Exploring Simulation Workflows, Tools, and Metrics for Beyond-Vision Effects in Multi-Objective Optimization: A Scoping Review

Lighting in the built environment affects different aspects, ranging from building performance in terms of costs and energy consumption to human well-being in terms of thermal comfort, visual effects, and beyond-vision effects. Buildings optimised for specific performance metrics rarely consider different aspects simultaneously, leading to sub-optimised, unbalanced, or non-trade-off solutions. Therefore, multi-objective optimisation has commonly been used to overcome conflicting performance objectives. Recently, light effects beyond vision gained more interest in building design but it is unclear if and how they are integrated with other existing building performance metrics and simulation workflows. A scoping review investigated the state-of-art in multi-objective lighting design optimisation regarding building performance and human well-being focusing on performance metrics, simulation workflows, and the overall information flow. Results show that metrics for beyond-vision effects are not integrated with other human well-being and building performance metrics. The simulation workflows included multiple steps and computational tools in multi-objective optimisation. This process has limitations such as a long simulation time, no ubiquitous integrated tool, and a reduced information flow.


Introduction
The concept of the built environment encompasses the physical and spatial surroundings that humans create for various activities such as habitation or work.The quality of the built environment profoundly influences human well-being, comfort, and overall quality of life.An essential element of this influence lies in the provision of adequate lighting.To create a sustainable, good-performing lighting environment, a broad spectrum of design variables must be considered, i.e., building envelope properties, building services' systems, occupants' dynamic behaviour, and life-cycle costs.Lighting is one of these variables and must be balanced with many other, sometimes conflicting, performance objectives [1,2].Earlier analysis and building performance simulations have shown that optimising building performance for energy and cost provides a great saving potential.However, it is accompanied by a great risk that it contributes to a sub-optimisation of the building's value at the cost of other values, e.g., user comfort and lighting quality [3].
Building optimisation can be performed using different multi-objective/multi-criteria optimisation methods which Yu et al. [4] have categorised into five different optimisation levels.Levels M1 and M2 investigate all aspects separately (M2 adds an overall comparison) and make no trade-offs.Level M3 ranks different criteria in a prioritised order, and only solutions that satisfy the first criteria continue for further analyses reducing the solution sample.Level M4 combines different criteria into a single metric.
The highest level M5: multi-objective optimisation (MOO) analyses all criteria simultaneously to balance all objectives and find trade-off solutions.Due to the high complexity of optimising different building design aspects simultaneously, solving this task without integrating computational tools is unfeasible.
Recent studies have increasingly used simulation-based multi-objective optimisation.It is a mathematical optimisation technique that explores solutions simultaneously satisfying multiple potentially conflicting performance objectives [5].Solutions that consider trade-offs between different objectives, i.e., not improving one objective without worsening the other, are called Pareto-front solutions [4].It can effectively solve dynamic building design optimisation problems by mimicking its evolution process [6].The MOO simulation workflows often include a design tool, a visual programming language, and a building performance simulation tool [7].Using a design tool provides the means to manipulate architectural elements and lighting parameters, while the visual programming language enables the creation of complex algorithms and design processes considering multiple objectives and constraints.Consequently, the information input and digital model complexity increase rapidly [8].Therefore, standardisation and digital support are required to optimise human-driven, multiobjective lighting design [9].
Optimising the built environment performance is not limited to managing energy consumption and life-cycle costs.It balances visual and non-visual light effects, a methodology now summarised by the International Commission on Illumination (CIE) as 'integrative lighting' [10].This integration must coexist with other, sometimes contradicting, design criteria.Light affects human physiology and psychology, including vision, circadian rhythm, sleep, melatonin suppression, heart rate, alertness, and (non-) seasonal depression [e.g., 11,12].The rods and cones (S, M, L) are the four classical photoreceptors for vision.A fifth type of photoreceptor, the intrinsically photosensitive retinal ganglion cells (ipRGCs), mainly controls the light's non-visual responses based on the photopigment melanopsin [13] and its sensitivity peaks at a shorter wavelength.However, as all five α-opic photoreceptors can contribute to both visual and beyond visual responses to light [14], it is more straightforward to distinguish between light effects related to 'vision' and 'beyond vision'.
Assessment of beyond-vision effects requires appropriate quantification.However, established photometric light measures such as photopic lux, are inadequate [15].Recently, recommendations were made proposing 'α-opic Equivalent Daylight Illuminance values' (in lux) as metrics to specify the impact of lighting on either of the five α-opic human photoreceptors.Metrics like 'melanopic equivalent daylight illuminance' (m-EDI) are based on the spectral sensitivity of the ipRGCs [14,16].The building certification institute WELL [17] recommended to use 'equivalent melanopic lux' (EML) based on the ipRGC spectral sensitivity curves [15].Other metrics are based on nocturnal suppression of the hormone melatonin such as 'circadian stimulus' (CS) which is a non-linear model based on the quantity and spectrum of light [18] and the 'non-visual direct response' (nvRD), a model that takes a time series of eye-level light stimuli as input to predict human alerting response [19].
Lighting simulation software is pivotal in architectural and lighting design optimisation.Conventional software simplifies the visible spectrum into RGB colour space, but recent spectral simulation tools, like Lark v2.0 [20] and ALFA [21], specialise in modelling effects beyond human vision.These innovative tools divide the visible spectrum into multiple channels, with Lark using nine and ALFA 81 channels and calculate metrics for beyond-vision effects directly.Both tools support static point-in-time simulations.Lark offers annual simulations with the daylight coefficient method for dynamic daylighting [20].Meanwhile, ALFA uses spectral ray tracing to predict light absorption by αopic photoreceptors based on location and view direction [21].
In light of the growing awareness of the profound impacts of beyond-vision effects on human health and well-being and according to the review done by Gkaintatzi-Masouti et al. [1], new simulation workflows have been introduced to better predict and integrate these effects into the design and optimization of architectural environment by considering luminous and temporal factors of light, specifically light quantity, spectrum, directionality, timing, duration and previous light history [22].These advanced simulation workflows provide practitioners and researchers with a detailed understanding of how lighting affects both the built environment and human occupants.They enable better decision-making and the creation of spaces that prioritize overall well-being.However, even though multiple studies may have combined, i.e., light-related indoor environmental quality with human comfort, energy efficiency and life-cycle cost factors, only recently, light effects beyond vision gained more interest in building design.Multi-objective optimization allows for the simultaneous consideration of various factors, but it is unclear if and how beyond-vision effects may interact with other building performance and human well-being aspects using the previously mentioned performance metrics, simulation tools and workflows.Therefore, a scoping review of the literature investigated if and how beyond-vision effects are integrated in multi-objective lighting design optimisation in terms of performance metrics and simulation tools and whether the integration of beyond-vision effects influences the simulation workflows.

Method
A scoping literature review was performed according to the PRISMA procedure for scoping reviews.The review process started with defining the scope and research question using the mnemonic Population, Concept, Context (PCC) framework [23].Population was not applicable since there was no focus on a certain group of people (i.e., age, gender).Three search concepts were applied, see Table 1.The search in Scopus and Web of Science databases was performed within title, abstract, and keywords and limited to scientific journal/conference publications after 2002 ((re-)discovery ipRGCs).Only English and available full text were included.Papers were excluded if they investigated building types with a specific (short stay) use (e.g., museums, airports), reported subjective evaluations and did not use computational modelling, investigated building performance aspects with less relevance for the thermal-visual-energy-cost balance (e.g., acoustics), or investigated effects on users/occupants with special conditions and needs (e.g., people with Alzheimer's Disease).Additionally, studies were excluded if the multiobjective/multi-criteria optimisation was lower than M5 [4].
The paper selection process is shown in Figure 1.From a total of 662 papers originally retrieved, 21 papers reached the highest optimisation level.The included papers were imported into NVivo 14 software for further (qualitative) analysis.The information extracted from each paper are presented in Table 2 and included the simulation level (building, floorplan, room), location, if available, and the type of building (office, educational, residential) to define the function of the building.Input properties include information about the case study in each paper and describe the fixed variables of the building studied like simulated light source (daylight, electric light), window, surface, envelope, occupancy, and HVAC properties.Design variables represent parameters that can be adjusted (independent variables) in the built environment or the test object.Different limitations about the aspects, design variables, simulation tools, or any presented columns in this table, were also mentioned.

Results
Focused on the issue of MOO, this review investigated if and how beyond-vision effects are integrated in multi-objective lighting design optimisation in terms of performance metrics and simulation tools and whether the integration of beyond-vision effects influences the simulation workflows.The information was categorised according to the building performance and human well-being aspects involved (energy consumption and life-cycle cost, visual effects both comfort and performance, beyond-vision effects, thermal comfort).The specific metrics from each aspect were reviewed, and the combinations of the aspects were investigated.The simulation workflow of multi-objective optimisation and software tools were presented.The main results are shown in Table 2, presenting the results of 21 papers in M5: MOO level.

Performance metrics and aspects combination
In Table 2, the aspects column shows which building performance and human well-being aspects are being optimised: energy consumption, life-cycle cost, visual performance, visual comfort, thermal comfort.Interestingly, beyond-vision effects were not investigated by any of the papers.The plus/minus +/-signs indicate if the objective is to maximise or minimise these aspects.In the papers, these aspects are presented in measurable metrics; most used ones are heating, cooling, lighting energy consumption (kWh), operational cost, useful daylight illuminance (UDI), daylight glare probability (DGP), and indoor air temperature, respectively.Visual performance was the most investigated aspect present in all studies.For daylight, there are static metrics like daylight factor (DF) and point-in-time illuminances, and dynamic ones like daylight autonomy (DA), useful daylight illuminance (UDI), or annual sunlight exposure (ASE).
Depending on the case study, the different aspects were not always conflicting with each other.For example, Lan et al. [33] investigated four aspects (visual performance, thermal comfort, energy consumption, and life-cycle cost) in net-zero energy buildings (NZEB) and found that the social aims do not conflict with the environmental and economical aims.The results show that designing for daylighting and natural cooling contributes to the improvement of energy efficiency and cost effectiveness as well as better visual performance and thermal comfort.In contrary, Rabani et al. [27] found that thermal comfort and energy efficiency have a positive correlation while they conflict with visual performance.

Simulation workflow
According to most reviewed studies, the simulation workflow in MOO starts with formulating the problem after preparing the building model and setting the required parameters in the available modelling, simulation and optimisation tools (see Figure 2) [8, 24-31, 33, 35, 37, 39, 42].Problem formulation includes defining the objective function (e.g., reduce total energy consumption, discomfort glare and thermal discomfort rate), defining the design variable(s), and defining the algorithm parameters or constraints (e.g., illuminance value > 300 lx for visual performance).Design variables can be active variables relying on the use of mechanical and electrical systems, or passive variables relying on natural phenomena to regulate the indoor environmental quality [33].
Next step includes the use of metaheuristics optimisation algorithms which stochastically generate an initial population ('chromosomes') to respond to an optimisation problem where parameters have been identified and encoded.The fitness of each member of the population or chromosome is then assessed using a fitness function or optimal target value for each algorithm parameter.The best fitted parents or chromosomes are then selected to generate a new generation, while poor performers are discarded.This procedure is then repeated using crossover of two or more chromosomes or mutation of random chromosomes and each time fitness values of the new offspring is evaluated until the optimal design solution is reached (see Figure 2).
The selection of an optimal solution could be done by the ranking method or Pareto front according to Sadegh et al. [42].Ranking method typically involve aggregating the objectives into a single score, and then ranking the solutions based on that score, while the Pareto front represents a set of nondominated solutions that capture the trade-offs between the objectives meaning that no other solution can improve one objective without worsening the other.Sadegh et al. [42] investigated the design and development of kinetic façades to optimise visual performance in terms of daylight autonomy (DA), useful daylight illuminance (UDI) using genetic algorithms and both ranking methods and Pareto front.They found that ranking method may provide a limited set of solutions while Pareto front provides a more comprehensive and diverse set of optimal solutions.The ranking method might be beneficial when one objective outweighs other objectives in terms of desired design parameters.The process of implementing genetic algorithms (GA) (adapted from Harish [31])

Metaheuristics optimisation algorithms
Two other studies have used hybrid Genetic Algorithm -Particle Swarm Optimisation (GA-PSO) [27,31], to optimise controlled ventilation system and energy control system, respectively.Ahmad et al. [25] used canonical generational Genetic Algorithm to optimise scheduled operation of window blinds.Kurian et al. [32] used ANFIS controller (Adaptive Neuro-Fuzzy Inference System) to optimise window blind controller and Tomažič et al. [38] used Fuzzy logic to optimise building-management systems (BMSs).Sun et al. [43] used stochastic dynamic programming (DP), rollout technique to optimise integrated shading and ventilation system.Fan et al. [30] used image density atlas (IDA) to provide decision-making basis for facade shading ratio (FSR) optimisation.
Two studies by Ayoub [40] and Kim [41] applied cellular automata (CA), which is a mathematical model consisting of a grid of cells that evolve over time according to a set of rules based on the states of neighbouring cells.They aimed to find the applicable cellular automata (CA) parameters and daylighting design criteria by using generative design and genetic algorithm to build an innovative design strategy to search the adaptive building facade.The quantitative evaluation method of CA rules controlled average opacity and perforation rate to ensure an optimal distribution of the natural daylight that enters the internal spaces, and to give an interesting visual appearance.
To ensure the validity of the optimal solution, some studies have tested the same design variables and measured metrics in real-life in comparison to simulation, e.g., [27].Lavin and Fiorito [29] and Mangkuto et al. [28] validated the simulation tool before generating the optimal solutions by comparing daylight factor in measurement and simulation.Others validated the results of one tool or framework with another simulation tool, e.g., [26].

Simulation tools
The simulation workflow included multiple steps and computational tools in multi-objective optimisation.Many studied cases started with parametric modelling of the building by using the Grasshopper and Rhinoceros 3D software [8,28,35,37,[39][40][41].Then multiple simulation tools and plug-ins were used for various purposes (e.g., Ladybug, Honeybee, DIVA).Kim [41], for example, used DIVA to perform daylight analysis on an existing architectural model via integration with Radiance and DAYSIM.Ayoub [40] conducted illuminance and energy computational simulations using Diva-for-Rhino and Archsim plugins that interface Radiance and EnergyPlus simulation engines respectively.Mangkuto et al. [28] also used DIVA-for-Rhino that works on validated simulation engines for daylighting (Radiance) and building energy use (EnergyPlus), both of which are open sources.To simulate the effect of adding shading on moderating glare and daylight illuminance, Sorooshnia et al. [39] used DIVA plugin and ClimateStudio.In general, Solemma based simulation tools were commonly used in MOO [21].
In the next step, building performance tools are coupled with optimisation tools to enable applying metaheuristic algorithms.For example, Lan et al. [33], coupled JEPlus+EA, a multi-objective optimisation tool based on NSGA-II, with EnergyPlus to explore optimised solutions and to seek insights into the design parametric space that leads to zero energy systems.Another example where no parametric modelling was applied was done by Rabani et al. [27].They used the dynamic building energy simulation software IDA-ICE to carry out daylight simulations through the Daylight-tab in IDA-ICE that uses backward raytracing and Radiance as a simulation engine.The software was coupled with GenOpt as the optimisation engine.Furthermore, a detailed thermal and visual comfort analysis of all scenarios were conducted through coupling of IDA-ICE with OpenFOAM, which is open source computational fluid dynamics (CFD) software [27].
In several studies, formulas were used to calculate some metrics in combination with the parametrised model and simulation tools [24,27,[31][32][33]43].In one of the studies by [43], only formulas in MATLAB were used to optimise integrated control shading and ventilation systems.A lot of statistical models, analysis techniques and mathematical framework were presented in the reviewed studies.For example, statistical models like regression analysis and polynomial functions offer insights into relationships between design variables and objectives.This understanding helps designers make informed decisions [32].Techniques like lexicographic ordering allow designers to prioritize objectives based on their significance.It can lead to more efficient optimization by focusing on critical objectives first [42].Techniques like Standardised Rank Regression Coefficients assist in determining appropriate objectives' weights.This ensures that objectives are appropriately balanced in the optimization process [33].Sensitivity analysis helps identify how variations in input variables affect objective responses.This information aids in designing solutions that are robust against uncertainties [43].

Discussion
The study aimed to investigate if and how beyond-vision effects are integrated in multi-objective lighting design optimisation in terms of performance metrics and simulation tools and whether the integration of beyond-vision effects influences the simulation workflows.By analysing the collected data and comparing it with previous research, insights into the broader implications of this work is gained.
The MOO in most studies focused on maximising visual performance and minimising energy consumption.Most studies included three or more aspects [26,27,[29][30][31][32][33][34][35][36][37] but mention in the limitations that more aspects with more metrics and design variables must be tested to achieve higher optimisation reliability and obtain a virtual environment that would accurately imitate the real-world conditions.The reviewed studies all lack an essential aspect: optimizing lighting design to incorporate beyond-vision effects alongside building performance and well-being considerations.Notably, these effects were not simulated at the user or product level in any of the studies.This gap highlights the need to comprehensively integrate beyond-vision effects for a more holistic lighting design approach.However, multi-objective optimisation allows adding more objectives and aspects with multiple metrics, meaning that beyond-vision effects can be added to the problem formulation by following the same workflow mentioned before, e.g., the objective could be reducing circadian stimulus at night to benefit beyondvision effects.Achieving this would entail advanced simulations to capture intricate human perceptions and lighting product attributes [44].Closing this gap could lead to designs that balance efficiency and occupant well-being more effectively.
Optimisation process often includes aspects with conflicting goals regarding metrics combinations.Many studies found that not all social, environmental, and economic aspects and their metrics are conflicting in results [27,29,30,[32][33][34][35]37].This agrees with Lan et al. [33] who found that adaptive shading could reduce glare, solar gains, and thermal temperature, and reduce energy and operational cost for cooling while balancing daylight level and electric lighting.Many studies used lighting quantity metrics like illuminance for evaluating visual comfort in combination with lighting quality metrics like glare and uniformity.Mixing the use of metrics in assessments of the incorrect aspect may lead to reduced quality of the optimised design solution.In addition, established photometric light measures such as illuminance, are inadequate for quantifying beyond-vision effects [15].Therefore, it is important to distinguish between visual and beyond-vision effects and use appropriate metrics [45,46].
The MOO simulation workflow included multiple steps and computational tools.Many studied cases started with parametric modelling of the building, then multiple simulation tools and plugins were used for various purposes, which were coupled with optimisation tools and algorithms [8, 24, 27-30, 32, 35, 37, 39-42].This process has some limitations like long simulation time, no single integrated tool that could investigate all aspects, a reduced information flow, and lost information between tools, software compatibility issues.According to Kebede et al. [9], these limitations would lead to double the work of recreating models, and manually inputting information into building models.In general, the information flow between product data or existing multi-disciplinary models and simulation tools are not discussed in these studies.Instead, authors described the built environment and input properties in the presented test case.
Some software packages such as Grasshopper, Rhinoceros3D, DIVA, open-source engines Radiance, EnergyPlus, and other plug-ins for Ladybug and Honeybee, were used commonly by many studies due to their compatibility within the simulation workflow.Other simulation tools that are compatible with these packages can be used for simulating beyond-vision effects like ALFA and LARK.Most simulation tools were developed over the years with having one aspect in focus (e.g., 'energy' or 'daylight') with one (set of) metrics as results.Later in time, researchers started to couple tools in order to calculate multiple (sets of) metrics at the same time [47].
MOO is a complex process combining multiple aspects and metrics, and, therefore, many studies avoided evaluating the existing environment first and jumped directly to optimisation.Some studies used a black-box or white-box model to simplify the environment and only focus on the investigated design variable.Many studies omitted validating the optimisation solutions in real-life afterwards.According to Maile et al. [48], omitting validation affects both reliability by optimising the incorrect variable and validity if measured and simulated results show significant deviation.In general, solutions that are optimal in simulations, do not necessarily perform as well in complex and dynamic real-life, therefore validation is an important step to ensure quality of the design solution and validity of the simulation tool.

Limitations
In this review only the highest level of optimisation M5 (multi-objective optimisation) is investigated.Multi-objective analysis includes multi-disciplinary aspects but, in this project, the author has major focus on light-related aspects and beyond vision effects.Building performance aspects with less relevance for the thermal-visual-energy-cost balance (e.g., acoustics) were not considered.Studies that solely focus on simulations of beyond-vision effects were not investigated in the scoping search as this review focused on multi-objective optimisation integrating beyond vision effects.
An overall limitation is the use of only simulation software to assess visual comfort and well-being.As Yusoff et al. [49] and Houser and Esposito [44] showed there are certain spatial aspects like special perception, multi-sensory experience and user engagement that cannot be predicted and evaluated if user-based assessment is missing.Therefore, simulations and on-site assessments complement each other and can be used at different design stages [49].

Conclusions
The review shows that none of the found papers combine visual and beyond-vison effects with other building performance aspects, and is, therefore, not integrated in the workflow of a full multi-objective optimisation simulation.Multiple steps and simulation tools are used starting with parametric modelling, building performance simulation tools and optimisation tools for generating algorithms.The simulation workflow is very affected by the compatibility of the software packages that hinders the choice of the most suited lighting metrics.Other factors like building type, light source, design variables are less represented in reviewed papers.Future research could focus on the less represented conditions in MOO to increase the scientific evidence and integrate beyond-vision effects in the MOO workflow.

Reference Light source a Simulation level, building type, location b Aspects c Inputs properties d Design variables Modelling/ simulation/ optimisation tools e Generative system f Main conclusions Limitations
BD, BO, E, S, W, WWR, WD shading v, h system angle NSGA-II all-purpose polygonal shade showed increase in VP and VC vs. no shade building conditions