Portal de Periódicos da CAPES

New reward and penalty scheme for electric distribution utilities employing load‐based reliability indices

2018; Institution of Engineering and Technology; Volume: 12; Issue: 15 Linguagem: Inglês

10.1049/iet-gtd.2017.1809

1751-8695

Bo Wang, Jorge Alexis Camacho, Gary Michael Pulliam, Arash Etemadi, Payman Dehghanian,

Electric Power System Optimization

IET Generation, Transmission & DistributionVolume 12, Issue 15 p. 3647-3654 Research ArticleFree Access New reward and penalty scheme for electric distribution utilities employing load-based reliability indices Bo Wang, Corresponding Author Bo Wang wangbo@gwu.edu orcid.org/0000-0001-7972-1633 Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this authorJorge Alexis Camacho, Jorge Alexis Camacho United States Department of Commerce, National Institute of Standards and Technology, Washington, D.C., USASearch for more papers by this authorGary Michael Pulliam, Gary Michael Pulliam Division of Infrastructure and System Planning, The Public Service Commission of the District of Columbia, Washington, D.C., USASearch for more papers by this authorAmir Hossein Etemadi, Amir Hossein Etemadi Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this authorPayman Dehghanian, Payman Dehghanian orcid.org/0000-0003-2237-4284 Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this author Bo Wang, Corresponding Author Bo Wang wangbo@gwu.edu orcid.org/0000-0001-7972-1633 Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this authorJorge Alexis Camacho, Jorge Alexis Camacho United States Department of Commerce, National Institute of Standards and Technology, Washington, D.C., USASearch for more papers by this authorGary Michael Pulliam, Gary Michael Pulliam Division of Infrastructure and System Planning, The Public Service Commission of the District of Columbia, Washington, D.C., USASearch for more papers by this authorAmir Hossein Etemadi, Amir Hossein Etemadi Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this authorPayman Dehghanian, Payman Dehghanian orcid.org/0000-0003-2237-4284 Department of Electrical and Computer Engineering, The George Washington University, Washington, D.C., USASearch for more papers by this author First published: 15 June 2018 https://doi.org/10.1049/iet-gtd.2017.1809Citations: 12AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Electric distribution utilities are required to continuously deliver reliable electric power to their customers. Regulatory utility commissions often practise reward and penalty schemes to regulate reliability performance of utility companies annually with respect to a desired performance targets. However, the conventional regulation procedures are commonly found based on the customer-based standard reliability indices, which are not able to discern the service characteristics behind the electric meters and, hence, fail to holistically characterise the actual impact of electricity interruption. This study proposes a new method to evaluate the load-based reliability indices in power distribution systems using advanced metering infrastructure data. Furthermore, the authors introduce a reward/penalty regulation scheme for utility regulators to provide a reliability oversight using the proposed load-based reliability metrics. The new load-based reliability metric and the reward/penalty scheme proposed bring about superior advantages as the distribution grids become further complex with a high penetration of distributed energy resources and enabled microgrid flexibilities. Numerical analyses on different settings with and without microgrid considerations reveal the applicability and effectiveness of the proposed approach in real-world scenarios. Nomenclature forecasted value of ASIDI annual severe weather impact factor for a regional distribution system benefit of increasing feeder reliability to avoid PF (in $) benefit of customer reliability premium (in $) cost of compensation for long outages CI composite index for evaluating feeder reliability interruption duration of outage event i ENS forecasted value of annual energy not supplied (ENS) to the feeder equivalent ENS to the feeder considering the impacts of microgrid ENS to the electric vehicles caused by failing to charge or swap the batteries ENS to the feeder during outage event i effective load control of non-critical loads during outages G effective generation supplied to the customer during outages i index of an interruption event IEAR interrupted energy assessment rate (estimated cost per unserved kWh during outage event i) IR incentive rate of utility regulation interrupted load in kVA for the outage event i total connected load served N time steps of load forecasting PF feeder penalty factor forecasted value of PF S annual energy supplied to electric vehicles by rescheduling the service t index of time intervals (1 to T) upper and lower limits of weight factors for ASIFI, ASIDI, and ASSDI, respectively real and forecasted value of interrupted load 1 Introduction 1.1 Problem description Electric utilities are continuously seeking solutions to engender a more reliable, cost-effective and interactive power distribution systems through advanced technologies and modernisation efforts supported by the regulatory commissions [1]. Smart grid technologies are deployed to accomplish this modernisation mission and meet the intensified sustainability goals with smart meters, smart appliances, electric vehicles (EVs) and distributed energy resources (DERs), among others [2]. Advanced metering infrastructure (AMI), which consists of smart meters, communication technologies, meter data management system (MDMS), and the associated software/hardware platforms, enables active interactions between the smart grid components. Each end user connected to a node and associated with a smart meter in an AMI system is characterised as a customer regardless of its load scale. Despite the undeniable advantages, smart meters generate data with high velocity and variety resulting in several challenges ranging from tremendous volumes of data to be processed and complicated AMI architectures that are not easy and practical to develop [3]. In a hierarchical AMI architecture, data is automatically collected from customer meters and communicated to the utility MDMS through data access points [4, 5]. AMI implementation enables visualisation of the distribution system assets, operating states, and prevailing conditions including outage events [6]. It also enables more accurate reliability assessments by updating and uploading outage information to the utility database and analytic platforms [7, 8]. Optimal set of maintenance strategies can be adopted based on the outage information corresponding to the utility-controlled territory to improve the system reliability performance requirements [9]. Most utility commissions solely track the system average interruption frequency index (SAIFI) and system average interruption duration index (SAIDI) metrics to evaluate system reliability and reward/penalise the electric utilities accordingly depending on their performance with regard to the desired targets and requirements. Reward and penalty schemes (RPSs) are, hence, designed to regulate the performance of the electric distribution companies based on the reported reliability performance metrics [10–13]. However, the aforementioned two customer-based reliability indices are dominated by residential customers [14]. For instance, based on the data provided by the local utility, the US District of Columbia features 99% penetration of AMI, residential customers in this area consume 17% of total load but accounting for 90% of the AMI customers. Some utilities have started migrating to a new decision paradigm by including the momentary average interruption frequency index (MAIFI) as part of their reliability performance evaluations but load-based reliability indices such as average system interruption frequency index (ASIFI) and average system interruption duration index (ASIDI) are still not widely used. The challenge to wide adoption of such load-based reliability metrics is acquiring information on the quantity of the interrupted load, which could be more challenging than the number of interrupted customers [14]. With the increasing trend in penetration of distributed renewable ERs (DERs), local storage units, and demand response programmes and load control mechanisms, real-time assessment of interrupted loads becomes more and more challenging than ever before. As a result, reliability regulatory policies should also go through a transformation to meet such emerging challenges in future. 1.2 Literature survey In exploring the existing literature, an automated reliability assessment mechanism is designed in [15] to calculate both customer-based and load-based key reliability indices, where pre-outage kVA is utilised to quantitatively assess the ASIFI and ASIDI metrics. In [16], the annual average number of connected loads is utilised to calculate the ASIDI metric and to design the RPS for electric utilities. However, the ASIDI metrics calculated in [15, 16] are unable to reflect load profile variations and DER spatio-temporal impacts. The SAIFI and SAIDI indices of reliability are modified in [17] to incorporate the priority and corresponding penalty factor (PF) for interrupted load of each customer when direct load controls of all consumers are enabled. However, the energy not supplied (ENS) during an outage event still needs to be assessed to account for the amount and duration of load interruptions. New metrics have been proposed to assess the reliability of microgrids in [18] and to optimise the DER allocation in [19], through value-based reliability planning approaches. The simulation results based on load point average failure rate and average outage duration in [20] are different from the true values captured during interruptions. Several techniques for analysing utility long-term investment plans are suggested in [21, 22], where the reliability indices and utility regulations are overlooked. A multi-agent system architecture for virtual power plants has been introduced to manage smart grids and forecast energy demand in [23], where a detailed model for low-level management of virtual power plants is introduced. With a lower load forecast error, virtual power plants are shown to achieve a decentralised intelligent management and communication with other agents through negotiation. Load forecasting (LF) at the feeder level and even the consumer level through AMI data is also approached in [24–28], where numerical results indicated acceptable short-term LF (STLF) performance with appropriate load aggregation levels. Note that the aforementioned references neither evaluated the predictability of load-based reliability indices nor calculated the load-based reliability metrics considering high penetration of DERs. Currently, public literature available specifically on studying the RPS for distribution utilities considering the impacts of microgrids is scarce and research efforts must be focused to address this emerging topic of interest from the perspective of a utility regulator. 1.3 Contributions The contributions of this paper are three-fold: (i) to explore the applicability of the load-based reliability metrics and calculate such reliability indices of ASIDI using the AMI-captured load data and LF techniques; (ii) to introduce a new reliability index and an AMI-assisted reliability assessment architecture to incorporate the impacts of microgrids; (iii) to propose a novel RPS to regulate the reliability performance of distribution utilities based on specific feeder characteristics and via employing the proposed load-based reliability metrics. This paper is organised as follows. In Section 2, we present the proposed AMI-assisted architecture for assessing the suggested load-based reliability metrics in power distribution systems. A new reliability regulation mechanism from the utility regulator perspective using the proposed load-based reliability metrics is introduced in Section 3. Numerical case studies based on both traditional and the proposed RPS schemes are conducted in Section 4, where the impacts of microgrids and DER penetrations are extensively explored. Moreover, finally comes the conclusions in Section 5 that summarises this paper contributions. 2 AMI-assisted reliability assessment Hourly customer load profiles for several years are usually uploaded and stored to MDMS by utilities through an AMI platform. Hence, we regard the mean total aggregated load as the average load, , in contrast with the transformer rated kVA, which has been the common practise in the past. In this section, hourly load data from smart meters is used to calculate the ASIDI. The missing data of smart meters is treated as 0 kW. 2.1 Calculation of ASIDI using AMI data Different outage events are characterised with different interruption durations and impose different system-wide impacts. Fig. 1 illustrates the distribution of interruption frequency and number of interrupted customers for the District of Columbia distribution system from 2011 to 2015. As one can see from Fig. 1, interruption duration and number of interrupted customers are highly correlated: 91.5% of the average outage durations are <6 h and affect 79.0% of the total interrupted customers. Hence, majority of the outage events can be forecasted through an STLF mechanism with the lead time of 1 h to 1 day ahead. The main uncorrelated observations are related to the long interruption duration outages with a large number of customers affected but with lower frequency. Medium-term LF (MTLF) can be applied for outage events that last more than 24 h with relatively large number of affected customers. Fig. 1Open in figure viewerPowerPoint Interruption frequency against restoration time (in percentage), and number of customers interrupted against restoration time (in percentage). Daily mean restoration time is utilised and the correlation coefficient is 0.95 Algorithm 1.Algorithm for calculating ENS 1: Import the hourly customer load profiles of the feeder for 5 years as well as the yearly outage report. Calculate the average load . 2: Calculate the outage duration for each outage event i. 3: For each outage event i, if , consider the pre-outage load as the interrupted load; otherwise, aggregate the load profiles of interrupted customers, forecast N -step ahead interrupted load. 4: Calculate for the outage event i. 5: Sum to get the ENS. We propose Algorithm 1 to calculate the annual of the feeders. Historical customer load profiles are aggregated to predict the hourly interrupted load during each outage event considering different chronological and weather conditions. Then, ENS metrics during outage events () are calculated considering outage start and end times, and then added up to evaluate ENS. In step 3, if outage duration is ≤1 h, we use pre-outage load profile as the load does not change much in an hour and load forecast with resolution of <1 h is hard to achieve in this model. In case where the interruption duration is longer than 1 h, we modified the neural networks (NNs) in [29] to forecast the interrupted load. LF time horizons vary from 1 h to 1 week and we forecast N as EndHour–StartHour + 1 h-ahead load. When N is ≤24 h, we use STLF; otherwise, MTLF is applied. The interrupted load profiles are aggregated utilising the AMI data. Then, the Levenberg–Marquardt approach with 22 hidden neurons are used to train the model. The ASIDI index of reliability is calculated as in the equation below: (1) We use mean absolute percentage error (MAPE) to measure the forecast performance [25]. The MAPE is defined as the ratio of the absolute forecast errors and the actual observed values (2) where are actual values, are corresponding forecast values, and is time series. Two feeders in the US District of Columbia are used here to demonstrate the effectiveness of the proposed algorithm. The anonymous load profile data and historical outage reports of the two feeders are provided by the Potomac Electric Power Company. Both feeders are overhead lines feeding residential loads and a few commercial customers. Feeder 1 supplies 549 customers with an average load demand of 1.47 MW, whereas Feeder 2 supplies 1476 customers with the average load demand of 2.46 MW. We used 3 years historical load data from June 2013 to May 2016, where the first 2 years of the data were used for training and estimating the model parameters, and observations from the past year were utilised for model evaluation and performance verification. We also employed historical temperature data of the Ronald Reagan Washington National Airport (DCA) from the National Oceanic and Atmospheric Administration [30]. We randomly select subsets of customers with different load scales from the two feeders, aggregate their load profiles, and sample ten times for each load scale to estimate the LF error of the interrupted load. The average MAPE for the forecasted interrupted load by STLF and MTLF techniques are demonstrated in Figs. 2 and 3, respectively. The MAPE corresponding to the forecasted interrupted load increases when the aggregated load decreases and interruption duration increases. According to Fig. 2, the average MAPE for a load scale of 50 kW is higher than 10% and even unstable when interruption lasts more than 4 h. However, the average MAPE of interrupted load with the load scale of more than 500 kW is relatively low and does not increase much even when the interruption lasts for several days as shown in Fig. 3. The MAPE of the forecasted load decreases through load aggregation, which is commonly expected in the industry practise. However, we demonstrate here the performance of the proposed Algorithm 1 in this application: the MAPE associated with the ASIDI metric is small when the feeder has medium or low reliability level. The reason lies in the facts that (i) the MAPE associated with a large amount of interrupted loads is low and less sensitive to interruption duration and (ii) the contributions of the severe outage events dominate the reliability index of ASIDI. The MAPE associated with the ASIDI further decreases as LF error is approximated by a normal probability distribution with mean value of 0. When the feeder is more reliable, less load will be interrupted and, hence, the ASIDI metric becomes smaller while the corresponding MAPE is relatively large. Fig. 2Open in figure viewerPowerPoint Average MAPE on the STLF results against the forecast lead time of 1, 2,…, 24 h ahead. Each line represents the aggregated load scale Fig. 3Open in figure viewerPowerPoint Average MAPE on the MTLF results against the forecast lead time of 1, 2,…, 7 days ahead. Each line represents the aggregated load scale 2.2 Proposed load-based reliability metrics considering penetration of DERs and EVs Microgrids can improve the reliability performance of the power distribution systems. Load interruptions can be reduced through an effective allocation and utilisation of the distributed generation supply and direct load control mechanisms. Despite its advantages, it also brings about difficult-to-manage challenges for public utility commissions (PUCs) to collect the high-resolution AMI data since many microgrids are owned by the customers and/or third parties. The utility may not have access to AMI data from the microgrids as utility commissions only regulate utility assets. In this paper, we introduce observation point to reflect the scope of data collection from the feeder. We define a smart meter as observable if the information behind the smart meter can be acquired. Consequently, a utility-owned microgrid is observable as the utility can gather all the information behind the microgrid substation meter. A single meter connected to a node is unobservable as there is no meter behind. If a third party or residential facility owns a microgrid and they are willing to share information behind the meter with the utility company, this smart meter connected to the microgrid is observable too. Hence, the distribution system reliability assessment will terminate on where the utility observation points end, which is in contrast with the traditional view of to the point where the utility assets are covered. The equivalent ENS of each feeder is calculated as follows in the equation below: (3) where ENS is the forecasted energy demand of the interrupted customers, is the effective load control applied to non-critical loads, and G are the loads partially supplied via DERs which are measured via the connected production meter. The procedure to calculate is presented in Fig. 4. For each feeder, we first check if each interrupted node is observable. If it is observable, we acquire the load profile, measured generation, and effective load control corresponding to the load being served at that node. If it is unobservable, we only gather the load profile from the connected smart meter. We then aggregate the load profiles of the customers that are interrupted during the same time interval, then forecast the load demand, and eventually calculate ENS as introduced earlier in Section 2.1. Fig. 4Open in figure viewerPowerPoint Implementation procedure of the proposed offline reliability assessment loop We assume EVs are charged under plug-in or battery swapping mode and their daily charging curves are stable. An aggregator is assumed to be responsible for charging the EVs so as to meet the customer demand and also to coordinate the operation of the energy management system under the plug-in mode. When an interruption occurs and is recovered before the EV departure time, the state of charge (SOC) of the EV battery is metered: if it is less than the expected SOC at the EV departure time, the difference between expected SOC and captured SOC actually reflects the ENS to the customer and can be assessed by the aggregator. If the aggregator charges the EV battery to expected SOC after the interruption but before the EV scheduled departure time, we regard the EV load as not interrupted. The extra energy supplied to the load demand during the interval between the restoration time and the EV departure time is the difference between the actual energy supplied to the load and the scheduled energy supplied in that time interval and is actually amounted equal to the shifted load. Under battery swapping mode, EVs can utilise the battery swapping stations (BSSs) connected to adjacent feeders without interruption, if customers are well informed and the batteries are not already depleted (i.e. the BSS connected to the interrupted feeder can supply power to the feeder). We assume the EVs belonged to the interrupted feeder can swap their batteries with a discount; hence, the battery swapping process can be recorded and the increased SOC of their batteries can be regarded as the shifted load. Note that the EV load interruption caused by a failure in battery swapping, leading to a travel delay, can be acquired via post-surveys. We define the EV service interruption S as the energy interrupted but re-dispatched via aggregators or BSSs. Rather than a direct outage, EV service interruption can lead to an intensified operational cost or customer inconvenience indirectly. With the low penetration of EVs, the utility may jointly forecast the customer and the EV loads, and, hence, the shifted load S can be regarded as the effective generation G during the interruption to calculate . With the high penetration of EVs, the utility may forecast the load and schedule the controllable loads separately, and therefore the total interrupted energy will be represented as . We still use pre-outage served (connected) kVA to calculate the ASIFI index of reliability, as introduced in (4). The concept and its calculation procedure are detailed in [15] and reflect the instantaneous load interruption. ASIDI in [31] is adjusted in order to reflect sustained interruptions and incorporate microgrid impacts as shown in (5). We propose the average system service disruption index (ASSDI) as a new load-based reliability index to reflect service interruption to flexible loads such as EVs, as introduced in the equation below: (4) (5) (6) 2.3 Proposed AMI architecture Fig. 5 demonstrates the proposed AMI architecture: a hierarchical system including smart meters, neighbourhood area networks (NANs), and the local area networks (LAN) within the utility domain. In contrast with the conventional AMI structures through which meter data is directly uploaded to MDMS, the proposed architecture contains two inter-connected loops: (i) the online loop represented by red lines is mainly focusing on grid monitoring and power flow controls. Since the main concerns for the LAN are system operational states and power flow constraints, this online computational platform can swiftly classify and process the measurements and upload the necessary information. The online loop normally transmits data every 2 s to 5 min; (ii) the offline loop represented by blue lines store all data in the NAN-level database. Having evaluated the index associated with the neighbourhood feeders as shown in Fig. 4, NAN then calculates the three introduced load-based reliability indices using (4)–(6) and upload the results to LAN. The LAN-level billing system then maps different reliability performance levels of system feeders and analyses the overall reliability of the distribution system. The utility can also send the locational marginal prices (LMPs) to the NAN and calculate the electricity pricings at the NAN-level billing system. Fig. 5Open in figure viewerPowerPoint Logical view of the proposed advanced meter infrastructure This new architecture is suitable for prosumer-oriented smart grids with highly densed penetration of DERs and can be employed even by electric utilities with high-speed data transfer requirements. The primary advantages of the suggested AMI architecture can be summarised as follows: It preserves and protects customer privacy in the physical layer of communication network systems. The customer load profiles are actually to be stored in the NAN networks and will not require to be uploaded into a centralised data centre. It shrinks the volume of data uploaded into the utility MDMS platform. With the envisioned online and offline computation loops, the measured data is classified and processed before an upload process starts. All the information can be recorded in the NAN and, hence, the communication capacity requirement from NAN to LAN significantly decreases. It brings about potentials for more accurate load forecast and feeder health diagnosis with local weather and temperature information as well as feeder-level data analytics. It offers opportunities for fast and efficient energy management in distribution systems. With the suggested online computation loop, wide area networks can acquire robust real-time system operational conditions from distribution feeders. Through a system-wide optimal power flow mechanism, the feeder-level energy management signal can be sent to NANs. As NANs gather smart meters data and monitor the feeder in real time, they can determine the operation points of each DER and load control signals for each connected node. 3 Proposed utility regulation model A distribution utility owns, monitors, and controls hundreds of feeders, the reliability performance of which is closely dependent on the reliability characteristics of its feeders. We classify feeders of the distribution system based on different reliability requirements, and we propose a mechanism for their regulation via contracts between the distribution utilities and the PUCs. The suggested utility regulation model accommodates the new load-based reliability metrics using the AMI data and captures well the utility requirements under high penetration of DERs and microgrid-enabled flexibilities. Details of the proposed regulation model are presented as follows. 3.1 Feeder PF For the distribution feeders with AMI infrastructure installed, the ASIFI, ASIDI, and ASSDI metrics can be assessed using AMI data. We propose the PF in (7) that integrates the aforementioned reliability indices and can be utilised to penalise the utilities if they do not meet the reliability performance requirements (7) where IR is the incentive rate to reflect the penalty and reward characteristics and is the impact factor driven by the annual severe weather conditions and their impact on the distribution system. We introduce a composite index (CI) to represent the difference between the expected and the realised reliability performance (8) where , , and are the weight factors for each reliability metric assigned by the distribution utility experts and they add to one. , , and are expected values set by utility regulators. If , the utility meets the customer requirements and provides extra high-quality reliability services and, hence, it should be rewarded. By improving