Designing electrically self-sufficient distributed energy systems under energy demand and solar radiation uncertainty

Available Available online online at at www.sciencedirect.com www.sciencedirect.com

Available online at www.sciencedirect.com Available online www.sciencedirect.com Available online atatwww.sciencedirect.com

ScienceDirect ScienceDirect

Energy Procedia Procedia 00 00 (2016) (2016) 000–000 000–000 Energy Energy Procedia 122 1027–1032 Energy Procedia 00(2017) (2017) 000–000 Energy Procedia 00 (2016) 000–000

www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia

CISBAT 2017 International Conference Future Buildings & Districts Energy Efficiency from CISBAT 2017 International Conference Future Buildings Districts Energy Efficiency CISBAT 2017 International Conference Future Buildings & & Districts Energy Efficiency from from Nano to Urban Scale, CISBAT 2017 6-8 September 2017, Lausanne, Switzerland Nano Nano to to Urban Urban Scale, Scale, CISBAT CISBAT 2017 2017 6-8 6-8 September September 2017, 2017, Lausanne, Lausanne, Switzerland Switzerland

Designing electrically self-sufficient distributed energy systems Designing self-sufficient distributed systems Theelectrically 15th International Symposium on District Heating energy and Cooling under energy demand and solar radiation uncertainty under energy demand and solar radiation uncertainty Assessing the feasibility of using the heat demand-outdoor a,b,∗ a,b a,b,∗, Kristina a,b , Jan a,c Georgios Mavromatidis Mavromatidis Orehounig Carmelieta,c Georgios a,b,∗, Kristina Orehounig a,b , Jan Carmeliet a,c Georgios Mavromatidis , Kristina Orehounig , Jan Carmeliet temperature functionChair for a long-term district heat demand forecast Chair of of Building Building Physics, Physics, ETH ETH Zurich, Zurich, Switzerland Switzerland aa a Chair of Building Physics, ETH Zurich, Switzerland bb Laboratory for for Urban Urban Energy Energy Systems, Systems, Empa Empa Duebendorf, Duebendorf, Switzerland Switzerland b Laboratory Laboratory for Urban Energy Systems, Empa Duebendorf, Switzerland

cc Laboratory

a b Duebendorf, Switzerland c for Multiscale Studies inaBuilding Physics, Empa c Laboratory I. Andrića,b,c *, A. Pina , P. Ferrão , J. Fournier ., B. Lacarrière , O. Le Correc Laboratory for Multiscale Studies in Building Physics, Empa Duebendorf, Switzerland a

for Multiscale Studies in Building Physics, Empa Duebendorf, Switzerland

IN+ Center for Innovation, Technology and Policy Research - Instituto Superior Técnico, Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal b

Veolia Recherche & Innovation, 291 Avenue Dreyfous Daniel, 78520 Limay, France Abstract c Abstract Département Systèmes Énergétiques et Environnement - IMT Atlantique, 4 rue Alfred Kastler, 44300 Nantes, France Abstract This paper paper examines examines the the design design of of autonomous autonomous Distributed Distributed Energy Energy Systems Systems (DES) (DES) under under energy energy demand demand and and solar solar radiation radiation This This paper examines the design of autonomous Distributedseeks Energy Systems (DES) under energy demand and solar radiation uncertainty. uncertainty. A A two-stage two-stage stochastic stochastic program program is is developed developed that that seeks cost-optimal cost-optimal DES DES designs designs considering considering probabilistic probabilistic scenarios scenarios uncertainty. A two-stage stochastic program is developed that seeks cost-optimal DES designs considering probabilistic scenarios for for the the uncertain uncertain parameters parameters to to represent represent possible possible operating operating conditions. conditions. Then, Then, energy energy autonomy autonomy constraints constraints are are imposed imposed to to each each for the uncertain parameters to represent possible operating conditions. Then, energy autonomy constraints are imposed to each Abstract individual scenario, ensuring the autonomy robustness of the DES. The model is applied to a Swiss office building and results individual scenario, ensuring the autonomy robustness of the DES. The model is applied to a Swiss office building and results reveal reveal individual scenario, ensuringDES the autonomy robustness of the DES. The model is applied to aonSwiss office building and results reveal that the the most most cost-effective solutions achieve achieve an electrical electrical autonomy of 20% 20% relying renewable PV electricity electricity generation. that cost-effective DES solutions an autonomy of relying on renewable PV generation. that the autonomy most cost-effective DES solutions achieve an electrical autonomy 20% onautonomous renewablesolutions PV electricity generation. District heating networks are commonly addressed in thea literature as of one of relying the most effective for decreasing the Higher levels require additional electricity from CHP engine, while a 100% system is achievable Higher autonomy levels require additional electricity from a CHP engine, while a 100% autonomous system is achievable but but Higher autonomy levels require additional electricity from systems a Finally, CHP engine, while a 100% autonomous system is through achievable but greenhouse gas emissions from the building sector. These require high investments which are returned the heat requires significant amounts of thermal and electrical storage. comparing the stochastic DES designs against deterministic requires significant amounts of thermal and electrical storage. Finally, comparing the stochastic DES designs against deterministic requires significant ofclimate thermal and electrical Finally, comparing the stochastic DES designs againstcould deterministic sales. Due to the amounts changed conditions andstorage. building renovation policies, heat in demand in the future decrease, ones reveal significant differences, illustrating the importance importance of uncertainty considerations in the design of autonomous autonomous DES. ones reveal significant differences, illustrating the of uncertainty considerations the design of DES. ones reveal significant differences, illustrating the importance of uncertainty considerations in the design of autonomous DES. prolonging the investment return period. c  2016 The Authors. Published by Elsevier Ltd. c 2016 The Authors. Published by Elsevier Ltd.  cThe  2016 Thescope Authors.this Published by Elsevier Ltd. main paper isby to assess the feasibility of using theCISBAT heat demand outdoor temperature function forBuildings heat demand © 2017 The Authors. Published Ltd. Peer-review under of the committee of 2017 Conference Future & Peer-review underofresponsibility responsibility ofElsevier the scientific scientific committee of the the CISBAT 2017 –International International Conference Future Buildings & Peer-review under responsibility of the scientific committee theCISBAT CISBAT 2017 Conference Future Buildings & Peer-review under responsibility of the scientific committee ofofthe 2017 Conference – Future Buildings forecast.Energy The district of Alvalade, in Scale. Lisbon (Portugal), was used asInternational aInternational case study. The district is consisted of &665 Districts Efficiency from to Districts Energy Efficiency from Nano Nanolocated to Urban Urban Scale. Efficiency Nano Districts –Energy Energy Efficiency from NanototoUrban UrbanScale. Scaletypology. Three weather scenarios (low, medium, high) and three district buildings that vary in bothfrom construction period and Distributed Energy System, hub, autonomy, Stochastic Programming Keywords: Distributed Energy System, Energy Energy hub, Energy Energy autonomy, Uncertainty, Uncertainty, Stochastic Programming Keywords: renovation scenarios were developed (shallow, intermediate, To estimate the error, obtained heat demand values were Keywords: Distributed Energy System, Energy hub, Energy autonomy,deep). Uncertainty, Stochastic Programming compared with results from a dynamic heat demand model, previously developed and validated by the authors. The results showed that when only weather change is considered, the margin of error could be acceptable for some applications (the error in annual demand was lower than 20% for all weather scenarios considered). However, after introducing renovation scenarios, the error value increased up to 59.5% (depending on the weather and renovation scenarios combination considered). 1. Introduction 1. Introduction value of slope coefficient increased on average within the range of 3.8% up to 8% per decade, that corresponds to the 1.The Introduction decrease in the number of heating hours of 22-139h during the heating season (depending on the combination of weather and Distributed Energy Systems (DES) (DES) can increase the efficiency efficiency ofincreased energy for supply to urban urban buildings and districts districts Distributed Energy Systems the supply to and renovation scenarios considered). On the can otherincrease hand, function interceptof 7.8-12.7% per buildings decade (depending on the Distributed Energy Systemsof(DES) can increase the efficiency of energy energy supply to urban buildings and districts by incorporating multitude renewable and efficient technologies to meet meet the buildings’ buildings’ heating, considered, cooling, and coupled scenarios).aa The values of suggested couldand be efficient used to modify the function parameters for the scenarios and by incorporating multitude renewable technologies to the heating, cooling, by incorporating a multitude of renewable and efficient technologies to meet the buildings’ heating, cooling, and and electricity their economic, improve thedemands. accuracy ofBesides heat demand electricity demands. Besides theirestimations. economic, environmental environmental and and technical technical benefits benefits [1], [1], DES DES can can also also unlock unlock the the

electricity demands. Besides their economic, environmental and technical benefits [1], DES can also unlock the potential for highly energy autonomous buildings and districts, which cover high shares of energy demands with local potential for highly energy autonomous buildings and districts, which cover high shares of energy demands with local potential for highly energy autonomous buildings and districts, which cover high shares of energy demands with local © 2017generation. The Authors.Autonomous Published by DES Elsevier Ltd. energy can be attractive for aa series of reasons including independence from energy generation. Autonomous DES can attractive for series of reasons including independence from fluctuating fluctuating energy generation. Autonomous can be be attractive forof aThe series ofInternational reasonsof including independence fluctuating Peer-review underbetter responsibility ofDES theenergy Scientific Committee 15th Symposium on District from Heating and energy markets, control over decisions and local consumption low-carbon generated energy [2]. energy markets, better control over energy decisions and local consumption of low-carbon generated energy [2]. energy markets, better control over energy decisions and local consumption of low-carbon generated energy [2]. Cooling. Given the complexity of designing autonomous DES, mathematical optimisation models are commonly developed Given the of autonomous DES, mathematical optimisation models commonly developed Givenwith the complexity complexity of designing designing autonomous DES, mathematical models are are commonly developed to assist the task. Milan [3] presented aa model for the design of aa optimisation residential building’s 100% renewable energy to assist with the task. Milan [3] presented model for the design of residential building’s 100% renewable Heatthe demand; Climate change a model for the design of a residential building’s 100% renewable energy toKeywords: assist with task.Forecast; Milan [3] presented energy ∗∗ ∗

Corresponding author. author. Tel.: Tel.: +41-58-765-4002. +41-58-765-4002. Corresponding Corresponding author. Tel.: +41-58-765-4002. E-mail E-mail address: address: [email protected] [email protected] E-mail address: [email protected] 1876-6102 © 2017 The Authors. Published by Elsevier Ltd. cc 2016 1876-6102  Authors. by Ltd. 1876-6102  2016 The The Authors. Published Published by Elsevier Elsevier Ltd. of The 15th International Symposium on District Heating and Cooling. Peer-review responsibility of the Scientific Committee c under  2016 by Elsevier Ltd. Ltd. 1876-6102 © 2017 The TheAuthors. Authors.Published Published by Elsevier Peer-review Peer-review under under responsibility responsibility of of the the scientific scientific committee committee of of the the CISBAT CISBAT 2017 2017 International International Conference Conference Future Future Buildings Buildings & & Districts Districts Energy Energy Peer-review under responsibility of the scientific committee the CISBAT 2017 International Conference – Future&Buildings Peer-review under responsibility of the scientific committee of theofCISBAT 2017 International Conference Future Buildings Districts & Energy Efficiency from Nano to Urban Scale. Efficiency from Nano to Urban Scale. Districts Energy from Nano to Urban Scale Efficiency–from NanoEfficiency to Urban Scale. 10.1016/j.egypro.2017.07.470

Georgios Mavromatidis et al. / Energy Procedia 122 (2017) 1027–1032 Mavromatidis et al. / Energy Procedia 00 (2016) 000–000

1028 2

system. Orehounig et al. [4] considered energy autonomy as a performance criterion for the comparison of energy system configurations for a Swiss neighbourhood. Finally, McKenna et al. [2] performed a comparative analysis of autonomous energy systems at different spatial scales ranging from the single building to the city district scale. All these studies, though, are performed deterministically assuming perfect knowledge of all the relevant model parameters required for the design. Nevertheless, many aspects involved in DES design (and their corresponding model parameters) are actually uncertain. Building energy demands are inherently stochastic and are influenced by weather and stochastic occupant behaviours. Additionally, solar radiation data required to predict the performance of solar technologies like photovoltaics (PV) can also be considered uncertain due to factors like measurement accuracy, inter-annual variability, the representativeness of the monitoring period etc. [5]. Neglecting the uncertainty of relevant model parameters could lead to suboptimal DES designs that do not meet the desired energy autonomy targets in cases when the actual parameters deviate from their design values. Therefore, in this paper, the aim is to present an optimisation model for the design of autonomous DES that will also directly incorporate the uncertainty associated with energy demands and solar radiation into the decision-making process. 2. Methodology 2.1. A two-stage stochastic model for the design of autonomous DES and a case study In this work, the design of an autonomous DES is investigated for an office building in Zurich, Switzerland with a total floor area of 1620 m2 and a total roof area of 162 m2 for solar installations, shown in Fig. 1a. The candidate energy technologies for the office’s DES shown in Fig. 1b include different types of boilers, heat pumps (air-, ASHP, and ground-source, GSHP), a cogeneration engine (CHP), PV panels, thermal storage and batteries in order to meet the building’s heating and electricity demands. The DES design task then includes the identification of the optimal DES configuration (technology selection and sizing) that will maximise the building’s energy autonomy, while minimising the total system cost. Investigating the trade-offs between the two optimisation objectives can be accommodated in multi-objective optimisation. This can equivalently be expressed as a single-objective, cost-minimisation problem, in which the autonomy-maximisation objective is treated within a constraint imposing a minimum autonomy requirement. By varying the levels of the autonomy requirement and performing multiple optimisation runs, different optimal DES configurations can be obtained that represent the trade-off between system cost and system autonomy.

a

b

Inputs

Distributed energy system

Biomass

Biomass boiler

Heating oil

Oil boiler

Thermal storage

Gas boiler Natural gas

ASHP

Grid electricity Solar radiation

GSHP

CHP engine

PV panels

Batteries

Outputs

Heating demand Electricity export Electricity demand

Biomass Oil Natural gas Solar radiation Thermal energy Electricity

Fig. 1. (a) Illustration of the office building used as a case study; (b) set of candidate technologies considered for the autonomous DES.

The optimal DES design model’s formulation for this paper is built on the basis of previous DES modelling efforts [6] that were based on the energy hub concept [7]. A requirement for this paper’s model, though, is that it should explicitly account for the uncertainty in the office’s energy demands and the incoming solar radiation. However, as the model in [6] is deterministic, the technique for Optimisation under Uncertainty (OU2 ) called Two-Stage Stochastic Programming (SP) is used to extend it towards a stochastic formulation that explicitly considers uncertainty. The main hypothesis of SP is that a probabilistic description of uncertainty is available in the form of scenarios s ∈ S each with a probability π s . In two-stage SP then, decisions are split in two-stages corresponding to decisions that need to be made before (here-and-now) and after uncertainty is revealed (wait-and-see), respectively. For the task of DES design under uncertainty, design aspects, namely the selection and sizing of the DES technologies, correspond



Georgios Mavromatidis et al. / Energy Procedia 122 (2017) 1027–1032 Mavromatidis et al. / Energy Procedia 00 (2016) 000–000

1029 3

to first stage decisions as they need to be made before knowing the actual values of the uncertain parameters. The operating aspects of the DES (e.g. when to import electricity from the grid and how much, when to store energy etc.) belong in the second stage and are made only after uncertainty is revealed. By explicitly including a set of scenarios s for the uncertain energy demand and solar radiation patterns, the possible conditions under which the DES will need to operate are considered. The aim of the model then, expressed in Eq. (1), is to minimise the first-stage cost, Inv, (i.e. the investment expenditure associated with first-stage decisions), and the expectation of the second-stage cost, E s∈S [Op s ], where Op s includes the operational expenditure of the system for scenario s. The total cost for a specific scenario s can then be defined as CPS s = Inv + Op s , ∀s ∈ S.      min Inv + E s∈S Op s = Inv + s∈S π s · Op s (1)

To reflect the scenario-dependence of the second-stage decisions, all model variables in [6] that are related to operating aspects of the DES (e.g. consumption of energy carriers, electricity imports and exports, energy flows to and from storage etc.) need to be expressed per scenario s. Similarly, all the model constraints in which second-stage variables are included (e.g. the energy balances) need to be expressed for each scenario s in addition to their current domain (e.g. for each time step t). On the other hand, first-stage variables and constraints should remain identical. The final aspect of the model pertains to the energy autonomy of the DES. In this paper, the term energy autonomy refers to the degree of energy self-sufficiency i.e. the percentage of energy demand that is covered by local, building-integrated generation. Given that heat demands can only be covered with local generation, energy autonomy corresponds to the building’s electrical self-sufficiency. The autonomy constraint for this paper is expressed in Eq. (2), which defines the degree of electrical autonomy as the part of the total electrical requirements that is not met with grid imported electricity. Electricity generated locally but exported does not contribute towards energy autonomy. Note that in its current form the constraint in Eq. (2) is nonlinear and it would need to be linearised by moving the denominator to the right-hand side. Its form is preserved like that, though, for clarity purposes.  grid Total grid elec. imports t∈T P s,t  >= Autonomy requirement [%], ∀s ∈ S = 1 −   ashp 1− gshp Total elec. requirement + P s,t + Lelec t∈T P s,t s,t

(2)

grid ashp gshp In Eq. (2), P s,t represents the grid imported electricity, P s,t and P s,t the electricity inputs to the ASHP and the elec GSHP, respectively, and L s,t represents the electricity demand of the building for lighting and equipment. All these indicators are indexed per time step t and scenario s. The constraint in Eq. (2) is enforced separately for each scenario s, meaning that the model will seek robust DES designs that ensure that the autonomy requirements will be fulfilled for every realisation of the uncertain energy demands and solar radiation.

2.2. Scenario generation and reduction In order to generate scenarios for the uncertain building energy demand and solar radiation patterns, an approach using the Building Performance Simulation (BPS) software EnergyPlus [8] is used. Initially, probability distributions are assigned to the BPS input parameters and then Monte Carlo BPS simulations are launched to generate multiple scenarios for the uncertain parameters. A discussion of the uncertain BPS input parameters is given as follows: Normal distributions are assigned to material properties, infiltration, ventilation rates and thermostat settings [9,10] using nominal values from [11] for the first two and from the Swiss norm SIA 2024 [12] for the latter two as the distributions’ mean. Triangular distributions are assigned to the occupancy density, the capacities for lighting and appliances, and the hot water demands [10], created using the min-nominal-max values in [12]. Moreover, to represent aspects like the stochastic occupancy patterns, appliance usage etc. additional variability is introduced to the schedules used in BPS. Starting from the nominal schedules of [12], each hourly value is varied by ±15%. Then the values of each profile are resampled with replacement within specified blocks in each day [13] to introduce variability in the course of actions that each schedule represents (e.g. when a device is used). Finally, to represent weather uncertainty, multiple future climate projections are sourced from the CORDEX project [14] for the system’s operating period (2021-2040) and transformed into BPS weather files using the ’morphing’ technique [15]. In total, approx. 1500 scenarios are generated for the uncertain energy demands and solar radiation with their annual total values illustrated in Fig. 2. If all scenarios are used in the optimisation problem though, its size would

Georgios Mavromatidis et al. / Energy Procedia 122 (2017) 1027–1032 Mavromatidis et al. / Energy Procedia 00 (2016) 000–000

1030 4

267

(+40%)

Annual global horiz. solar radiation (kWh/m2)

Annual heat demand (MWh)

lead to computational tractability issues. Therefore, a scenario reduction technique is needed to select a subset of the most representative scenarios and calculate their probabilities, ensuring little or no sacrifice to the model’s accuracy. For this task, a feature-based clustering approach is applied [16]. First, statistical features (mean, variance, max, kurtosis and skewness) are extracted for the energy demand and radiation time series of each scenario to represent the original dataset. Then, the k-medoids clustering is applied to this statistical feature set in order to identify the most representative scenarios. External clusters are also introduced to represent scenarios with extreme low and high total annual energy demands, which might not be selected by the clustering algorithm, and scenarios whose peak demands correspond to the desired values for accurate system sizing (e.g. a chosen percentile of the peak energy demand distributions). The probability of each representative scenario is calculated as the percentage of all scenarios that belong in each representative scenario’s cluster. In this paper, the selected subset consists of 20 scenarios, which are illustrated in Fig. 2, showing the effective coverage of all the regions in the uncertain parameter space. Fig. 2 shows also the deterministic values for the energy demands and solar radiation, which are calculated using the nominal values of the BPS parameters and a Typical Meteorological Year (TMY). As it can be seen, the deterministic values represent ”average” conditions that do not reflect the range of possibilities for the uncertain parameters.

229

(+20%)

190 (0%)

152

(-20%)

114

(-40%)

1263 (+15%)

1181

(+7.5%)

Variable scenarios Reduced scenario set Deterministic point

1098 (0%)

1016

(-7.5%)

934

(-15%)

37

(-40%)

49

(-20%)

62

(0%)

74

(+20%)

87

(+40%)

Annual electricity demand (MWh)

99

(+60%)

37

(-40%)

49

(-20%)

62

(0%)

74

(+20%)

87

(+40%)

99

(+60%)

Annual electricity demand (MWh)

Fig. 2. Variable annual heat and electricity demands and incoming solar radiation, subset of most representative scenarios, and deterministic values (the percentage values in each axis indicate the deviation from the deterministic value)

3. Results The results of this paper are summarised in Fig. 3. In Fig. 3a, the economic performance of the optimal system configurations is presented in terms of the mean total system cost (see Eq. (1)) and the range of the individual scenario costs, CPS s . The lowest mean costs are observed when the system autonomy requirement is set at 0%-20%. The mean cost then increases at a relatively constant rate until the 90% point, while for the 100% autonomous system it reaches 60 kCHF. Besides the mean costs, the information regarding the CPS s range can also be valuable for decision-makers in selecting the desired autonomy level. For instance, in the case of the 100% autonomous system, the cost for the most unfavourable scenario exceeds 70 kCHF, which, if considered extreme, could render the system an unfavourable choice. An additional performance metric presented in Fig. 3a is the system’s CO2 emissions. It can be seen that when autonomy requirements increase, the mean CO2 emissions as well as their corresponding scenario range increase with it. These patterns can be better understood by examining Fig. 3b and Fig. 3c, which present the selected technologies and their sizes for the different systems. For the configurations that result in the lowest total system cost (0%-20% autonomy range), the technology selection includes an oil boiler, a GSHP, PV and thermal storage. This also indicates that even when the autonomy requirement is set at 0% or 10%, the autonomy per scenario is actually higher because of PV and reaches 20%. Increasing the autonomy requirement, the first change observed is that a CHP engine is introduced in the DES configuration and its size gradually increases until it reaches 55 kWth at the 100% point. The increased CHP capacities are initially accompanied by similar decreases of the GSHP size, while the oil boiler’s capacity is decreased only after the 70% point. At the same time, the capacity of PV remains constant for all configurations. With regards to energy storage systems, the thermal storage tank sizes remains constant until the 70% point and is then increased almost exponentially. Thermal storage is used to store excess heat at times when electricity is generated by the CHP and the CHP’s heat output exceeds the heat demand. Increased CHP sizes at higher autonomy levels increase the DES on the CHP

Georgios Mavromatidis et al. / Energy Procedia 122 (2017) 1027–1032 Mavromatidis et al. / Energy Procedia 00 (2016) 000–000

a

80

Mean total cost Total cost scenario range Mean CO2 emissions CO2 scenario range

60

100 80 60

40

40

20

20

0

80

Capacity (kW)

Total cost (kCHF)

100

Total emissions (tCO2)



0 0% 10%20%30%40%50%60%70%80%90%100%

20

200

c

Stochastic designs Deterministic designs PV PV Therm. Storage Therm. Storage Battery Battery

150 100 50 0

0% 10%20%30%40%50%60%70%80%90%100%

Autonomy requirement

Electricity flow (MWh)

Capacity (m 2 | kWh)

250

Stochastic designs GSHP CHP Oil Boiler Deterministic designs GSHP CHP Oil Boiler

40

Autonomy requirement

300

b

60

0

1031 5

150 100

d

Grid imports PV elec CHP elec Elec exports

50 0

0% 10%20%30%40%50%60%70%80%90%100%

Autonomy requirement

0% 10%20%30%40%50%60%70%80%90%100%

Autonomy requirement

Fig. 3. (a) Mean value and scenario ranges for total system cost and CO2 emissions; (b) Stochastic and deterministic capacities of energy generation technologies; (c) Stochastic and deterministic capacities of renewable and energy storage technologies; (d) Average electricity flows in the DES for different degrees of energy autonomy.

and, thus, need to be accompanied by higher thermal storage sizes. Finally, batteries are only introduced at a small capacity at the 90% point and reach a size of 45 kWh for the 100% autonomous system. Besides the system configurations resulting from the two-stage stochastic program, Fig. 3b and Fig. 3c also present the optimal DES designs when the analysis is performed deterministically. These results are obtained using the same two-stage stochastic model but with only one scenario corresponding to the deterministic demand and radiation profiles. Starting from the generation technologies, it can be seen that the oil boiler tends to be undersized in the deterministic compared to the stochastic case, which can be attributed to the deterministic model’s reduced knowledge about “extreme” demand scenarios. The implication of selecting the deterministic boiler capacities for the DES is that for some days in a year the DES might be incapable of covering completely the building’s thermal needs. On the other hand, the deterministic GSHP capacities are mostly higher than in the stochastic case. Given the uncertainty of the building’s electricity demand, the stochastic model reduces the GSHP size to lower the total electricity requirements of the building. The CHP capacities are aligned between the two models, apart from the case of the 100% autonomous system, for which given the extreme scenarios included in the stochastic case, the CHP is sized at a higher capacity. Similarly, in both cases, the maximum roof area is fitted with PV. Finally, for the thermal storage modules, the deterministic capacities are lower than the stochastic ones. Especially, at the highest autonomy requirements, the thermal storage’s size is 54% smaller and the battery’s 37% smaller. As a result of these differences, if the deterministic design is adopted instead of the stochastic one and given the variable energy demand and solar radiation patterns, there is the risk of not meeting the autonomy requirements of the building. This would be then attributed to both the higher electricity demands resulting from higher GSHP capacities and the smaller CHP and energy storage capacities. Overall, thus, these results illustrate the importance of stochastic analysis in the design of autonomous DES since by examining the range of possible uncertain parameter outcomes better-informed decisions are made. The final set of information given in Fig. 3d pertains to the total annual electricity flows, averaged over all scenarios, for different energy autonomy levels. It can be seen that for the 0%-20% autonomy range, the dominating source of electricity is the electrical grid supplemented by photovoltaic generation. As the autonomy requirements increase, the grid imported electricity is gradually replaced by CHP-generated electricity until, finally, no electricity is imported for the 100% autonomous system. Overall, given the higher reliance on the CHP engine to cover the building’s energy demands, the reducing GSHP capacities in Fig. 3b, and the much lower Swiss grid emission factor compared to the one for natural gas [17], these patterns can also explain why the CO2 emissions in Fig. 3a are increasing.

1032 6

Georgios Mavromatidis et al. / Energy Procedia 122 (2017) 1027–1032 Mavromatidis et al. / Energy Procedia 00 (2016) 000–000

Overall, it is shown that electrical self-sufficiency for the examined office building is possible even until 100%. Nevertheless, the system needs to rely heavily on natural gas for electricity generation via the CHP engine, which makes it dependent on another non-renewable resource that needs to be imported. This finding could be thus used as an indication that autonomous DES are perhaps more suitable for whole communities rather than single buildings where resources like hydro power, biomass and wind can be more easily harnessed. 4. Conclusions In this paper, a two-stage stochastic model is presented for the design of autonomous DES under uncertain energy demand and solar radiation patterns. The model seeks optimal DES designs considering a range of scenarios for the uncertain demand and radiation patterns, which are calculated using Building Performance Simulation. By enforcing the energy autonomy requirement for each individual scenario, robust DES are obtained, which will reach the desired energy autonomy levels for any realisation of the uncertain parameters. The model is applied to the task of designing an autonomous DES for an office building. Results show that when cost-optimality is sought, 20% of electrical energy autonomy is possible using PV panels. Increasing the autonomy levels requires additional electricity generation by a CHP engine, which in turn increases both the costs and the CO2 emissions. Finally, a comparison between stochastic and deterministic solutions prove the superiority of stochastic analysis and illustrate the risks of suboptimal DES designs when the analysis is performed deterministically. As future work, additional uncertain parameters like the energy carrier prices will be included in the model. Moreover, the design of autonomous DES will be investigated at the whole-community scale, where more diverse resource portfolios that include hydro or wind energy can contribute towards energy autonomy. Acknowledgements The research presented in this paper is supported in part by funds from the Competence Center Energy and Mobility (CCEM) (IDEAS4cities project) and the Swiss Competence Centres for Energy Research (SCCER) Future Energy Efficient Buildings & Districts (FEEB&D). References [1] Mancarella P. MES (multi-energy systems): An overview of concepts and evaluation models. Energy 2014;65:1-17. [2] McKenna R, Merkel E, Fichtner W. Energy autonomy in residential buildings: A techno-economic model-based analysis of the scale effects. Applied Energy 2017;189:800-15. [3] Milan C, Bojesen C, Nielsen MP. A cost optimization model for 100% renewable residential energy supply systems. Energy 2012;48:118-27. [4] Orehounig K, Evins R, Dorer V. Integration of decentralized energy systems in neighbourhoods using the energy hub approach. Applied Energy 2015;154:277-89. [5] Schnitzer M, Thuman C, Johnson P. Reducing uncertainty in Solar Energy Estimates. AWSTruepower; 2012. [6] Mavromatidis G, Evins R, Orehounig K, Dorer V, Carmeliet J. Multi-objective optimization to simultaneously address energy hub layout, sizing and scheduling using a linear formulation. In: Engineering Optimization (ENGOPT) 2014, Lisbon, Portugal: 2014. [7] Geidl M, Andersson G. Optimal Coupling of Energy Infrastructures. In: Power Tech 2007 IEEE Lausanne: 2007. [8] EnergyPlus. EnergyPlus Simulation Software. Available from: https://energyplus.net/. 2017. [9] Macdonald IA. Quantifying the effects of uncertainty in building simulation. PhD thesis. University of Strathclyde, 2002. [10] Gang W, Wang S, Shan K, Gao D. Impacts of cooling load calculation uncertainties on the design optimization of building cooling systems. Energy and Buildings 2015;94:1-9. [11] Landolt J. A bottom-up modelling approach to address sustainable transformation strategies for the Swiss building stock. MSc thesis. ETH Zurich, 2016. [12] SIA. SIA 2024 Standard-Nutzungsbedingungen fuer die Energie- und Gebaeudetechnik. Swiss society of Engineers and Architects; 2006. [13] Mavromatidis G, Orehounig K, Carmeliet J. Evaluation of photovoltaic integration potential in a village. Solar Energy 2015;121:152-68. [14] Giorgi F, Jones C, Asrar GR. Addressing climate information needs at the regional level: the CORDEX framework. World Meteorological Organization (WMO) Bulletin 2009;58:175. [15] Belcher S, Hacker J, Powell D. Constructing design weather data for future climates. Building Services Engineering Research and Technology 2005;26:49-61. [16] Warren Liao T. Clustering of time series data?a survey. Pattern Recognition 2005;38:1857-74. [17] IEA. CO2 Emissions from Fuel Combustion 2015. International Energy Agency. OECD Publishing; 2015.