Montorng large-scale power dstrbuton grds D. Gavrlov, M. Gouzman, and S. Lury Center for Advanced Technology n Sensor Systems, Stony Brook Unversty, Stony Brook, NY 11794 Keywords: smart grd; sensor network; autonomous sensor Abstract Power grds are dstrbuted over vast geographcal areas and have sophstcated multlayered archtecture. The structure of the grd dstrbuton layer s often poorly documented and sometmes unknown, presentng addtonal challenges to the development of systems for automated montorng of power delvery to consumers. The proposed system performs the smultaneous functons of estmatng the power grd topology (map) and montorng of the grd operaton. The core of the system s the dstrbuted network of sensors nstalled at the branchng ponts of electrcal conductors. The sensors perodcally measure the RMS current n the conductor, and the phase shft between current and voltage. Localzaton and tme synchronzaton of sensors are performed usng GPS modules. The sensors communcate over the powerlne conductor. Transformers block communcaton sgnals, separatng the network nto clusters. The maps of the grd segments are reconstructed for each network cluster and then combned nto the full grd map. The map s used for real-tme montorng of nconsstences n the grd behavor to detect conductor breakage, powerlne overload and possbly electrcty theft. The autonomous sensors are nductvely powered; auxlary solar cells are nstalled as backup power source. Introducton Smart electrcal power dstrbuton grds represent rapdly developng feld. When mplemented, smart grds wll provde more effcent transmsson of energy, quck restoraton of power after breakdown, smooth ntegraton of renewable energy sources, and reduced cost of management and operaton for utlty companes. Achevng those benefts s possble only f there exsts a developed system for real-tme montorng of the grd. Montorng of energy flows n the grd becomes partcularly mportant as the role of dstrbuted energy producton from renewable sources (e.g. solar and wnd power generaton statons) s contnuously ncreasng, contrbutng to complexty of the network. A montorng system must collect, store and process the detaled nformaton on the grd operaton. The results of processng are used for control and
management of the grd: some of the control functons may be automated, some functons are performed by human operators. Currently, grd montorng s mplemented n hgh voltage transmsson layers of the grd. The montorng s performed by usng expensve sensors, whch requre complcated nstallaton procedures and use dedcated data transmsson channels. Several manufacturers are offerng such sensors. For example, the sensors manufactured by Aclara/Tollgrade Inc., are desgned for montorng medum voltage powerlnes, wth voltage n the range from 4 KV to 46 KV and current rangng from 0 to 17,000 RMS amperes. The sensor optons nclude cellular and W-F communcaton modules. The sensor s powered by energy, harvested from electromagnetc feld of the host conductor, wth mnmum operaton current as low as 3 amperes. Sensor nstallaton requres ground connecton for montorng the lne voltage. Smlar sensors are manufactured by GE, Cooper Power Systems, and Sentent. Unfortunately, the exstng technology s not sutable for montorng local dstrbuton layers of the network. The typcal dstrbuton layer represents a complcated system of nterconnected conductors, connectng large number of ndvdual consumers to transmsson layer of the network. Detaled montorng of such a system requres large number of nexpensve sensors. The cost and complexty of nstallaton of the sensors s also mportant: the sensors should be nstalled wthout dsconnectng the lne voltage. Ths s possble only f there s no galvanc connecton between sensors and the powerlne. For example, the sensor can be snapped around the powerlne conductor on top of the nsulaton. Such method of nstallaton lmts the montorng capabltes of the sensor: all the montored parameters should be estmated by analyzng the alternatng electromagnetc feld around the conductor. Current technology allows to equp each sensor wth nexpensve and accurate GPS module, whch determnes locaton of each sensor, as well as allows global tme synchronzaton of all sensors. Organzaton of the sensor network s also mportant. Most of the proposed montorng systems are based on wreless ad-hoc networks, n whch sensors autonomously determne optmal paths for delvery of packets to nearest Ethernet connected hub. Another common opton s to nstall a network Ethernet connected communcaton modules along the path of power dstrbuton, wth each module communcatng to one or several sensors, located n ts vcnty. The overvew of wreless networks for grd montorng may be found n [1]. Some exstng solutons are usng cellular networks to communcate the collected data to the central server. An example of such network for montorng long dstance transmsson lnes s presented n [2]. In ths network, sensors, nstalled on a sngle support form a cluster and communcate wrelessly wthn the cluster. Only one sensor n the cluster, equpped wth a cellular network modem, transmts ntegrated data from the whole cluster. Applcatons of wreless communcaton technology for grd montorng may be found n [3]. Harvestng of energy from stray electromagnetc feld around the powerlne for applcaton n smart grd sensors s well known n the ndustry. The dscusson of nductve and capactve power harvesters may be found n [4] and [5]. We use nductve energy harvestng as
the man source of power for the sensor. The sensor s also equpped wth auxlary solar cells, for emergency operaton durng prolonged powerlne falure In ths paper we dscuss aspects of mplementaton of a system for montorng of dstrbuton layer of the grd. In Secton 1 we descrbe the proposed method for grd montorng. The method s based on measurement of real-tme parameters of ac current n branches of the network, and subsequently usng those parameters to estmate the topology of the network and montor the network ntegrty. The GPS locaton nformaton may be used together wth the estmated topology to create geographcal map of the network. In Secton 2 we dscuss the desgn of sensors. The method of measurement of ampltude and phase of ac current n a conductor, as well as the drecton of energy flow, usng stray electromagnetc feld around the conductor s dscussed. We also explore power harvestng optons, useful for long-term autonomous operaton of the sensor. The deployed sensors are self-organzed nto a network for communcaton and transmsson of collected data. The archtecture of the network s dscussed n Secton 3. The sensors are usng the carryng conductor as the prmary communcaton medum. We propose a method for clusterzaton of the network, reducng the communcaton load and smplfyng data collecton. In Secton 4 we dscuss the methods for collectng and processng of data from the sensor network. The method for estmaton of grd topology and generaton of the grd map s outlned. The approach to grd montorng for detecton and dagnostcs of power outages, and detecton of electrcty theft s dscussed. 1. Prncples of Power Grd Montorng. Montorng of smart power grd requres real-tme acquston of complete nformaton about power flow n all branches of the network. Collectng data from local dstrbuton grds s complcated by the fact that the topology of such grds s often unknown and orderly plannng of the sensor network s rarely possble. Ths can be mtgated by deployment of dense sensor network, whch collects suffcent nformaton to recover the topology and buld the map of the grd. The mplementaton of dense sensor network requres collecton and processng of large quanttes of real-tme data. Collecton of such amount of data s possble only f the communcaton channel s properly organzed. Our method for grd montorng employs a dense network of nexpensve sensors nstalled at each branchng pont of the powerlne. Ths network splt nto clusters such that sensors wthn each cluster communcate by sendng hgh-frequency sgnals over the powerlne tself. The archtecture of the dstrbuton grd s shown n Fg.1. The grd conssts of conductors, electrcally connected va step-down transformers to the conductors of upper layers. The conductors of the lowest layer are termnated wth transformers, convertng lne medum voltage to low voltage, dstrbuted to ndvdual consumers. The sensor network s usng powerlne as the communcaton medum. The powerlne communcaton s usng frequences n the range 150-500 khz, whch s much hgher than powerlne frequency 60 Hz. Due to hgh mpedance of step-
down transformers to communcaton carrer frequences, communcaton between sensors nstalled on dfferent conductors becomes mpossble. Therefore, we can consder the sensor network to be separated nto clusters, wth the sensors nstalled on the same conductor belongng to the same cluster. The sensors may easly self-organze nto clusters at the tme of nstallaton by dentfyng themselves to the network. The most complete data on the grd may be obtaned by nstallng sensors around each branchng pont. A segment of a grd conductor wth two branchng ponts and nstalled sensors are shown n Fg.2. Each sensor s measurng several parameters that nclude RMS of the current n the conductor, phase of current wth respect to global synchronzaton pulse and phase shft between voltage and current. The estmated phase shft between current and voltage may be used to determne the drecton of energy flow through the conductor. The prncples of measurement of phase shft are demonstrated n Fg.3. Phase shft s computed as dfference between the tmes of zero crossng for current I(t) and voltage V(t) sgnals (see Fg.3a). The precson of measurement of phase shft s lmted by the accuracy of detecton of zero crossng. Snce voltage sgnal s nosy, phase shft may be estmated wth any reasonable precson by averagng the results of many measurements. More accurate estmaton of phase of current I(t) may be obtaned by measurng tme between global synchronzaton pulse and the pont of zero crossng of I(t) (see Fg.3b). Global synchronzaton pulse s generated by GPS modules n all sensors of the network smultaneously. Therefore, the phases of I(t) measured by each sensor may be used to accurately analyze phase shfts of current between any branches n the grd. The method of grd montorng s based on verfyng match between currents at branchng nodes usng Krchhoff s Current Law (KCL), whch may be wrtten n the phasor form as follows j Ie 0, where I and are the magntude and phase of the current flowng nto the node from the -th branch. Verfcaton of KCL for each node wll ndcate ntegrty of the grd. The currents may also be montored over tme, and the phasors of branch currents may also be compared wth those measured by sensors nstalled at the other nodes of the network. The consstent match between currents reported by sensors ndcates that those sensors are nstalled on the same branch of the network. If two sensors that belong to the same branch are located around two dstnct nodes, t may be consdered that the branch s connectng those two nodes. Ths prncple, appled to the whole network, may be used to estmate the grd topology.
Fg.1. Dstrbutons layers of the power grd represent the set of conductors, electrcally connected by transformers. Due to the hgh mpedance of the transformers to communcaton sgnals, the sensors are naturally separated nto clusters, composed of the sensors nstalled on sngle-lne branched conductor. Fg.2. The sensors are nstalled around each branchng pont (node) of the conductor and montor the RMS and the phase of ncomng and outgong currents.
Fg.3. Illustraton of two optons for measurement of the phase of current I(t). The phase can be measured wth respect to the phase of voltage sgnal V(t) (blue curve) (a) or global synchronzaton pulse (b) provded by GPS. Due to sgnfcant amount of nose n voltage sgnal, usng global synchronzaton provdes better accuracy of measurement. (Plotted data are obtaned by smulaton.) Multple measurements of the phase of voltage sgnal are repeated contnuously both for mprovng the accuracy by averagng and for tracng ts tme varaton.
2. Desgn of Sensors. The proposed montorng method requres nstallaton of a large number of sensors. The sensors must satsfy the followng requrements: Low cost of manufacturng. Avalablty of nexpensve sensors facltates deployment of large and dense sensor networks. Inexpensve nstallaton. The sensors must employ contactless sensng technologes for measurng the electrcal parameters. Then sensors could be snapped around electrcal wres on top of the nsulaton: the nstallaton wll take seconds and wll not requre dsconnectng the lne power. Autonomous operaton. Sensors must be desgned for long term operaton wthout mantenance, harvestng the necessary energy from magnetc feld surroundng the host conductor. Some sensors may be equpped wth optonal solar elements for smooth operaton durng power outages. Perodc frmware should be upgraded remotely, wthout physcal access to the sensor. Robust communcaton features. Installaton can be greatly smplfed, f sensors wll automatcally connect to communcaton network by usng host conductor as communcaton medum. Each segment of the network must contan dedcated Communcaton Unts (at least one per conductor), whch collect data from sensors and wrelessly transmt them to the remote server for processng. Sensors are also equpped wth auxlary wreless communcaton modules, used when data packets cannot be delvered over the powerlne. In order to satsfy those requrements, the sensor should mplement non-destructve sensng technology: all the parameters should be measured wthout galvanc connecton to the power lne. The parameters nclude RMS value and phase of current n the conductor, and shft between phases of current and voltage at the pont of measurement. Measurement of RMS value and phase of current s performed non-destructvely by observng changes n electromagnetc feld around the host conductor. Ths may be acheved by nstallng Hall sensor n the vcnty of the conductor (see Fg.4). The Hall sensor s nstalled at known dstance from the surface of the conductor. When the AC current s flowng n the conductor, the varable voltage wth proportonal to nstant value of the current, s observed at the output of the sensor. RMS of the conductor current may be estmated from measured RMS of the voltage. Snce Hall sensor ntroduces very low angle delay, the phase of ts output voltage s very close to the true phase of the conductor current. Therefore, the phase of the Hall sensor voltage may be used as an accurate estmate of the current phase. Lne voltage s dffcult to measure accurately wthout connectng to the powerlne. Therefore, our sensor s lmted to measurng the phase of the voltage. The sensor s equpped wth capactve sensor, generatng the varable voltage, proportonal to change of the electrc feld around the conductor (see Fg.5). Snce change of the electrc feld s proportonal to the
dervatve of lne voltage, the output of the capactve sensor may be used to estmate the phase of the voltage. The output voltage of the capactve sensor beng nosy, accurate estmaton of ts phase s possble only by averagng over large number of measurements. The sgnal representng AC lne voltage, obtaned from the capactve sensor, s too nosy to serve as relable reference for measurement of the phase of the lne current. We are measurng the phases of lne current and voltage as phase angles between the global synchronzaton pulse and the ponts of zero crossng. The prncple of measurement of phase s demonstrated for lne current n Fg.3(b). Global synchronzaton pulse s produced by GPS smultaneously n all sensors of the network and serves as relable reference for accurate phase measurement. If a sensor s nstalled at the locaton where GPS sgnal s weak and global synchronzaton pulse s not receved, the sensor s swtched to the mode of drect measurement of phase shft between current and voltage. Block dagram of the sensor s shown n Fg.6. The sensor s powered by energy, harvested from electromagnetc feld around the host conductor. The harvester s based on the ferrte splt core. As a sensor clamps on the host conductor, the halves of the core connect. The core wth the wndng around ts lower half acts as a current transformer. The power collected by the transformer depends on the current n the host conductor. The sensor s mplementng smart power management scheme: the mode of operaton s selected dependng on avalable power. The sensors placed on the root lnes, constantly carryng sgnfcant current, collect and communcate data more frequently and are more lkely to be used as gateways to retransmt wreless data from dsconnected segments of the network (e.g. n case of power falure). The excess power s stored n the rechargeable battery for use n case of dramatc decrease of current n the host conductor or powerlne falure. Each sensor s equpped wth auxlary solar cells for emergency operaton durng long-term power outage, when current n the host conductor s very low or non-exstent. Powerlne communcaton s performed by njectng current n the powerlne usng hghfrequency current transformer.
Fg.4. Non-destructve measurement of current I(t) n the host conductor usng Hall sensor. The sensor s nstalled at known dstance d from the surface of the conductor. Varable magnetc feld H(t) s causng voltage V Hall (t) at the output of the sensor, that s proportonal to current I(t). Fg.5. Non-destructve measurement of voltage V(t) n the host conductor usng capactve sensor. The sensor s nstalled at known dstance l from the surface of the conductor. Varable electrc feld E(t) s causng voltage V Cap (t) at the output of the sensor, that s proportonal to the voltage V(t).
Fg.6. Block dagram of the sensor. The sensor s usng Hall and Capactve Sensor to measure characterstcs of current and voltage n the host conductor. GPS data s used for global synchronzaton of measurements. GPS also provdes locaton of the sensor, used for grd map estmaton. The Communcaton Unt s sendng the processed measurement data over the powerlne (Communcaton Col). In case of powerlne falure, the sensor may attempt to establsh communcaton va wreless nterface (Wreless Antenna). The sensor s powered by energy, harvested from the magnetc feld around the host conductor (Current Transformer). The excess energy s stored n the rechargeable Battery. Auxlary Solar Cells are used to replensh the Battery n case of long-term powerlne falure.
3. Archtecture of the Sensor Network. The archtecture of the network s desgned to provde optmal throughput of communcaton channel for collecton of measurement data. The sensors are usng powerlne as the preferable medum for communcaton. Indvdual conductors of the grd are separated from each other for communcaton sgnal and may be consdered as clusters or segments of the grd. The sensors nstalled on one segment may communcate to each other, but they are solated from sensors of other segments by hgh mpedance of the transformers. Each cluster contans at least one dedcated Communcaton Unt (CU), whch s equpped wth wreless or wred nterface (GPRS, W-F, Ethernet etc.), and acts as nterface between the powerlne network and the external Processng Server (PS). Multple CUs may be nstalled n the same cluster for mproved relablty. Snce only a few CUs are requred for montorng of the grd, they could be powered from the grd and equpped wth large processng capabltes. The typcal block dagram of the network cluster s shown n Fg.7. The network s desgned to self-organze nto clusters. If addtonal CU s nstalled on one segment of the grd, the system can automatcally assgn t to an exstng cluster, snce t wll communcate over the powerlne to already exstng CU. The sensors are also assgned to one of the clusters at the tme of nstallaton: the sensor dentfes tself to the network by sendng ts ID, whch s receved by cluster CU(s). The sensor also receves and stores ID(s) of CU(s) of the cluster to whch t communcates. Each sensor s equpped wth auxlary wreless communcaton unt. If CU s not accessble, the sensor attempts to communcate wth sensors or CUs of neghborng clusters. Ths stuaton may arse f the sensor was nstalled on the grd segment whch does not contan CU, the exstng cluster CU malfunctons or the powerlne s damaged. For example, the sensors nstalled on short segments, runnng from a step-down transformer to a few houses, wll form a small cluster, whch does not justfy nstallaton of a separate CU. Those sensors may share the CU of the upper-level cluster by perodcally reportng ther measurements wrelessly across the transformer to sensors of the upper-level cluster. The data s then retransmtted to CU(s) of the upper-level cluster over the powerlne. If sensors are cut off from the CU due to CU falure or powerlne damage, they attempt to communcate wrelessly wth sensors from other clusters that have access to CU. If a sensor n cut-off segment succeeds, then the wreless communcaton lnk s establshed between the cutoff segment and the other segment. Ths lnk s used to transfer acqured data to the accessble CU and then to central PS. Such mode of operaton s ntended to be temporary, and may quckly dran the batteres of the sensors, transferrng data between two segments. Therefore, only the data, whch s necessary for dagnostc of the grd malfuncton, should be transmtted by the sensors of the cut-off segment.
Fg.7. Archtecture of a cluster of the sensor network. Each cluster conssts of sensors nstalled on the same host conductor. The sensors are usng powerlne communcaton to transmt data. The cluster contans at least one Communcaton Unt (CU), whch collects data from sensors of the cluster and transmts t to remote Processng Server. Addtonal CUs may be nstalled to mprove relablty of communcaton.
4. Sensor Data Collecton and Processng; Grd Topology Estmaton. Sensors are performng perodc synchronous measurements of RMS value, dfferences n phases of current and voltage at the pont of measurement, as well as the phase of current wth respect wth global synchronzaton mpulse. Synchronzaton of measurements s acheved by usng the GPS tme. The results of measurements are then transmtted by the sensors over the powerlne conductor to CU. The data s transmtted n the form of packets. The communcaton throughput may be mproved by combnng the results of multple measurements by the sensor n one packet. Snce the consecutvely acqured data ponts from the same sensor are lkely to be correlated, compresson of data may reduce the total amount of transmtted data. Each measurement should be marked by the tme stamp that ndcates the tme of data acquston. If multple perodcally obtaned data ponts are combned n one packet, only one tmestamp for the frst data pont may be ncluded n the packet. In ths case, tme for the remanng data ponts may be computed from tme stamp of the frst data pont and known perod between measurements. The data from all sensors that belong to one cluster are collected by CU, whch converts data to the form, sutable for transmsson to PS, and performs data transfer usng sutable communcaton method, such as Ethernet, W-F and cellular wreless network. The communcaton method s selected based on the facltes avalable at the locaton of CU nstallaton, takng nto account cost of nstallaton and data transfer fees. Dependng on mplementaton, CU may also perform data preprocessng, ncludng cluster map generaton or valdaton of the exstng cluster map, thus reducng the computatonal load on PS. In the computer descrpton of the cluster map, the connectons between the branchng ponts of a conductor, whch s carryng the sensors that belong to one cluster, are performed usng the data on RMS values and phases of current synchronously measured by all the sensors n the cluster and GPS coordnates of those sensors. The method desgnates one sensor of the cluster, whch s nstalled on the conductor at the locaton where t s connected to the output of the step-down transformer, as the root sensor. The computaton procedure starts from the root sensor and attempts to fnd a sensor or a combnaton of sensors, such that the current or the sum of currents reported by those sensors s equal to the current reported by the root sensor. The currents should be added accordng to Krchhoff Current Law (KCL), takng nto account the synchronously measured phase of each current. Such sensor or a group of sensors s assumed to be drectly connected to the root sensor. To reduce the complexty of computatons durng search of connected sensors, the geographcal area can be lmted to the vcnty of the root sensor by usng GPS coordnates of the sensors. The same procedure s teratvely repeated for all the sensors, connected to the root sensor, and then contnued untl connectons for all sensors n the cluster are found. Snce the accuracy of current measurements s lmted, the frst generated map may be only an estmate, and the procedure should be repeated multple tmes wth dfferent dstrbuton of currents, untl accurate network map s generated.
The computaton procedure s llustrated usng smple, but non-trval network, shown n Fg.8. The network conssts of 6 nodes, numbered from I to VI, and contans 10 sensors, numbered from 1 to 10. The arrows next to sensors are showng sensor orentaton, whch s used when wrtng KCL equatons. The sensors are always nstalled accordng to the followng rules: sensors placed at the splt node are pontng towards the node; sensors at the load are pontng towards the load, unless those sensors are at the splt node; sngle sensors nstalled on a wre could be arbtrarly orented. The topology estmaton starts from Sensor 1, whch s placed at the root of the network. An example of numercal nformaton obtaned from the sensors s shown n Table 1. The nformaton conssts of the sensor number (sensor ID), magntude and phase of current and coordnates of the sensor (GPS coordnates). For ths smulaton, the table was generated by selectng arbtrary current magntude and phase values at each load, and then computng the remanng currents usng the dagram n Fg.8. Intally, the topology estmaton algorthm does not have nformaton on connecton wthn the network or whch sensors are placed at the loads. The only avalable data ncludes the locaton of root sensor, the values of currents measured by all sensors and the coordnates of all sensors. In the followng dscusson we wll denote the magntude of current measured by the -th sensor as I, phase of the current as and the coordnates as x, y. The topology estmaton algorthm wll execute the followng steps: 1. The system starts at the root Sensor #1 and attempts to fnd a set of sensors that ncludes Sensor 1, such that KCL s satsfed. In the followng example, we wll wrte the magntude and phase of the currents n the phasor format: I, where I represents the magntude of current and represents the phase of current measured by -th sensor. The current 0 measured by Sensor #1 can be found from the table n Table 1 as 43.2064 4.8275. The algorthm explores two possbltes. Sensor #1 can be nstalled at the transformer and connected to the next sensor by the conductor wthout splttng. In ths case, the next sensor wll measure the same current, but, snce ts orentaton s unknown, the phase may be shfted by 180 o. The algorthm s checkng the table to see whether there s any other 0 sensor that reported current 43.2064 4.8275 or 180 0 shfted current 0 43.2064-175.1725. Quck verfcaton of Table 1 shows that there s no such sensor. Then the algorthm explores the second possblty, that the sensor s nstalled at the splt node. Snce the sensors on the splt node are always orented towards, the node, the followng KCL equaton must hold: I e j 1 j 1 Ie 0 Expandng exponental as e I cos 1 1 I sn 1 1. I cos. I sn j cos j sn yelds
By tryng combnatons of currents from the table, t could be found that the followng combnatons of sensors satsfy the above equatons: Combnaton 1: Sensor #2 and Sensor #3 Combnaton 2: Sensor #5 and Sensor #3 In order to decde whch combnaton to select, the algorthm s computng the sum of dstances from Sensor #1 to all the sensors n the combnaton. The dstance to sensors n Combnaton 1: D d d 1 12 13 2 2 d x x y y j j j and the dstance to sensors n Combnaton 2: D d d 2 15 13. Substtutng numercal values yelds D1 4.47 and D2 6.71. The Combnaton 1 that conssts of the sensors wth closest geographcal locaton s selected. It s recorded that Sensors #1, #2 and #3 are located at the splt node. The algorthm marks those sensors as already connected and does not use them n further search. 2. Search for the combnaton of connected sensors contanng Sensor #3. The search s also performed n two stages. Frst, a sngle sensor that reported the same current as Sensor 0 0 #3, 15 160 or the current 15 20, whch s shfted by 180 0. The search of the table of 0 currents shows that Sensor #4 reported the current 15 20, whch means that those sensors are nstalled on the same conductor but orented n opposte drecton. Sensor #4 s marked as connected. 3. Search for possble connectons for Sensor #4 returns negatve result, so Sensor #4 s marked as located at the node. 4. Search for the connectons for Sensor #2 s performed. Sensor #2 reported the current 0 28.9963 177.0461. The table s searched for the dentcal current, or the current 0 28.9963 2.9539, whch s shfted by 180 0. Searchng through the table, the algorthm fnds two possble connectons: Sensor #5 returnng dentcal current, and Sensor #6 returnng the current shfted by 180 0. The closest sensor s selected. The dstance between Sensor #2 and Sensor #5 d25 2.24 and the dstance between Sensor #2 and Sensor #6 s d26 3.61. The Sensor #5 s selected as the closest and s marked as already connected. 5. Search for the connectons for Sensor #5 s performed. Searchng the table shows, that Sensor #6 returned the same current as Sensor #5 but shfted by 180 0. Snce ths s the only matchng result, the connecton between Sensor#5 and Sensor #6 s recorded, and Sensor #6 s marked as already connected.
6. Search for the connectons for Sensor #6 s performed. The search for sensors that reported dentcal or 180 0 -shfted current fals. Then the opton, that the Sensor #6 s placed at the splt node s explored. The search returned that for currents reported by Sensor#7 and Sensor #8 the KCL holds. Snce there are no other combnatons, t s recorded that Sensors #6, #7 and #8 are located at the splt node. The sensors are marked as already connected. 7. Smlarly, Sensor #9 returns the same, but 180 0 -shfted current as Sensor #7, so the connecton between those sensors s recorded. 8. Sensor #10 returns the same, but 180 0 -shfted current as Sensor #8, so the connecton between those sensors s recorded. As a result of ths algorthm, the graph of connectons between all the sensors s created. Ths establshes the network topology. Plottng the graph of connectons on a geographcal map wll gve the map of the cluster. The maps of multple connected clusters are joned to obtan the map of the dstrbuton grd. The presented algorthm may be drectly appled to grd confguratons represented as a tree graph. Some grd confguratons may nclude loops,.e. there may exst multple paths from the root to some nodes. The algorthm may be adapted to work wth loops at the expense of computatonal complexty by ncludng already connected sensors n search f approprate combnaton of unconnected sensors cannot be found. Relablty of the network s an mportant ssue. The presented algorthm wll succeed n estmatng the grd map f some sensors n the dense network fal or produce naccurate readngs. For example, f Sensor #8 fals, the algorthm wll establsh connectons between Sensors #6, #7 and #10. In some cases, nstallaton of dense sensor network may be unfeasble. In ths case, dfferent algorthm for grd reconstructon may be mplemented, based on analyss of correlatons between varatons of currents measured by sensors. For example, let us assume that Sensors #6, #7 and #8 are not nstalled (or faled). Then, varatons of currents measured by Sensors #9 and #10 are uncorrelated: t s hghly unlkely that two homeowners wll synchronously flp lght swtches on consstent bass. At the same tme, varatons of currents regstered by each of the Sensors #9 and #10 s correlated wth readngs from Sensor #5. The correlatons ndcate exstence of connectons between Sensors #5 and #9, and Sensors #5 and #10. The algorthm based on correlatons wll successfully recover approxmate grd map f sensor network s not dense (or even sparse). The accuracy of the map mproves as addtonal sensors are nstalled. It s also expected that analyss of correlatons s not nfluenced by naccuraces n current and phase measurements, snce n most practcal cases of based, ncorrectly scaled or nosy data, the relatonshps between correlatons wll be preserved. In the rare case of meanngless data readngs due to sensor falure, all data from the sensor may be completely gnored durng the analyss.
Fg.8. Illustraton of the cluster topology estmaton and the generaton of cluster map. The sensors are nstalled accordng to the followng rules: f the sensors are nstalled at the branchng node, the postve drecton of current (used n wrtng the Krchhoff s Current Law, shown by arrows) s always towards the node (sensors 1,2 and 3 at node I, sensors 6,7,8 at node III); f the sensor s nstalled at the transformer, the postve drecton of current s always towards the transformer (sensors 4, 9 and 10), unless the sensor s at the branchng node (sensor 1); f a sensor nstalled n the mddle of the conductor, the drecton of current may be selected arbtrarly (sensor 5). Sensor 1 s located at the pont of connecton to upper-level conductor of the network and desgnated as the root sensor of the cluster.
Table 1. Exemplary data, collected from the sensors of the network llustrated n Fg.8. Sensor Current (Magntude) Current (Phase) Geographcal Coordnates (X,Y) #1 (root) 43.2064 4.8275 0 (1, 10) #2 28.9963 177.0461 0 (2, 12) #3 15-160 0 (2, 8) #4 (load) 15 20 0 (3, 6) #5 28.9963 177.0461 0 (3, 14) #6 28.9963-2.9539 0 (5, 14) #7 17-170 0 (6, 16) #8 13 160 0 (6, 12) #9 (load) 17 10 0 (7, 17) #10 (load) 13-20 0 (7, 10) Fg.9. Detecton of powerlne falure. The locaton of powerlne falure may be detected by analyzng connectvty between sensors. Sensors are capable of communcatng over the powerlne between each other nsde the dsconnected segment, but they need to use auxlary wreless module to reach the Communcaton Node. 5. Montorng the grd operaton. The constructed map may be used for montorng of the grd. Power outages may be quckly detected by montorng connectvty between branches of the grd. Damage to lne conductors may be detected by montorng the currents n the network and comparng the results wth nomnal values. Severe damage, such as short crcut or lne breakage, s nstantly detected as loss of powerlne communcaton between sensors of the cluster (see Fg.9). The precse locaton of the breakage may be accurately and nstantly pnponted by determnng the number and
locatons of separated sensors. The conductors, connectng separated segments of the grd are consdered damaged, and repar crews are drected to precse locatons. Detecton of electrcty theft s an mportant problem n the feld of grd montorng. The descrbed network allows to perform montorng true power consumpton n all branches of the grd ndependently from resdental power meters. The data collected from the network may be compared to the data reported by the meters. The systematc dscrepancy between true and reported data ndcates possble theft of electrcty or damaged power meter. Further nvestgaton may be ntated to fnd the reason for excessve power consumpton. Unauthorzed taps nto the powerlne may be detected by verfyng KCL for all nodes of the network. Suspcous branches of the grd may be detected by comparng power consumpton and supply data. The precse locaton of tap-n may be pnponted by fndng volatons of KCL around the ndvdual grd nodes. Concluson. In the paper we presented the archtecture of a montorng system for applcaton n Smart Grds. The modern power dstrbuton grds are ncreasngly employng dstrbuted generaton of electrcty: substantal part of electrcty s generated locally by the consumers usng solar, wnd and other sources. In such grd, accurate real-tme montorng of all elements of the system becomes ncreasngly mportant. The proposed novel archtecture s employng a dense network of sensors for collectng real-tme data about the state of the network. Montorng on every element of the grd s acheved by usng autonomous nexpensve sensors, whch could be nstalled n large quanttes to form dense network. The sensor desgn must provde capablty of autonomous operaton over the lfecycle. Ths may be acheved by usng energy harvestng for powerng the sensors and flexble communcaton capabltes for ad-hoc organzaton of sensors n the network, whch s used to transfer the montorng data from the sensors to the Processng Server, as well as to remotely control the sensors and update sensor frmware. The communcaton medum s optmally selected by the sensors, whch gve preference to communcaton over the powerlne, swtchng to wreless communcaton only when necessary. The maxmum utlzaton of communcaton over powerlne allows to mnmze power consumpton and used wreless bandwdth. The cost of sensor nstallaton may be substantally lowered by usng non-destructve sensng technologes for measurement of the grd parameters. Non-destructve measurement of some parameters may be challengng. It was shown, that all requred parameter may be estmated by montorng varatons of the electromagnetc feld around conductors at the ponts of nterest f sensors of the network are synchronzed n tme. In order to acheve global tme synchronzaton, each sensor s equpped wth GPS recevers, whch also provde sensors wth ther locaton data.
We demonstrated, that the proposed montorng system s capable of estmatng the topology and generate the map of the dstrbuton grd. Our novel approach s usng communcaton over powerlne data to automatcally splt the network nto clusters of sensors, so that all the sensors n the cluster are nstalled on the same physcal conductor. The connectons between the ponts of sensor nstallaton may be estmated wth hgh probablty from the collected real-tme electrcal parameters receved from the sensors. The topology s frst estmated for ndvdual clusters of the network, whch allows to reduce the amount of requred computatons. The estmated topology together wth GPS locatons of the sensors are used to generate the grd map. References [1] Lm, H.B., Teo, Y.M., Mukherjee, P., at al., Sensor Grd: Integraton of Wreless Sensor Networks and the Grd., Proc. of 30 th IEEE Conference on Local Computer Networks, Sydney, Australa, 2005., pp. 91-98. [2] Ln, J., Zhu, B., Zeng, P. et al. Montorng power transmsson lnes usng a wreless sensor network, Wreless Communcatons and Moble Computng, 15, 2015: 1799-1821. [3] Chhaya, L., Sharma, P., et al., Wreless Sensor Network Based Smart Grd Communcatons: Cyber Attacks, Intruson Detecton System and Topology Control., Electroncs 2017, 6, 5. [4] Khan, F.U., Energy Harvestng from the Stray Electromagnetc Feld around the Electrcal Power Cable for Smart Grd Applcatons, The Scentfc World Journal, ID 3934289, 2016. [5] Roscoe, N. M. and Judd, M. D., Harvestng energy from magnetc felds to power condton montorng sensors, IEEE Sensors Journal, vol. 13, no. 6, pp. 2263 2270, 2013.