An edge-computing-based fault diagnosis and analysis method and system for transformer area collection

Abstract

The application discloses an edge-computing-based fault diagnosis and analysis method and system for a transformer area, and the method comprises the following steps: a fault diagnosis device acquires working condition information in real time, wherein the working condition information comprises real-time power utilization working condition information of a power utilization device, real-time electric meter working condition information of an intelligent electric meter and real-time concentrator working condition information of a concentrator, and the fault diagnosis device actively diagnoses the working condition information; a cloud computing platform establishes an LSTM model to predict faults through the working condition information; and a user terminal receives diagnosis results of the fault diagnosis device and prediction results of the LSTM model in predicting faults. Through the edge-computing-based fault diagnosis and analysis method and system for the transformer area, the efficiency of solving faults can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology, specifically to a method and system for fault diagnosis and analysis of transformer substations based on edge computing. Background Technology

[0002] The electricity information collection system is a crucial component of the State Grid's smart grid, providing substantial data support for various functions such as power supply voltage monitoring, power quality management, line loss management, fee control management, remote meter reading, electricity inspection, and value-added services for users. The current collection system infrastructure has largely achieved its goal of "full coverage and full collection," and the focus is gradually shifting from system construction to system operation and maintenance. Due to the large number and variety of collection devices, their wide distribution, and complex operating environments, the following main problems exist during the collection and maintenance process:

[0003] (1) There are many types of data acquisition failures. Many factors can affect the success of data acquisition, including the main station, communication network (GPRS public network, RS485, power line carrier), concentrator, data acquisition device, and electricity meter. Among these, there are many brands of data acquisition equipment with varying quality, resulting in frequent abnormal data acquisition situations and a variety of failure types.

[0004] (2) Lack of technical means for fault diagnosis. Due to the lack of effective diagnostic technology, fault location is difficult and mostly relies on equipment manufacturers. This often results in equipment manufacturers concealing equipment quality problems or passing the buck among themselves, leading to insufficient timeliness and accuracy in fault handling.

[0005] (3) Fault analysis and handling rely too much on manual methods. Repetitive mechanical work takes up a long time, resulting in a waste of human resources. The current efficiency of fault handling can no longer meet the increasing workload of operation and maintenance.

[0006] (4) The lack of system information and fault prediction and analysis capabilities for the status of terminal equipment in the entire region makes it difficult to improve the level of system management and equipment operation and maintenance.

[0007] Existing technology, specifically invention patent CN114019299A, describes a method and system for proactively assessing and reporting power outage information for dedicated transformer users. Based on 5G modules, it collects power outage data from dedicated transformer users, efficiently uploading structured and unstructured data collected by IoT sensing terminals. Through multi-source data fusion technology and edge computing modules, it achieves power outage information feature analysis, efficient fault assessment, and precise fault location. Finally, 5G technology enables fault information reporting, achieving second-level transmission. Upon receiving the power outage information, the main station personnel can quickly implement emergency repairs. However, this existing technology has limited application scenarios, identifies a limited number of fault types, and cannot provide early warnings. Summary of the Invention

[0008] The technical problem to be solved by this invention is to address the issues of multiple types of faults in data acquisition equipment, a lack of technical means for fault diagnosis, low efficiency in fault analysis and processing, and a lack of fault prediction for the status of terminal equipment in the entire region.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0010] A method for fault diagnosis and analysis of transformer substations based on edge computing includes the following steps:

[0011] The fault diagnosis equipment acquires operating condition information in real time, including real-time power consumption information of electrical equipment, real-time meter operating condition information of smart meters, and real-time concentrator operating condition information of concentrators. The fault diagnosis equipment actively diagnoses the operating condition information.

[0012] The cloud computing platform uses the aforementioned operating condition information to build an LSTM model to predict faults;

[0013] The user terminal receives the diagnostic results from the fault diagnosis device and the fault prediction results from the LSTM model.

[0014] The LSTM model predicts faults through the following steps:

[0015] Extract current data, voltage data, and temperature and pressure data of the corresponding smart meter from the meter's operating condition information to form a four-dimensional feature vector at any time point;

[0016] Based on the historical data of the electricity meter's operating condition, four-dimensional feature vectors of n consecutive time points are arbitrarily extracted to form a training set, and each set of n consecutive four-dimensional feature vectors is recorded as a training data point.

[0017] The first n-1 four-dimensional feature vectors of each training data point are the input of the LSTM model, and the nth four-dimensional feature vector is the output of the LSTM model. The LSTM model is then trained.

[0018] For each smart meter, the meter operating condition information is sampled at any time point t, and the first n-1 consecutive four-dimensional feature vectors at sampling time point t are obtained. The LSTM model then predicts and outputs the nth four-dimensional feature vector.

[0019] A preset prediction threshold is set, and the true four-dimensional feature vector at sampling time point t is obtained. If the sum of the absolute values ​​of the differences between the current data, voltage data, temperature data, and pressure data in the true four-dimensional feature vector and the current data, voltage data, temperature data, and pressure data in the nth four-dimensional feature vector predicted by the corresponding LSTM model is greater than the prediction threshold, then the smart meter is determined to have a data acquisition failure.

[0020] Advantages: The fault diagnosis equipment of this invention can support active diagnosis of various faults, realize active diagnosis of acquisition failure, in-depth analysis of abnormal acquisition data, establish LSTM model to predict acquisition faults, improve the efficiency of on-site fault handling, reduce company operation and maintenance costs, and improve the level of intelligent operation and maintenance.

[0021] In one embodiment of the present invention, the fault diagnosis device supports fault diagnosis types including concentrator tube faults and smart meter faults, wherein the concentrator tube faults include carrier module non-networking faults and offline faults; the fault diagnosis device diagnoses the carrier module non-networking fault of the concentrator in the following manner:

[0022] The address correspondence between the concentrator and the smart meter is preset;

[0023] For each message sent by each concentrator to the fault diagnosis device, the addresses of all the smart meters are parsed to form a set of addresses where communication was successful.

[0024] Compare the preset address mapping with the set of addresses that constitute successful communication;

[0025] If a certain address information is missing, it is assumed that the concentrator corresponding to that address is not communicating with the smart meter, and the carrier module of that concentrator is not networked; otherwise, the information is reversed.

[0026] The fault diagnosis equipment diagnoses the concentrator as offline using the following methods:

[0027] If the fault diagnosis device does not receive a message from a concentrator for 5 to 15 minutes, it determines that the concentrator is suspected to be offline.

[0028] The fault diagnosis device sends a ping command to the concentrator. If there is no response, the concentrator is determined to be offline; if there is a response, the concentrator's signal output function is determined to be faulty.

[0029] If the fault diagnosis device receives garbled messages from a concentrator for 50 to 70 consecutive minutes, it determines that the signal quality of that concentrator is problematic.

[0030] In one embodiment of the present invention, when it is determined that the signal quality of the concentrator is problematic, the fault diagnosis device diagnoses the signal strength of the concentrator through the following steps:

[0031] Obtain the autocorrelation function of the initial signal received by the concentrator;

[0032] The central part of the autocorrelation function is selected, and noise outside a certain window is excluded to obtain a noise-reduced received signal that reduces the noise of the initial signal. The noise-reduced received signal includes an analysis signal.

[0033] The noise-reduced signal is subjected to Fourier transform to obtain the power density of the noise-reduced received signal after time-domain filtering; the power density of the noise-reduced received signal includes the power density of the analysis signal;

[0034] The power density of the analyzed signal is segmented, and the average value of each segment is calculated and compared with the threshold value to determine the signal strength of the concentrator.

[0035] In one embodiment of the present invention, the concentrator fault also includes a fault of not collecting cross-collection power and a fault of not returning daily frozen data. The fault diagnosis device diagnoses the concentrator as having a fault of not collecting cross-collection power by the following method:

[0036] If the message data sent by the concentrator is not contained in the data collection data, it is determined that the concentrator is not collecting the data collection power.

[0037] The fault diagnosis equipment diagnoses the concentrator as having a fault where it does not return daily frozen data using the following methods:

[0038] The concentrator uploads the daily frozen data and its timestamp to the fault diagnosis device;

[0039] After the daily frozen data return time, a fixed query period is scheduled. During the fixed query period, the fault diagnosis device queries daily whether it has received the daily frozen data sent by each concentrator.

[0040] If no data is found, it is determined that the corresponding concentrator has experienced a daily data freeze failure; if the daily data freeze is found, the timestamp is then subjected to a second check.

[0041] Compare whether the timestamp corresponding to the daily frozen data is within the valid time range corresponding to the fault diagnosis device on that day;

[0042] If the timestamp is not within the valid range, the corresponding concentrator is determined to have experienced a daily data freeze and not return a fault; if the timestamp is within the valid range, the corresponding concentrator is determined to have experienced a daily data freeze and return a normal value.

[0043] In one embodiment of the present invention, the concentrator fault also includes a recall failure fault, and the fault diagnosis device diagnoses the concentrator as having a recall failure fault in the following manner:

[0044] The fault diagnosis device sends a call command to the concentrator, and the concentrator sends an hourly data call command to the smart meter according to the call command.

[0045] If a concentrator fails to obtain the real-time meter status information of the smart meter it is communicating with, it is determined that the concentrator has experienced a call failure; otherwise, it returns.

[0046] In one embodiment of the present invention, the smart meter faults include the meter not recording electricity, the meter recording reverse electricity, and the meter having a negative positive reactive power value; the fault diagnosis device diagnoses the fault of the meter not recording electricity in the following ways:

[0047] The fault diagnosis device uses a first-order forward differential method to determine whether the smart meter is not moving.

[0048] If the smart meter stops moving, it is suspected to be a fault where the meter is not registering electricity; otherwise, the smart meter is normal.

[0049] When it is determined that the fault is suspected to be the fault of the meter not moving, the fault diagnosis device obtains the difference between the total power of other smart meters in the local network area containing the smart meter suspected to be the fault of the meter not moving and the power collected by all collectors covering the area.

[0050] If the difference continues to increase over time, it is determined that the smart meter has a fault where the meter reading is not being recorded; otherwise, it is determined that the user corresponding to the smart meter is not using electricity.

[0051] The fault diagnosis equipment diagnoses reverse power generation faults in the electricity meter using the following methods:

[0052] The fault diagnosis device subtracts the real-time power consumption information of each smart meter uploaded by each concentrator from the real-time power consumption information of the previous moment; if a negative number is obtained, it is determined that the current smart meter has a reverse power consumption fault; if a positive number is obtained, it is determined that the current smart meter has not a reverse power consumption fault.

[0053] The fault diagnosis equipment diagnoses a fault where the positive reactive power of the electricity meter is negative using the following methods:

[0054] The fault diagnosis device determines the reactive power data contained in the real-time concentrator operating status information uploaded by the concentrator, and determines whether the reactive power is less than 0; if it is less than 0, it is determined that the smart meter has a fault where the meter's positive reactive power is negative; if it is greater than 0, it is determined that the smart meter has not a fault where the meter's positive reactive power is negative.

[0055] In one embodiment of the present invention, the smart meter fault also includes a phase power failure fault, and the fault diagnosis device diagnoses the reverse power failure of the meter in the following manner:

[0056] Power on and initialize the smart meter, and clear the zero-line flag.

[0057] The real-time voltages of phases Ua, Ub, and Uc, the first angle between phase Ub and phase Uc, and the second angle between phase Uc and phase Ua are sampled and obtained.

[0058] The neutral wire failure indicator is checked. If the neutral wire failure indicator is not set, it indicates that the current smart meter has not experienced a neutral wire failure. Further, a phase power failure fault is determined. If the neutral wire failure indicator is set, it indicates that the current smart meter has experienced a phase power failure.

[0059] When the voltages of phase Ua, phase Ub, and phase Uc are all greater than 5V; when the first included angle is less than 110° or the second included angle is greater than 250°; when the voltage imbalance rate of the three phases is in the range of 95% to 100%, the voltage of phase Ub is greater than 120% of the rated voltage, the voltage of phase Uc is greater than 120% of the rated voltage, and the voltage of phase Ua is greater than 0V and less than 10% of the rated voltage; when all three conditions are met simultaneously, the smart meter is considered to have experienced a phase power failure.

[0060] If, under three-phase conditions, the voltage of one phase is 0V and the voltage of one of the other two phases is greater than 5V but less than the rated voltage, then if any one of the three conditions is met under this condition, the smart meter is considered to have experienced a phase power failure.

[0061] In one embodiment of the present invention, the smart meter fault also includes a meter flyaway abnormality fault, and the fault diagnosis device diagnoses the meter flyaway abnormality fault in the following manner:

[0062] The historical daily frozen data sequence of each smart meter is converted from time series format to supervised learning format data;

[0063] The supervised learning format data of the past T days is used as the training data input for the gradient boosting algorithm to predict the daily frozen data of T+1 day.

[0064] If the difference between the actual daily frozen data on day T+1 and the predicted daily frozen data is greater than the threshold, then the smart meter is considered to have a meter flying away abnormal fault.

[0065] In one embodiment of the present invention, the smart meter fault also includes a meter clock error fault, and the fault diagnosis device diagnoses the meter clock error fault in the following manner:

[0066] Obtain the clock data sequence by parsing the messages from each smart meter;

[0067] The clock difference sequence is obtained by subtracting each clock data value in the clock data sequence from the system clock value corresponding to the receiving time of the fault diagnosis device.

[0068] The clock difference sequence is subjected to multiple curve fitting operations to obtain multiple curve parameters;

[0069] Based on the multiple curve parameters, obtain multiple curve parameter features;

[0070] The center point features are obtained based on the characteristics of multiple curve parameters;

[0071] Obtain the Euclidean distance between the clock difference sequence and the center point feature, and compare the Euclidean distance with a preset distance threshold;

[0072] If the Euclidean distance is greater than a preset distance threshold, the smart meter is determined to have a clock error; otherwise, the opposite applies.

[0073] The center point feature is obtained using the following formula:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] In the formula, Represented as the central point feature, and Represented as multiple curve parameter characteristics, and It is represented by multiple curve parameters, where N represents the total number of smart meters, and i∈[1,N].

[0080] This invention also provides a system for fault diagnosis and analysis of transformer substations based on edge computing, comprising:

[0081] Electrical equipment;

[0082] A smart meter is communicatively connected to the electrical equipment to obtain real-time power consumption information of the electrical equipment.

[0083] A concentrator, which is communicatively connected to the smart meter, is used to obtain real-time meter operating status information of the smart meter;

[0084] The fault diagnosis device is communicatively connected to the concentrator and is used to obtain real-time concentrator operating condition information and perform active diagnosis based on the operating condition information.

[0085] A cloud computing platform, communicatively connected to the fault diagnosis equipment, is used to establish an LSTM model to predict faults based on the operating condition information. The LSTM model predicts faults through the following steps: extracting current data, voltage data, and corresponding temperature and pressure data of the smart meter from the meter's operating condition information to form a four-dimensional feature vector at any given time point; arbitrarily selecting four-dimensional feature vectors from n consecutive time points based on historical data of the meter's operating condition information to form a training set, with each n consecutive four-dimensional feature vectors representing a training data point; the first n-1 four-dimensional feature vectors of each training data point serve as the input to the LSTM model, and the nth four-dimensional feature vector is the LST (Learning Set for LSTM). The output of the M-model is used to train the LSTM model. For any sampling time point t of the meter operating information of each smart meter, the first n-1 consecutive four-dimensional feature vectors of sampling time point t are obtained. The LSTM model predicts and outputs the nth four-dimensional feature vector. A preset prediction threshold is set, and the true four-dimensional feature vector of sampling time point t is obtained. If the sum of the absolute values ​​of the differences between the current data, voltage data, temperature data, and pressure data in the true four-dimensional feature vector and the current data, voltage data, temperature data, and pressure data in the corresponding nth four-dimensional feature vector predicted by the LSTM model is greater than the prediction threshold, then it is determined that the smart meter has a data acquisition failure.

[0086] The user terminal is communicatively connected to the cloud computing platform and is used to receive the diagnostic results of the fault diagnosis equipment and the prediction results of the fault prediction by the LSTM model.

[0087] Compared with the prior art, the beneficial effects of the present invention are: (1) The fault diagnosis method and equipment of the present invention can support active diagnosis of 15 types of faults of the transformer area concentrator and smart meter, improve the efficiency of on-site fault handling, and reduce the company's operation and maintenance costs. (2) In addition to timely and accurate diagnosis of faults that have already occurred, the present invention also performs in-depth analysis of anomalies in the transformer area data, establishes a fault early warning model based on LSTM transformer area data collection, realizes fault early warning, and improves the level of intelligent operation and maintenance. (3) The method of the present invention is a non-intrusive active diagnosis method, that is, it does not affect the normal operation of the existing transformer area data collection system. Attached Figure Description

[0088] Figure 1 This is a flowchart illustrating the LSTM model for predicting faults in this invention.

[0089] Figure 2 This is a flowchart illustrating the process of determining whether the concentrator carrier module is not networking, as described in this invention.

[0090] Figure 3 This is a flowchart for determining the offline fault of the concentrator according to the present invention.

[0091] Figure 4 This is a flowchart illustrating the process of determining the signal strength of a concentrator according to the present invention.

[0092] Figure 5 This is a flowchart illustrating the process for determining the failure of frozen data to be returned in the present invention.

[0093] Figure 6 This is a flowchart for determining the fault of the electricity meter not registering power consumption according to the present invention.

[0094] Figure 7 This is a flowchart for determining a phase power failure in a smart meter according to the present invention.

[0095] Figure 8 This is a flowchart for determining the abnormal fault of the electricity meter flying away, as described in this invention.

[0096] Figure 9 This is a flowchart for determining the fault of an electricity meter clock out of tolerance according to the present invention.

[0097] Figure 10 This is a system block diagram of a fault diagnosis and analysis method for transformer substations based on edge computing, according to the present invention. Detailed Implementation

[0098] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0099] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0100] See Figure 1 As shown, this invention provides a method for fault diagnosis and analysis of transformer substations based on edge computing, comprising: a fault diagnosis device acquiring real-time operating condition information, the operating condition information including real-time power consumption information of electrical equipment, real-time meter operating condition information of smart meters, and real-time concentrator operating condition information of concentrators; the fault diagnosis device actively diagnosing the operating condition information; a cloud computing platform establishing an LSTM model to predict faults based on the operating condition information; and a user terminal receiving the diagnosis results of the fault diagnosis device and the prediction results of the faults predicted by the LSTM model.

[0101] The LSTM model predicts faults through the following steps:

[0102] S111, extract current data, voltage data, and temperature and pressure data of the corresponding smart meter from the meter's operating condition information to form a four-dimensional feature vector at any time point.

[0103] S112, Based on the historical data of the electricity meter's operating condition information, arbitrarily extract four-dimensional feature vectors from n consecutive time points to form a training set, and each n consecutive four-dimensional feature vectors is recorded as a training data point.

[0104] S113, the first n-1 four-dimensional feature vectors of each training data are the input of the LSTM model, and the nth four-dimensional feature vector is the output of the LSTM model, thereby training the LSTM model.

[0105] S114, for any sampling time point t of the meter operating information of each smart meter, obtain the first n-1 consecutive four-dimensional feature vectors of sampling time point t, and the LSTM model predicts and outputs the nth four-dimensional feature vector.

[0106] S115, preset the prediction threshold, obtain the true four-dimensional feature vector at sampling time point t, if the sum of the absolute values ​​of the differences between the current data, voltage data, temperature data and pressure data in the true four-dimensional feature vector and the current data, voltage data, temperature data and pressure data in the nth four-dimensional feature vector predicted by the corresponding LSTM model is greater than the prediction threshold, then it is determined that the smart meter has a data acquisition failure.

[0107] See Figure 1 As shown, in one embodiment of the present invention, the LSTM model adopts the TensorFlow framework and utilizes Tencent's NCNN neural network inference library when deployed in the cloud. In step S111, the four-dimensional feature vector at any time point is [voltage, current, temperature, pressure]. In step S112, for example, arbitrarily 100 consecutive time points are selected as four-dimensional feature vectors, and every 100 consecutive four-dimensional feature vectors are recorded as one training data. In steps S114 and S115, the LSTM model predicts and outputs the 100th four-dimensional feature vector, i.e., the output voltage data, current data, temperature data, and pressure data, denoted as v, i, t, and p, respectively. The voltage data, current data, temperature data, and pressure data actually sampled at sampling time point t are denoted as v^, i^, t^, and p^, respectively. The absolute value of the difference between the two is...

[0108] Referring to Table 1, in one embodiment of the present invention, the fault diagnosis device supports fault diagnosis types including concentrator faults and smart meter faults. The fault diagnosis device supports active diagnosis of 15 fault types, as shown in Table 1.

[0109] Table 1. List of Active Fault Diagnosis Types Supported by Fault Diagnosis Equipment

[0110]

[0111] Please see Figure 2 As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the carrier module non-networking fault of the concentrator in the following manner:

[0112] S121, Preset the address correspondence between the concentrator and the smart meter.

[0113] S122, for each message sent by each concentrator to the fault diagnosis device, parse the addresses of all the smart meters to form a set of addresses where communication was successful.

[0114] S123, compare the preset address mapping relationship with the set of addresses that constitute successful communication.

[0115] S124 If a certain address information is missing, it is assumed that the concentrator corresponding to that address is not communicating with the smart meter, and the carrier module of this concentrator is not networked; otherwise, the value is inverted.

[0116] Please see Figure 2 As shown, in one embodiment of the present invention, the smart meter can be a three-phase meter or a single-phase meter. In this embodiment, a three-phase meter is used as an example.

[0117] Please see Figure 3 As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the concentrator as offline by means of the following method:

[0118] S131, if the fault diagnosis device does not receive a message from a concentrator for 5 to 15 minutes, it determines that the concentrator is suspected to be offline;

[0119] S132, the fault diagnosis device sends a ping command to the concentrator. If there is no response, the concentrator is determined to be offline; if there is a response, the signal output function of the concentrator is determined to be faulty.

[0120] S133, if the fault diagnosis device receives garbled data in the message sent by a certain concentrator for 50 to 70 consecutive minutes, it is determined that the signal quality of the concentrator has a problem.

[0121] Please see Figure 3 As shown, in one embodiment of the present invention, in steps S131 to S133, the fault diagnosis device monitors the messages uploaded by each concentrator in real time and records the time of receiving each message. When it is determined that the signal quality of the concentrator has a problem, the present invention quantitatively analyzes the quality of the concentrator communication network using a signal strength method based on frequency domain processing and a signal-to-noise ratio estimation method based on signal spectrum self-normalization.

[0122] Please see Figure 4 As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the signal strength of the concentrator through the following steps:

[0123] S141, Obtain the autocorrelation function of the initial signal received by the concentrator.

[0124] S142, Select the central part of the autocorrelation function, exclude some noise outside the window, and obtain a noise-reduced received signal that reduces the noise of the initial signal. The noise-reduced received signal includes an analysis signal.

[0125] Where, x(t1) = s(t1) + n(t1);

[0126] In the formula, x(t1) represents the noise-reduced received signal, s(t1) represents the analyzed signal, and n(t1) represents the noise signal.

[0127] S143, perform Fourier transform on the denoised signal to obtain the power density of the denoised output signal after time-domain filtering; the power density of the denoised received signal includes the power density of the analysis signal.

[0128] Among them, P x (w)=P s (w)+P n (w);

[0129] In the formula, P x (w) represents the power spectral density of the noise-reduced received signal, P s (w) represents the power spectral density of the analyzed signal, P n (w) represents the power spectral density of the noise signal.

[0130] S144, the power density of the analyzed signal is segmented, and the average value of each segment is calculated and compared with the threshold value to determine the signal strength of the concentrator.

[0131] Please see Figure 4As shown, in one embodiment of the present invention, in steps S141 to S144, since the power density of Gaussian white noise is uniformly distributed throughout the bandwidth of the smart meter, the effective value of the power spectral density of the noise-reduced received signal is mainly concentrated at the center of the power spectral density of the analytical signal. Therefore, the calculated power spectral density can be segmented, and the average of each segment is calculated and compared with a threshold value to determine the strength of the acquired HPLC communication message signal. Specifically, the threshold value can be set based on manual experience. If the average value is greater than the threshold value, the signal strength acquired by the concentrator is stronger; otherwise, the value is reversed.

[0132] Please see Figure 4 As shown, in one embodiment of the present invention, when it is determined that the signal quality of the concentrator has a problem, the fault diagnosis device diagnoses the quality of the concentrator's communication network through the following steps. Based on the symmetrical characteristics of the signal spectrum, and according to the basic idea that the power of the signal spectrum is equal to that on both sides of the center frequency, the position of the carrier center frequency is obtained by iteratively using a moving average and self-normalized spectrum, and then the frequency deviation value is calculated. If the signal is after ideal carrier recovery, then the center frequency of the baseband signal should be near zero frequency. In actual reception, the signal spectrum is symmetrical about the zero frequency of the analysis bandwidth. Therefore, in practice, if the signal has a frequency deviation, then the offset should be equal to the frequency difference between the center position of its signal spectrum and the center frequency of the analysis bandwidth. After obtaining the signal bandwidth, the frequency band of the signal is known, and the frequency band of the noise can be determined. After determining the frequency range of the signal, the signal-to-noise ratio can be obtained by calculating the average power of the noise and the average power of the signal. The signal quality of the concentrator is judged based on the signal-to-noise ratio.

[0133] In one embodiment of the present invention, the fault diagnosis device diagnoses the concentrator as having a failure to collect cross-collection power data by parsing the message data sent by the concentrator. If the message data does not contain cross-collection data, the concentrator is deemed to be experiencing a failure to collect cross-collection power data. Since the message sent by a concentrator with normal cross-collection should contain cross-collection information (voltage, current, power, power factor, etc., as required by clause 4.9.2.4 of Q / GDW 374.3—2009), the fault diagnosis device considers a failure to collect cross-collection power data to have occurred when it parses the message data of the concentrator and finds that it does not contain cross-collection data.

[0134] Please see Figure 5 As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the concentrator as having a fault of not returning daily frozen data by means of the following method:

[0135] S151, the concentrator uploads the daily frozen data and its timestamp to the fault diagnosis device.

[0136] S152, after the daily frozen data return time, a fixed query time is scheduled. During the fixed query time period, the fault diagnosis device queries daily whether it has received the daily frozen data sent by each concentrator.

[0137] S153, if no data is found, it is determined that the corresponding concentrator has a daily frozen data non-return fault; if the daily frozen data is found, the timestamp is judged a second time.

[0138] S154, compare whether the timestamp corresponding to the daily frozen data is within the valid time range corresponding to the fault diagnosis device on that day.

[0139] S155, if the timestamp is not within the valid range, it is determined that the corresponding concentrator has experienced a daily data freeze failure and is not returning data; if the timestamp is within the valid range, it is determined that the corresponding concentrator has experienced a daily data freeze and is returning data normally.

[0140] Please see Figure 5 As shown, in one embodiment of the present invention, the fault diagnosis device first checks whether it has received the daily frozen data of each meter sent by each concentrator ten minutes after the predetermined daily frozen data return time. If no data is found, it is directly determined that the current concentrator has a fault of not returning daily frozen data. If daily frozen data is found, the timestamp is checked a second time to compare whether the timestamp corresponding to the daily frozen data is within the valid range of the system time of the day corresponding to the fault diagnosis device. If the timestamps are inconsistent, it is also determined that the current concentrator has a fault of not returning daily frozen data. Only when both daily frozen data and a valid timestamp exist is it determined that the concentrator's daily frozen data return is normal.

[0141] In one embodiment of the present invention, the fault diagnosis device diagnoses the concentrator as having a fault of returning only part of the daily frozen data, which is basically the same as the method for determining that the concentrator is having a fault of not returning daily frozen data. The difference is that if it is found that some of the daily frozen data of the smart meters are returned normally, while the other part is not returned, then it is determined that the concentrator has a fault of only returning only part of the daily frozen data.

[0142] In one embodiment of the present invention, the concentrator fault also includes a recall failure fault. The fault diagnosis device diagnoses the concentrator as having a recall failure fault in the following manner: the fault diagnosis device sends a recall command to the concentrator, and the concentrator sends an hourly data recall to the smart meter according to the recall command. If a concentrator fails to successfully obtain the real-time meter status information of the smart meter it communicates with, it is determined that the concentrator has a recall failure fault; otherwise, it returns.

[0143] The fault diagnosis device sends a call command to all concentrators every 24 hours. The command action is that each concentrator sends a data call to the smart meters it is connected to at the top of the hour. The specific top of the hour is randomly specified by the program, that is, each time the data of a certain top of the hour stored in the smart meter within the past 24 hours is randomly called.

[0144] Please see Figure 6 As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the fault of the electricity meter not registering electricity usage in the following manner:

[0145] S211, the fault diagnosis device uses a first-order forward differential method to determine whether the smart meter is not moving.

[0146] S212, if the smart meter stops moving, it is determined to be a suspected fault of the meter not consuming electricity; otherwise, the smart meter is normal.

[0147] S213, when it is determined that the fault is suspected to be the fault of the meter not moving, the fault diagnosis device obtains the difference between the total power of other smart meters in the local network area containing the smart meter suspected to be the fault of the meter not moving and the power collected by all collectors covering the area.

[0148] S214, if the difference continues to increase over time, it is determined that the smart meter has a fault where the meter power is not running; otherwise, it is determined that the user corresponding to the smart meter is not using electricity.

[0149] Please see Figure 6 As shown, in one embodiment of the present invention, there are two main reasons why the smart meter is not registering electricity: one is that the user is indeed out of power, and the other is that the smart meter is malfunctioning. Therefore, it is necessary to diagnose the specific cause.

[0150] In one embodiment of the present invention, the fault diagnosis device diagnoses the reverse power consumption fault of the electricity meter in the following manner: the fault diagnosis device subtracts the real-time power consumption information corresponding to each smart meter uploaded by each concentrator from the real-time power consumption information of the previous moment. If a negative number is obtained, it is determined that the current smart meter has the reverse power consumption fault; if a positive number is obtained, it is determined that the current smart meter does not have the reverse power consumption fault.

[0151] In one embodiment of the present invention, the fault diagnosis device diagnoses a negative positive reactive power fault in the smart meter by: determining the reactive power data contained in the real-time concentrator operating status information uploaded by the concentrator; and determining whether the reactive power is less than 0. If it is less than 0, the smart meter is determined to have a negative positive reactive power fault; if it is greater than 0, the smart meter is determined not to have a negative positive reactive power fault. The real-time concentrator operating status information directly contains reactive power data.

[0152] Please see Figure 7 As shown, in one embodiment of the present invention, existing three-phase smart meters do not have the function of detecting the loss of neutral wire, but the data acquisition unit can accurately obtain the three-phase voltage of the smart meter. Based on the principle of three-phase load imbalance, it can be determined whether the three-phase smart meter has experienced a loss of neutral wire fault.

[0153] Under normal circumstances, there is no potential difference between the neutral point "O" of a three-phase four-wire power supply and the neutral point O of a three-phase smart meter. If the neutral wire of the three-phase smart meter is interrupted, the three-phase load of the meter actually forms a star connection with the neutral point not connected to the neutral wire. At this time, an asymmetrical star connection will occur, resulting in a neutral point O offset phenomenon. That is, the line voltages of the three-phase loads are equal, but the phase voltages are not equal. The greater the difference in the three-phase loads, the greater the O offset, and the greater the difference in phase voltages. The phase with a larger load has a smaller phase voltage, while the phase with a smaller load has a larger phase voltage. For example, if the load of the three-phase smart meter module is on phase A, which is larger than the other two phases, then when the neutral wire is lost, the phase voltage Ua will decrease, while the phase voltages Ub and Uc will increase. The fault diagnosis equipment diagnoses the reverse power fault of the meter in the following ways:

[0154] S221, Power on and initialize the smart meter, and clear the zero-line flag;

[0155] S222, sample and acquire the real-time voltages of phases Ua, Ub, and Uc, the first angle between phase Ub and phase Uc, and the second angle between phase Uc and phase Ua;

[0156] S223, Check the neutral wire failure flag. If the neutral wire failure flag is not set, it indicates that the current smart meter has not experienced a neutral wire failure. Further determine if a phase power failure has occurred. If the neutral wire failure flag is set, it indicates that the current smart meter has experienced a phase power failure.

[0157] S224, when the voltage of phase Ua, phase Ub, and phase Uc are all greater than 5V; when the first included angle is less than 110° or the second included angle is greater than 250°; when the voltage imbalance rate of the three phases is in the range of 95% to 100%, the voltage of phase Ub is greater than 120% of the rated voltage, the voltage of phase Uc is greater than 120% of the rated voltage, and the voltage of phase Ua is greater than 0V and less than 10% of the rated voltage; when all three conditions are met simultaneously, the smart meter is considered to have experienced a phase power failure.

[0158] S225. When the voltage of one phase in the three phases is 0V and the voltage of one of the other two phases is greater than 5V but less than the rated voltage, then if any one of the three conditions is met under this condition, the smart meter is considered to have experienced a phase power failure.

[0159] Please see Figure 7 As shown, in one embodiment of the present invention, in steps S221 to S225, these three cases include:

[0160] 1. When the voltage of phase Ua, the voltage of phase Ub, and the voltage of phase Uc are all greater than 5V, i.e., Ua>5V&&Ub>5V&&Uc>5V.

[0161] 2. When the first included angle is less than 110° or the second included angle is greater than 250°, that is, the first included angle φbc < 110° || the second included angle φca > 250°.

[0162] 3. When the voltage imbalance rate of the three-phase voltage is between 95% and 100%, the voltage of phase Ub is greater than 120% of the rated voltage, the voltage of phase Uc is greater than 120% of the rated voltage, and the voltage of phase Ua is greater than 0V and less than 10% of the rated voltage, i.e., 100% > voltage imbalance rate > 95% && (Ub > 120% Un) && (Uc > 120% Un) && (10% Un > Ua > 0), where Un represents the rated voltage (220V for residential electricity and 380V for commercial electricity). If all three conditions are met simultaneously when all three phase voltages of the three-phase smart meter are present, then the smart meter experiences a phase power failure.

[0163] When one phase of a three-phase smart meter is de-energized, but the other two phases are energized (e.g., phase A is de-energized, but phases B and C are energized), if the voltage of one phase is 0V and the voltage of one of the other two phases is greater than 5V but less than the rated voltage, this can be expressed as (Ua==0)&&(5V) <Ub<Un||5V<Uc<Un)。

[0164] Please see Figure 8As shown, in one embodiment of the present invention, the fault diagnosis device diagnoses the meter flying away abnormal fault in the following manner:

[0165] S231 converts the historical daily frozen data sequence of each smart meter from time series format to supervised learning format data.

[0166] S232 uses the supervised learning format data from the past T days as the training data input for the gradient boosting algorithm to predict the daily frozen data for day T+1.

[0167] S233, if the difference between the actual daily frozen data and the predicted daily frozen data on day T+1 is greater than the threshold, then the smart meter is considered to have a meter flying away abnormal fault.

[0168] Please see Figure 8 As shown, in one embodiment of the present invention, according to the "Online Monitoring and Intelligent Diagnostic Analysis Model for Meters," a meter reading "flying away" is defined as a daily electricity consumption significantly exceeding the normal value, with the data source being daily frozen data. The calculation method is that the ratio of the meter's daily electricity consumption to the user's maximum possible daily electricity consumption is greater than a threshold K, with a suggested threshold value of 60%. This anomaly is monitored daily. While directly using this definition for meter flying away fault detection can identify meters experiencing this fault, interference from remote transmission signals often leads to misjudging data anomalies caused by signal interference as meter flying away faults, impacting on-site work efficiency. By modeling the energy data of each smart meter using a Kalman filter based on historical daily frozen data, the current energy value can be predicted. The difference between the actual and predicted current energy values ​​can be calculated to determine if the difference is within a predetermined threshold, thus identifying whether the smart meter has experienced an energy flying away anomaly. This avoids misjudgment of faults and improves the efficiency of on-site fault handling.

[0169] Please see Figure 9 As shown, in one embodiment of the present invention, a clock error in a smart meter refers to a malfunction in its independently operating timing system, resulting in a discrepancy between the recorded electricity consumption time and the user's actual electricity consumption time, severely affecting the calculation of time-of-use tiered pricing. In actual electricity consumption area data collection scenarios, the number of smart meters is extremely large; for example, a typical residential community may have thousands of smart meters. Real-time active testing to diagnose clock error faults in all smart meters is impractical from both a time and bandwidth perspective.

[0170] The fault diagnosis device sends clock information active recall commands to the smart meters every ten minutes according to a preset address sequence via a concentrator. It then calculates the difference between the returned clock information and the current standard Beijing time. If the difference in seconds is within a preset threshold, the meter is considered not to have a clock error; otherwise, it is judged to have a clock error fault. However, as mentioned earlier, this process is extremely time-consuming, resulting in a significant lag in the clock error judgment. For example, if there are 1000 smart meters in a distribution area, one round of active recall takes 167 hours. Therefore, while conducting active recall, this invention designs a clock error prediction model based on polynomial fitting to determine whether a smart meter has a clock error fault based on historical and current data uploaded to the fault diagnosis device by any smart meter. The reason for using data analysis to determine clock errors is that the data automatically uploaded by the power grid distribution area acquisition system has the following two characteristics: 1) Within a time period, the proportion of smart meters with clock error faults is relatively low. 2) Although it is unknown whether the actual data acquisition time of the smart meter corresponding to each message from the concentrator exceeds the tolerance, the order of sending acquisition commands is determined when the entire acquisition system is established. Therefore, the time sequence distribution of the overall clock error data, i.e., the time series of the difference between the sampling time corresponding to each message and the current system time, should satisfy an approximate statistical law. Therefore, the fault diagnosis device diagnoses the meter clock tolerance fault in the following way:

[0171] S241, obtain the clock data sequence based on the message parsing of each smart meter;

[0172] S242, Subtract each clock data value in the clock data sequence from the system clock value corresponding to the receiving time of the fault diagnosis device to obtain a clock difference sequence;

[0173] S243, Perform multiple curve fitting operations on the clock difference sequence to obtain multiple curve parameters;

[0174] S244, Based on the multiple curve parameters, obtain multiple curve parameter features;

[0175] S245, obtain the center point features based on multiple curve parameter characteristics;

[0176] S246, Obtain the Euclidean distance between the clock difference sequence and the center point feature, and compare the Euclidean distance with a preset distance threshold;

[0177] S247, if the Euclidean distance is greater than the preset distance threshold, then the smart meter is determined to have a clock error; otherwise, the opposite is true.

[0178] The center point feature is obtained using the following formula:

[0179]

[0180]

[0181]

[0182]

[0183]

[0184] In the formula, Represented as the central point feature, and Represented as multiple curve parameter characteristics, and It is represented by multiple curve parameters, where N represents the total number of smart meters, and i∈[1,N].

[0185] In one embodiment of the present invention, the fault diagnosis device diagnoses meter reading failure faults in the following ways: There are three main reasons for smart meter reading failures: communication failure between the meter and the concentrator (i.e., the concentrator's carrier module is not networked); communication failure between the concentrator and the main station (i.e., the concentrator is offline); or damage to internal components of the meter. The first two fault diagnosis methods have already been described. Therefore, an elimination method is used here to determine whether the meter reading failure is caused by damage to internal components of the smart meter. Specifically, firstly, the device analyzes whether the concentrator sends any meter reading data. If not, it determines whether communication failures between the meter and the concentrator, and between the concentrator and the main station, have occurred; otherwise, the meter reading failure is caused by damage to internal components of the meter.

[0186] In one embodiment of the present invention, the fault diagnosis device diagnoses the fault of changing meter parameters in the following way: since the concentrator itself can record changes in terminal parameters, it directly parses the message data uploaded by the concentrator to determine whether the meter parameters have changed. If there is no predetermined change record, it is determined that the corresponding meter has a parameter change fault.

[0187] In one embodiment of the present invention, the fault diagnosis device diagnoses the power outage and power restoration fault of the electricity meter in the following manner: for a smart meter that has sent a power restoration command, the device checks the status information of the smart meter transmitted by the concentrator half an hour after the command is sent. If the smart meter is still in a power outage state, the device determines that the meter has a power outage and power restoration fault; otherwise, the device reverses the process.

[0188] Please see Figure 10As shown, this invention also provides a system for fault diagnosis and analysis of transformer substations based on edge computing, including an electrical device 100, a smart meter 200, a concentrator 300, a fault diagnosis device 400, a cloud computing platform 500, and a user terminal 600. The smart meter 200 is communicatively connected to the electrical device 100 and is used to acquire real-time power consumption information of the electrical device 100. The concentrator 300 is communicatively connected to the smart meter 200 and is used to acquire real-time meter condition information of the smart meter 200. The fault diagnosis device 400 is communicatively connected to the concentrator 300 and is used to acquire real-time concentrator condition information of the concentrator 300 and perform active diagnosis based on the condition information. The cloud computing platform 500 is communicatively connected to the fault diagnosis device 400 and is used to establish an LSTM model to predict faults using the condition information. The LSTM model predicts faults through the following steps: extracting current data, voltage data, and corresponding smart meter temperature and pressure data from the meter condition information to form a four-dimensional feature vector at any time point. Based on the historical data of the smart meter's operating condition information, four-dimensional feature vectors are arbitrarily extracted from n consecutive time points to form a training set. Every n consecutive four-dimensional feature vectors are recorded as one training data point. The first n-1 four-dimensional feature vectors of each training data point are the input of the LSTM model, and the nth four-dimensional feature vector is the output of the LSTM model, thus training the LSTM model. For any sampling time point t of the smart meter's operating condition information, the first n-1 consecutive four-dimensional feature vectors of sampling time point t are obtained, and the LSTM model predicts and outputs the nth four-dimensional feature vector. A preset prediction threshold is set, and the true four-dimensional feature vector at sampling time point t is obtained. If the sum of the absolute values ​​of the differences between the current data, voltage data, temperature data, and pressure data in the true four-dimensional feature vector and the current data, voltage data, temperature data, and pressure data in the corresponding LSTM model's predicted nth four-dimensional feature vector is greater than the prediction threshold, then the smart meter is determined to have a data acquisition failure. User terminal 600 is communicatively connected to cloud computing platform 500 and is used to receive diagnostic results from fault diagnosis device 400 and fault prediction results from LSTM model.

[0189] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0190] The above embodiments are merely examples of implementation methods of the invention. The scope of protection of the present invention is not limited to the above embodiments. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. An edge-computing-based fault diagnosis and analysis method for transformer area acquisition, characterized in that, The application comprises: A fault diagnosis device acquires working condition information in real time, which includes real-time power consumption working condition information of a power consumption device, real-time electric meter working condition information of a smart electric meter, and real-time concentrator working condition information of a concentrator. The fault diagnosis device actively diagnoses the working condition information; A cloud computing platform establishes an LSTM model to predict faults through the working condition information; A user terminal receives diagnosis results of the fault diagnosis device and prediction results of the LSTM model predicting faults; The LSTM model predicting faults comprises the following steps: Extracting current data, voltage data, and temperature data and pressure data corresponding to a smart electric meter from the electric meter working condition information to form a four-dimensional feature vector at any time node; According to historical data of the electric meter working condition information, any n continuous four-dimensional feature vectors are intercepted to form a training set. Each n continuous four-dimensional feature vector is recorded as a training data; The first n-1 four-dimensional feature vectors of each training data are input into the LSTM model, and the n four-dimensional feature vector is output from the LSTM model. The LSTM model is trained; For any sampling time point t of the electric meter working condition information of each smart electric meter, the first n-1 continuous four-dimensional feature vectors at the sampling time point t are obtained. The LSTM model predicts the n four-dimensional feature vector; A preset prediction threshold is set. The real four-dimensional feature vector at the sampling time point t is obtained. If the sum of the absolute values of the differences between the current data, voltage data, temperature data, and pressure data in the real four-dimensional feature vector and the current data, voltage data, temperature data, and pressure data in the n four-dimensional feature vector predicted by the LSTM model is greater than the prediction threshold, it is determined that the smart electric meter has a collection fault; The smart electric meter fault includes an electric meter clock deviation fault. The fault diagnosis device diagnoses the electric meter clock deviation fault in the following manner: Clock data sequences are obtained according to message analysis of each smart electric meter; The clock difference sequence is obtained by subtracting the system clock value corresponding to the receiving time of the fault diagnosis device from each clock data value in the clock data sequence; A plurality of curve parameters are obtained by fitting the clock difference sequence; A plurality of curve parameter features are obtained according to the plurality of curve parameters; A center point feature is obtained according to the plurality of curve parameter features; The Euclidean distance between the clock difference sequence and the center point feature is obtained, and the Euclidean distance is compared with a preset distance threshold; If the Euclidean distance is greater than the preset distance threshold, it is determined that the smart electric meter has an electric meter clock deviation fault, otherwise, it is not; The center point feature is obtained by the following formula: ； ； ； ； ； wherein are represented as center point features, , , and are represented as multiple curve parameter features, , , and are represented as multiple curve parameters, N are represented as the total number of smart meters, .

2. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 1, characterized in that, The fault diagnosis device supports fault diagnosis types including concentrator pipe faults and smart electric meter faults. The concentrator pipe faults include carrier module non-networking faults and offline faults. The fault diagnosis device diagnoses the carrier module non-networking faults of the concentrator in the following manner: The address correspondence between the concentrator and the smart electric meter is preset. For each message sent by each concentrator to the fault diagnosis device, analyze the addresses of all smart meters to form a set of addresses that successfully communicate; Compare the pre-set address correspondence with the set of addresses that successfully communicate; If some address information is missing, it is considered that the concentrator and the smart meter corresponding to the address do not communicate, and it is diagnosed that the carrier module of the concentrator does not network; otherwise, the opposite is true. The fault diagnosis device diagnoses the concentrator as an off-line fault by the following method: If the fault diagnosis device does not receive a message sent by a certain concentrator within 5-15 minutes, it is determined that the concentrator is suspected to be off-line; The fault diagnosis device sends a ping command to the concentrator, and if there is no response, it is determined that the concentrator is off-line; if there is a response, it is determined that the signal output function of the concentrator has a problem; If the fault diagnosis device receives garbled data from a certain concentrator within 50-70 minutes, it is determined that the signal quality of the concentrator has a problem.

3. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 2, characterized in that, When it is determined that the signal quality of the concentrator has a problem, the fault diagnosis device diagnoses the signal strength of the concentrator by the following steps: Obtain the autocorrelation function of the initial signal received by the concentrator; Select the central part of the autocorrelation function, exclude noise outside the window, obtain a noise-reduced received signal that reduces the noise of the initial signal, and the noise-reduced received signal includes an analysis signal; Perform Fourier transform on the noise-reduced received signal to obtain the power spectral density of the noise-reduced received signal after time domain filtering; the power spectral density of the noise-reduced received signal includes the power spectral density of the analysis signal; Segment the power spectral density of the analysis signal, and compare the average value of each segment with a threshold value to determine the signal strength of the concentrator.

4. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 2, characterized in that, The concentrator fault also includes a failure to collect interchange power and a failure to return daily frozen data, and the fault diagnosis device diagnoses the concentrator as a failure to collect interchange power by the following method: Parse the message data sent by the concentrator, and if it does not contain interchange data, it is determined that the concentrator does not collect interchange power; The fault diagnosis device diagnoses the concentrator as a failure to return daily frozen data by the following method: The concentrator uploads daily frozen data and its timestamp to the fault diagnosis device; After the daily frozen data return time, a fixed query time is predetermined, and the fault diagnosis device queries whether it receives daily frozen data sent by each concentrator every day within the fixed query time; If not, it is determined that the corresponding concentrator has a failure to return daily frozen data; if the daily frozen data is found, the timestamp is judged again; Compare whether the timestamp corresponding to the daily frozen data is within the effective range of the time corresponding to the fault diagnosis device on the same day; If the timestamp is not within the effective range, it is determined that the corresponding concentrator has a failure to return daily frozen data; If the timestamp is within the effective range, it is determined that the corresponding concentrator has a failure to return daily frozen data.

5. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 3, characterized in that, The concentrator failure also includes a polling failure, and the fault diagnosis device diagnoses the concentrator as having a polling failure by the following method: The fault diagnosis device sends a polling instruction to the concentrator, and the concentrator sends a polling of the hourly data to the smart meter according to the polling instruction; If a concentrator fails to successfully obtain real-time meter operating condition information of a connected smart meter, it is determined that the concentrator has a polling failure; otherwise, it returns.

6. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 2, characterized in that, The smart meter failure includes a meter power failure, a meter reverse power failure, and a meter positive reactive power negative failure; and the fault diagnosis device diagnoses the meter power failure by the following method: The fault diagnosis device uses a first-order forward difference method to determine whether the smart meter has a failure to move; If the smart meter has a failure to move, it is determined to be a suspected meter power failure; otherwise, the smart meter is normal; When it is determined to be a suspected meter power failure, the fault diagnosis device obtains the sum of the power of other smart meters in the local network area of the suspected smart meter and the difference between the sum and the collection power of all collectors covering the area; If the difference continues to increase over time, it is determined that the smart meter has a meter power failure; otherwise, it is determined that the user corresponding to the smart meter has no power consumption; The fault diagnosis device diagnoses the meter reverse power failure by the following method: The fault diagnosis device subtracts the real-time power consumption condition information of each smart meter uploaded by each concentrator from the real-time power consumption condition information of the previous time; If a negative number appears, it is determined that the current smart meter has a meter reverse power failure; if a positive number appears, it is determined that the current smart meter does not have a meter reverse power failure; The fault diagnosis device diagnoses the meter positive reactive power negative failure by the following method: The fault diagnosis device determines whether the reactive power data contained in the real-time concentrator operating condition information uploaded by the concentrator is less than 0; if it is less than 0, it is determined that the smart meter has a meter positive reactive power negative failure; if it is greater than 0, it is determined that the smart meter does not have a meter positive reactive power negative failure.

7. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 2, characterized in that, The smart meter failure also includes a phase power failure, and the fault diagnosis device diagnoses the meter reverse power failure by the following method: The smart meter is powered on and initialized, and the zero-line drop flag is cleared; The real-time voltages of Ua phase, Ub phase, and Uc phase, the first angle between Ub phase and Uc phase, and the second angle between Uc phase and Ua phase are sampled and obtained; The zero-line drop flag is checked; if the zero-line drop flag is not set, it indicates that the current smart meter does not have a zero-line drop state, and further phase power failure determination is performed; if the zero-line drop flag is set, it indicates that the current smart meter has a phase power failure. When the voltage of the Ua phase, the voltage of the Ub phase and the voltage of the Uc phase are all greater than 5V; when the first included angle is less than 110° or the second included angle is greater than 250°; when the voltage unbalance rate of the three-phase voltage is in the interval of 95% to 100%, the voltage of the Ub phase is greater than 120% of the rated voltage, the voltage of the Uc phase is greater than 120% of the rated voltage, and the voltage of the Ua phase is greater than 0V and less than 10% of the rated voltage; when the three conditions are met at the same time, it is considered that the smart meter has a phase power failure fault; When the voltage of one phase in the three-phase is 0V and the voltage of one phase in the other two phases is greater than 5V and less than the rated voltage, then under this condition, if one of the three conditions is met, it is considered that the smart meter has a phase power failure fault.

8. The edge computing-based fault diagnosis and analysis method for a transformer area acquisition system according to claim 2, characterized in that, The smart meter fault also includes a meter flying away abnormal fault, and the fault diagnosis device diagnoses the meter flying away abnormal fault by the following method: Convert the historical daily frozen data sequence of each smart meter from a time series format to a supervised learning format data; Input the supervised learning format data of the past T days as the training data of the gradient boosting algorithm to predict the daily frozen data of T+1 day; If the actual daily frozen data of T+1 day and the predicted value of the daily frozen data differ by more than a threshold value, it is considered that the smart meter has a meter flying away abnormal fault.

9. An edge computing-based system for collecting and diagnosing faults of a transformer area according to any one of claims 1-8, comprising: an electric device; a smart meter, which is in communication connection with the electric device, and is configured to acquire real-time electric working condition information of the electric device; a concentrator, which is in communication connection with the smart meter, and is configured to acquire real-time meter working condition information of the smart meter; a fault diagnosis device, which is in communication connection with the concentrator, and is configured to acquire real-time concentrator working condition information of the concentrator, and to actively diagnose according to the working condition information; A cloud computing platform is in communication connection with the fault diagnosis device, and is configured to predict a fault through an LSTM model based on the working condition information. The LSTM model predicts a fault through the following steps: extracting current data, voltage data, temperature data and pressure data corresponding to the smart meter from the working condition information of the smart meter to form a four-dimensional feature vector at an arbitrary time node; according to the historical data of the working condition information of the smart meter, four-dimensional feature vectors at any n consecutive time points are intercepted to form a training set, and each n consecutive four-dimensional feature vector is recorded as a training data; the first n-1 four-dimensional feature vectors of each training data are input to the LSTM model, and the n four-dimensional feature vector is output from the LSTM model, and the LSTM model is trained; for each smart meter, a sampling time point t is randomly selected from the working condition information of the smart meter, and the first n-1 consecutive four-dimensional feature vectors at the sampling time point t are obtained, and the LSTM model predicts the n four-dimensional feature vector output; a preset prediction threshold is set, the real four-dimensional feature vector at the sampling time point t is obtained, and if the sum of the absolute values of the differences between the current data, voltage data, temperature data and pressure data in the real four-dimensional feature vector and the corresponding current data, voltage data, temperature data and pressure data in the n four-dimensional feature vector predicted by the LSTM model is greater than the prediction threshold, it is determined that the smart meter has a collection fault; A user terminal is in communication connection with the cloud computing platform, and is configured to receive the diagnosis result of the fault diagnosis device and the prediction result of the LSTM model.

Images