Case Study and Feasibility Analysis of Multi-Objective Life Cycle Energy System Optimization in a Nordic Campus Building

: Due to the high energy consumption of buildings, there is a demand for both economically and environmentally effective designs for building energy system retroﬁts. While multi-objective optimization can be used to solve complicated problems, its use is not yet widespread in the industry. This study ﬁrst aims to develop an efﬁcient and applicable multi-objective building energy system optimization method, used to dimension energy production and storage retroﬁt components in a case campus building in Lahti, Finland. Energy consumption data of the building are obtained with a dynamic energy model. The optimization model includes economic and environmental objectives, and the approach is found to function satisfactorily. Second, this study aims to assess the feasibility and issues of multi-objective single-building energy system optimization via the analysis of the case optimization results. The results suggest that economically beneﬁcial local energy production and storage retroﬁts could not always lead to life cycle CO 2 -eq emission reductions. The recognized causes are high life cycle emissions from the retroﬁt components and low Nordic grid energy emissions. The performed sensitivity and feasibility analyses show that correctness and methodological comparability of the used emission factors and future assumptions are crucial for reliable optimization results. PV installation becomes feasible in the − 50% scenario, resulting in maximum possible installation. The analysis also shows that correct component life cycle emission values are essential for reliable environmental optimization results.


Introduction
Buildings and their construction consume over one-third of the total global energy consumption and cause almost 40% of the global carbon dioxide (CO 2 ) emissions [1], and 36% of the CO 2 emissions in the European Union (EU). The EU Energy Performance of Buildings Directive (EPBD) already requires new buildings to be nearly zero-energy buildings (NZEB); however, the EU building stock is also rather old and slowly renovated. Therefore, retrofits and modernizations to existing properties offer large energy performance and sustainability potential [2]. Thus, there is a need for reliable methods to identify the best-performing retrofit and modernization targets.
In a typical Finnish design, on-site renewable energy production, especially photovoltaic (PV) generation, is dimensioned to maximize self-utilization of the produced energy in the building, and therefore, to minimize the excess production, usually exported to the grid [3]. To increase peak renewable energy production, exported energy needs to be increased, demand flexibility needs to be implemented, or storage technologies need to be utilized. If energy storage is added to the system, dimensioning becomes complicated because of component interdependency. The system state is defined not only by current energy flows but also by past energy flows that manifest in storage state-of-charge. Therefore, energy systems containing storage components cannot be easily dimensioned with traditional methods, and the design must often be assisted with simulation.
Dynamic building energy simulation software, such as IDA ICE [4], EnergyPlus [5], and TRNSYS [6], couple the indoor zone conditions and energy system state at each and environmental life cycle optimization is enabled by obtaining component life cycle emission data from literature. Second, this study aims to assess the feasibility of economicenvironmental multi-objective single-building energy system optimization by analyzing result reliability and comparability. The process is enabled by sensitivity, feasibility, and qualitative analyses of the case optimization results, from which potential actions and best practices are derived and discussed.
The article is organized as follows: first, the applied modeling approaches are presented, including the model validation and used parameters. Second, the optimization results are presented along with sensitivity analyses and further insights. Last, the results are discussed along with future research needs and conclusions.

Methodology
The case building in Lahti, Finland is a repurposed furniture factory, currently accommodating production, office, storage, and campus spaces. Floor plans of the building are available in Appendix A. The building was originally built in the 1950s and partly modernized in the late 2010s, when 111 kW photovoltaic (PV) solar panels and a 1 MW ground source heat pump (GSHP) system with ground boreholes were installed. The building is connected to the local district heat (DH) network.
As the building consists of several parts and is partially renovated, it contains structural parts from different eras, differing in heat conductivity. Original windows have double glazing, i.e., inner and outer frames with single panes, whereas modernized windows have triple glazing, i.e., inner frames with two pane insulating glass elements and outer frames with single panes. Large glazing areas have solar control glass.
Due to the northern location of Finland, the achievable PV production profile is distinct with high insolation in the long summer days, and low in the short winter days [30]. Hence, utilizing excess PV electricity production in the summer to drive heat pumps and storing the generated heat seasonally could be one solution to balance this characteristic. Combining PV production and heat pump with battery [31] or thermal storage [32] have been studied; however, the integration of these systems in Nordic conditions still needs further research. The feasibility of this concept is investigated in the results.
As this study focuses exclusively on the building energy system, other means of energy performance improvement, such as renovation, insulation improvement, occupant coaching, or schedule optimization, are not considered.

Dynamic Building Energy Model and Validation
Energy demand data for the building were obtained with dynamic building simulation software IDA Indoor Climate and Energy (IDA ICE) [4]. The building geometry was derived from floor plans and by site inspection. Thermodynamic properties, such as thermal conductivities, were sourced from structural design documentation and by inspection, and they are presented in Table 1. The model floor area is 79,184 m 2 and the volume 374,643 m 3 . The default energy plant model of IDA ICE was used, containing simple heating and cooling heat exchangers with unlimited capacities. Heating distribution fluid temperature was set to adjust linearly based on the ambient temperature, from 70 • C at −30 • C ambient to 20 • C at +20 • C ambient, and the zones were heated with water radiators with 21 • C setpoints. Indoor cooling was performed mainly via chilled supply air with 24 • C zone setpoints. The standard IDA ICE air handling unit model was used, implementing a heat recovery exchanger with 50% efficiency and supply air heating and cooling. Constant supply air temperature was set at 17 • C and fan operation schedule to 6:00-19:00 on workdays. Office and learning spaces were configured for temperature-controlled variable air flows from 2 L/(s m 2 ) to 4 L/(s m 2 ), and standard spaces with constant air flows of 2 L/(s m 2 ). Occupancy, internal load, and indoor lighting schedule was set to 7:00-19:00 on workdays. ASHRAE IWEC2 weather data [33] for Lahti, Finland, and the default IDA ICE urban wind profile were used. The model construction, visible in Figure 1, was simplified by combining zones and windows to ensure practical simulation times with satisfactory resolution. The simplified model contains 111 window and 70 opening or door elements. The model was validated by comparing monthly simulated space heating and cooling energy consumptions with measured data from the building, which includes GSHP heating and cooling and imported district heat energies for the year 2020. In Finland, annual variations of space heating energy consumption data are standardized according to [34]: where Q std is the standardized space heating energy consumption, S N the average reference value for the location, S R the realized reference number for the location, Q R the realized energy consumption for the current year, and Q DHW the domestic hot water energy consumption to be excluded from the standardization. Since the available measured data are consolidated with no possibility to separate the energy used for DHW heating, the correction was performed inversely on the simulated data to convert them to the year 2020. Therefore, Q R is solved by Equation (1), resulting in: where Q std is now the energy consumption input from the simulated data. The relative reference values for Lahti, Finland for the whole year of 2020 are S R = 3438 and S N = 4392 [35], indicating a warmer year than usual. In the model validation, the monthly heating degree days values for the year 2020 [35] were used. Monthly comparison of measured energy consumption data and corrected simulated values for model validity assessment is presented in Table 2. Table 2. Corrected simulated monthly energy consumption data from IDA ICE is compared to the measured data from 2020. The simulated values are corrected to represent the year 2020 through the heating degree days correction. The data are used for the validation of the energy consumption model.

Month
Measured Although some monthly and seasonal variance is visible in the data in Table 2, the annual difference of −4.2% and the similar trend shapes between measured and simulated data were considered to validate the energy consumption model for use in this study.

Modular Energy System Model
The building energy system model was constructed with an open-source linear programming energy system modeling framework Calliope [29]. The framework enables multi-objective optimization with a linear programming solver of choice; in this case, COIN-OR Branch-and-Cut (CBC) solver was used. The problem was defined as linear programming (LP) instead of mixed integer linear programming (MILP), since values such as storage capacity can reduce to zero to represent an absent equipment, and no binary investment costs are included.
The model is based on energy balances of each modeled component, fulfilling the energy balance at each timestep for each separate location, technology, and energy carrier. In this single-building case, there is only one location, and the balance is written as follows [36]: where for a specific technology in an energy carrier (tech :: carrier), E prod is the energy produced in or brought to the node, E con is consumed, and E export is exported. In this case, three main energy carriers were included in the model: electricity, heating, and cooling. The heat carrier is divided into high-and low-temperature subcarriers: hightemperature heat is always imported from district heat, while low-temperature heat can be sourced from GSHP or storage. At least 50% of DHW heating power is always sourced from the high-temperature subcarrier. The overall structure of the model components and carriers is presented in Figure 2. Diagram of the energy system optimization model for this specific case. The main component enabling flexibility is the ground source heat pump, which links separate energy carriers through its energy conversion ability. In this model, energy demands are pre-defined from the dynamic building model data.
The model configuration optimizes three parameters: PV panel peak power, heat storage capacities, and electricity storage capacities. The system's lifetime is 25 years and the real interest rate is 5%. Emissions are considered with no depreciation rate. The following subsubsections describe the individual components modeled in the energy system model, as well as their mathematical formulations.

GSHP Model
The GSHP is a central component in the model, as it enables power-to-heat conversion when excess PV production is available. It was simplified as an energy conversion component with constant conversion ratios, following Equation (4) [36] in each timestep. Using the constant coefficient of performance (COP) for heating and energy efficiency ratio (EER) for cooling is reasonable, as the heat pump evaporator and condenser temperatures stay quite constant in operation.
where R is the set carrier conversion ratio. Here, the ratios correspond to COP and EER and include the conversion efficiency η(tech). Thus, this specific GSHP is modeled as: The measured GSHP production data from the site data showed a combined productionweighted average COP/EER 3.8, calibrated IDA ICE GSHP model COP 3.3 and EER 4.8, and system datasheets COP 3.3 and EER 4.3 at design operating points. The datasheet values were selected and resulted in approximately 50 • C condensing temperature both in the IDA ICE model and datasheets.
In GSHP systems, too intense heat removal can lead to freezing of borehole water and the surrounding soil [37], and therefore, the borehole heat extraction power must be limited. Real systems are often controlled with pre-set operation modes for different ambient temperatures. In the location, geothermal potential is approximately 7 kW per one 300 m borehole [38], which scales to 368 kW for the entire borehole field. Maximum GSHP compressor power of 300 kW fulfilled the condition on annual level.

Solar Generation and Energy Storage Technologies
The modeled PV capacity was constrained to a range from the present 111 kW to the maximum that was estimated to fit the building roof, i.e., 555 kW. Widely commercially available mono-or polycrystalline silicon panels were chosen, for which price and carbon dioxide-equivalent (CO 2 -eq) life cycle emission data are widely available. The used parameters for the PV production component are listed in Table 3. Table 3. Boundaries and specific parameter for photovoltaic systems.

Parameter Value
Maximum total peak power [kW peak ] 555 Specific cost [€/kW peak ] 1000 [39] Specific emission [kg CO 2 -eq/kW peak ] 1200 [40,41] Thermal storage was modeled as a water tank sensible heat storage, as the other alternative, phase change material (PCM) latent heat storage, is not yet technologically mature enough. Furthermore, the available cost and life cycle emission data of PCM storages is scarce and uncertain. Electrical storage was modeled as a lithium-ion battery due to its wide commercial availability and data availability. The simulated storage energy states are defined according to Equation (6) [36], in which the current charge state is defined via the previous timestep state: where E storage is the charge of the observed storage and η(tech) is the technological charge and discharge efficiency including all losses. Relevant constraints and values for the thermal and electrical storage systems are listed in Table 4.

Cost and Emission Factors of Energy Grid Connections
Emission factors for imported energies can be defined by accounted, average, marginal. or other values. Although electricity and district heat can be bought as accounted carbonfree products from the utilities, the actual delivered energy can be from mixed sources. In this model, average values for the connected grids were used. The current national electricity emission factor is approximately 72 g CO 2 /kWh [47], and the local district heat emission factor is 57 g CO 2 /kWh [48]. Emission factors for electricity and district heat are assumed to decrease linearly and reach zero after 25 years. Energy import costs are the total empirical values for the case, including energy, transmission, and tax components. Exported electricity compensation includes only the energy component, and thus, it is lower than the total cost of imported energy. The values are consolidated in Table 5. Life cycle costs and emissions were calculated with Calliope, following Equation (7) [36]. Emissions from maintenance operations were not considered separately, as they are not a significant part of the total life cycle emissions in this case. However, they can be included in other models or applications, if necessary.
where C(cost, tech) represents the cost for a single cost type (cost, such as monetary or CO 2 -eq) and for a single technology (tech). C inv is the present value of investment and C var is the present value of variable costs for each considered timestep. The total system cost or emission for the entire lifetime is calculated by summing all individual technology cost values calculated with Equation (7).

Results
The photovoltaic panel, thermal storage, and electrical storage retrofits were optimally dimensioned for the case building for three objective scenarios, which minimize the objective functions for the system lifetime of 25 years: • Economic: cost 100%. • Balanced: cost 50%, emissions 50%. • Environmental: emissions 100%.
The optimization results are presented in Table 6. Compared to the reference scenario, in all modified scenarios, the GSHP can be at least 5% better utilized in heating due to more available renewable electricity or the added thermal storage component. Electricity storage (lithium-ion battery) is not deployed on any objective due to high life cycle costs and emissions.
As the data in Table 6 shows, in the economic scenario, maximum possible PV generation capacity and some thermal storage is installed. This setup provides a total energy import reduction of over 5% annually, but considering life cycle emissions from the equipment, it causes 6.5 tons more annual CO 2 -eq emissions compared to the reference scenario. This converts to a 2.6% life cycle increase. The large increase of PV production capacity leads to more electricity exports than in the reference scenario, which is especially noticeable on the self-used PV electricity ratio of 90% (approximately nine percentage points lower than in the reference results). The rather small compensation of exported electricity 0.03 €/kWh keeps the arrangement profitable.
The balanced scenario increases annual costs slightly while reducing emissions roughly by 1%, enabled by thermal energy storage at 2.5 MWh capacity. It is notable that high CO 2 -eq emissions from PV panel production block additional installations even in this half-environmental objective. DH imports are replaced with electricity imports due to more extensive use of GSHP heating.
In the environmental scenario, the thermal storage capacity is increased to the cap of 10 MWh, with an annual cost increase of almost 10 k€. This option seems the least desirable for the building owner, even though the GSHP can be very well utilized.

Sensitivity from Imported Energy Prices and Emissions
Result sensitivity was assessed by varying the uncertain inputs and assumptions of the energy system model: future imported energy price, grid emission factors and their projections. The assessment was performed on three scenarios, simulated for each optimization objective to enable evaluating effects from both monetary and emission factors: All imported electricity and district heat is considered zero-emission for the whole simulation period. Procurement of carbon-neutral energies raise the specific costs of imported energy by 20% in the model. It is acknowledged that full life cycle emissions are neglected, and actual emissions could be compensated in accounting. B.
Instead of carbon-neutral electricity and DH networks after 25 years, the specific emissions will only reduce linearly to 50% from current levels. In the model, the emission factors for imported energy are increased by 50%. C.
This modification assumes that the cost of imported energy will increase linearly by 100% in 25 years, instead of staying at current inflation-corrected levels. In the model, the imported energy prices are increased by 50%. Figure 3 shows simulation results from the listed scenarios. Overall, the optimization results are very sensitive to assumptions about future energy prices and emissions. Lithiumion battery storage capacity remains at zero even in all the above scenarios.
The most apparent remark from Figure 3 is that in scenario A with environmental optimization objective, the optimized system obviously prefers importing all energy from the networks instead of installing any local production or storage equipment. This is clear, as installing new local equipment would produce life cycle emissions. However, it well represents the discrepancy of the available carbon emission values and the possibilities of creative CO 2 accounting.
In scenarios B and C, the modifications cause the largest effect in the balanced optimization objective (50% monetary, 50% emission), in which the installed PV power and heat storage capacities are greatly increased due to less desirable imports. However, even in the environmental objective in scenario B, the PV power capacity is not increased due to its high specific emissions.

Feasibility of Component Life Cycle Emissions
As no additional PV or electricity storage capacity were introduced in the environmental optimization results, a feasibility study for component life cycle emissions was conducted to determine the magnitude of necessary life cycle emission reductions to make the component installations environmentally feasible. At the same time, the specific component life cycle emission values sourced from literature are a major source of result uncertainty. Specific emission values for lithium-ion batteries vary greatly in the literature and depend on the manufacturing location. The current values range approximately 80-250 kg CO 2 -eq/kWh [45,46], and similarly, broad ranges are present in the PV and heat storage values. Therefore, sensitivity from the life cycle emission factors was assessed by varying the emission factors of all components, as in Table 7. The optimization runs were performed for the environmental optimization objective, as the modifications cause no monetary impacts. The simulations resulted in configurations presented in Table 8. In this specific case, a lithium-ion battery becomes environmentally feasible only in the −75% scenario, with a specific life cycle emission of 37.5 kg CO 2 -eq/kWh. According to the literature, this level is not currently achievable. PV installation becomes feasible in the −50% scenario, resulting in maximum possible installation. The analysis also shows that correct component life cycle emission values are essential for reliable environmental optimization results.

Power-to-Heat and Storage Interoperation
As heat production with GSHP is cheaper and causes lower emissions than DH even with imported electricity due to the high COP, the model utilizes the GSHP to proactively store heat in the thermal storage for space and DHW heating. This power-to-heat operation mode is especially beneficial in the timesteps where PV production exceeds local electricity demand, which is often the case on weekends. For assessment, the economic scenario optimization result was used because of the maximum available PV generation. The optimized storage capacities are not suitable for seasonal storage, but sufficient for interday buffer. Figure 4 shows system operation for the first week (Monday-Friday) of July including the leading weekend (Saturday-Sunday), when excess PV production is available.  Figure 4 shows the thermal storage charging to 100% state-of-charge (SOC) during the weekend using heat generated with GSHP, when the local photovoltaic generation exceeds the total electricity demand. The storage is then optimally discharged during the week by utilizing the heat mainly for DHW heating.

Discussion and Conclusions
In this study, a new building energy system optimization approach was developed. The approach was tested for a case campus building, for which energy consumption data were generated with a dynamic energy model. The consumption data were used to develop an energy system optimization model, which in this case was used to dimension photovoltaic (PV) generation and storage equipment. The optimization objectives were both monetary cost and carbon dioxide equivalent (CO 2 -eq) emissions. The developed method was identified to function satisfactorily in the case study, and the optimization results served as the material for the feasibility analysis of multi-objective optimization. The developed energy system optimization model sees input data from dynamic building simulation or measurements as "black box", taken as provided. The black box nature of the input data reduces the computational load and sensitivity in the optimization phase. For existing buildings, measured data could also be used. The approach prevents considering energy demand flexibility, and possible thermodynamic effects to the building structures are also neglected. Comprehensive modeling of these factors would require a more integrated optimization solution, potentially implemented directly in the dynamic building simulation software. Integrated optimization, however, requires considerably more computational resources, while the industry needs an easily applicable and computationally affordable approach. The developed method fulfills this requirement.
The power-to-heat concept manifested naturally from the available component mix of PV panels, thermal storage, and heat pump. In this case, it was demonstrated as the capability to convert excess solar electricity production to heat with the heat pump and store it in the thermal storage for later consumption. This concept could turn out as an essential function in future carbon-free energy systems. Studying this approach more specifically, however, is outside the scope of this study and requires further research.
The retrofit dimensioning results of the case campus building are significant and puzzling. Due to high emissions from manufacturing of photovoltaic panels and lithiumion batteries in addition to low CO 2 factors of imported energy, no life cycle emission reductions appear to be gained from installing additional PV panels or batteries for the 25-year lifetime in this case. This result significantly diverges from the presumption that local renewable energy production and storage provide environmental life cycle emission benefits. The conclusion is not directly generalizable, however, as it depends on the specific building and its equipment, local CO 2 emission factors, and projections that electricity and district heat imported from the energy networks will be carbon neutral after 25 years. Future research needs to cover more buildings in different locations to study differences between geographical areas.
The sensitivity and feasibility analyses of the case results indicate that correct imported energy and component life cycle emission values are crucial for reliable environmental optimization. Sourcing of correct life cycle emission data for components such as lithiumion batteries requires knowledge of the manufacturing location and detailed specifications. Furthermore, emission factors of imported grid energy are rapidly decreasing, as electricity and DH networks in Finland are converting to carbon-neutral production, and the reported values do not always represent the full life cycle. This complicates the sourcing of reliable input values for environmental optimization and causes the results to expire rapidly.
The presented optimization approach considers only a single building. For cost optimization, this seems feasible, as the incentives are clear for the property owner in the form of cost savings. However, emission minimization has currently no tangible incentive for the property owner and may thus remain unimplemented. Furthermore, the environmental optimization result with no installed PV panels seems to conflict with the fact that small-scale distributed renewable energy generation including PV is generally considered a requirement for carbon-neutral electricity networks. Applicable results would probably require optimization on several system levels and coordination between them. In the future, frameworks such as emissions trade could be further extended to translate these considerations into the monetary domain.
Based on the previous points, it would seem feasible that especially in future renewable energy systems, reported specific energy emissions included life cycle emissions from system manufacture, deployment, operation, decommissioning, and other considerations, instead of only direct emissions from combustion. The current methodological discrepancies in emission factors complicate emission optimization and cause uncertainty in the results. Naturally, the situation might change with future regulations or incentives, but the field still needs further research. Currently, for a building owner, single-objective cost optimization could be the most sensible approach due to its clear incentives and robustness.
Funding: This study received funding from the European Regional Development Fund (ERDF) under project A75174.

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Appendix A
Appendix A contains floor plans of the modeled case building. Variable z represents the vertical coordinate.