which is most suitable and efficient for condensing boilers [27]. Other attempts to reduce the environmental impact, whilst achieving satisfactory combustion performance, include the use of different combustion catalysts [28]. Several European manufacturers offer domestic gas boilers that are able to burn gases of different compositions with the automatic adjustment of the excess air ratio. One of the cases examined is a mix of natural gas and hydrogen. Xin et al. performed simulations to determine the best hydrogen to natural gas volume ratio during combustion and concluded that the hydrogen mixing technique can help increase the combustion temperature and rate and reduce flue gas and CO2/NOx emissions [29]. The presence of hydrogen, which is highly flammable, requires increased control over the combustion process; the types of systems used to control the combustion process in natural gas fired residential boilers are “Flue gas analysis” and “Flame ionization” [30].

4.2. Boiler Operation Manual Improvement Techniques

Manual optimization is used to describe any process that can improve a boiler’s functionality from the design and production phase to the installation and commencement of operation. From the first day of their conception, boilers are designed and built with the target of maximum efficiency. Nowadays, there are guidelines and practices manufacturers can deploy when designing a boiler regarding their efficiency and safety ^[31][32][31,32]. These practices are crucial in ensuring that the boiler operates in its optimal state; neglecting them can cause significant performance issues. However, there is still room for improvement. As mentioned above, natural gas boilers are divided into two categories depending on whether they recapture escaping heat from flue gases: conventional and condensing. It is worth noting that a conventional boiler can be retrofitted to a condensing one even after its installation [33]. This results in the significant improvement of the efficiency of the boiler by a simple addition of a condensing heat exchanger, and it can be applied after the construction phase. Another important procedure that is commonly overlooked and plays a crucial role in a boiler’s operation, wear, and life expectancy is maintenance. Many users neglect the fact that their boiler is a machine that requires regular tuning to perform nominally. The importance of maintenance is such that research has been devoted to finding models to predict the required maintenance on buildings considering user discomfort [34]. Additionally, frequent maintenance increases the systems reliability, reduces boiler hazards, and potentially keeps down costs [35], while at the same time mitigating any potential health risks to the residents of the building [36]. Automatic early fault detection can help the process of proper maintenance by alerting the residents when a possible operational issue is detected. Achieving sufficient accuracy in the fault detection process is significant in this context. Shohet et al. tested a variety of machine learning algorithms (K-nearest neighbor, decision tree, random forest, and support vector machine) in a simulated environment with 14 different boilers [37]. The results were impressive, displaying an accuracy of over 95% in the models trained for each boiler, but generalization was not possible. This indicates that, with the described methodology, it is difficult to create a single robust model that can be used as a generalized approach for fault detection in all boilers, but, instead, models can be trained for each boiler specifically, deployed during the tuning phase, and installed before the operational phase to inform the user of possible faults, which would then require manual fault fixes to be applied. Finally, the functional parameters of the boiler’s operation can play a crucial role in its efficiency during its normal daily operation. Wu et al. [38] optimized the boiler’s efficiency by employing an artificial bee colony (ABC) algorithm in order to determine the functional parameters (exhaust gas temperature, volume percentage of O₂, combustible material in fly ash, and boiler load percentage) that minimize the system’s heat losses, based on the model of boiler combustion efficiency. The resulting parameters can be used to fine-tune the boiler before its operational phase and after its construction to ensure optimal efficiency. The test case displays that the ABC methodology performs better than a genetic algorithm (GA), achieving quicker convergence and increased robustness. Nevertheless, despite the fact that the literature is mainly focused on the optimization of individual boilers’ efficiency, the authors of [39] introduced an interesting conclusion about the zonal controlling of domestic heating, where zonal controlling means heating the rooms of the residence only when they were ‘occupied’. In a pilot study of an 8-week winter test period in the UK, a house with zonal control used 11.8% less gas, despite a 2.4 percentage point drop in average daily boiler efficiency, compared to conventionally controlled heating. This minor parametrization technique indicates the significance of the added value that a smart heat system could have and the huge potential of automated solutions.

4.3. Automated Optimization during the Boiler Operational Phase

As discussed above, achieving higher efficiency levels of energy usage, particularly for natural gas boilers that constitute the main subject of this work, and minimizing greenhouse gas emissions constitute primary objectives of the EU Energy policy [13]. The methods discussed in the previous sections mainly focus on well-known practices, improvements during the design phase, or the addition/upgrade of various compartments that can boost a boiler’s performance. However, the rapid growth of data science and the development of smarter algorithms has opened new opportunities that allow for further improvements in machine operations through the analysis and application of software tools. These new capabilities can help tous tune the machines to operate optimally, but they also provide the added value of offering real-time automated solutions that require minimal outside/human intervention. In order to create energy efficient buildings and deploy solutions regarding automatic control, one must firstly gain a better understanding of the various factors affecting their energy consumption and efficiency performance. The first step to this process is the development of an evaluation model that performs effectively for multiple types of buildings. Such a model was implemented, using multi-scale analysis, and tested by Tronchin et al. [40]. Among the various parameters used for such a system, a crucial parameter is the building’s energy rating, since it holds significant information regarding its thermal behavior. Aiming for a specific building energy profile and being able to account for it during its design phase can prove to be a significant advantage [41], with the derived data being potentially helpful in the selection and tuning of both manual and automated efficiency optimization solutions deployed later on. A boiler is designed and built in an environment that has quite different conditions from the one it is called to operate in. This fundamental difference calls for additional actions to ensure that every boiler performs optimally depending on the environment it is placed and installed in. Weather is an important aspect when attempting to provide heating for a building/home. It is constantly changing and can be quite unpredictable. Thus, any system responsible for heating should be quick to adapt to weather changes and be able to predict, to a certain degree, its future behavior. Weather compensation is the ability of a system to account for the weather variations and tune itself to operate in the most efficient way, providing a remarkable base for automated optimization solutions. An interesting approach was proposed by Ping et al. [42], who developed a model predictive control (MPC) technique for the control of the heating process based on weather forecast compensation. MPC constitutes an advanced method of process control that is used to control a process while satisfying a set of constraints. The main advantage of MPC is the fact that it allows the current timeslot to be optimized, while taking future timeslots into account. The proposed system receives, in real-time, the indoor and ambient temperatures, as well as an ambient temperature forecast for future time-intervals, and appropriately adjusts the heat supply based on thermal comfort constraints and the modelling of the building’s thermodynamics. Simulation results indicate that the delay of heating and the overheating of the space due to thermal inertia are limited, providing an overall better user experience regarding thermal comfort and eliminating the wasted energy consumed when the space is heated above the comfort level, which is most often the case when using more classical control approaches. In most commercial systems using weather compensation, the temperature of the heating fluid is calculated as a function of some predetermined relation to the outdoor temperature called heating curve. This approach, however, often fails to capture the building’s physics and conditions and cannot compute for future outdoor conditions, thus leading to an excess of energy consumption to maintain the users’ thermal comfort. A convenient system of a non-invasive add-on module that can connect to existing heat controllers was developed using MPC to control the building’s heating requirements [43]. The system was deployed during the 2013–2014 heating season in several locations, with the results being quite encouraging, achieving positive energy savings for all test sites. A different approach is the use of data predictive control (DPC) methodologies, such as neural network prediction models and relevant machine learning algorithms. Data predictive control is a framework designed to combine the simplicity of model-based methods with the predictive capability of data-driven control. Using DPC algorithms, one can synthesize finite-horizon predictive control decisions after learning dynamical system models based on historical data. Of course, not all models are suitable for all problems, but they can be easily modified to serve similar purposes. In recent years, the introduction of controllers that are designed to house neural network models has increased the ability of locally processing information and decision making, allowing for more autonomy, advanced capabilities, and reliability. An example of such a controller was proposed in 2015 by Meng et al. [44]. The research team presented an improved version of the conventional PID neural network (PIDNN) control algorithm with additional momentum, which is used to improve its learning efficiency and solve the problem of local minimums, along with the introduction of an improved particle swarm optimization that helps initialize the weights of the neural network. The simulation was conducted via a multi-variable nonlinear coupling system and showed that the proposed algorithm displayed improvements in terms of regulating time and controlling precision compared to the original algorithm. Despite the fact that the presented controller approach is neither developed for, nor directly tested in, a boiler optimization scenario, the presented control methodologies can potentially be applied as an alternative for conventional boiler control methodologies. Smarra et al. [45] developed a data-driven control methodology based on random forests and regression trees, where an on-off biomass boiler was controlled in real-time, along with other sources of energy. The test cases, which contain both a simulated case and a real-world house scenario, displayed positive results, especially in the case of random forests. Macarulla et al. [46] introduced an adaptive control strategy targeted at commercial buildings with the help of feed-forward neural networks. The proposed system sends on/off commands to boilers at specific time intervals, attempting to minimize total consumption and the loss of user comfort at the same time. Tsoumalis et al. proposed another DPC methodology where LSTM neural networks are used to predict the change in indoor temperature and the load of the boiler in the short-term (30-min look ahead), and a genetic algorithm (GA) was employed to obtain the optimal boiler configuration with regards to user thermal comfort and gas consumption minimization [47]. Results acquired from the trial in four real-world houses indicated a significant reduction in natural gas consumption with minimal comfort loss. summarizes the automated optimization/control methods presented in the literature. Since automatic optimization usually involves minimal or no user interference, one aspect of significant importance is monitoring, which helps ensure that the system operates in nominal conditions and detects any problems or errors related to safety and performance. A paper published by a team at DELFT University describes a set of fault detection and diagnostic tools for condensing boilers [48]. The system was designed to use real-time measurements in order to evaluate performance degradation, making it ideal for building energy management systems that can store limited amounts of data. Through extensive simulations, the effectiveness of those tools was verified both in terms of quick fault detection, but also in isolating the source of the problem. Such systems contribute to making sure that boilers remain at an optimal operational level, and they can be integrated in an optimization system in order to provide more complete and production-ready solutions. Though there are significant advancements in IoT technologies and quite a variety of methodologies have been explored, the transition to commercial applications seems to be quite slow. The transfer of research results to the market is a field of its own, which opens numerous subjects for investigation, including the validation of the derived conclusions in real life, as well as the impact from their massive application.