Agent-based modelling of urban district energy system decarbonisation – A systematic literature review

: There is an increased interest in the district-scale energy transition within interdisciplinary research community. Agent-based modelling presents a suitable approach to address variety of questions related to policies, technologies, processes, and the different stakeholder roles that can foster such transition. This state-of-the-art review focuses on the application of agent-based modelling for exploring policy interventions that facilitate the decarbonisation (i.e., energy transition) of districts and neighbourhoods while considering stakeholders’ social characteristics and interactions. We systematically select and analyse peer-reviewed literature and discuss the key modelling aspects, such as model purpose, agents and decision-making logic, spatial and temporal aspects, and empirical grounding. The analysis reveals that the most established agent-based models’ focus on innovation diffusion (e.g., adoption of solar panels) and dissemination of energy-saving behaviour among a group of buildings in urban areas. We see a considerable gap in exploring the decisions and interactions of agents other than residential households, such as commercial and even industrial energy consumers (and prosumers). Moreover, measures such as building retroﬁts and conversion to district energy systems involve many stakeholders and complex interactions between them that up to now have hardly been represented in the agent-based modelling environment.


Introduction
Deep decarbonisation of the building sector in the EU is one of the key prerequisites for becoming climate neutral by 2050, as buildings account for around 40% of final energy consumption [1]. In this regard, "zero energy" building concepts, which largely rely on reduced energy demand and on-site renewable generation, have recently gained considerable interest in both scientific literature [2][3][4][5][6][7] and in practice [8]. However, some researchers argue that dense and compact buildings on small plots have a small potential for an on-site renewable generation [2,9] and can hardly achieve zero energy balance. Thus, the expansion of building-level "zero energy" concept to the scale of neighbourhoods, districts and communities is a potential alternative solution. With this motivation, several concepts that aim to acheve zero or positive energy balance, such as Net-Zero Energy Neighbourhoods (or Districts) [2,4,10], Plus-Energy Quarters [11,12], and Positive Energy Districts [13][14][15] are being implemented currently.
Increased interest in such neighbourhood or district-level concepts as a solution for energy and climate issues, raise a multitude of new questions, the most generic of them being: "what socio-techno-economic conditions support the transition of urban districts towards zero and positive energy districts?". More concretely, what policies, technologies and processes can foster this transition? In this context, it is becoming even more critical to understand the perspectives and roles of various stakeholders, including households, firms and public institutions, as their participation (e.g. via energy conservation, prosumption and energy trading, infrastructure development) in transitioning to a decarbonised society can be supported by well-designed and inclusive policies and programs [16,17].
Within a broad selection of models used in energy system analysis [18], Agent-based Modelling (ABM) approach is distinguished by its ability to represent individual decisionmaking of heterogeneous actors, as well as interactions between them [19,20]. Moreover, it is a simulation-type model that allows defining micro-level action and interaction rules, leading to macro-level emergent insights [21]. Hence, it is deemed suitable for exploring policy-related "what-if" questions and incorporating actors' perspectives in the energy system [16,17,22].
This article aims to obtain an overview of how ABM has been used to model policy interventions that facilitate the decarbonisation (i.e. energy transition) of building-related urban district energy systems and consider stakeholders' social characteristics and interactions. We use systematic literature review (SLR) to select the studies and discuss critically the important aspects of ABMs, such as modelling choices and agent characterisation. Hence, this SLR serves as a starting point for those who want to understand how ABM can simulate urban district-level energy transition and contributes with: • A detailed insight on how ABM has been used in modelling urban district's (buildingrelated) energy systems while considering stakeholders and policies; • A discussion of modelling choices and methodologies; • Identification of research gaps and potential application streams.
This paper is structured as follows. Section 2 provides the context to this research topic by defining urban district energy systems and summarising the previous research on applying ABM to model energy systems. It is followed by the description of our approach for the systematic selection and review of the articles in Section 3. The main results of the review are presented in Section 4 and organised in different thematic subsections related to essential aspects of ABMs of urban district energy systems, namely: model purpose and outputs, agents, their decision-making and interaction rules, technologies and policies covered, spatial and temporal aspects, as well as experimental setup of simulations, use of empirical data, and implementation platform used. The paper is finalised with synthesised observations and future research suggestions in Section 5.

Background and Definitions
In this section, we lay down the foundations for the topic of our focus. Namely, we want to refer to the existing literature and define the urban district energy system. Secondly, we discuss the state-of-the-art of ABM's application in the energy systems research.

Urban district energy systems and models
The energy system is defined by the IPCC [23] as: "all components related to the production, conversion, delivery, and use of energy." The energy system is also seen as a socio-technical system, comprised of more than just technical components, but also markets, institutions, consumer behaviours and other factors that affect the construction and operation of technical infrastructures [24].
The differentiation of energy systems into "urban" and "district" is generally about defining the system's scope. In Europe, "urban areas" refer to cities (i.e. densely populated areas), towns and suburbs (i.e. intermediate density areas), as opposed to rural (i.e. sparsely populated) areas [25]. According to the motivation and purpose of this work, we look at the studies that address the energy system challenges of densely populated urban areas.
Depending on various national contexts, "districts" and "neighbourhoods" can denote different administrative and non-administrative areas of cities or countries. Like [6,7,26], we do not refer to certain juridical or administrative areas, but as part of an urban area. Hence, everything from a small to a large group of buildings is considered a "district" within this work. Due to the inconsistent use of the similar terms in the literature, the  [44] synonyms of "district" such as "neighbourhood", "quarter", "block" and "community" are included in the analysis.
It is important to note though, that the search for "district energy systems" brings to district-scale energy systems, be that traditional district-level thermal and hybrid energy systems (e.g. cogeneration) [27][28][29][30] or distributed energy systems such as PV, solar thermal, battery storage [31][32][33][34]. However, consistent with the above-mentioned definitions of [23,24], we keep the scope of "district energy system" broader and do not limit it to the technical components only.
There are various energy system modelling approaches and tools that can be or are used at the district-scale for different purposes [24,31,34]. As [31] conclude about the numerous urban district-level energy models and tools: "some tools aim to provide a single simulation that addresses many issues, while others give detailed results regarding specific parts of the system". Although the advantages of ABM in studying complex systems and enabling the analysis of policies are acknowledged [17,24,[35][36][37], its role in studying district energy systems, to the authors' knowledge, has not yet been explored in detail.

Agent-based modelling in energy systems research
ABM is a modelling approach that can be seen as one of the applications of a software engineering paradigm named "Multi-agent systems" (MAS) [38]. (Some application fields of MAS are represented in Figure 1). There is an ambiguity between MAS and ABM. However, the general understanding is that MAS is an overarching architecture or paradigm, which, when applied for simulating various phenomena by abstracting real-life systems (e.g. human, animals, organisations) is usually called ABM or Agent-Based Simulation (ABS). Whereas MAS-based engineering deals with applying the MAS architecture to create a software or control system, ABM applies MAS paradigm to draw implications about other systems (e.g. human settlements, stock markets, etc.). The common point between MAS-based engineering and ABM is in the desire to understand a complex system by assuming a distributed or autonomous behaviour instead of centralised or equation-governed behaviour of system elements (e.g. like in System Dynamics approach). Hence, the terms "multi-agent-system", "multi-agent-based-modelling" and "agent-based modelling" are sometimes used interchangeably in the literature [39][40][41][42]. However, the difference of these two approaches, namely that ABM sets up agents with characteristics of real-world analogy to see what happens when they act, while in a multi-agent system, agents are defined with certain characteristics, connections and choices, such that they achieve specified emergent states [43].
ABM can, thus, be more specifically understood as a computer simulation of an artificial world populated by agents -discrete decision-making entities (individual, household, firm, etc.) -whose behaviours and rules of different complexity can govern interactions. One of the main reasons for choosing ABM over traditional equation-based modelling approaches in energy systems analysis (i.e. system dynamics, optimisation models, computable general equilibrium models) is its ability to incorporate heterogeneity and adaptivity of energy consumers [45]. In the energy system research, this strength has been exploited for: (a) analysing the demand side of energy system [17], e.g. incorporating occupant behaviour in buildings [46,47]; (b) better-informing policy-making and infrastructure planning [22,36], e.g. determining target groups for interventions [48,49] or recommendations specific to the adoption of particular renewable energy or energy-efficient technologies [3,50,51].
As the number and publication date of review papers indicate (see Table A1 in Appendix B), the first applications of ABM in energy research were for representing wholesale electricity markets to analyse market structures [45]. The possibility of using ABM for questions related to smart electricity grids and markets, such as the integration of demand response and distributed generation in local or centralised markets, is explored by [44]. The potential of ABM to improve our understanding of consumer energy demand, by allowing to account for social, behavioural, economic, technological, and market and policy factors that influence energy demand is presented by [17]. Questions that interest energy economists and policymakers are how consumers adopt energy-efficient technology and how to encourage them. The benefit that ABM can bring to this stream of research, as well as barriers and incentives for the adoption of energy-efficient measures in the residential sector are addressed by [22,36]. Though our review topic overlaps with theirs, we do not focus on the ABMs of "innovation diffusion" only and explore a wider range of approaches.

Methods
This work is based on the literature review type originating in biomedical and healthcare research and becoming prominent in energy system research too [35,36] -systematic literature review (SLR).
The current SLR is carried out on the 13th of September, 2021 in the Scopus database only. The main research question thereby is: "how ABM has been applied in studying the urban district (building-related) energy systems?". Accordingly, the search string provided in the PRISMA Flow diagram in Figure 2 reflects this question. First, the literature suggests many variations of agent-based concepts -simulations, models, approaches, as well as "multi-agent" and "multi-agent-based" simulations, models, and approaches. Although there are differences between MAS and ABM (see Section 2.2), they are sometimes used interchangeably in the literature. Therefore, the studies referring to "multi-agent-based" simulations were not excluded automatically but carefully checked. Second, the search term "energy OR heat*" ensures that all studies mentioning energy or heat are captured. Urban district energy systems are defined here as a group of buildings, heating and cooling infrastructure, distributed energy resources (PV, battery, solar thermal, heat pump, CHP), electricity distribution network, and energy producers, consumers, prosumers and other relevant stakeholders in a given district or city. Hence, we exclude, for example, transportrelated studies, which returned 92 additional records in Scopus. Third, as explained in Section 2.1, "district" is used interchangeably with "neighbourhood, quarter, block, community. Moreover, sometimes city or town-level models are applicable to a smaller scale too. Hence, we considered the article with at least one of the terms.
After a rigorous identification in the Scopus database and removing duplicated records, further screening was performed using Scopus automatic filtering, reading titles and abstracts. Journal and conference articles, written in English, accessible either openly or the research institution's library, and relevant to the energy research were filtered out. Finally, full-text analysis has been applied to ensure the selected studies match the aim of this review. The exact reasons for exclusion together with the full SLR process are presented in Figure 2.
After the papers have been selected, they are qualitatively analysed based on the 4.1. Model purposes and outputs 29 The review by [17] highlights that ABM is well-suited to answer two kinds of energy- 30 demand questions: those related to policy design and evaluation and those related to 31 system design and infrastructure planning. The review process reflects the existence of 32 these two motivations for modelling, of which we only focus on those that are relevant for 33 policy design. These studies evaluate the agents' behavioural response to external stimuli 34 in the form of a policy, regulation, observation or feedback, and peer influence. Rai

39
A model's purpose or objective must be "clear, concise and specific" [52], which is 40 essential for others to understand why some aspects of reality are included in a model 41 while others are omitted. It is because each a model should be a "purposeful" abstraction 42 of reality [55]. The purposes of the 25 selected models are diverse. However, we identified 43 two main thematic clusters: diffusion and exploratory ABMs (see Figure 3). One large thematic cluster is the exploration of technology adoption that has its 45 foundations in innovation diffusion theories [56]. This type of ABM is often named "agent-46 based diffusion model" [22,36,56,57]. They aim to analyse adoptions of energy-efficient or 47 renewable energy technology by households, firms and other entities, often due to certain 48 policy interventions [3,51,[58][59][60][61][62][63][64]. Usually, such models' outputs are the number of adopters 49 or adopted units, energy or emissions saved over time (see Table 1). This approach allows 50 us to observe what factors affect the adoptions of technologies in which ways. The term 51 "diffusion" encompasses concepts like social learning and dissemination [65]. Thus, this 52 approach is also well-suited to represent the dissemination of energy-related practices and 53 behaviours, such as energy-saving [47,49], energy-efficient ventilation behaviour [66,67], 54 user learning (i.e. energy saving) after authoritative smart meter adoption [68], building 55 renovation behaviour [69], weatherisation (i.e. making apartments weather-proof) [70], 56 buying energy-efficient appliances and switching an energy provider [71]. Similar to 57 technology adoption, these studies investigate how energy-related behaviours are adopted 58 and how much energy can be saved. Three models [66][67][68] [47] Explore the effect of social-network characteristics on the diffusion process of energy conservation % energy savings from different feedback methods with various social network characteristics [50] Examine the impact of information diffusion algorithm on residential PV adoption in city neighbourhoods Number of new and total adopters over time [58] Test alternative policy scenarios for PV adoption in a neighbourhood Predict the consumer adoption of different renewable energy models and to determine the resulting impacts on energy system performance Utility and solar installer revenues, total power added to the grid, total number of adopters, number of rooftop PV and community solar adopters over time [62] Determine the effect of PV diffusion on the profitability of utilities % of buildings with installed PV, % of new installations per year, % of demand met by PV, spatial representation of building adoption. [71] Observe the impact of socioeconomic heterogeneity, social dynamics, and carbon pricing on individual energyrelated decisions CO2-emissions over time; avoided CO2emissions by each type of behaviour (investment, conservation, switching supplier) [51] Test the effect of solar rebates on PV adoption Cumulative number of PV systems over time; thematic maps with spatial distribution and density of PV systems adopted [63] Determine the diffusion rate of the green technologies under uncertainties caused by climate change, characteristics of adopters, and their interactions Number of installed technologies over time, under six different policies [64] Assess the impact of switching from the self-consumer paradigm to a jointly acting renewable community on adoption rate of rooftop PV in a city district kW installed over time, number of new adopters per year, spatial distribution, typical daily production-consumption profile [68] Study user learning in authoritative technology adoption based on the case of smart meter deployment in Leeds Average daily electricity load curve (kW), number of experienced users, agents' attitude and energy-saving awareness over time instead of instantaneous decisions (e.g. to adopt, to invest). In these studies, the output 67 metrics are very specific to the purpose and subject studies (see Table 2). 68

69
Agent is a key element in this modelling approach. Many previous studies highlight 70 that there is no common definition of an agent [44,78], as its properties depend on the 71 model's purpose and application area. Nevertheless, many authors refer to the following 72 basic definition presented by [79]: "Agent is an encapsulated computer system that is 73 situated in some environment, and that is capable of flexible, autonomous action in that 74 environment in order to meet its design objectives". In the ODD protocol, agents are one 75 of the model's "entities", along with spatial units and the overall environment [54]. It is 76 due to the parallels between the agent-based modelling approach and Object-Oriented 77 Programming (OOP) (i.e. the 'classes' or its instances in OOP could be equivalent to 78 'entities' in ABM). It might lead to confusion among readers who are new to Agent-based 79 modelling or use different implementation tools. In the current article, we differentiate 80 between agents and other entities, where we refer to "agents" as autonomous entities that 81 can make decisions (i.e. implement certain algorithms) and interact (i.e. obtain information 82 from its environment or other agents) in order to reach its objectives. 83 Most of the agents in the selected studies are "households" (15 out of 25) and three 84 studies also denote them as "energy consumers" [3,68,71] (see Table 2). Since most of these 85 studies model the adoption of PV or other technologies, "households" are most common 86 decision-makers in this regard. Majority of these models limit their agent population to the to represent group decision-making regarding heating system, insulation or RE system 93 installation in multi-family houses. In other models, building (or building block) owner 94 [60,69] and building agents [59] can make building-level decisions, i.e., adopting PV or 95 renovation. The rationale of these models is that there is only one building owner that can 96 make such a decision. 97 While the above-mentioned studies focus predominantly on one type of stakeholder, 98 there are few models that involve different types of stakeholders as agents [73]. For 99 example, in [73], instigator agents (i.e. local authorities, commercial, and community-based 100 developers) are driving the development of projects, whereas "projects" are management 101 agents responsible for carrying out actions on behalf of their parent instigators [73]. In 102 models with multiple types of stakeholders, it is becoming more challenging to draw a 103 line between agents and other entities, e.g. as in [47], as all of them are essentially realised 104 as classes. However, one can observe the tendency to call human-like entities "agents", 105 e.g. instigator agents, and passive entities like grid cells and projects [73] as just "entities".
106 Figure 4 summarises the types of agents we identified in the reviewed models.

107
The essential part of ABMs is decision rules that govern the actions of agents. Decision 108 rules are realised with the help of attributes that describe agents [43]. Moreover, interaction 109 and social influence play a significant role in agent's decision making. Hence, the following  translate into agent action [43]. Behaviour is the overall sum of agent actions and state 116 changes [43]. However, authors often use the terms "actions", "behaviours" and "decisions"

117
interchangeably [80]. The ODD protocol suggests to include a detailed description of 118 Number of heating systems adopted at certain combination of time horizon for all, changes in natural gas price and electricity price, fraction of households that is able to compare combined investments [76] Explore how group decision-making in strata buildings could affect the heat transition in the owner-occupied share of the housing sector in the Netherlands Individual preferences for thermal systems at the beginning of the simulation, group lock out (when the Homeowner Association can't agree on the decision), cumulative heating costs over time [69] Explore the development of the renovation state of the building stock based on renovation behaviour of different types of homeowners Development of overall heat demand (GWh/a) and number of buildings renovated in the city over time [77] Analyse the effect of behavioural outcomes in different policy situation due to the influence of energy-saving behaviour and intentions Descriptive statistical mean values of different situational factors [48] Find the near-optimum targets among a social network of households in order to participate in a typical Energy Efficiency Program (EEP) Energy Index that changes due to the EEP or the social interactions [49] Investigate participants' related factors that can affect short-term and long-term effects of these programs Short-term (right after the eco-feedback program) and long-term (after interactions with other agents) efficiencies of the program individual decision-making [81]. The information such as identifying subjects and objects, 119 the method, the uncertainty, and other aspects must be part of this documentation [81]. 120 However, in practice, such protocols are rarely adhered to by the authors.

121
The articles describing the diffusion ABMs are more explicit about the decision-making 122 algorithms. In such models, agents decide to adopt or not adopt (i.e. to invest or not invest 123 in a certain technology or to perform a certain energy-related action) based on specific rules 124 or algorithms. Decision rules range from simple ad-hoc rules to most elaborate models, 125 such as psychosocial or cognitive models [43]. The classification of existing decision models 126 has been previously done by [80] (for human agents in ecological ABMs), [56] and [57] (for 127 agents in ABMs innovation diffusion) and [43] (for ABMs of socio-technical systems). The

128
ODD+D by [81] clusters agent decision algorithms based on the nature of the underlying namely, psychosocial (also called "socio-psychological" or "cognitive") and microeconomic 137 models. Psychosocial models are based on social psychology theories that assume that 138 human decisions are based on psychological rules, rather than on rational economic 139 rules. The most frequently used psychosocial theory in the selected models is the Theory 140 of Planned Behaviour (TPB) by [82]. It states that human behaviour results from the 141 intention to perform the behaviour; individual attitudes, subjective norms, and perceived 142 expectations can influence the agent to perform such behaviour [83]. Usually, the more 143 favourable these three aspects of human psychology are, the stronger is the person's 144 intention to perform a certain behaviour [83]. The standard form of TPB is static, i.e. it 145 describes how these three components are translated into intention and action at a given 146 time. The models by [51,66,67] are examples of implementing this theoretical model. Other 147 psychosocial models including "consumat" model by [84] in [68], Norm Activation theory 148 by [85] in [71], the goal-framing theory by [86] in [74], and Influence, Susceptibility, and 149 Conformity Model by [87] in [49], are also used. Several models rely on models from Ansari [47] draw on the opinion dynamics models by [88][89][90] to represent the effect of 152 energy feedback interventions among building residents.

153
Another class of frequently used agent decision-making model is the empirical-based 154 heuristic models. They are described as models "not built on any grounded theories" and 155 "having the impression of being ad-hoc" [57]. Agents are often assigned rules derived from 156 empirical data, and also model parameters are selected such that results match simulated 157 output against a real-life observation [57,80]. They might not represent the process of 158 agent decision-making very accurately or realistically, but have the advantage of being 159 easy to implement and to interpret [57]. Heuristic decision rules can be implemented 160 in various ways. Several modellers favour data-driven approaches, thus, implementing 161 machine learning algorithms, such as logistic regression models [59] and artificial neural 162 networks [77]. In this approach, several sets of factors that can affect the adoption of PV 163 or energy-saving behaviour, given that data about those factors are available, are tested.

164
The more qualitative approach is followed by [72] and [73], who created the decision rules 165 relying on the stakeholder's expertise.

166
Some models rely on ad-hoc rules without any validated theory or empirical ground-  Table   176 3. in Figure 6). two studies have not considered agent interactions in any way [62,69]. In [73] and [77], 193 interactions are considered as important, however, treated in an abstract and implicit way.
194 Table 3 shows how interactions are represented in each reviewed study.

195
The majority of studies which include agent interaction agents are often placed in a 196 network structure, often called "social network", that imitates the relationship between on each other's behaviour than strangers [66,67]. In some cases, agents interact based on 214 similarity (also called 'homophily') [3] or geographical proximity [51] ('neighbour effect').

215
Another choice that a modeller should take is regarding the frequency of interactions. 216 Huang et al [70], for example, let agents that are linked with each other interact every time   Table 3. Agents, decision frameworks and representation of agent interaction [47] occupants and buildings

Study Agents Decision framework Interaction
Theory-based: several opinion dynamics models Opinion dynamic models (information exchange within own social networks with following topologies: small world, scale free, and random) [50] households    Table 4). Majority of these studies consider the diffusion of a single technology: 233 rooftop PV [50,59,60,62,64], feedback device (CO2 meter) [66,67]. In some cases, there could 234 be several options are available for agents: [3,61] let agents choose between buying PV via 235 cash payment of a loan, adopting community solar (i.e. renewable energy community) or 236 opting for green electricity; [63] make agents choose the optimal solution for their rooftops 237 -either rooftop PV or green roof; [76] introduces the combinations of technologies as 238 "technology state" of a household (i.e. combination of heating system, insulation level, and 239 appliances). Building insulation or renovation is addressed in three studies [69,70,75]. Most  of the first type of policies are those that stimulate PV system investments [51,60,63,64], 261 assistance programs for weather-proofing [70], and promotional campaigns for feedback 262 devices [67]. The examples of the interventions for stimulating energy-saving are energy 263 feedback mechanisms [47]. Beyond these clusters, [71] introduces several carbon emission 264 price scenarios to see how it affects the emissions caused by household energy consumption.  Identifying the spatial and temporal scale of the models is important in order to un-279 derstand the system modelled. Moreover, certain patterns and processes can be dependent 280 on the scale [94] and, thus, they need to be clearly stated. By spatial scale, we mean "geo-281 graphic scale", defined as a research area's spatial extent in a study [94]. The geographic 282 scale of the models considered range from "group of buildings" [47] to an entire city, such 283 as Hamburg [69]. 16 studies describe community, or district, or neighbourhood-scale mod-284 els, while nine studies are in city-scale [51,[67][68][69]73,77]. Although these articles present 285 the models as having been applied to specific geographic scales (i.e. via case studies), it is 286 difficult to say if they can be scaled up or down, as it might depend on many factors.

287
The chosen scale in ABM usually determines the number of entities (i.e. agents) 288 covered [33]. This can be limited by computers' processing capacity, especially if decision 289 algorithms are sophisticated, much data is used, or a considered city is very large, e.g. 290 like in [59]. Therefore, the majority of selected models opt for district or neighbourhood 291 scale. Those whose models are in city-scale focus on smaller cities of about 100-150,000 292 [62,66,67]. Only one model has modelled a city of approx. 174,000 households and the 293 simulation had to be carried out on a supercomputer [51]. There are also such models 294 whose scale depend on the topic of research. For example, DH network development is 295 usually city-scale phenomena [73], the development or properties of energy communities 296 are explored on a neighbourhood or district level [3,72].

297
Although traditionally ABMs have not focused on the geographic environment and 298 spatial representation, more and more models are striving to represent space explicitly 299 and realistically (e.g. using GIS techniques) [95]. According to [95], models can have three 300 levels of spatial explicitness: 1) implicit and non-geographic representation of space (e.g. 301 social networks that are only partially tied to space); 2) explicitly represented but abstract 302 in how it maps onto reality (e.g. Schelling's segregation model); 3) explicit and realistic 303 spatial representation. Among the reviewed models, only a few are spatially explicit and 304 realistic. For instance, [51,58,59,64]   Empirical grounding of ABMs is becoming more important, especially for models that 323 aim to reflect a specific real-world situation and provide decision support for policymakers 324 and stakeholders [57,96]. As opposed to hypothetical or theoretical (or highly abstract) 325 ABMs, empirical ABMs use real-life data to parameterise models, initialise simulations, and 326 evaluate model validity [57]. Modellers try to improve the realism of agent decision-making 327 algorithms by consulting with system-relevant actors [72,73] or relying on empirical data 328 [59,64,66], e.g. geospatial information on buildings. It is becoming more feasible due 329 to the contemporary trends we observe the availability of high-resolution data sets, the 330 spread of open data culture in science, advances in data analytics, machine learning, and 331 computational power. Therefore, we aim to assess for what purpose, what kind of, and 332 how empirical data is used in the selected ABMs of district energy systems. By empirical 333 data, we mean both qualitative and quantitative data based on observation or experiment.

334
The review by [36] highlights that empirical data in ABMs are used for two general information about surveys or stakeholder interviews [73,74]. 353 In general, there are three processes in model building where the use of empirical data 354 make models more reliable and realistic: parametrisation, calibration and validation [37]. 355 The parameterisation is the process of connecting model and target system (i.e. the real 356 system being modelled) via assigning the set of parameters and their values to enable sim-357 ulation [96]. In line with observations of [37], only a few modellers explicitly differentiate 358 their modelling process into these three phases. Moreover, if calibration and validation 359 are somewhat known to data-driven modellers, the process of parameterisation is not 360 recognised as much. Among the selected models, only [66,67] describe parameterisation in 361 more detail: they select the parameter values to reflect the empirical patterns of ventilation 362 behaviour adoption derived from survey data.

363
Calibration is the adjustment of parameters to ensure that model output matches the 364 relevant empirical data, e.g. in a specific location and application [37]. The difference 365 to validation is that the parameters are tuned to match a specific context (i.e. location, 366 time), which does not necessarily mean that the model will exhibit accurate results and be 367 predictive upon application in another context. To achieve that it has to be first validated on 368 a separate set of data independent of data used for calibration [57]. The following models 369 describe how they calibrated their models: [62] calibrates the parameters of the logistic datasets; [66] provides an indirect calibration with three empirical patterns, the same 374 used for parameterisation in [67]. As for the remaining models, some do not differentiate 375 between validation and calibration [60], some call calibration "model fitting" [51], but 376 the majority do not mention calibration at all. Often authors mention the lack of data for 377 calibration as their limitations [63,73].

378
Validation aims to achieve the matching between the observations of the models and 379 reality. It should not be confused with "verification", which is the process of making sure 380 the model implementation is carried out correctly with respect to the conceptual model 381 [97]. As ABM is a highly multi-disciplinary and flexible framework, its validation is a 382 highly debated topic. For more detail, we suggest referring to the works of [98] and [57]  adoption with empirical adoption level for the period starting after the last date in the 396 calibration dataset. Also, they carry out temporal, spatial, and demographic validation 397 [51]. Another group of modellers [47,49,73] pay more attention to the validation of social 398 processes and, by drawing on the work of [99], offer conceptual, operational or structural, 399 and technical validation (by this, [47] refer to verification). Conceptual validation is the 400 process of determining that the theories and assumptions underlying the conceptual model 401 are correct [99] and usually achieved by basing the model on validated concepts [47,49] or 402 the insights from stakeholder workshops [73].  Such adaptations are often study-specific, and therefore, some essential modelling details 421 may get lost or unclear to the audience without careful and standardised documentation.

422
In this regard, the ODD protocol provides an essential standardised framework for model 423 documentation.

424
Our analysis shows tremendous potential in ABMs to help policymakers make better   The following abbreviations are used in this manuscript: Technologies studied, barriers to the adoption of energy efficiency, policy measures that are explored using the ABMs, theories used to describe decision making of households and the use of empirical data Modelled policies: subsidies, regulation and taxation, technology ban, household adoption obligation and various information campaigns. Many of the models are rooted in the TPB, use utility functions, and/or use empirical data. [22] Application of ABM for understanding technology diffusion of residential energy efficient technologies and to evaluate policies' effects on adoption.
selective -Types of ABM approaches (both theoretical and empirical); applicability and limitations of ABM for modelling of the uptake of en-eff tech-s in energy sector Key components of ABM for describing the adoption and key decision when intending to model the uptake of energy-efficiency technologies. ABM can model technology diffusion with at least the same accuracy as equation-based modelling when appropriately parameterised based on empirical data, calibrated based on macro-level data, and validated using sensitivity analysis.  [17] ABM work in the area of consumer energy choices, with a focus on the demand side of energy to aid the design of better policies and programmes selective, critical about 60 Limitations of non-ABM approaches, framework for describing the essential features of ABM, use of ABM in practice Two major types of energy-demand questions that ABM is well-suited to answer: those related to policy design and evaluation, and those related to system design and infrastructure planning. [44] Application of ABM for analysing smart grids from a systems perspective selective 23 How ABM can be used to analyse electricity systems; typology of agent-based research of electricity systems; review of literature specifically studying smart grids using ABMS techniques is reviewed ABM is still a limited field of research, but can deliver specific insights about how different agents in a smart grid would interact and which effects would occur on a global level. Valuable input for decision processes of stakeholders and policy making. [45] Overview of AB electricity market models and present the most relevant work in detail.
selective 31 Comparison of current AB electricity models, Methodological questions: Agent learning behavior, Market dynamics and complexity, calibration and validation, Model description and publication.
Choice of specific learning algorithms, more careful and well documented validation and verification procedures as well as the appropriate publication of details of concrete simulation models are crucial for the further development of AB electricity market modeling. [101] Study of the ABM simulation packages for electricity markets selective 4 Overview of electricity markets, general-purpose ABS tools to introduce some background of ABS, detailed study of four popular ABS packages for Electricity Markets (SEPIA, EMCAS, STEMT-RT, NEMSIM).
ABS packages are divided into 2 types: toolkit (Netlogo, Repast) and software (AnyLogic, AgentSheets) Rebates for low-income households (i.e. households in the bottom quartile of wealth, proxied by home value).

Appendix B Technologies and Policies
[63] PV, green roof Adoption Investment Tax Credit, promotional campaigns [64] PV Adoption Self-consumption scheme (PV electricity is sold at market price) and Citizen/Renewable Energy Community scheme (share the electricity produced by a single PV unit with many citizens, e.g., in a condominium) Electric appliances, insulation purchase No policy described, but the model is capable [70] Weatherproofing ("weatherization" for winter) technology Adoption Publicly funded Weatherization Assistance Programs that are intended to help lowresource residents improve the energy efficiency of their homes [75] Insulation, renewable heating investments in new technology No policy; changes in natural gas price and electricity price are taken as proxies for market forces and policies [76] insulation, renewable heating investments in new technology Fiscal policy (i.e. linear growth of natural gas taxes, taxes on electricity, and regulated price of heat from networks) and disconnection from gas network. [69] Renovation technology renovation decision No policy [77] No technology energy-saving behaviour Range of external situational factors are tested: social norms related to energy saving, popularization of economic energysaving policies, etc. [48] No technology energy-saving behaviour No policy; insights for EEP [49] No technology energy-saving behaviour No policy; insights for normative interventions (ecofeedback programs)