A method and device for prediction and validation of meter-transformer connections in a power grid

Abstract

A method is described for prediction and validation of meter-transformer connections in a radial power grid. The method uses location data for the meters and transformers in mapping data, known connections between these meters and transformers, and voltage data representative of voltage as a function of time for the meters in the mapping data. Are also described a device implementing the method and a system comprising a power grid and the device.

Description

[0001] A method and device for prediction and validation of meter-transformer connections in a power grid

[0002] Technical field

[0003] A method and device for prediction and validation of meter-transformer connections in a power grid is described. The method and device can be used in the frame of power grid management, in particular a radial power grid.

[0004] Background

[0005] In a power grid, a distribution transformer is connected to several meters, for example smart meters in the frame of an advanced metering infrastructure (‘AMI’) system. In the context of a radial topology as used for example in the United States of America, up to fifty meters may typically be connected to a single transformer, as a function of the transformer’s kVA rating, i.e. its capacity.

[0006] Typically, a transformer does not maintain a record of the meters it is connected to, and does not have circuitry, such as sensors, to obtain such information.

[0007] Meters may be disconnected or reconnected dynamically. E.g. a transformer-meter connection may be changed manually by operators to meet load demands. For example, during a truck roll, a meter connected to a first transformer may be disconnected from that transformer, and possibly reconnected to a second transformer of the same power distribution network. This information is generally not updated immediately on maintenance records - or is not recorded at all - which may lead to lower voltages for consumers already connected to the second transformer.

[0008] Information on which meters are connected to a given transformer, if any such information exists, may thus be incomplete or wrong.

[0009] As a result, maintaining an adequate real-time power demand / supply balance may prove difficult, loading models for predicting overloading and underloading of transformers may perform poorly, and knowledge of transformer usage may be incorrect or incomplete, ultimately leading to outages.

[0010] Most of the times, it is not practical to schedule truck rolls for surveying transformermeter connections. Such actions would, to the least, increase operational and maintenance costs for the utilities.

[0011] An efficient method for improving transformer-meter connection data reliability is consequently desirable.

[0012] Summary

[0013] A first aspect concerns a method for prediction and validation of meter-transformer connections in a radial power grid, said method being carried out by a device comprising a processor, said method comprising: obtaining mapping data comprising a list of meters and transformers and indicative of a connection of meters to transformers, wherein each transformer is connected to at least one meter; location data for the meters and transformers in the mapping data; voltage data representative of voltage as a function of time for the meters in the mapping data; predicting a connection of a given meter to a transformer based on at least one among location data and an evaluation of a similarity criterion of evolution of voltage data for the given meter and voltage data of other meters; validating a connection of the given meter to a transformer in the mapping data if it matches the predicted connection. According to one or more embodiments, the method comprises identification of a given connection between a meter and a transformer in the mapping data as incorrect, when a predicted connection does not match the given connection.

[0014] According to one or more embodiments, predicting a connection comprises determining as a predicted connection a connection of a meter with its closest transformer, based on the location data.

[0015] According to one or more embodiments, predicting a connection further comprises, for a given transformer with a validated connection to its closest meter: for a given meter indicated as connected to the given transformer in the mapping data and other than the closest transformer, identifying a validated connection meter with voltage data similar to voltage data of the given meter based on said similarity criterion; and obtaining the predicted connection as a connection between the given meter and a transformer to which the identified meter is connected to according to the identified validated connection.

[0016] According to one or more embodiments, the evaluation of the similarity criterion comprises comparing the voltage data for the given meter with the voltage data of another meter by applying one of a dynamic time warping model or an ensemble of a dynamic time warping model and an elastic net model.

[0017] According to one or more embodiments, the method further comprises: obtaining a confidence value for the similarity evaluation criterion; and determining whether the confidence value is above a threshold and using a predicted connection for validation only if said confidence is above the threshold.

[0018] According to one or more embodiments, the method further comprises, when said confidence threshold is not met, reducing volatility of the voltage data at least by removing voltage data samples considered volatile and carrying out again the similarity criterion evaluation. According to one or more embodiments, the method further comprises, in case volatility cannot be further reduced, outputting said predicted connection and an information representative of a probability that the predicted connection is correct.

[0019] According to one or more embodiments, volatility is evaluated using a volatility function such as

[0020] Volatility function = ( f — Vf ) A (Vf — Vf) where VTiis the voltage of a meter under study at time Ti, VTiis the voltage of a predicted meter at time Ti and V t) is the voltage of a meter in the recorded mapping data at time Ti.

[0021] According to one or more embodiments, predicting a connection further comprises, for a given transformer without a validated connection to its closest meter: if a single meter is indicated in the mapping data as being connected to the given transformer, determining that the single meter is connected to a transformer part of a validated connection and with the least distance to the single meter; if more than one meter is indicated in the mapping data as being connected to the given transformer, for a given meter among the more than one meter, determining a meter with the most similar voltage data and, if the meter with the most similar voltage data is validly connected to a transformer, determining that the given meter is connected to the latter transformer; else determining that the given meter is connected to a transformer part of a validated connection and with the least distance to the single meter.

[0022] Another aspect concerns a device comprising a processor and memory containing software code, the processor, when executing the code, causing the device to perform a method disclosed herein.

[0023] Another aspect concerns a system comprising a power grid with radial architecture and a device as disclosed herein. List of drawings

[0024] Further embodiments, features and advantages will become apparent from the detailed description of one or more embodiments described below, which are to be considered in conjunction with the accompanying drawings.

[0025] Fig. 1 is a schematic block diagram of a power distribution network and of a device implementing a method according to one or more non-limiting embodiments;

[0026] Fig. 2 is a flowchart of a method according to one or more non-limiting embodiments;

[0027] Fig. 3 is a flowchart of a method according to one or more non-limiting embodiments and that can be used within the method of Fig. 2 to evaluate a similarity criterion between a voltage signal of a meter under consideration and voltage signals of nearby meters;

[0028] Fig. 4 is a flowchart illustrating a voltage data improvement method based on a voltage data volatility evaluation according to one or more non-limiting embodiments;

[0029] Fig. 5 is a flowchart of a part of the method of Fig. 2 according to one or more nonlimiting embodiments;

[0030] Fig. 6 is a schematic diagram of a simplified power distribution network used as example for illustrating the application of one or more of the above methods.

[0031] Detailed description of at least one embodiment

[0032] Figure l is a schematic block diagram of a simple power distribution network which will be used for illustrative purposes. The exemplary network comprises two transformers T1 and T2. T1 is connected to meter Ml, and T2 is connected to meters M2, M3 and M4. A meter, which can also be referred to as an ‘Advanced metering infrastructure’ or ‘AMI’ provides information on the power consumption and instantaneous voltage over time (for example at 15 min or 1 hour frequency), also called ‘time series voltage data’ in the present description. In the example of Fig. 1, meter M3 is geographically the meter closest to transformer T1 and meter M4 is the meter closest to transformer T2.

[0033] A utility server 100 is schematically illustrated as receiving the time series voltage data from the meters. A processing device 101 comprises a processor 102, a working memory 103, a communication interface 104 and a long term memory 105. The long term memory holds software code which, when executed by the processor 102, causes processing device 101 to carry out one or more of the methods described herein. Communication interface 104 is used to receive data provided by the meters. The data may be provided to the processing device 101 in various ways, for example by one or more intermediate data collector devices such as server 100.

[0034] In an actual network, the number of meters per transformer may obviously be larger. E.g. in a network with a radial topology as used in the United States, between 1 and 50 meters would be connected to single transformer.

[0035] According to one or more non-limiting embodiments, input data for the method comprises: meter-transformer connection mapping data, as known ('recorded mapping data’); the geographical coordinates or other data characterizing the respective locations of meters and transformers (‘GIS’ or Geographical Information System data); time series voltage data of meters.

[0036] The example of Fig. 1 shows that individual location data of meters and transformers cannot be used alone in order to predict the connection between a meter and a transformer for all meters, e.g. on the basis of a meter being connected to the closest transformer. Indeed, meter M3 is connected to transformer T2, but is geographically closer to transformer Tl. This issue increases with the density of the population in a given area. According to one or more embodiment, depending on each meter, use can be made of both the geographical location and the time series voltage data of that meter to either validate recorded mapping data, provide a prediction as to which transformer the meter is connected to, or identify a wrong connection of the meter in the recorded mapping data. Meter-transformer connection mapping data can be based on utility installation and maintenance records, truck rolls or any other sources. It may be incomplete and / or incorrect. GIS data, i.e. location information is used to evaluate distances between the various assets. Such information may for example be derived from physical addresses associated with each meter and transformer in a database and provided to processing device by the power distribution network operator through a device such as server 100.

[0037] Voltage signals comprise time series voltage data, e.g. periodic voltage amplitude samples provided by meters. Periodicity may vary and may be, for the sake of providing an illustrative non-limiting example, 15 minutes or 1 hour. It is supposed that the voltage data available for the various meters covers a same time period, to allow determining whether the voltages of any two meters are similar over this time period.

[0038] Optionally, the input data may comprise historical and present load profiles of consumers.

[0039] According to one or more embodiments, if a connection predicted by the process matches the recorded mapping data, the latter is supposed correct. If at the end of the process, no prediction matches the recorded mapping data, the mapping data is determined to be incorrect, a corresponding report is generated. Thus, the process validates recorded mapping data and identifies incorrect recorded mapping data.

[0040] In addition, the method can provide a prediction as to which transformer a meter is connected to even in the absence of recorded mapping data, e.g. in case a new meter is present.

[0041] It is to be noted that the method is arbitrarily split into three ‘stages’ for the clarity of the detailed description only. The method could be described using more or less than the three stages described herein. Certain steps could be included in a different stage than the one presented herein. E.g. the steps presented as being part of the third stage in fact constitute a loop within the second stage.

[0042] Fig. 2 is a flowchart of a method according to a non-limiting embodiment.

[0043] Data preprocessing is shown at 201 in Fig. 2 but it is optional if data is already in a form usable by the method, e.g. it may have been appropriately preprocessed earlier. Preprocessing may for example include removal of meters which do not have time series voltage data available for certain periods. The number of such meters may be relatively important, especially in case of a radial topology. Indeed, meters may have been disconnected from their transformer, but this information may not necessarily be reflected in the recorded mapping data.

[0044] Unconnected transformers are removed from the recorded mapping data during preprocessing.

[0045] The following process assumes that each meter is connected to a single transformer and each transformer is connected to at least one meter.

[0046] First stage

[0047] The first stage processes a set of meters comprising each meter (among the meters under consideration) that is closest in distance to each transformer. As an example, the distance may be a Manhattan distance, but other types of distances may be used.

[0048] At stage 1, it is initially assumed that a meter closest to a transformer is indeed connected to this transformer, since this would be most efficient and logical connection with regard to the fact that losses increase with distance. In an exemplary actual power grid, the metertransformer distance would typically be under approx. 350 ft / 107 meters. A prediction is formed at 202 for each meter closest to a transformer that this meter is connected to this transformer. The prediction based on this assumption is however validated (at 203) by checking whether it matches the recorded mapping data.

[0049] The meters for which the meter-transformer connection predictions in set 1 are validated are considered correctly mapped (205) and are not further checked. The corresponding set of meters is labelled ‘Set 1A’ in Fig. 2. A corresponding report or another type of output can be generated. The meters present in the ‘Set 1A’ are termed as ground truth meters for their respective transformer.

[0050] Any further meters connected to a given transformer for which the prediction above is validated (i.e. for which a ground truth meter is available) are determined at 204. These meters are processed in the second stage, which uses the fact that there is a meter with a validated connection for that given transformer. This set is called ‘Set 2’ in Fig. 2.

[0051] Any meters connected to a transformer for which the above prediction is not validated, i.e. a transformer for which no validated connection is available in step-1, are considered in a separate process 206. The corresponding set of meters is labelled ‘Set IB’ in Fig. 2.

[0052] Second stage

[0053] The second stage processes meters in ‘Set 2’.

[0054] The second stage checks for similar voltage signal behavior between a meter under consideration and voltages of nearby meters, for which a connection to a transformer has been validated in stage 1 (‘ground truth meters’).

[0055] This is based on the assumption that the input voltages of a meter connected to a transformer and the output voltage of the transformer present similarities- hence by measuring the similarities, a prediction can be formed (at 207) that a meter is connected to a transformer.

[0056] Various methods exist for evaluating similarity between the voltage signal of the meter under study and the voltage signals of nearby meters with validated connections (i.e. ‘ground truth meters’). Any machine learning or deep learning or heuristics-based method can be used. In the frame of the present example, either a dynamic time warping (‘DTW’) algorithm or an ensemble of a DTW algorithm and an elastic net algorithm is used, as further detailed later. DTW has the advantage of taking into account phase shifts in the compared signals.

[0057] If the output of the method yields a predicted connection to a transformer with sufficient confidence (at 208), e.g. it is above a threshold, and the prediction is confirmed by the recorded mapping data (at 209), then the connection in the recorded mapping data is validated (205) and considered as ‘ground truth’, in addition to the validated connections from stage 1.

[0058] Meters for which the predicted connection with sufficient confidence is not validated are considered incorrectly mapped in the recorded mapping data (at 210). A corresponding report or another type of output can be generated, along with the predicted connection. Arbitrarily, these meters are assigned a rank called ‘Rank 1 ’ .

[0059] If the confidence provided by the similarity evaluation is considered insufficient, e.g. it is below a threshold, a process for increasing the confidence is carried out (at 211) when possible. This process comprises increasing voltage data quality by reducing data volatility.

[0060] Third stage

[0061] For certain meters, the output of the method of the second stage can initially yield predictions below a required confidence level. In such a case, voltage data used as input for the similarity evaluation is processed and / or supplemented in order to increase the confidence of the evaluation (at 211). The voltage signals are processed by removing data samples based on a volatility criterion and if necessary to add further data samples. Volatility is an indicator of the standard deviation of the data samples. Data samples with high volatility are considered outliers and are discarded. Stage 2 is then performed again based on the new set of data samples.

[0062] The data improvement process can be performed iteratively, either until the confidence level at the output of the second stage is sufficient, or until it is judged that the data samples available are insufficient to allow the method of stage 2 to be carried out. In the latter case, a signal indicating that more data samples are required is output.

[0063] Voltage data similarity evaluation

[0064] Fig. 3 is a flowchart of a method for evaluating the similarity of the voltage signal of a meter under study and the voltage signal of one or more other meters, according to a nonlimiting embodiment. The method of Fig. 3 is one possible implementation of step 207 in Fig. 2. The specific implementation shown in Fig. 3 is based on a dynamic time warping (DTW) algorithm or model. Preprocessing is shown at 301 in Fig. 3 but it is optional if data is already in a form usable by the DTW algorithm, e.g. it may have been appropriately preprocessed earlier. M meters are then selected (at 302), with M>1. The M meters have a validated connection to the M transformers closest to the meter under study and for which a connection to a transformer has been validated. Voltage data for the meter under study is processed to obtain sliding windows (at 303). For the sake of providing an illustrative and non-limiting example, if the data sample rate is 15 minutes, windows of 10 samples and a step of 5 samples from one window to the next can be considered. The period under consideration can for example be a few days, e.g. three days. The DTW algorithm is then applied. The DTW algorithm is applied for N windows (304). N can be chosen as a function of a specific application context, e.g. N can be chosen relatively small, e.g. around 100, for example as to as to allow a real time or near-real time processing. A voting mechanism is applied to the DTW results for each window, with the vote going to the transformer of the meter among the M meters with the minimum DTW distance. The transformer with the highest number of votes is selected as the transformer to which the meter under study is connected, i.e. the predicted transformer (305) is one output (306) of the process of Fig. 4 in addition to the confidence value.

[0065] The confidence associated with the prediction can for example be obtained by calculating the ratio between the number of windows associated with the predicted transformer divided by the total number of windows. The prediction and confidence can be used as input to step 208.

[0066] M depends on the context, in certain contexts M may be chosen to be equal to 10.

[0067] According to a variant embodiment, a cap is put on the distance between the meter under study and the transformers for the selection of the closest transformers. Indeed, above a certain distance, e.g. 500 meters in certain countries, the likelihood of a connection is zero or near zero given the low efficiency that a connection above such a distance would provide.

[0068] According to a variant embodiment, the DTW algorithm is a DTW bagging algorithm. As a variant embodiment, an ensemble of several models providing an output using a voting mechanism can be used. For example, a two-model ensemble can be used, with the first model being a DTW model and the second model being a linear regression model.

[0069] The DTW model can be implemented as described in conjunction with Fig. 4.

[0070] The linear regression model can for example be an elastic net model. Within the frame of the variant embodiment, a simplified form may be used, in which both ridge and lasso regression coefficients are set to 0. For example, the regression may be performed using the voltage signals of Q (Q>1) meters of the Q transformers closest to the meter under study and which have validated connections. The corresponding equation is as follows:

[0071] Vu= + ajVj + a2V2+ a3V3+ ••• + aQVQ(Eq. 1) where Vuis the voltage signal of the meter under study, is a constant intercept value, Vtare the voltages of the Q meters selected for comparison and atare the regression coefficients. The meter for which the coefficient atis the highest is the meter whose voltage signal is considered the most similar to the voltage signal of the meter under consideration. The meter under study is predicted as being connected to the transformer to which the meter with the highest coefficient is also connected. Q = 10 has been found to yield good results in a specific experimental context, but other values can be selected as appropriate.

[0072] According to one embodiment, a median value ('8') of the distances between any voltage signals of any pair of meters compared by the DTW algorithm is obtained. The median value of all absolute voltage differences obtained for all windows is considered. This median value is used later in the method.

[0073] Similarity evaluation confidence increase / Voltage data quality processing

[0074] Fig. 4 is a flowchart of a method for processing data samples used to carry out the second stage according to a non-limiting embodiment. Other methods can also be applied. The goal is to improve the data sample quality used for voltage similarity evaluation in order to increase the confidence of the prediction.

[0075] The method of Fig. 4 is an example of a possible implementation of step 211 of the method illustrated by Fig. 2. It considers volatility between the voltage of the meter under study and the previously predicted meter on one hand, and on the other hand between the voltage of the meter under study and the meter validly connected to the same transformer to which the meter under study is connected according to the recorded mapping data.

[0076] Volatility hampers the performance of the model or models used for the similarity evaluation in stage 2. It is not initially removed during data preprocessing because its evaluation and understanding at that stage may be difficult due to wrong connections listed in the recorded mapping data and because volatility at certain times may be normal (e.g. at meter shutdowns), and hence such information should not be removed.

[0077] The method of Fig. 4 evaluates voltage data volatility and, when possible, removes data samples considered volatile. Additional data samples may be added when necessary and when such additional data samples are available. If volatility cannot be further reduced and when available data allows to do so, an evaluation of the probability that the predicted transformer is correct is output (arbitrarily characterized by a ‘Rank’ in the present embodiment).

[0078] According to the non-limiting embodiment illustrated by Fig. 4, the following volatility function is applied (at 401 in Fig. 4):

[0079] Volatility function = (Vff — V j A (Vff — Vf') (Eq. 2) where Vf is the voltage of a meter under study at time Ti, Vtis the voltage of a predicted meter at time Ti and is the voltage of a meter in the recorded mapping data at time Ti. Each of the two differences in equation 2 has their own distribution.

[0080] The same voltage data samples as used for the similarity evaluation that were considered for determining the differences. A check is carried out for each time instant whether each of these two differences at that time instant lie within a same deviation from the mean of their respective distribution. If this is the case, the corresponding sampling time instant is considered as falling within the intersection of equation 2. Sampling time instants for which the condition is not verified are considered as corresponding to outliers. The given distance can be varied.

[0081] The non-limiting embodiment of Fig. 4 uses e.g. three consecutively tested percentage thresholds, i.e. 68%, 95% and 99.7%, corresponding for a normal distribution respectively to a distance away by one, two or three standard deviations compared to the mean value.

[0082] Based on the sampling time instants in the intersection, volatility is estimated. Arbitrarily, a Rank can be assigned. In the example of Fig. 4, the Ranks range from Rank 2 to Rank 5, with a decreasing probability of a correct prediction from Rank 2 to Rank 4. On one end of the spectrum, a Rank 2 indicates that the data is estimated as not volatile and that the confidence cannot be improved by removing outliers. Rank 5 on the other end of the spectrum indicates that voltage data is too volatile. The intermediary Ranks 3 and 4 allow for outlier removal and, if required by the voltage similarity evaluation, addition of further data samples (e.g. from past periods).

[0083] If all the differences lie within one standard deviation of the mean value of their respective distribution (tested at 402), then it is considered that the data samples are not volatile, and that the volatility cannot be reduced further to improve the confidence of the prediction as to which transformer the meter under study is connected to. A Rank 2 is assigned (at 403). A corresponding output is generated (at 404), e.g. informing a user that further analysis of the meter under study is required, for example through a field validation of the connection present in the recorded mapping data. A probability of 80% that the prediction from stage 2 is correct can also be provided.

[0084] If all the differences lie within two standard deviations of the mean value of their respective distribution (tested at 405), then it is considered that certain data samples are volatile and that the quality of the data samples can be improved. A rank 3 is assigned (at 406), outliers are removed (at 407) and it is checked (at 408) whether the remaining samples are sufficient to carry out the voltage similarity evaluation. If yes, the process loops back to the evaluation at 207, else further samples are added (at 414), e.g. from a past period. Generally, a probability of 60% that the prediction from stage 2 is correct can also be provided.

[0085] If all the differences lie within three standard deviations of the mean value of their respective distribution (tested at 409), then it is considered that certain data samples are volatile and that the quality of the data samples can be improved. A rank 4 is assigned (at 410), outliers are removed (at 411) and it is checked (at 408) whether the remaining samples are sufficient to carry out the voltage similarity evaluation. If yes, the process loops back to the evaluation at 207, else further samples are added (at 414), e.g. from a past period. A probability of 40% that the prediction from stage 2 is correct can also be provided.

[0086] If all sampling time instants do not lie within 99.7% of the mean value (negative output of test 409), then it is considered that the data samples are too volatile to allow improvement. A rank 5 is assigned (at 412). A corresponding output is generated (at 413), e.g. informing a user that further analysis of the meter under study is required. A probability of 20% is associated with this Rank concerning the correctness of the prediction from stage 2.

[0087] Meters connected to transformers for which the connection with the closest meter was not validated in stage 1

[0088] The processing of the meters connected to transformers for which no valid connection was obtained in stage 1 is performed after stage 2. This allows for additional validated connections on which to rely in the processing. In Fig. 2, this corresponds to reference 206.

[0089] The processing of these meters relies on the geographical location of the meters and transformers, a voltage similarity evaluation and the output of stages 1 and 2.

[0090] Fig. 5 is a flowchart of a non-limiting embodiment of a method for processing these meters. The method is carried out for each transformer (‘predicted transformer) for which no connection to the closest meter was shown in the recorded mapping data. At 501, a check is carried out to verify whether the recorded mapping data shows one - or more than one - meter being connected to the predicted transformer.

[0091] If only one meter is connected to the predicted transformer, a prediction of the transformer connected to the one meter is determined, the determination comprising considering the set of transformers for which no connection with a meter was previously validated and selecting a transformer in this set based on a least cost function, with the cost being the distance between the meter under study and the transformers in the set. Indeed, each transformer of this set has to be validly connected to at least one meter. The connection between the meter under study and the newly predicted transformer is considered valid. For example, a Hungarian algorithm may be used for this purpose. A new transformer prediction is thus obtained for the meter under study.

[0092] The transformers present in set IB with more than one meter now require prediction, at 503. Every transformer must have at least one meter connected to it. All the meters connected to a given transformer with more than one meter connected to it according to the recorded mapping data are checked.

[0093] Voltage similarity is checked between each pair of meters of the given transformer by checking, for a period under consideration, whether the median of all differences of the voltage data time series samples for the two meters lies below a threshold of the median distance value '5' as previously determined. If the check is positive, then both meters of the pair are considered connected to a same transformer. A least-distance cost function is applied to connect these meters to their nearest transformer at 506, similar to what was described in conjunction with 502, e.g. using a Hungarian algorithm.

[0094] Connections of meters for which the check is negative are predicted as follows. If a given meter’s voltage is not similar to the voltage of any of the other meters of the given transformer, the given meter’s voltage is compared to that of meters with a validated connection in the vicinity and the predicted transformer is the one with the validly connected meter with the most similar voltage, using e.g. DTW. This means that the given meter is incorrectly mapped to the given transformer in the recorded mapping data. This is indicated as an output at 505, along with the predicted transformer. An illustrative example is as follows: three meters Ml, M2 and M3 are connected to the given transformer. Voltage similarity within the threshold ‘6’ is found to exist between Ml and M2. No voltage similarity is found to exist between M3 and either Ml or M2. For M3, a transformer is predicted that has a validated connection to a meter that matches best the voltage signal of M3.

[0095] According to another illustrative example, in case only two meters are connected to the given transformer, if the voltage similarity between the two meters is not within the threshold 'S’, a check is carried out for each of these two meters to determine, based on voltage similarity, the nearest match among nearby meters with validated connections to their transformers. The meter among the two with the best match to any of the nearby meters is predicted to be connected to the transformer of the matching nearby meter. The other meter is considered to be connected to the given transformer, i.e. the transformer it is connected to in the recorded mapping data.

[0096] Fig. 6 is a schematic diagram of an exemplary network that can be used to illustrate how the method of Fig. 5 is applied. The exemplary network of Fig. 6 comprises transformers Tl, T2 and meters Ml, M2, M3 and M4. Meter M3 is closest to transformer T1 and meter M4 is closest to transformer T2. In the recorded mapping data, meter Ml is connected to transformer Tl, meter M2 (wrongly) is connected to transformer Tl and meter M3 is connected to transformer T2. In the field, meter M2 is connected to transformer T2.

[0097] During stage 1, the connection between meter M4 and transformer T2 is validated since M4 is closest to T2 and the recorded mapping data shows this connection.

[0098] Meter M3 being closest to transformer Tl and there being no connection between meter M3 and transformer Tl in the recorded data, meters Ml and M2 are moved to set IB and are not processed in stage 2.

[0099] Meter M3 is however processed during the second stage. A voltage similarity evaluation shows that the voltage data of meter M4 is similar to that of meter M3. The connection between transformer T2 and meter M3 in the recorded mapping data is thus validated. Since two meters are connected to transformer T1 in the recorded mapping data, meters Ml and M2 are submitted to a voltage similarity evaluation with meters in their vicinity.

[0100] It is found that the voltage signals of meters M2 and M4 are similar. It is determined on that basis that a connection exists between meter M2 and transformer T2 (i.e. the transformer to which M4 is validly connected). The connection of M2 with T1 shown in the recorded mapping data is identified as wrong.

[0101] It is found that no meter has a voltage signal similar to that of meter Ml, thus the least-distance function is applied to meter Ml. It is found that transformer T1 is the transformer without validated connection closest to meter Ml . It is determined on that basis that a connection exists between meter Ml transformer Tl.

Claims

Claims1. A method for prediction and validation of meter-transformer connections in a radial power grid, said method being carried out by a device comprising a processor, said method comprising: obtaining (201) o mapping data comprising a list of meters and transformers and indicative of a connection of meters to transformers, wherein each transformer is connected to at least one meter; o location data for the meters and transformers in the mapping data; o voltage data representative of voltage as a function of time for the meters in the mapping data; predicting (202, 207, 503) a connection of a given meter to a transformer based on at least one of the location data and an evaluation of a similarity criterion of evolution of voltage data for the given meter and voltage data of other meters; validating (203, 209, 505) a connection of the given meter to a transformer in the mapping data if it matches the predicted connection.

2. The method of claim 1, further comprising identification (210, 505) of a given connection between a meter and a transformer in the mapping data as incorrect, when a predicted connection does not match the given connection.

3. The method of one of the claims 1 or 2, wherein predicting a connection comprises determining (202) as a predicted connection a connection of a meter with its closest transformer, based on the location data.

4. The method of claim 3, wherein predicting a connection further comprises, for a given transformer with a validated connection to its closest meter (204):for a given meter indicated as connected to the given transformer in the mapping data and other than the closest transformer, identifying a validated connection meter with voltage data similar to voltage data of the given meter based on said similarity criterion; and obtaining (207) the predicted connection as a connection between the given meter and a transformer to which the identified meter is connected to according to the identified validated connection.

5. The method according to one of the claims 1 to 4, wherein the evaluation of the similarity criterion comprises comparing the voltage data for the given meter with the voltage data of another meter by applying one of a dynamic time warping model or an ensemble of a dynamic time warping model and an elastic net model.

6. The method according to any of the claims 1 to 5, further comprising obtaining (306) a confidence value for the similarity evaluation criterion; and determining (208) whether the confidence value is above a threshold and using a predicted connection for validation only if said confidence is above the threshold.

7. The method according to claim 6, further comprising, when said confidence threshold is not met, reducing volatility of the voltage data at least by removing voltage data samples considered volatile and carrying out again the similarity criterion evaluation.

8. The method according to claim 7, further comprising, in case volatility cannot be further reduced, outputting said predicted connection and an information representative of a probability that the predicted connection is correct.

9. The method according to one of claims 7 or 8, wherein volatility is determined using a volatility function such asVolatility function = (Vf — Vf ) A (Vf — Vf') where Vff is the voltage of a meter under study at time Ti,the voltage of a predicted meter at time Ti and Vft) is the voltage of a meter in the recorded mapping data at time Ti.

10. The method according to claim 3 combined with any of the claims 4 to 9, wherein predicting a connection further comprises, for a given transformer without a validated connection to its closest meter (204): if a single meter is indicated in the mapping data as being connected to the given transformer, determining (502) that the single meter is connected to a transformer part of a validated connection and with the least distance to the single meter; if more than one meter is indicated in the mapping data as being connected to the given transformer, for a given meter among the more than one meter, determining (503) a meter with the most similar voltage data and, if (504) the meter with the most similar voltage data is validly connected to a transformer, determining (505) that the given meter is connected to the latter transformer; else determining (506) that the given meter is connected to a transformer part of a validated connection and with the least distance to the single meter.

11. Device (101) comprising a processor (102) and memory (105) containing software code, the processor, when executing the code, causing the device to perform a method according to one of the claims 1 to 10.

12. System comprising a power grid with radial architecture and a device according to claim 11.

Images