IMPACT OF PV DISTRIBUTED GENERATION ON EDP DISTRIBUIÇÃO LV GRID LOSSES

In this paper we detail the method used to build a set of LV typical grids, which are able to capture the variability of assets and operating conditions in order to assess the non-linear impact of distributed generation on losses. Based on simulations carried out over such typical networks, the impact of PV distributed generation on LV grid losses is evaluated for (i) different PV penetration levels, up to 100% of secondary substation peak load, and (ii) for different PV geographic penetration scenarios, from urban to rural.


INTRODUCTION
There has been an increase of self-consumption and/or photovoltaic (PV) micro-generation connected to the lowvoltage (LV) network and the prospects are expected to remain given the recent developments in PV technology.Thus, assessing the impact of PV onto LV grids is considered a priority, especially the impact on losses, as previous studies have concluded that the most significant energy losses are expected at the LV networks [1].In order to assess both the technical losses in LV and the impact of PV generation in such losses, a set of networks typifying LV grids of EDP Distribuição was developed.Since losses are a nonlinear function of consumption, these typical networks cannot be built based solely on average network characteristics.In this paper we will give details of the method used to build such a set of LV typical grids, which are able to capture the variability of assets and operating conditions in order to assess the nonlinear impact of distributed generation on losses.The main components of variability considered are: (i) grid equipment, (ii) network topology, (iii) aggregate consumption profile, (iv) individual consumption profiles and (v) unbalances between conducting phases.

GRID CHACARTERIZATION
EDP Distribuição has about 66 thousand LV grids, which can be divided according to the installed capacity of the associated secondary substation: 50 kVA, 100 kVA, 160 kVA, 250 kVA, 400 kVA and 630 kVA.Other secondary substation types are represented within these categories.Therefore, we divided the LV grids into six major categories, where we included into the 50 kVA grid category all the secondary substations (SSE) with an installed capacity below 50 kVA.For the 100 kVA grids, the secondary substations included have a capacity between 50 and 100 kVA and so on, ending with the 630 kVA secondary substations.For each of the categories mentioned above, we have determined the average characteristics of the corresponding networks.The average characteristics led to a total of five LV networks, since 100 kVA and 160 kVA networks have shown to have similar characteristics (number of customers and contracted power).Since losses are a nonlinear function of consumption, average values for grid characterization are not sufficient to characterize grid losses within each category.Grid characterization has to be changed from average characterization in order to represent the existing variability within each category.To accomplish this, we had to take into account the Grid Equipment, Network Topology, Load pattern (aggregate and individual consumption profiles) and Load Unbalance of each grid category.The methods used to characterize these variability sources are hereafter described.

Grid Equipment
The grid impedance has a linear influence on losses and, for that reason, the average impedance of the typical networks considered should be coherent with the average impedance of existing LV networks.Typical networks are not required to explicitly model the impedance variability.However, that variability is also included in these models in order to consider the combined effect of the impedance variability with load variability.Traditionally, LV feeders are planned to have higher cross sections near the secondary substation.MV/LV transformers with higher installed capacities tend to be located in the urban areas, while the smaller ones tend to be located in rural areas.Feeder cross sections were characterized using these concepts and using the criterion of optimal planning, as described in [2].

Grid Topology
Part of the topology variability is represented through the categorization of the LV grids into the five different groups.Let us call such variability the inter-grid variability.Such variability is necessary but is not sufficient to represent topology variability for losses estimation.Within each category, one needs to represent additional topology variability in order to estimate losses.We call such variability the intra-grid topology variability.We represent the intra-grid topology  Load Pattern

Aggregate substation profiles
The substation load pattern depends on the grid category, since some categories are mostly urban and others mostly rural.Based on yearly telemetry data of 40,000 MV/LV power transformers, we have sampled daily profiles by category, with 15 min resolution, to estimate average load patterns and their corresponding load factors (LF), as defined below: max Where P is the average active power metered in period t.
By analysing the sampled data, we found that daily LFs did not change substantially with season within each category and decided to characterize load pattern variability by characterizing the daily-average power variability only.To characterize daily-average power variability, we estimated load patterns with RMS profiles instead of with average profiles.RMS profiles are computed as follows:

𝑁
Where    is the average active power in time period t in profile i of the sample and N is the number of load profiles in the sample.

Individual customer profiles
To estimate losses, the aggregate load profile cannot be evenly distributed by the individual consumers, as that would lead to a significant under-estimation of losses.To obtain an adequate distribution of load among the consumers, individual load profiles were generated through a non-stationary Markov process [3].The non-stationary Markov process was parameterized based on individual consumer's telemetry data and on the aggregate RMS profiles, obtained for each substation and grid category.This method of distributing the aggregate load profile among the consumers allowed for the representation of the load dynamics of each consumer while ensuring that the sum of the individual profiles matched the pattern of the aggregated load profile, as estimated for each substation.Fig. 2 shows an example of one load profile generated through the Markov process and how it differs from the aggregate load profile (scaled to the contracted power of the same consumer).

Load Unbalance
About half of the LV grid customers in Portugal are single-phase connected customers.The remaining costumers are three-phase connected customers whose per-phase load is also very unbalanced in time.The variability of LV customer per-phase consumption has a significant impact on losses as it increases phasecurrents variability (and consequently also the neutral current) and in each time period.To represent load unbalance, single-phase customers were assigned randomly to a specific phase in each grid category.

LV LOSSES ESTIMATION BY CHRONOLOGICAL SIMULATION
The evaluation of losses for each grid category was done by simulation with the chronological mode of DPlan (software used by EDP to model, analyse and simulate electrical networks).To perform the simulation, individual daily load profiles were generated according to the assumptions described in the previous section.These profiles were then assigned to each customer site under an arbitrary connection phase, when single-phase customers, or as perfectly balanced load when threephase customers.After assigning the load profiles, simulation is performed to run a 4-wire unbalanced power-flow for each period of 15 minutes, estimating the grid power losses by integrating the 15 min losses along the whole day.After having properly defined the consumption scenarios, solar photovoltaic (PV) panels with a defined generation profile were added to the customers.The panels were added successively to simulate an increasing penetration of distributed solar photovoltaic production according to the following guidelines:  Customers with higher contracted power are given priority in the assignment of solar panels.
Paper 0306 CIRED 2017 4/4  The priority in panel assignment is renewed after all clients being assigned a 500 Wp panel.l. The solar panel is always connected to the phase of the customer, except for three-phase customers where the installed panel is also considered to be three-phase.
Other reasonable criteria for distributing panels by customers have been analysed, having obtained similar results for the evolution of losses.
The results of LV losses with PV for homogeneous penetration are shown in Fig. 3, where we present the evolution of the losses as a function of the PV penetration for losses referred to energy supplied (blue) and for losses referred to energy consumed (grey).We can conclude that the LV losses referred to energy consumed decrease with PV penetration.However, the LV losses referred to the energy supplied (reference value so far) increase with PV penetration.Actually, the PV penetration is not expected to happen homogeneously within the territory and it is therefore important to know the effects of possible variations of such penetration, considering different evolutions of PV production in the different grid categories, e.g., a higher penetration in the urban networks or a higher penetration in the rural networks.The results for 100 combinations of penetration [rural]x[urban], are shown in Fig. 4. Results have shown that the impact of PV distributed generation onto the value of losses is significant and tends to reduce the absolute value of losses: this reduction has been estimated at about 19% for a PV penetration level of 100% of the peak load.However, such impact can be interpreted very differently if losses are measured as a relative value.When referred to the energy supplied to the secondary substation, results have shown that PV penetration tend to increase relative losses.Conversely, when referred to the energy consumed, relative losses tend to decrease with PV penetration (see Fig. 4).

CONCLUSION
In this paper we gave details of the method used to build a set of LV grids, which are able to capture the variability of assets and operating conditions in order to assess the non-linear impact of distributed generation onto losses.
Based on simulations carried out over such networks, the impact of PV distributed generation on LV grid losses was assessed for (i) different PV penetration levels, up to 100% of secondary substation peak load, and (ii) for different PV geographic penetration scenarios, from urban to rural.Results have shown that the impact of PV distributed generation on the value of losses is expected to be significant and tends to reduce the absolute value of losses.However, such impact can be interpreted very differently if losses are measured as a relative value.When referred to the energy supplied to the secondary substation, results have shown that relative losses tend to increase.Conversely, when referred to the energy consumed, relative losses tend to decrease.

2 . 2 .
the topology of the average grid of each category.Modifications are undertaken by reconfiguring adjacent feeders in order to load one feeder with part of the load of the other.To be representative of losses, such part needs to correspond to one standard deviation of the feeder sizes (and their assigned loads).This result derives from the definition of the variance,   2 , of a random variable x with expected value,   , as expressed in the following:  2 = ( 2 ) −   2 ( 2 ) =   2 +   2 = (  +  ) 2 +(  −  ) 2Therefore, the expected value of a quadratic function of x (as the losses function is), () = 2 , can be given by the average value of two loss values calculated for two values of consumption that are distant from the average value of one standard deviation each, i.e.,̂= ( ̂+  )+( ̂−  )The LV Grids of Fig.1(a-e) depict the lossrepresentative topology of each category, in which the triangle represents the secondary substation and the circles represent the LV connection sites.The different line colours identify the different substation feeders.

Fig. 2
Fig. 2 Example of one load profile generated through the Markov process and how it differs from the aggregate load profile (scaled to the contracted power of the same consumer).

Fig. 3
Fig. 3 Evolution of LV losses with PV penetration for a homogeneous geographic penetration.

Fig. 4
Fig. 4 Evolution of LV losses with PV penetration considering different scenarios for geographic penetration.