Hybrid European MV-LV Network Models for Smart Distribution Network Modelling
HHybrid European MV–LV Network Modelsfor Smart Distribution Network Modelling
Matthew Deakin,
Member, IEEE , David Greenwood,
Member, IEEE ,Sara Walker,
Senior Member, IEEE , Phil C. Taylor † , Senior Member, IEEE
School of Engineering, Newcastle University, Newcastle-upon-Tyne, UK † Department of Electrical and Electronic Engineering, University of Bristol, Bristol, UK
Abstract —A pair of European-style, integrated MV–LV circuitsare presented, created by combining generic MV and real LVnetworks. The two models have 86,000 and 113,000 nodes, and aremade readily available for download in the OpenDSS file format.Primary substation tap change controls and MV–LV feedersare represented as three-phase unbalanced distribution networkmodels, capturing the coupling of voltages at the MV level.The assumptions made in constructing the models are outlined,including a preconditioning step that reduces the number ofnodes by more than five times without affecting the solution. Twoflexibility-based case studies are presented, with TSO–DSO andpeer–peer-based smart controls considered. The demonstration ofthe heterogeneous nature of these systems is corroborated by theanalysis of measured LV voltage data. The models are intendedto aid the development of algorithms for maximising the benefitsof smart devices within the context of whole energy systems.
Index Terms —Unbalanced distribution network modelling,distribution network analysis, TSO–DSO, flexibility services.
I. I
NTRODUCTION T HE release of the IEEE 8500 Node feeder in 2010[1] coincided with an unprecedented increase in smartcontrol system development for distribution network opera-tions. Robust test-beds such as this are necessary so thatthese controls can designed to scale efficiently to the sizesrequired for implementation by distribution system operators(DSOs) or aggregators [2]. Smart controls are often designedto maximise the benefits of low-carbon Distributed EnergyResources (DERs), which can affect networks in both positiveand negative senses. Domestic-scale DERs (such as heatpumps and electric vehicles) often have large power andenergy requirements, and have therefore been proposed toprovide valuable flexibility services at scales well above theLV level at which they are connected [2].Whilst the radial operation of distribution networks is almostubiquitous in most localities (with the exception of denseurban areas), the size and number of customers fed by LV cir-cuits is very different in European– and North American–stylecircuits. The former is characterised by the use of extensiveLV circuits (with hundreds of customers fed from a single LVtransformer). Conversely, North American-style circuits havemore extensive MV feeders and smaller LV sections, usuallywith just a few customers. As a result, LV circuits in North
This work was funded by the Engineering and Physical Sciences ResearchCouncil through grant no. EP/S00078X/1 (Supergen Energy Networks Hub2018). Email: [email protected].
American-style networks can be modelled reasonably usingservice cable model templates (as in [1]). This is in starkcontrast to European-style LV network models–for example,modelling just a single LV network resulted in a networkmodel of over 4,500 nodes in [3]. Additionally, the numberof customers fed by a European-style MV primary circuit isoften much larger, with over ten thousand customers fed by asingle primary substation being a common occurrence.Because of their size and data requirements (and DSOsreluctance to share potentially sensitive information), thereare few readily accessible European MV–LV models. Oneapproach that has been proposed to overcome the DSO datasensitivity issue is to build synthetic European test cases basedon geographical data [4], with a similar approach taken in aUS-based context in [5] and a Central American case study in[6]. A set of non-synthetic, unbalanced LV circuits fed from acommon MV primary substation are given in [7]. Other worksuse a ‘network allocation’ approach, whereby LV networks areallocated to selected loads in an MV network [3], [8], [9]. Thisapproach is attractive as it allows for the use of thoroughlyvalidated, real LV circuit models (it can be challenging toconvert DSO geographic databases to clean electrical networkmodels). For a more detailed review of distribution networkmodels, we refer the interested reader to [10].To our knowledge, however, there are no full-scale unbal-anced European style MV–LV networks that are openly avail-able for researchers to work from. This view is mirrored by arecent review [2], which highlights the importance of a three-phase representation of MV–LV test circuits, whilst notingthe scarcity of models of this type. The overhead involvedwith constructing and validating these types of circuits is non-trivial, and so this represents a significant gap.In this work we present a pair of hybrid European un-balanced MV–LV models to address this gap. The modelsare neither fully synthetic nor fully physical, with over onehundred real LV feeders allocated to the MV loads. Thecircuits have both urban (underground) and rural (overhead)MV feeders, with the MV–LV construction allowing for thecoupling between LV loads on adjacent networks to be cap-tured. The models are available from https://github.com/deakinmt/uk-mvlv-models , and are designed to be a readily available test bed for the devel-opment of scalable controls that take advantage of the highly a r X i v : . [ ee ss . S Y ] S e p eterogeneous behaviour of LV circuits whilst accounting forMV level coupling.The rest of this paper is structured as follows. SectionII outlines both MV and LV circuit preprocessing stagesand the load-network allocation steps, to clearly outline theassumptions made in the construction of the models. A numberof validation steps are discussed in Section III, to check thatthe networks behave as expected. Some possible use casesof the models are presented in Section IV, to highlight howthese models could be used to study issues in the wider energysystems context. Salient conclusions are drawn in Section V.II. MV–LV C IRCUIT P RE - PROCESSING AND N ETWORK A LLOCATION
In this work we use a hybrid approach for creating the MV–LV test cases, in which real LV networks are allocated to MVcircuit models (as illustrated in Fig. 1). There are three stepsin the creation of the models. First, MV and LV test circuitsare preprocessed; then, the LV networks are allocated to theMV loads; finally, the MV–LV model is validated by studyingload flow solutions under high and low demand conditions.
A. LV Network Preprocessing
The Low Voltage Network Solution (LVNS) test cir-cuits [11] were developed for studying the impacts of low-carbon technologies on European-style distribution networks.These circuits have been thoroughly validated and are well-documented (one of the feeders became the European 906-Node Low Voltage Test Feeder). There are 25 LV networks,disaggregated into 128 feeders. One of the networks, consist-ing of four individual feeders, is plotted in Fig. 1a.Three specific challenges were considered while preprocess-ing these circuits for use in the integrated MV–LV model.Firstly, there are a relatively large number of nodes percustomer in the LVNS networks, which (due to the lackof modelled cable capacitance) can be spliced together toform individual branch models without changing the load flowsolution [12]. The steps taken to validate this step are describedin the Appendix, with the number of nodes in each LV networkbeing reduced by between 5 and 18 times.Secondly, it is noted in the LVNS documentation thatsome feeders have unusually heavy unbalance [13]. Therefore,individual LV feeders with more than 5% zero sequenceunbalance are removed, as are feeders explicitly mentionedas being unusually heavily unbalanced in [13]. Additionally,as shall be noted in the next section, some LVNS networks arelarger than the largest of all of the UK Generic DistributionSystem (UKGDS) MV loads. For example, ‘Network 17’ has883 loads which is (assuming a conservative 1 kW per load)much larger than the largest load of 436 kW on the UG circuit(Table I). Therefore, in the three largest networks (networks2, 15 and 17) one feeder is removed, so that the total numberof loads is fewer than 500 in all circuits used for allocation.In total, 112 of the 128 feeders are available for allocation tothe MV circuit. (a) LVNS ‘Network 1’ topology(b) UKGDS ‘HV-UG’ topology ...
MV Model: UKGDSLV Models: LVNS (c) Integrated MV–LV ModelsFig. 1. The LVNS network models (e.g., (a)) are allocated to the loads ofthe UKGDS models (e.g., (b)) to create the integrated MV-LV models (c).
1) Transformer Sizing:
All transformers in the LVNS setare modelled as 800 kVA transformers, which in many casesrepresents an inefficient allocation of resource (the apparentoverrating of these transformers has been noted in previousworks such as [7]). Whilst there are reasons why they may bethe true sizes in reality [14], following [7] it is assumed thatthese transformers are unlikely to be broadly representative ofsecondary substation ratings across European LV networks.Therefore, once any modifications have been made (interms of excising unwanted feeders described previously) atransformer rating is specified according to after diversitymaximum demand (ADMD) estimates from [15] and thenthe transformer size is specified according to a UK-basednetwork operator’s domestic transformer sizing specifications.Specifically, the ADMD is calculated from the number of LVnetwork loads N Lds as ADMD = N Lds × max (cid:8) S N − α Lds , S ∞ (cid:9) , (1)where S = 5 . kVA is the single-load rating, α = 0 . is the ADMD coefficient, and S ∞ = 1 . kVA is the mini-mum per-load ADMD [15] (to ensure reasonable capacity isallocated when there are large numbers of loads). With thisADMD level, the smallest acceptable transformer rating wasselected from [14, Table 8]. Transformer impedance values ABLE IT WO UKGDS
CIRCUITS FORM THE BASIS OF THE
MV-LV C
IRCUITS , WITH UP TO EIGHTEEN THOUSAND LOADS ESTIMATED ( BASED ON ANASSUMPTION OF K VA PER LV LOAD [16]).Ntwk. ID Tot. Demand,MVA MVBuses Est. LVLoads Load stats., kW(Min., Med., Max.)UG 25.3 78 18991 (100, 344, 436)UG/OH, A 21.4 399 15727 (4, 28, 420) of 1 +
5% pu were assumed for the secondary transformers(with the pu base as the transformer rating), following [16].
B. MV Network Preprocessing
The UKGDS test systems define a set of meshed subtrans-mission (EHV) networks and a set of radial 11 kV MV models(referred to in the UKGDS documentation as ‘HV’ models)[17]. These MV models are useful for the purposes describedhere as the circuits are of the correct voltage level and are ofradial topology (the topology of one of the circuits is plottedin Fig. 1b).From the seven MV UKGDS circuits, two circuits (the UGand UG/OH-A circuits) were chosen as the MV circuits forthe test networks. These two circuits were selected becausethe sizes of the feeders are typical of UK-style circuits, butthey show very different distributions of loads (see Table I).In fact, the UG circuit has a demand 20% greater than that ofthe UG/OH-A circuit but less than one quarter the number ofLV circuits (Table II).As with the LV networks, we make modifications to thetransformer impedance values to ensure the models are realis-tic. The models in the UKGDS MV networks have a per-unit transformer reactance between 2.5% and 10% on thetransformer rating, whilst industry-approved simplified models[16] and data from industry [18] both show impedancescloser to 20% on the transformer rating. A more broadlyrepresentative value was therefore chosen by calculating themedian resistance and reactance of all transformers from [18]that had a low-voltage rating of 11 kV. This approach ledto a per-unit impedance of 0.88 + C. MV Load–LV Network Allocation
The LV Network allocation step was approached with theassumption that the peak loads on the MV network correspondclosely to the designed peak load of LV customers, where eachcustomer is assumed to have a demand of 1.3 kVA [16]. Toensure that small loads would have a network allocated tothem (the smallest LV network has 42 loads, whilst there aresome small loads of just 4 kVA in the UG/OH circuit), thefirst feeder (‘Feeder 1’) from each of the LV networks wasalso considered for allocation.The LV network allocation procedure consisted of two steps,with the goal of ensuring good coverage of LV networks across
TABLE III
NTEGRATED
MV-LV C
IRCUIT P ARAMETERS
Ckt. ID No. Nodes No. Lds. No. LV Ckts.UG 112,887 19,031 75UG/OH, A 86,448 15,166 308 the MV–LV networks whilst ensuring MV branch power flowsdid not change too much. • For each of the loads on the MV circuit, a random numberwas drawn from a normal distribution with unity meanand a standard deviation of 0.175; • Then, the network with a kVA rating closest to themultiplication of this random number and the MV loadwas allocated.With this approach, all 25 of the full LV networks were pickedbetween the two MV–LV circuits, as well as twenty of theindividual LV feeders (for smaller loads). A summary of keycircuit parameters is given in Table II.It is worth noting that the models in this work are focused ondomestic-style LV networks, with no allocation of industrial orcommercial load. As a result, there are likely to be more loadsin Table II than there would be customers allocated to typicalEuropean-style primary substations of the power ratings givenin Table I. Nevertheless, even with this taken into account therewould still be a much larger number of loads than there are in,say, the IEEE 8500-node circuit, which has 1,177 individualloads modelled.III. I
NTEGRATED
MV–LV M
ODEL V ALIDATION
The final stage in the model development was the modelvalidation, ensuring that the circuits that have been built arereasonable in terms of the voltages and powers found at loadflow solutions under realistic conditions. For this purpose, theload flow at a ‘High’ demand and ‘Low’ demand conditionwere considered, with individual LV customers having loadsof 1.3 kVA and 0.16 kVA at unity power factor, based on [16].The voltages under these conditions are plotted as boxplotsin Figs. 2a, 2b. Both circuits tend to have voltages that arehigh during Low demand conditions, with a wide spreadof voltages during High demand conditions. The load flowsolution’s voltages do not violate the UK statutory limits of(0.94, 1.06) pu for MV circuits and (0.94, 1.1) pu for LVcircuits [19].In addition, the powers seen at the primary substation arealso calculated to ensure that the High demand condition doesnot violate thermal constraints, as plotted in Fig. 2c. The LVnetwork allocation has resulted in the MV–LV model powersremaining close to the powers of the UKGDS models fromwhich they are taken. This is particularly important in theUG/OH case, for which the total combined rating of theprimary substation transformers is very close to that of theUKGDS circuit’s power. * 8 * 2 + $ 9 R O W D J H S X ' H P D Q G / R Z + L J K (a) Voltages (MV) 8 * 8 * 2 + $ 9 R O W D J H S X (b) Voltages (LV) 8 * 8 * 2 + $ 3 U L P D U \ 6 X E 3 R Z H U 0 9 $ 5 D W L Q J 8 . * ' 6 0 9 / 9 (c) PowersFig. 2. Checking the MV and LV spread of voltages and the total powers of thetwo MV–LV integrated models. Voltages are calculated for both High and Lowdemand conditions (with 1.3 kW and 0.16 kW at all customers, respectively),whilst the substation powers are calculated during High demand conditions.The boxplots show the range, interquartile range and median voltages. / 9 &