Evaluation of Decentralized Voltage Harmonic Mitigation through DRES converter active filtering capability

The increased penetration of Converter-Interfaced Distributed Renewable Energy Sources (CI-DRES) has posed power quality challenges into the distribution system. In the future CI-DRES may have the responsibility to cope simultaneously with several issues. One function that has already been presented in the literature, but has not yet been included in Standards, is the operation of the DRES converters as active harmonic filters. This paper firstly develops an active filtering control for the CI-DRES, taking into account its thermal limit. This parameter is important, since DRES are already prescribed by Standards to participate in the voltage regulation process by providing reactive power. If the CI-DRES operates also as active filter, it may exceed its thermal capacity. Secondly, the decentralized operation of CI-DRES is evaluated in the CIGRE Benchmark LV distribution system by considering several scenarios of DRES/Non-linear loads penetration and mixture. Time-domain (TD) simulations are carried out in PowerSim Software to demonstrate the contribution of each CI-DRES in the active filtering process. The derived conclusions will serve as a basis for the development of a new coordinated algorithm, so as the CI-DRES can mitigate properly voltage harmonic distortion (HD) in the most efficient techno-economic way. In this way, the active filtering can be treated as a new ancillary service to be introduced in respective markets.


Introduction
The increased penetration of Converter-Interfaced Distributed Renewable Energy Sources (CI-DRES) has caused several problems into the distribution system, e.g. reverse power flows, voltage regulation issues, power quality problems, [1]. Power electronic devices are sources of both low-and highfrequency current components, and pose several challenges on distribution system equipment and loads, [1], since the current harmonics lead to voltage harmonics. According to [1], type and severity of harmonic problems depend -amongst also other parameters-on the loading conditions of the host distribution feeder. To address these problems, several standards (Stds), e.g. IEEE Std 519, [2], and IEC 61000-3-6, [3], define limits for individual and total harmonic distortion (HD) at the Point of Common Coupling (PCC) voltages and currents of a CI-DRES. Until now, the research community has focused into several categories with respect to the presence of harmonics. In general, the studies concern: (1) the impact of the CI-DRES penetration and non-linear loads (NLLs) on the power systems, e.g. posing a limit to the DRES hosting capacity (HC); (2) the optimal placement of active and passive filters considering the limits on the system HC; (3) the control of active filters as well as the operation of CI-DRES as active filters. With respect to the 1 st category, in [2], [4] it has been discussed that, CI-DRES are considered as high-frequency harmonic sources that impact the current quality in the network leading to limited HC in power systems. Some studies [5], [6], have investigated the impact of the CI-DRES generated HD units on the HC of distribution systems taking into account the limits imposed by the aforementioned Stds. The importance of including HD limits in the HC assessment is presented in [7], while in [8] the Harmonic-Constrained HC term is introduced to highlight that the power system HC can be determined by considering only the voltage HD limits.
With respect to the 2 nd category, different methods of optimal placement and sizing of passive filters have been proposed for managing the grid-HC, [9], [10] and maximize the CI-DRES penetration, [11]. A specific passive filter is suggested to be placed centrally, in an industrial system with load nonlinearities through the harmonic-constrained HC assessment considering several limits, e.g. over-voltages. Moreover, total and individual HD percentages are the performance indices that may be violated first. The optimization is solved as mixed integer optimization problem. Furthermore, different active filtering techniques have been proposed considering that HD is a locational variable (decentralized approaches). The installation of additional filters is not a cost-effective choice, therefore, it is proposed in [12] that a CI-DRES -in addition to its major power delivery function-can provide harmonic mitigation Ancillary Services (AS) without additional costs if its apparent power is sufficient. The CI-DRES usually does not exhaust the inverter limits for its primary function, therefore, this AS is technically feasible. In [12] a review is performed to classify the CI-DRES active filtering control schemes and two main approaches are recognized: virtual impedance-based method, [13], and active harmonic filtering-based method. Virtual impedance can emulate the effect of physical impedance without the need to connect a real component to a system and it can be either at the fundamental or harmonic frequencies, [13]. In the active filtering approach, the main control objective is to generate an appropriate compensating current in the inverter, which is produced by comparing the load current with a predetermined reference current. The operation of grid-connected PV converters to provide this AS is studied in [14], [15] and [16]. With respect to wind turbines, a review is conducted in [17], an active damping technique is proposed in [18] to mitigate harmonic resonances, while in [19] a double tuned PI-R based Although there are several scientific papers on the aforementioned topics, there yet exist several gaps. In the future CI-DRES may have the responsibility to perform multiple functions, e.g. provision of virtual inertia and reactive power for voltage regulation, etc. Some of these functions are already included in recently issued Stds, e.g. [20], [21]. The operation of the CI-DRES as an active filter needs to be coevaluated with the other functions of the CI-DRES keeping in mind that when performing various functions, the CI-DRES may exceed its thermal capacity. To the authors' best knowledge there exist no such control scheme. Moreover, up to now, the harmonic mitigation by CI-DRES has been treated in a decentralized manner. Similar to the voltage rise, high voltage HD is also a locational phenomenon, and is co-related mostly with the presence of non-linear (NL) loads, which emit low-order harmonics. The sensitivity of a CI-DRES location with respect to the low-order harmonics has not been studied yet. Therefore, when performing a decentralized algorithm for harmonic mitigation, the contribution of a CI-DRES to the process needs to be examined with respect to its location and in terms of its thermal capacity. These investigations may lead to an optimized harmonic mitigation process by CI-DRES, considering the above parameters and the associated costs. This is missing from the current technical literature. This paper contributes to the raised issues in the following manner: Firstly, a method for harmonic filtering is proposed, which is activated only when the 1 st harmonic current is lower that the nominal CI-DRES current. A harmonic conductance Gk per harmonic k is calculated, so that the CI-DRES injects a proper harmonic current. Secondly, this method is evaluated in the CIGRE Residential LV Network under different CI-DRES/ NLLs under several penetrations and mixtures, in order to investigate the contribution of each CI-DRES with respect to its location and thermal limit. Indices for the contribution are selected to be achieved reduction of the average Total HD (THD) of the voltage and the THD per node. In addition, some thoughts on the optimized provision of this AS are presented together with the possible associated costs. These results will serve as a basis for the development of a coordinated voltage regulation/harmonic mitigation algorithm in future work.

Harmonics Control Algorithm
In this section the new control scheme of the grid-connected CI-DRES is described. The examined topology of the CI-DRES is a 3-phase 3-wire Voltage Source Converter (VSC) with an LC filter (depicted in Fig. 1), which ensures THDV<2.5% at normal conditions, i.e. when the CI-DRES injects active and reactive power at fundamental frequency. This is the actual filter topology of the CI-DRES prototype constructed at University of Sevilla, which is going to perform multiple AS as a part of the EASY-RES H2020 project. The input DC voltage of the DRES inverter was set vDC=730V and the switching frequency is 10kHz. Firstly, the voltages and currents at the CI-DRES PCC ( , , , ) and the current at the inductor of the LC filter, , , are measured and transformed to a synchronous reference frame, while the angle θ and angular frequency are detected through a PLL. This control scheme is illustrated in Fig. 1, where the reference signals are denoted with stars (*). The currents , and , are used to define errors between actual and desired values of currents are the PCC. The reference values of , at the 1 st harmonic are calculated by the reference active and reactive power. Then, the 1 st harmonic dq current is summed with the harmonic dq reference current, , * , generated by the Harmonics Control illustrated in Fig. 2, so as to derive the total , * , which in turn are compared with the actual output dq current. The error , is passed through PI controllers to generate the reference inductor current , * . The , * is then compared with the actual inductor current , and the error , between them is inserted into PI controllers. The modulation signals of the CI-DRES are calculating taking into account the inductor currents in dq reference frame, which are transformed into the abc frame. The proportional and integral gains of the current PI controllers are specified from the time constant of the current control, which is = 0.25 , as , = and , = respectively, according to [22]. With respect to the harmonics control in Fig. 2, both , and , are analysed per harmonic k via dq transformation considering as rotation angle with equal to k·θ. The dq voltages and currents are passed through a moving-average filter. The dq harmonic references voltages, , * , for each harmonic k are transformed again into the abc frame, , * , and are multiplied by Gk to generate the reference harmonic currents. In this control scheme it is considered that the low-order voltage harmonics 5 th , 7 th , 11 th and 13 th are compensated via the proper injection of respective harmonic currents, which are shifted 180° from the k voltage harmonic. The calculation of appropriate Gk per harmonic and the thermal capacity check with respect to the 1 st harmonic are depicted in Fig. 3. This block requires as inputs the voltages and currents in dq per harmonic. Firstly, this block checks if there is availability with respect to the 1 st harmonic. In case there is not, Gk is set zero. In case there is capacity, the total RMS current is calculated to check if there is available capacity with respect to the injection of harmonic currents and G control is activated. The value ei is equal to − . In case there is capacity (ei>tol), Gk is increased. The term tol denotes that there is a tolerance zone around and has been taken equal to 0.5A. In case ei lies within the tolerance zone, the value of Gk remains equal to the previous value (previous measuring period=50us) Gk_z. By adding this zone, it is ensured that the harmonic currents and power do not oscillate. The initial value of Gk is set equal to a factor fk per harmonic. This factor is equal to Where hk are the EN50160 limits of voltage distortion at nominal voltage, i.e. h5=6%, h7=5%, h11=3.5% and h13=3%.

Evaluation at LV grid level
The proposed scheme has been tested in PowerSim Software in the TD. The examined network is a modified version of benchmark European LV network Cigre Task Force C6.04, [23]. The NLLs are modelled as a 3-phase diode bridge with PNL=40kW at VNL=0.4kV, while the DC output of the bridge is a capacitance CNL=1mF parallel to a resistance, RNL=7.05Ω. This type of NLL has been chosen on purpose, since it is causes high THDV at harmonics 5 th , 7 th , 11 th and 13 th . The distortion power of the loads is measured based on the stationary frame, [24], and at nominal voltage it is equal to 34kVAr, therefore, it is comparable with their active power. The distortion power, Dh, of the CI-DRES is calculated for 5 th , 7 th , 11 th and 13 th harmonics according to the definitions of IEEE 1459 Std. The simulations concern the Cases depicted in Fig. 4. In all simulations the reactive power is set to 0. Firstly, several investigations where performed with the Case 1 configuration to examine the following issues: (i) which CI-DRES is most efficient when only 1 or 2 of them is activated; (ii) how does the size of the inverter affect the active filtering process. All studied scenarios of Case 1 appear in Table 1 and Fig. 5. This case is considered as base case, since all CI-DRES are in the main feeder. Table 1 provides the injected active Fig. 4 Examined Topology of the CIGRE LV network powers per node (P) and the respective apparent power (S). In order to study issue (i) the following scenarios are defined: In case denoted as "No Control" it is assumed that there is no active filtering control performed by the CI-DRES. In case denoted as "With Control" it is assumed that all CI-DRES perform G-control. In cases denoted as "R11", "R16" and "R18" it is assumed that only one CI-DRES performs G-control at the respective node. In cases "R11+R16", "R11+R18" and "R16+R18" it is assumed that two CI-DRES perform G-control in the respective nodes. In order to test issue (ii) regarding also the location of the CI-DRES, Sinv at nodes 11 or 18 is changed at either 10kVA or 15 kVA. The cases denoted as "S18=10k" and "S11=10k" means that the CI-DRES at node 18 or 11 respectively, has Sinv=10kVA and operates also at Pref=10kW, therefore, G-control is not activated because there is no capacity with respect to the 1 st harmonic. In the other two nodes there is remaining capacity and the CI-DRES perform G-control. In scenarios denoted as "S18=15k" and "S11=15k", all CI-DRES perform G-control and the CI-DRES apparent power at the respective node is increased. Therefore, this node has enhanced G-performance. respect to CI-DRES/NLLs penetration and mixture. The purpose of this examination is to check how the penetration and mixture of non-linearities affect the LV grid performance with and without the G-control and how the location of the NLL affects the nearby CI-DRES. The cases have been build considering each time that a CI-DRES or a NLL is introduced in the LV grid, and preferably next to an existing element. Therefore, the examined combinations include the cases of Fig. 4. The power quality index that is studied is the average THDV,% of the 18 nodes within the grid. In all cases that have been simulated with G control all the CI-DRES inject currents up to − when they have available capacity. It has been observed via these test cases that the CI-DRES with the lowest voltage senses also the highest THDV. This is justified by the fact that the NLL causes voltage drop to the nearby CI-DRES and simultaneously, it increases the THDV sensed by the CI-DRES. Therefore, these CI-DRES (9,18) always reach first the thermal capacity compared to the other CI-DRES nodes. Moreover, the Dh of each CI-DRES depends on the voltage value, hence for CI-DRES that operate at − with lower voltage exhibit larger amounts of Dh. With respect to both issues, it can be observed from Fig. 5 (c) that the G-control does not affect the voltage values, but only the THDV (evident also in Fig. 6(c)). With respect to issue (i), from Fig. 5 (a) and Table 2 it is noticed that when only one CI-DRES has active filtering capabilities, lowest THDV appears in "R18", while "R11" has the highest one, therefore, better compensation is achieved when G control is activated at 18. If two CI-DRES have this ability, the most efficient combination is "R16+R18". However, for the same apparent and active power, if all three CI-DRES are activated, they can achieve the lowest average THDV (4.32%). With respect to issue (ii) aggregated results are depicted in Fig. 5 (b). It can be deduced that by de-activating G-control (through no space of thermal capacity) at 18 leads to higher THDV with respect to "S11=10k". The lowest THDV appears when all three CI-DRES are activated and CI-DRES at 18 has more available capacity ("S18=15k"). Therefore, it is much better when Gcontrol is allocated at all the available CI-DRES, but the most efficient solution is that the most remote nodes have more available capacity. Table 3 contains the results with respect to the 2 nd round of investigations. Case 2, 4 and 6 have almost the same average THDV under no control, since all concern 3 NLLs. Including more CI-DRES with G-control is more efficient, e.g. Case 6 regards 5 CI-DRES with G-control and the THD drops at 5.2%. The same is true for Case 3 with respect to Case 1. Moreover, Cases 3, 4 and 5 concern 4 CI-DRES and different number of NLLs. The highest THDV appears in Case 5, where the NLLs are more than the other cases. Therefore, it can be deduced that the THDV is heavily affected by the presence of the NLLs, and not the CI-DRES. Between Cases 7 and 8, Although the average THDV is almost identical for these cases, G control is more effective when the NLL is placed at 13, because the branch has lower R/X ratio than the main feeder. Case 8 was selected to demonstrate the effectiveness of the G control and the sensitivity of the NLL/CI-DRES location, since under no control this case exhibits the highest average THDV. As it can be noticed in Fig. 6, individual Harmonics are compensated in the most critical nodes with the applied control and in all cases the individual HD remains below the respective hk. Although there is load at node 13, highest THD appears at 15, 17 and 10 (NLL nodes) and 18, 9, (CI-DRES nodes). It is noted that although in the branch 4-15 there are two NLLs, the THDV is not as high as at nodes 9,10,17,18 of the main feeder. This is because the R/X ratio is much smaller at the branch than in the main feeder. Since the nodes with the highest THDV are the remote nodes with the NLLs and their respective nearby CI-DRES, the voltage value at these CI-DRES nodes is also low, due to the existence of the NLLs. Similar to the voltage rise, the most remote nodes have a Voltage THD per node, %

Nodes 3 Inverters-2 Loads
No control With G Control S18=10KVA S11=10KVA S18=15KVA S11=15KVA  R11  R16  R18  R11+R16  R11+R18 R16+R18 S18=10KVA S11=10KVA S18=15KVA S11=15KVA higher sensitivity with respect to others. In a scenario where both voltage regulation and active filtering process are about to be performed simultaneously, these two AS will probably not coincide for the same CI-DRES, since the CI-DRES with higher THDV will have also lower voltage (closer to a NLL), while CI-DRES with higher voltages will be the ones to be assigned with reactive power provision. In any case, for both voltage regulation and active filtering, the CI-DRES at most remote nodes need to have an oversized converter with respect to the primary source. The optimal allocation of the harmonic mitigation among the CI-DRES has not been examined yet in the technical literature. A preliminary harmonic voltage sensitivity analysis performed by the authors with respect to the location, the amount of harmonic current by the CI-DRES at specific harmonic, and the phase of this current has shown that the value of the harmonic voltage depends heavily on the phase of the injected current. More specifically, the harmonic voltages are minimized when the downstream CI-DRES injects a current shifted 180° with respect to the respective harmonic current by the NLL. In this paper, the injected harmonic currents are shifted 180° with respect to the harmonic voltages, which have a small angle (0-3°). Of course, more mathematical analysis is required, but those preliminary cases revealed that in order to sense the phase of a specific harmonic current of the upstream NLL, highly qualified measuring devices need to be employed at both NLLs and CI-DRES. Consequently, in the optimization process several parameters need to be taken into account: i) the operational losses on the feeders and the CI-DRES: According to [25], [26] the operation of the PV converters at relatively low power or with low power factor can cause operational power and energy losses; ii) the thermal capacity of the CI-DRES when performing this AS: In the case of over-voltage mitigation in distribution systems, it has been proved that the CI-DRES located at the most remote nodes shall provide reactive power (since they have higher sensitivity), therefore, it is recommended that the DRES nominal power is oversized up to SN,DRES/PPV=1.41, [27]. Similar to reactive power, this AS needs an additional oversizing, especially at the most remote nodes close to a NL load, hence, there is an additional involved purchase cost. iii) costs related to the CI-DRES controller implementation, e.g. in order to receive a signal related to other harmonics than the fundamental from a central controller, the CI-DRES is required to possess a highly qualified DSP; iv) for a centralized implementation of a coordinated harmonic mitigation, each consumer with high harmonic pollution needs to possess a specialized measuring device to detect the voltage harmonic distortion in each low-order harmonic. In [28] an extension for PMUs is proposed, so that they include measuring capabilities for proper calculation of harmonic currents and voltages as defined by the IEEE 1459:2010 Std. However, including PMUs at distribution system level is rather an expensive approach. v) In order to prove the effectiveness of an optimized voltage harmonic mitigation by CI-DRES, in the optimization procedure, the constraints with respect to the individual harmonic currents proposed in [2] and used in [9], [10] need to be relaxed. Special attention needs to be paid also in the objective function to be minimized. In this paper the average THDV was studied, however, probably the average (with respect to nodes) individual HD per harmonic needs to be taken into account or the RMS THDV at all nodes. vi) A central controller within the distribution network will probably need to perform sequential harmonic power flows per harmonic to achieve an optimized result. The involved costs need to be considered for this implementation. Therefore, it is crucial to include all costs involved for this AS in an optimized procedure, since it is yet not an established one by the Grid Codes. As an example, it is noted that even reactive power, which is already prescribed as system support function, is not remunerated for CI-DRES, [29].

Conclusions
In this paper, a voltage harmonic mitigation control strategy is developed for the CI-DRES, so that it does not exceed its thermal capacity. Secondly, several scenarios with respect to different penetration and mixture of CI-DRES and NLLs in the CIGRE Residential LV Network have been examined via TD simulations, to demonstrate the effectiveness of the proposed method vs. low-order harmonics, while the CI-DRES operate close to their thermal capacity. Furthermore, this paper comes to provide a further insight on the evaluation of the CI-DRES contribution for providing this AS with respect to its location within the grid and its available capacity. The proposed control is efficient for mitigating individual voltage harmonics. In all cases it has been shown that: 1) the average THDV is reduced; 2) the CI-DRES downstream an NLL mitigates harmonics more efficiently: The location of the CI-DRES plays an important role in a similar manner as in the voltage regulation process, i.e. the DRES placed at the most remote nodes may be required to inject higher amounts of distortion power, therefore, the CI-DRES may require converter oversizing. In the future, CI-DRES may be required to perform several functions simultaneously, harmonic mitigation needs to be coevaluated with respect to other AS, as well, both at DRES and system level. Since in the state-of-the art the harmonic filtering is performed in a decentralized manner, the evaluation presented in this paper will serve as a basis for the development of a distributed coordinated algorithm in future work, allocating properly two AS: active harmonic filtering and reactive power exchange without exhausting the DRES converters' thermal limit. This algorithm will be based on techno-economic criteria, taking into account the analysed costs and losses, so as to enable the proper allocation and remuneration of the DRES owners and make the trading of these AS in respective markets feasible.

Acknowledgements
This work is part of the EASY-RES project that has received funding from the European Union's Horizon 2020 Research & Innovation programme under Grant Agreement No 764090.