ALTERNATIVE MODELS FOR BUILDING ENERGY PERFORMANCE ASSESSMENT

The research activity carried out during the three years of the PhD course attended, at the Engineering Department of the University of Palermo, was aimed at the identification of an alternative predictive model able to solve the traditional building thermal balance in a simple but reliable way, speeding up any first phase of energy planning. Nowadays, worldwide directives aimed at reducing energy consumptions and environmental impacts have focused the attention of the scientific community on improving energy efficiency in the building sector. The reduction of energy consumption and CO2 emissions for heating and cooling needs of buildings is an important challenge for the European Union, because the buildings sector contributes up to 36% of the global CO2 emissions [1] and up to 40% of total primary energy consumptions [2]. Despite the ambitious goals set by the Energy Performance of Buildings Directive (EPBD) at the European level [1], which states that, by 2020, all new buildings and existing buildings undergoing major refurbishments will have to be Nearly Zero Energy Buildings (NZEB) [3,4], the critical challenge remains the improvement of the efficiency when upgrading the existing building stock to standards of the NZEB level [5]. The improvement of the energy efficiency of buildings and their operational energy usage should be estimated early in the design phase to guarantee a reduction in energy consumption, so buildings can be as sustainable as possible [6]. While a newly constructed NZEB can employ the “state of the art” of available efficient technologies and design practices, the optimization of existing buildings requires better efforts [7]. One way or the other, the identification of the best energy retrofit actions or the choice of a better technological solution to plan a building is not so simple. It has become one of the main objectives of several research studies, which require deep knowledge in the field of the building energy balance. The building thermal balance includes all sources and sinks of energy, as well as all energy that flows through its envelope. More in detail, the energy demand in buildings depends on the combination of several parameters, such as climate, envelope features, occupant behaviour and intended use. Indeed, the assessment of building energy performance requires substantial input data describing structures, environmental conditions [8], thermo-physical properties of the envelope, geometry, control strategies, and several other parameters. From the first design phases designers and researchers, which are trying to respect the prescriptions of the EPBD directive and to simultaneously ensure the thermal comfort of the occupants, must optimize all possible aspects that represent the key points in the building energy balance. As will be shown in Chapter A, the literature offers highly numerous complex and simplified resolution approaches [9]. Some are based on knowledge of the building thermal balance and on the resolution of physical equations; others are based on cumulated building data and on implementations of forecast models developed by machine-learning techniques [10]. Several numerical approaches are most widespread; these have undergone testing and implementing in specialised software tools such as DOE-2 [11], Energy Plus [12], TRNSYS [13] and ESP-r [14]. Such building modelling software can be employed in several ways on different scales; they can be simplified [15,16] or detailed comprehensively by different methods and numerical approaches [17]. Nevertheless, they are often characterised by a lack of a common language, which constitutes an obstacle for making a suitable choice. It is often more convenient to accelerate the building thermal needs evaluation and use the simplified methods and models. For example, a steady state approach for the evaluation of thermal loads is characterised by a good level of accuracy and low computational costs. However, its main limitation is that some phenomenon, such as the thermal inertia of the building envelope/structure, may be completely neglected. On the other hand, the choice of a more complex solution, such as the dynamic approach, uses very elaborate physical functions to evaluate the energy consumption of buildings. Although these dynamic simulation tools are effective and accurate, they have some practical difficulties such as collecting detailed building data and/or evaluating the proper boundary conditions. The use of these tools normally requires an expert user and a careful calibration of the model and do not provide a generalised response for a group of buildings with the same simulation, because they support a specific answer to a specific problem. Meanwhile the lack of precise input can lead to low-accuracy simulation. Anyway, in all cases it is necessary to be an expert user to implement, solve and evaluate the results, and these phases are not fast and not always immediately provide the correct evaluation, conducting the user to restart the entire procedure. In the field of energy planning, in order to identify energy efficiency actions aimed at a particular context, could be more convenient to speed up the preliminary assessment phase resorting to a simplified model that allows the evaluation of thermal energy demand with a good level of accuracy and without excessive computational cost or user expertise. The aim of this research, conducted during the three years of the PhD studies, is based on the idea of overcoming the limits previously indicated developing a reliable and a simple building energy tool or an evaluation model capable of helping an unskilled user at least in the first evaluation phase. To achieve this purpose, the first part of the research was characterised of an in-depth study of the sector bibliography with the analysis of the most widespread and used methods aimed at solving the thermal balance of buildings. After a brief distinction of the analysed methods in White, Black and Grey Box category, it was possible to highlight the strengths and weaknesses of each one [9]. Based on the analysis of this study, some alternative methods have been investigated. In detail, the idea was to investigate several Black-Box approaches; mainly used to deduce prediction models from a relevant database. This category does not require any information about physical phenomena but are based on a function deduced only by means of sample data connected to each other and which describes the behaviour of a specific system. Therefore, it is fundamental the presence of a suitable and well-set database that characterise the problem, so that the output data are strongly related to one or more input data. The completely absence of this information and the great difficulty in finding data, has led to the creation of a basic energy database which, under certain hypotheses, is representative of a specific building stock. For this reason, in the first step of this research was developed a generic building energy database that in a reliable way, and underlining the main features of the thermal balance, issues information about the energy performances. In detail, two energy building databases representative of a non-residential building-stock located in the European and Italian territory have been created. Starting from a well-known and calibrated Base-Case dynamic model, which simulates the actual behaviour of a non-residential building located in Palermo, it was created an Ideal Building representative of a new non-residential building designed with high energy performances in accordance whit the highest standard requirements of the European Community. Taking into consideration the differences existing in the regulations and technical standards about the building energy performance of various European countries, several detailed dynamic simulation models were developed. Moreover, to consider different climatic characteristics, different locations were evaluated for each country or thermal zone which represent the hottest, the coolest and the mildest climate. The shape factor of buildings, which represents the ratio between the total of the loss surfaces to the gross heated volume of a building, was varied from 0.24 to 0.90. To develop a representative database where the data that identify the building conditions are the inputs of the model linked to an output that describes the energy performances it was decided to develop a parametric simulation. In detail different transmittance values, boundary conditions, construction materials, and energy carriers were chosen and employed to model representative building stocks of European and Italian cities for different climatic zones, weather conditions, and shape factor; all details and the main features are described in Chapter B.   These two databases were used to investigated three alternative methods to solve the building thermal balance; these are: • Multi Linear Regression (MLR): identification of some simple correlations that uses well known parameters in every energy diagnosis [18–20]; • Buckingham Method (BM): definition of dimensionless numbers that synthetically describe the relationships between the main characteristic parameters of the thermal balance [21]; and • Artificial Neural Network (ANN): Application of a specific Artificial Intelligence (AI) to determine the thermal needs of a [22] building. These methods, belonging to the Black-Box category, permit solving a complex problem easier with respect to the White-Box methods because they do not require any information about physical phenomena and expert user skills. Only a small amount of data on well-known parameters that represent the thermal balance of a building is required. The first analysed alternative method was the MLR, described in Chapter C. This approach allowed to develop a simple model that guarantees a quick evaluation of building energy needs [19] and is often used as a predictive tool. It is reliable and, at the same time, easy to use even for a non-expert user since an in-depth knowledge in the use phase is not needed, and computational costs are low. Moreover, the presence of an accurate input analysis guarantees greater speed and simplicity in the data collection phase [23]. The basis for this model is the linear regression among the variables to forecast and two or more explanatory variables. The feasibility and reliability of MLR models is demonstrated by the publication of the main achieved results in international journals. At first, the MLR method was applied on a dataset that considered heating energy consumptions for three configurations of non-residential buildings located in seven European countries. In this way, it was developed a specific equation for each country and three equations that describe each climatic region identified by a cluster analysis; these results were published in [19]. In a second work [18], it was applied the same methodology to a set of data referring to buildings located in the Italian peninsula. In this case, three building analysed configurations, in accordance to Italian legislative requirements regarding the construction of high energy performance buildings, have been employed. The achievement of the generalised results along with a high level of reliability it was achieved by diversifying each individual model according to its climate zone. It was provided an equation for each climate zone along with a unique equation applicable to the entire peninsula, obviously with different degrees of reliability. An improved version of the latest work concerning the Italian case study appeared in the paper published in [20]. The revised model provided an ability to predict the energy needs for both heating and cooling. Furthermore, to simplify the data retrieval phase that is required for the use of the developed MLR tool, an input selection analysis based on the Pearson coefficient has been performed. In this way the explanatory variables, needful for an optimal identification of thermal loads, have been identified. Finally, a comprehensive statistical analysis of errors ensured high reliability. The second analysed alternative method represents an innovative approach in developing a flexible and efficient tool in the building energy forecast framework. This tool predicts the energy performance of a building based some dimensionless parameters implemented through the application of the Buckingham theorem. A detailed description of the methodology and results is discussed in the Chapter D and is also published in [21]. The Buckingham theorem represents a key theorem of the dimensional analysis since it is able to define the dimensionless parameters representing the building balance [24]. These parameters define the relationships between the descriptive variables and the fundamental dimensions. Such a dimensional analysis guarantees that the relationship between physical quantities remains valid, even if there is a variation of the magnitudes of the base units of measurement [25]. The dimensional analysis represents a good model to simplify a problem by means of the dimensional homogeneity and, therefore, the consequent reduction in the number of variables. Therefore, this model works well with different applications such as forecasting, planning, control, diagnostics and monitoring in different sectors. The application of the BM for predicting the energy performance of buildings determined nine ad hoc dimensionless numbers. The identification of a set of criteria and a critical analysis of the results allowed to immediately determine thought the dimensionless numbers and without using any software tool, the heating energy demand with a reliability of over 90%. Furthermore, the validation of the proposed methodology was carried out by comparing the heating energy demand that was calculated by a detailed and accurate dynamic simulation. The last Black-Box examined model was the application of Artificial Neural Networks. The ANNs are the most widely used data mining models, characterised by one of the highest levels of accuracy with respect to other methods but generally have higher computational costs in the developing phase [26]. The design of a neural network, inspired by the behaviour of the human brain, involves the large number of suitably connected nodes (neurons) that, upon applications of simple mathematical operations, influence the learning ability of the network itself [27]. Also in this case, as described in Chapter E, this methodology was applied at the two different energy databases. In [22], the ANN was used to predict the demand for thermal energy linked to the winter climatization of non-residential buildings located in European context, while in another work under review, the ANN was used to determine the heating and cooling energy demand of a representative Italian building stock. The validation of the ANNs was carried out by using a set of data corresponding to 15% of the initial set which were not used to train the ANNs. The obtained good results (determination coefficient values higher than 0.95 and Mean Absolute Percentage Error lower than 10%) show the suitability of the calculation model based on the use of adaptive systems for the evaluation of energy performance of buildings. Simultaneously, a deep analysis of the investigated problem, underlines how to determine the thermal behaviour of a building trough Black-Box models, particular attention must be paid to the choice of an accurate climate database that along with thermophysical characteristics, strongly influence the thermal behaviour of a building [9]. In detail, to develop a predictive model of thermal needs, it is also necessary to pay close attention to the climate aspects. In the literature, many studies use the degree day (DD) to predict building energy demand, but this assessment, through the use of a climatic index, is correct only if its determination is a function of the same weather data used for the model implementation. Otherwise, the predictive model is generally affected by a greater evaluation error; all these aspects are deeply discussed analysing a specific Italian case study in Chapter F, and the main results are published in [8]. The results achieved during the three years of PhD research, make it possible to affirm that each model can be used to solve thermal building balance by knowing merely a few parameters representative of the analysed problem. Nonetheless, some questions may be asked: Which of these models can be identified as the most efficient solution? Is it possible to compare the performances of these models? Is it possible to choose the most efficient model based on some specific phase in the evaluation? To attempt to answer these questions, during the research period it was decided to compare the three selected alternative models by applying a Multi Criteria Analysis (MCA), that explicitly evaluates multiple criteria in decision-making. It is a useful decision support tool to apply to many complex decisions by choosing among several alternatives. The idea rising thanks to the scientific collaboration with the VGTU University of Vilnius, Lithuanian, in the person of Prof. A. Kaklauskas and Prof. L. Tupènaitè, experts in the field of multi-criteria analysis. At the first time a multi-criteria procedure was applied to determine the most efficient alternative model among some resolution procedures of a building’s energy balance. This application required extra effort in defining the criteria and identifying a team of experts. To apply the MCA, it was necessary to identify the salient phases of the evaluation procedure to explain the most sensitive criteria for acquiring conscious, truthful answers that only a pool of experts in the field can provide. Details of this work were carried out during the period of one-month research in Vilnius, from April to May 2019, where it was possible to improve the application of the Multiple Criteria Complex Proportional Evaluation (COPRAS) method for identifying the most efficient predictive tool to evaluate building thermal needs. These results are collected in Chapter G and the main results are explained in a paper under review in the Journal “Energy” from September. The identification of the most efficient alternative model to solve the building energy balance through the application of a specific MCA, allowed to deepen the identified methodology and improve research. In particular, the most efficient alternative resolution model was the subject of the research that took place during the research period at the RWTH in Aachen University, Germany with Prof. M. Traverso, Head of the INaB Department, from September 2018 to March 2019. The experience in the field of LCA and the possibility of identifying the environmental impacts linked to the building system, has led the research to investigate neural networks for a dual and simultaneous environmental-energy analysis. The results confirm that the application of ANNs is a good alternative model for solving the energy and environmental balance of a building and for ensuring the development of reliable decision support tools that can be used by non-expert users. ANNs can be improved by upgrading the training database and choosing the network structure and learning algorithm. The results of this research are collected in Chapter H and published in [28].