A multi-data fusion distribution network line transfer identification method based on neural network

Abstract

The present invention discloses a multi-data fusion distribution network line transfer identification method based on a neural network. The method includes obtaining the ledger and measurement data of the distribution transformer and line, the position change signal and measurement data sent by the switch, the line fault and plan-related records, and the line topology association relationship data; generating a trigger line set based on the acquired data, and extracting possible package combinations based on the trigger line set and the line contact ledger; establishing 14 characteristic values ​​such as the number of high-loss lines, the number of negative-loss lines, and the full-line loss rate of high-loss lines, and adding historical package status data to form a sample set and train a neural network model; using the trained model to perform batch transfer identification on line data in the distribution network. The present invention uses a data model to learn the data change characteristics of line transfer, judge whether the line has actually transferred, and achieve accurate judgment of the line transfer situation in the medium-voltage distribution network.

Description

Technical Field

[0001] The present invention belongs to the field of medium-voltage distribution network line transfer identification and analysis, and relates to a multi-data fusion distribution network line transfer identification method based on a neural network. Background Art

[0002] Due to planned maintenance, operator adjustments, and emergency repairs, distribution networks frequently perform power transfers, transferring power to affected users from one feeder to another to avoid unnecessary outages. Existing technology primarily involves manually recording operation tickets in the OMS system. While planned power transfers are recorded in the OMS, some on-site emergency repairs are not recorded in the OMS or dispatch logs. Furthermore, due to issues with terminal online rates and signal transmission, distribution network switch telesignaling cannot be fully monitored. All of these factors hinder the ability to fully, timely, and accurately monitor power transfer operations.

[0003] In the prior art, such as patent number 202010922151.8, entitled A method and device for identifying medium-voltage distribution network transfer operations based on machine learning, can effectively identify medium-voltage distribution network transfer operations. However, because this method relies on manually recorded operation ticket data, if the operation ticket data is missed or entered incorrectly, the accuracy and completeness of the transfer identification results will be seriously affected; because the operation ticket data is delayed in being uploaded, feature selection is also in days, which will lead to a high delay in the recognition of the method results and the inability to timely guide emergency repairs and other auxiliary decision-making; in addition, in terms of feature selection, the line loss rate indicator has a serious and direct impact on the measurement of the line transfer status, but this method does not involve this, and this patent can only identify the load transfer situation between two lines.

[0004] There is also patent number 202010565701.5, named "A short-term low-voltage distribution network transfer determination method based on the 6sigma principle", which can effectively identify low-voltage distribution network transfer operations. However, due to the large differences in medium and low voltage data characteristics, this method cannot be used to effectively identify medium-voltage distribution network transfers. Moreover, this method only relies on data for transfer analysis and judgment, and there may be data anomalies that may cause incorrect transfer identification. Summary of the Invention

[0005] In response to the problem that the current traditional distribution network transfer operation management relies on manual records and it is difficult to timely and comprehensively grasp the distribution network transfer information, the present invention proposes a multi-data fusion distribution network line transfer identification method based on a neural network. The method can be used to identify medium-voltage distribution network transfer events of any scale. The calculation is simple and the principle is clear, which can help distribution network operators to timely discover and locate real transfer events.

[0006] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0007] The present invention provides a multi-data fusion distribution network line transfer identification method based on a neural network, comprising:

[0008] Obtain records and measurement data of distribution transformers and lines, position change signals and measurement data sent by switches, line fault and plan-related records, and line topology correlation data;

[0009] Generate a trigger line set based on the acquired data, and extract possible packaged line combination sets based on the trigger line set and line contact ledger;

[0010] Based on the packaged line combination set, 14 eigenvalues ​​are established, including the number of high-loss lines, the number of negative-loss lines, the loss rate of the entire antenna of high-loss lines, the variance of the loss rate of the entire antenna of high-loss lines, the loss rate of the entire antenna of negative-loss lines, the variance of the loss rate of the entire antenna of negative-loss lines, the packaged line loss rate of all lines throughout the day, the number of tie switch changes, the number of lines involved in the changed tie switch, the number of packaged lines, the number of suspected package time points, the packaged line loss rate during the suspected package period, the packaged power loss during the suspected package period, and the correlation coefficient between the power loss of negative-loss and high-loss lines during the suspected package period.

[0011] Add historical packaged status data of the line to form a sample set, and build and train the neural network model;

[0012] The trained model is used to batch identify the actual transfer data in the distribution network.

[0013] As a further improvement of the present invention,

[0014] Optionally,

[0015] The records and measurement data of the distribution transformer and the line include: basic information of the line and the distribution transformer, power of the line and the distribution transformer, and power factor value data;

[0016] The switch position change signal and measurement data sent up include: all switch position change information of the main distribution network, data related to the line and switch type, and power measurement data;

[0017] The line fault and plan related records include: 95598 power outage information release related line fault and plan information, weekly maintenance plan related line plan information;

[0018] The line topology association relationship data includes: line topology data, including line topology paths and tie switch opposite end line information and other data;

[0019] Optionally, a trigger line set is generated based on the acquired data, and a possible packaged line combination set is extracted based on the trigger line set and the line contact ledger, including:

[0020] According to the change from value to value of all switch measurements on the line, the position change information sent by all switches on the line, line fault and plan information, all lines are extracted as the trigger line set L = [l1, l2, ..., l i ,…,l a ], a is all trigger lines;

[0021] Based on the trigger line set, take the trigger line l i A list of all contact lines is obtained, and two or more potential packaged line combinations are performed on the lines in the list. The above potential packaged line combination data are extracted for all lines in the triggered line set, and finally a possible packaged line combination set is formed;

[0022] Optionally, the feature values ​​established based on the packaged line combination set include:

[0023] By calculating the line loss rate within the packaged line combination, we can obtain three features: the number of high-loss lines in the packaged line combination, the number of negative-loss lines in the packaged line combination, and the packaged line loss rate of all lines throughout the day. By calculating the high / negative line loss rate within the packaged line combination, we can obtain four features: the full-day loss rate of high-loss lines, the variance of the full-day loss rate of high-loss lines, the full-day loss rate of negative-loss lines, and the variance of the full-day loss rate of negative-loss lines.

[0024] By counting the position change information sent by all the tie switches connected to the lines in the packaged line combination, we can obtain two features: eight is the number of tie switch position changes, and nine is the number of lines involved in the changed tie switches; ten is the number of packaged lines, which is obtained by counting the number of lines in the packaged line combination; by calculating the line loss rate of all lines at each time point, taking the intersection of the set of all high-loss time points and the set of all negative-loss time points as the suspected packaged time point, we can obtain four features: eleven is the number of suspected packaged time points, twelve is the packaged line loss rate during the suspected packaged period, thirteen is the packaged power loss during the suspected packaged period, and fourteenth is the correlation coefficient between the power loss of negative loss and high-loss lines during the suspected packaged period.

[0025] Optionally, the adding of historical line packaging status data to form a sample set and establishing and training a neural network model includes the following steps:

[0026] Integrate the line transfer status and extracted feature data to form a transfer identification sample set, establish and train the neural network model: build an RNN neural network model, randomly divide the training set and test set, and train the neural network model.

[0027] Optionally, batch recognition of actual transfer data in the distribution network is performed using the trained model, including the following steps:

[0028] After the model prediction accuracy is high, the trained model is used to generate suspected power transfer information and identify power transfer for batch lines based on the operating data in the actual medium-voltage distribution network. That is, a set of packaged line combinations is generated in batches for each line and the required feature values ​​are extracted. The sets are then sent to the model for calculation, and finally the true or false power transfer status results corresponding to each packaged line combination are output.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] (1) The present invention learns the trend of massive historical data changes in a data-driven manner, combines multi-source data such as fault / line records, switch position changes, topological connections, line and distribution variable measurements, extracts a total of 14 relevant characteristic values ​​related to line transfer, and uses a neural network model to accurately determine whether the line transfer has actually occurred, avoiding the problem of incomplete and untimely transfer records recorded manually.

[0031] (2) The present invention comprehensively considers all situations involved in the suspected transfer of the line, the line loss rate of the entire line, the time-sharing line loss rate and its changes, the screening of suspected packaging time periods, the packaged line loss rate of suspected packaging time periods, the loss situation of suspected packaging time periods and the correlation coefficient, and other data characteristics, and comprehensively screens out the characteristic information of the packaged time period transfer, which can identify multiple packaging situations with high accuracy.

[0032] (3) The present invention has simple calculation and clear principle, and can help distribution network operators to timely discover the power transfer situation and the lines involved in the power transfer event, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments and in conjunction with the accompanying drawings, wherein:

[0034] Figure 1 This is a flow chart of a method for identifying power distribution network line transfer according to an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of packaging records in an embodiment of the present invention;

[0036] Figure 3 This is a green horizontal line power stacking diagram in an embodiment of the present invention;

[0037] Figure 4 This is a power stacking diagram of the Hefei line in an embodiment of the present invention;

[0038] Figure 5 This is a diagram of the power stacking of two circuit combinations in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0040] The application principle of the present invention is described in detail below with reference to the accompanying drawings.

[0041] The embodiment of the present invention provides a method for identifying the power supply transfer of a medium voltage distribution network line based on massive data analysis, such as Figure 1 As shown, the specific steps include:

[0042] Step (A) is to obtain the records and measurement data of the distribution transformer and the line, the position change signal and measurement data sent by the switch, the line fault and plan-related records, and the line topology correlation data;

[0043] In a specific implementation of the embodiment of the present invention, the specific implementation process of step (A) is:

[0044] In the energy management system (which is a system existing in the prior art), a medium-voltage distribution network to be processed is selected, and basic data of all switches and distribution transformers stored in the existing system is read, including switch ledger data, line topology paths, and associated relationship data. Active power measurement data of the distribution transformer sampled every 15 minutes is exported from the power consumption information collection system, and active power measurement data of the switch sampled every 15 minutes is exported from the real-time measurement center system. The aforementioned sampling frequency of every 15 minutes can be modified according to actual conditions.

[0045] The automatic switch position change information and manual position change information obtained in the D5200 system;

[0046] Obtain the provincial telecommunications mid-week maintenance plan, 95598 related fault and planned event records;

[0047] Step (B), generating a trigger line set according to the acquired data, and extracting a possible packaged line combination set based on the trigger line set and the line contact ledger;

[0048] In a specific implementation of the embodiment of the present invention, step (B) specifically includes the following sub-steps:

[0049] Step 1: Generate a switch position information trigger event:

[0050] The switch position information in the medium-voltage distribution network includes directly obtainable automatic switch position information and manual position information, as well as switch position information that needs to be extracted from the data layer. The switch position information mainly includes basic information of the position switch, position switch type, position change time, and position change status type data;

[0051] The specific steps to obtain the switch position information that needs to be extracted from the data layer are as follows:

[0052] A switch S k Hooked on line f, switch S k The active power measurement data is P k =[p k,1 ,p k,2 ,…,p k,i ,…,p k,m ], m represents the number of sampling points of switch active power measurement data;

[0053] If switch S k Active power measurement data P k In the continuous data segment [p k,e ,p k,f ,…,p k,i ,p k,j ], the power value changes from on to off at time f and from off to on at time j, and the time periods f and j are greater than 1 hour, then the switch S can be extracted. k At data points f and j, a state change occurs. The switch measurement changes from value to no value at time f, which is a type 1 state change. The switch measurement changes from no value to value at time j, which is a type 2 state change.

[0054] The line to which the switch belongs is the trigger line, and the switch position change time is the trigger moment, forming a switch position change information type trigger event set;

[0055] Step 2: Generate a trigger event for a planned / fault event:

[0056] Planned / fault events include the 95598 power outage information pool and recorded events in the weekly maintenance plan. The line where the event is recorded is the trigger line, and the recorded start time is the trigger moment, forming a set of trigger events for planned / fault events.

[0057] Step 3: Generate a continuous negative / high loss trigger event:

[0058] Calculate the time-based line loss rate for each line. If a line experiences negative / high loss for more than one hour, the negative / high loss line is used as the trigger line, and the time when the negative / high loss begins is used as the trigger time. A continuous negative / high loss trigger event is generated. The formula for calculating the line loss rate is as follows:

[0059] If the power value flowing into the line is set to positive and the power value flowing out is set to negative, then the input integral power Q of the line in time period T is CB+ And output integral power Q CB- They are:

[0060] For P(tk )>0;

[0061] For P(t k )<0;

[0062] Where P(t k ) represents a line at sampling time t k The power of , k is the sampling point number, K is the number of sampling points;

[0063] Correspondingly, for all distribution transformers connected to the line at sampling time t k The power sum is P a (t k ), the input integral power Q in time period T DT+ And output integral power Q DT- They are:

[0064] For P a (t k )>0;

[0065] For P a (t k )<0;

[0066] Then the line loss rate L in the statistical time period T T The calculation method is as follows:

[0067]

[0068] Among them, L T is the line loss rate value of the line in the period T; Q CB+ is the input integrated power of the line in time period T; Q CB - is the output integrated power of the line in time period T; Q DT+ is the output integral power of the distribution transformer in time period T; Q DT- It is the output integral power of the distribution transformer in the topological area in time period T, and the denominator is the power loss;

[0069] Step 4: Integrate the trigger line set:

[0070] The line trigger event sets extracted in the above three steps are merged to form a trigger line set L = [l1, l2, ..., l i ,…,l a ], a is all trigger lines;

[0071] Step 5: Extract possible packaged line combination sets based on the trigger line set and line contact ledger:

[0072] Based on the trigger line set L, take the trigger line l i List of all contact lines[l i1 ,l i2 ,...,l iN ], where N is the line l i The number of contactable lines. Combine 2 or more lines in the list to form a potential package line combination. Note that the combination must include the trigger line l i ;

[0073] Repeat the above steps to extract potential packaged line combination data for all lines in the triggered line set, and finally form a possible packaged line combination set;

[0074] Step (C) establishes 14 eigenvalues ​​based on the packaged line combination set, including the number of high-loss lines, the number of negative-loss lines, the full-day loss rate of high-loss lines, the variance of the full-day loss rate of high-loss lines, the full-day loss rate of negative-loss lines, the variance of the full-day loss rate of negative-loss lines, the full-day packaged line loss rate of all lines, the number of tie switch position changes, the number of lines involved in the changed tie switch, the number of packaged lines, the number of suspected package time points, the packaged line loss rate during the suspected package period, the packaged power loss during the suspected package period, and the correlation coefficient between the power loss of negative-loss and high-loss lines during the suspected package period;

[0075] In a specific implementation of the embodiment of the present invention, step (C) specifically includes the following sub-steps:

[0076] Extract the characteristic data of a certain combination in the possible packaged line combination set as follows:

[0077] Step 1: Calculate the total antenna loss rate for all lines in the combination according to the above line loss rate calculation formula, and obtain two characteristic values: the number of high-loss lines in the combination and the number of negative-loss lines in the combination;

[0078] Step 2: Add up the input and output of all lines and use the above line loss rate calculation formula to obtain the packaged line loss rate value characteristics of all lines in the combination;

[0079] Step 3: Add the input and output of the negative-loss line and the high-loss line respectively, and use the above line loss rate calculation formula to obtain four eigenvalues: the high-loss line full antenna loss rate, the negative-loss line full antenna loss rate, the variance of the high-loss line full antenna loss rate, and the variance of the negative-loss line full antenna loss rate. The variance calculation formula is:

[0080]

[0081] Where n is the number of time points at which measurements are collected within the time period T. is the average value of line loss rate in period T, X i is the line loss rate value at each time point in the period T;

[0082] Step 4: Obtain the position change information sent by all tie switches connected to the lines in the packaged line combination, and count the number of position changes and the number of lines involved. This will yield two features: the number of tie switch position changes and the number of lines involved in the position change tie switches.

[0083] Step 5: Count the number of lines involved in the combination to obtain the packaged line number feature;

[0084] Step 6: Calculate the time-sharing line loss rate of each line in the combination at each time point. If a line has a high time-sharing line loss rate for more than one hour, add the line to the time-sharing high-loss line set and take the union of all time points with high time-sharing line loss rates in the high-loss line set. Similarly, obtain the union of negative loss time points and extract the intersection of the union of the high loss time points and the union of the negative loss time points as the suspected packaging period. Count the number of sampling points involved in the suspected packaging period and use it as the characteristic value of the suspected packaging time point.

[0085] Step 7: Calculate the input and output power of all lines during the suspected packaging period, calculate the combined packaging line loss rate during the suspected packaging period, and obtain the package line loss rate characteristic value and the power loss characteristic value during the suspected packaging period. The calculation formula for the package line loss rate is as follows:

[0086]

[0087] Among them, L AT is the package line loss rate of the combined line in time period T, N is the line l i Number of contactable lines; Q nCB+ is the input integrated power of a certain line in time period T; Q nCB- Q is the output integral power of a line in time period T; nDT+ Q is the output integral power of the distribution transformer under a certain line in time period T; nDT- It is the output integral power of the distribution transformer in the topological area under a certain line in the time period T, and the denominator is the loss power;

[0088] Step 8: Based on the high-loss line set obtained during the suspected packaging period, calculate the input and output power of the line at each time throughout the day, and use the above-mentioned packaging line loss rate calculation formula to calculate the power loss of the high-loss line. Similarly, based on the negative-loss line set, calculate the power loss of the negative-loss line, and calculate the correlation coefficient characteristic value of the power loss of the high-loss line and the power loss of the negative-loss line.

[0089] Finally, repeat steps 1 to 8 above to extract the feature values ​​of all combinations in the possible packaged line combination set;

[0090] Step (D), adding historical packaged status data of the line to form a sample set, and establishing and training a neural network model;

[0091] In a specific implementation of the embodiment of the present invention, step (D) specifically includes:

[0092] Combined with the actual historical power transfer combination, the line power transfer status and the extracted feature data are integrated to form a power transfer identification sample set, and a neural network model is established and trained: an RNN neural network model is constructed, and the training set and test set are randomly divided to train the neural network model.

[0093] Step (E) uses the trained model to batch identify the actual transfer data in the distribution network.

[0094] In a specific implementation of the embodiment of the present invention, step (E) specifically includes the following steps:

[0095] After the model prediction accuracy is high, the trained model is used to generate suspected power transfer information and identify power transfer for batch lines based on the operating data in the actual medium-voltage distribution network. That is, a set of packaged line combinations is generated in batches for each line and the required feature values ​​are extracted. The sets are then sent to the model for calculation, and finally the true or false power transfer status results corresponding to each packaged line combination are output.

[0096] In a specific application, the distribution network operation data of a certain province in May is taken as an example to further illustrate the multi-data fusion distribution network line transfer identification method based on neural network of the present invention.

[0097] According to the historical packaging records filled in by the prefecture-level cities, the line loss rate of each line in the package and the line loss rate of the packaged lines in the package are calculated. If there is an abnormal line loss rate of the line in the package and the packaged line loss rate is normal, it will be used as the historical transfer sample data. If one of the packaging records is Figure 2 As shown in the figure, it can be seen that the Green Flat Line and the Hefei Line were bundled together due to maintenance on June 8. On that day, the Green Flat Line was in a negative loss state and the Hefei Line was in a high loss state. The specific line loss rate of the lines is shown in Table 1. The line loss rate calculated after the two lines are bundled is 3.75, which is within the normal line loss range. In addition, from the power stacking diagram of the lines (i.e. Figure 3-5 ), we can see that the time-sharing line loss rate is abnormal during the period of 5:15-19:00, which is the period of power transfer:

[0098]

[0099] Table 1

[0100] By calculating the line loss rate within the packaged line combination, we can obtain three characteristics: the number of high-loss lines in the packaged line combination, the number of negative-loss lines in the packaged line combination, and the packaged line loss rate of all lines throughout the day; by calculating the high / negative line loss rate within the packaged line combination, we can obtain four characteristics: the full-day loss rate of high-loss lines, the variance of the full-day loss rate of high-loss lines, the full-day loss rate of negative-loss lines, and the variance of the full-day loss rate of negative-loss lines; by counting the position change information sent by all the tie switches on the lines within the packaged line combination, we can obtain eight characteristics: the tie switch position change Nine is the number of times, and nine is the number of lines involved in the position-changing interconnecting switch; ten is the number of packaged lines, which is obtained by counting the number of lines in the packaged line combination; by calculating the line loss rate of all lines at each time point, taking the intersection of the set of all high-loss time points and the set of all negative-loss time points as the suspected packaged time point, four features can be obtained: eleven is the number of suspected packaged time points, twelve is the packaged line loss rate during the suspected packaged period, thirteen is the packaged power loss during the suspected packaged period, and fourteenth is the correlation coefficient between the power loss of negative loss and high-loss lines during the suspected packaged period. A total of 14 feature values ​​are extracted.

[0101] The province's packaged data for May was extracted, resulting in a total of 7,199 packaged combinations as a sample set for transfer, involving 16,286 lines. 53,000 combinations of lines not included in the packaged combinations and their interconnecting lines were randomly selected as historical non-transfer sample data. The dataset underwent data preprocessing, including data cleaning, missing outlier processing, and feature extraction. The dataset was then partitioned into a 7:3 ratio, with 70% used as the training set and 30% as the validation set to test model training and optimization.

[0102] Using the training set, we built and trained a recurrent neural network (RNN) model. Using the validation set data, we used accuracy, precision, recall, F1 score, and receiver operating characteristic (ROC) value as evaluation indicators to evaluate the trained model. The evaluation results are shown in the table below. All indicators are high, indicating that the model training effect is good:

[0103] Accuracy F1 score ROC value Precision Recall rate 0.993 0.962 0.991 0.997 0.945

[0104] After all indicators meet the requirements, the trained neural network model is obtained.

[0105] In actual use: all the triggered lines in the cities and prefectures in June were integrated step by step, and 953 package combinations that may have power transfer were extracted based on the triggered lines. The 14 feature values ​​of each possible power transfer package combination were input into the trained neural network model for automatic identification. Finally, a total of 325 power transfer combinations were identified in which power transfer occurred on the lines within the combination. After verification by the distribution network operation personnel of the city, 316 of the 325 identified power transfer combinations were confirmed to be power transfer operations, indicating that the model has strong generalization ability and can be put into online use.

[0106] The present invention combines multi-source real-time data such as switch signals, line plans and fault records, and equipment measurements to extract data change characteristics related to line transfers from multiple dimensions. It is also compatible with data upload anomalies, missed or delayed signal record uploads, and can identify real line transfer events from a data-driven perspective, ultimately diagnosing accurate, real-time, and comprehensive distribution network transfer information. By fully, promptly, and accurately identifying transfer operations occurring in the distribution network, it can effectively and promptly provide auxiliary decision support for distribution network scheduling, inspections, and emergency repairs, with low development costs and significant results.

[0107] The present invention can help operators to timely discover power transfer events and locate their locations, identify weak risk points in the grid and make relevant response decisions in a timely manner.

[0108] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying power supply transfer in distribution network based on multi-data fusion of neural network, characterized in that: include: Obtain records and measurement data of distribution transformers and lines, position change signals and measurement data sent by switches, line fault and plan-related records, and line topology correlation data; Generate a trigger line set based on the acquired data, and extract possible packaged line combination sets based on the trigger line set and line contact ledger; Based on the packaged line combination set, 14 eigenvalues ​​are established, including the number of high-loss lines, the number of negative-loss lines, the loss rate of the entire antenna of high-loss lines, the variance of the loss rate of the entire antenna of high-loss lines, the loss rate of the entire antenna of negative-loss lines, the variance of the loss rate of the entire antenna of negative-loss lines, the packaged line loss rate of all lines throughout the day, the number of tie switch changes, the number of lines involved in the changed tie switch, the number of packaged lines, the number of suspected package time points, the packaged line loss rate during the suspected package period, the packaged power loss during the suspected package period, and the correlation coefficient between the power loss of negative-loss and high-loss lines during the suspected package period. Add historical packaged status data of the line to form a sample set, and build and train the neural network model; The trained model is used to batch identify the actual transfer data in the distribution network.

2. The method for identifying power supply transfer in distribution network based on multi-data fusion of neural network according to claim 1, characterized in that: The records and measurement data of the distribution transformer and the line include: basic information of the line and the distribution transformer, power of the line and the distribution transformer, and power factor value data; The switch position change signal and measurement data sent up include: all switch position change information of the main distribution network, line association and switch type ledger data, and power measurement data; The line fault and plan related records include: publicly released relevant line fault and plan information, and weekly maintenance plan related line plan information; The line topology association relationship data includes: line topology data, including line topology paths and tie switch opposite end line information data.

3. The method for identifying power supply transfer in distribution network based on multi-data fusion using a neural network according to claim 1, characterized in that: The process of extracting the possible set of packaged line combinations includes: According to the change from value to value of all switch measurements on the line, the position change information sent by all switches on the line, line fault and plan information, all lines are extracted as the trigger line set L = [l1, l2, ..., l i ,…,l a ], a is all trigger lines; Based on the trigger line set, take the trigger line l i A list of all contact lines is obtained, and more than two potential packaged line combinations are performed on the lines in the list. The above potential packaged line combination data are extracted for all lines in the triggered line set, and finally a possible packaged line combination set is formed.

4. The method for identifying power supply transfer in distribution network based on multi-data fusion of neural network according to claim 1, characterized in that: The characteristic values ​​established based on the packaged line combination set include: By calculating the line loss rate within the packaged line combination, we can obtain three features: the number of high-loss lines in the packaged line combination, the number of negative-loss lines in the packaged line combination, and the packaged line loss rate of all lines throughout the day. By calculating the high / negative line loss rate within the packaged line combination, we can obtain four features: the full-day loss rate of high-loss lines, the variance of the full-day loss rate of high-loss lines, the full-day loss rate of negative-loss lines, and the variance of the full-day loss rate of negative-loss lines. By counting the position change information sent by all the tie switches connected to the lines in the packaged line combination, we can obtain two features: eight is the number of tie switch position changes, and nine is the number of lines involved in the changed tie switches; ten is the number of packaged lines, which is obtained by counting the number of lines in the packaged line combination; by calculating the line loss rate of all lines at each time point, taking the intersection of the set of all high-loss time points and the set of all negative-loss time points as the suspected packaged time point, we can obtain four features: eleven is the number of suspected packaged time points, twelve is the packaged line loss rate during the suspected packaged period, thirteen is the packaged power loss during the suspected packaged period, and fourteenth is the correlation coefficient between the power loss of negative loss and high-loss lines during the suspected packaged period.

5. The method for identifying power supply transfer in distribution network based on multi-data fusion of neural network according to claim 1, characterized in that: Add the historical packaged status data of the line to form a sample set, build and train the neural network model, including the following steps: Integrate the line transfer status and extracted feature data to form a transfer identification sample set, establish and train the neural network model: build an RNN neural network model, randomly divide the training set and test set, and train the neural network model.

6. The method for identifying power supply transfer in distribution network based on multi-data fusion of neural network according to claim 1, characterized in that: Using the trained model to batch identify actual transfer data in the distribution network includes the following steps: The packaged line combination sets for each line are batch generated and the required characteristic values ​​are extracted, which are sent to the model for calculation, and finally the true and false transfer status results corresponding to each packaged line combination are output.

Images