Maintenance and testing method for metering box

Abstract

The present invention relates to the technical field of meter box detection, and in particular to a maintenance and detection method for meter boxes. The present invention obtains the training weight of the data to be analyzed based on the fault possibility of the meter box to which the data to be analyzed belongs, and the data abnormality of the performance parameters contained in the data to be analyzed and the performance parameter time series data adjacent to the time series; the neural network model is trained based on all the performance parameter time series data in the historical database and their training weights, and the performance parameter time series data of the sampling period to be tested is input into the trained neural network model to perform fault detection and maintenance on the meter box to be tested. The present invention assigns different training weights to the performance parameter time series data of different abnormal situations to improve the detection ability of the trained neural network model for the performance parameter time series data caused by the meter box failure, thereby improving the detection ability of the neural network model for the meter box failure and the maintenance accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of meter box detection, and in particular to a maintenance detection method for a meter box. Background Art

[0002] Meter boxes usually refer to smart energy meter boxes developed with advanced technology. They have functions such as data collection, remote monitoring and automatic control to meet the development needs of smart grids. As a key device for energy metering, the accuracy of meter boxes directly affects fair transactions in the electricity market and the electricity costs of users. However, with the long-term use of meter boxes, the electrical components and lines in the meter boxes may have safety hazards such as aging and short circuits. In order to ensure the accuracy of the electricity consumption data of the meter boxes, it is necessary to detect abnormalities in the meter boxes.

[0003] The prior art uses the data collected by the meter box in each historical sampling period in the historical database to train the neural network, and then inputs the data collected by the meter box to be tested in the sampling period to be tested into the trained neural network model to perform abnormal detection on the meter box to be tested. When performing abnormal detection on the meter box, the prior art usually uses the same training weights for all collected data to train the neural network, making the trained neural network model insensitive to abnormal data caused by the meter box's own faults, making it difficult to accurately detect the abnormality of the meter box and repair it. Summary of the invention

[0004] In order to solve the technical problem that it is difficult to accurately detect abnormalities in the meter box in the prior art, the present invention aims to provide a maintenance and detection method for the meter box. The technical solution adopted is as follows:

[0005] A maintenance and detection method for a meter box, the method comprising:

[0006] Obtaining the performance parameter time series data of each meter box in each historical sampling period and the performance parameter time series data of the meter box to be tested in the sampling period to be tested in the historical database; the performance parameter time series data includes the performance parameters at each sampling moment;

[0007] In the historical database, the data abnormality of the performance parameter is obtained according to the deviation of the performance parameter from the preset performance standard value at the same sampling time; the fault possibility of the meter box is obtained according to the change of the data abnormality of all the performance parameters of the meter box with time series; any performance parameter time series data is used as the data to be analyzed, and the training weight of the data to be analyzed is obtained according to the fault possibility of the meter box to which the data to be analyzed belongs, and the data abnormality of the performance parameter contained in the performance parameter time series data adjacent to the data to be analyzed in time series;

[0008] The neural network model is trained based on all the performance parameter time series data and their training weights in the historical database to obtain a trained neural network model; the performance parameter time series data of the sampling period to be tested is input into the trained neural network model to perform fault detection and maintenance on the meter box to be tested.

[0009] Furthermore, the method for obtaining the data abnormality includes:

[0010] At any sampling moment, according to the difference between each performance parameter and the preset performance standard value, a local deviation parameter of each performance parameter is obtained;

[0011] The data abnormality of the performance parameter is obtained according to the prominence of the local deviation parameter of the performance parameter compared with the overall deviation parameter.

[0012] Furthermore, the method for obtaining the data abnormality of the performance parameter according to the prominence of the local deviation parameter of the performance parameter compared with the overall deviation parameter includes:

[0013] Calculate the mean of the local deviation parameters of all performance parameters corresponding to the performance parameters at the sampling time to obtain the overall deviation parameter at the sampling time;

[0014] For any performance parameter at a sampling moment, the absolute value of the difference between the local deviation parameter of the performance parameter and the overall deviation parameter corresponding to the sampling moment is calculated and normalized to obtain the data anomaly degree of the performance parameter.

[0015] Furthermore, the method for obtaining the fault probability includes:

[0016] For any meter box in the historical database, the period anomaly of each historical sampling period is obtained according to the overall distribution of the data anomaly corresponding to all the performance parameters of each historical sampling period; the first abnormal parameter of the meter box is obtained according to the difference between each historical sampling period of the meter box and the period anomaly corresponding to the previous historical sampling period in the time series dimension;

[0017] Obtaining a second abnormal parameter of the meter box according to the overall distribution of the abnormality of the time period in all historical sampling periods of the meter box;

[0018] The first abnormal parameter and the second abnormal parameter of the meter box are forwardly integrated to obtain the fault possibility of the meter box.

[0019] Furthermore, the method for obtaining the first abnormal parameter includes:

[0020] The difference between the abnormality of each historical sampling period and the corresponding period in the previous historical sampling period in the time series dimension is calculated and normalized to obtain the abnormal rising trend of the historical sampling period; the average of the abnormal rising trends of all historical sampling periods of the metering box is used as the first abnormal parameter of the metering box.

[0021] Furthermore, the method of forward fusing the first abnormal parameter and the second abnormal parameter of the meter box to obtain the fault possibility of the meter box includes:

[0022] The mean values ​​of the first abnormal parameter and the second abnormal parameter of the metering box are calculated to obtain the fault possibility of the metering box.

[0023] Furthermore, the method for obtaining the training weights includes:

[0024] The mean of the data anomaly degree corresponding to all performance parameters in the performance parameter time series data is used as the data segment anomaly index of the performance parameter time series data; the difference between the data to be analyzed and the data segment anomaly index of the previous performance parameter time series data in time series is calculated and normalized to obtain the abnormal rise index of the data to be analyzed;

[0025] The data segment abnormality index of the data to be analyzed, the abnormal rise index and the fault possibility of the meter box to which the data to be analyzed belongs are forwardly integrated to obtain the training weight of the data to be analyzed.

[0026] Furthermore, the method of forward fusing the data segment abnormality index, the abnormal rise index and the fault possibility of the meter box to which the data to be analyzed belongs to obtain the training weight of the data to be analyzed includes:

[0027] The data segment abnormality index, the abnormal rise index and the sum of the fault probability of the meter box to which the data to be analyzed belongs are calculated and normalized to obtain the training weight of the data to be analyzed.

[0028] Furthermore, the method for performing fault detection and repair on the meter box to be tested includes:

[0029] The performance parameter time series data of the sampling period to be tested is used as the input of the trained neural network to obtain the fault detection result of the meter box to be tested; and the meter box to be tested is repaired based on the fault detection result of the meter box to be tested.

[0030] Furthermore, the preset performance standard value is 30.

[0031] The present invention has the following beneficial effects:

[0032] Considering that unstable power supply to users in the same power supply area and meter box failure will cause fluctuations in performance parameters, the data anomaly degree is first used to preliminarily reflect the degree of abnormality of the performance parameters caused by the meter box failure. Considering that it is inaccurate to judge abnormalities based on a single performance parameter, the degree of abnormality of the meter box is reflected by the failure probability of the meter box. Considering that the performance parameter time series data corresponding to the meter box when an abnormal failure occurs, it is necessary to focus on it. Combined with the failure probability of the meter box to which the data to be analyzed belongs, and the data anomaly degree of the performance parameters contained in the performance parameter time series data adjacent to the performance parameter time series data in the time series, the training weight of the data to be analyzed is obtained. The larger the training weight, the greater the possibility that the performance parameter time series data corresponding to the meter box itself causes data abnormality. By assigning different training weights to the performance parameter time series data of different abnormal situations, the ability of the trained neural network model to detect the performance parameter time series data caused by the meter box failure is improved, thereby improving the detection ability of the neural network model for the meter box failure and the maintenance accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A flow chart of a maintenance and detection method for a meter box provided by an embodiment of the present invention;

[0035] Figure 2 A flow chart of a method for obtaining data abnormality provided by an embodiment of the present invention;

[0036] Figure 3 A structural diagram of a maintenance and detection system for a meter box provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the maintenance and detection method for a meter box proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0038] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0039] A specific solution of a maintenance and detection method for a meter box provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0040] The present invention provides a maintenance and detection method for a meter box. Figure 1 , which shows a flow chart of a maintenance and detection method for a meter box provided by an embodiment of the present invention, the method comprising the following steps:

[0041] Step S1: Obtain the performance parameter time series data of each meter box in the historical database in each historical sampling period and the performance parameter time series data of the meter box to be tested in the sampling period to be tested; the performance parameter time series data includes the performance parameters at each sampling moment.

[0042] The purpose of the present invention is to train a neural network model through data in a historical database, so as to use the trained neural network model to detect the data of the sampling period to be tested, evaluate the fault condition of the metering box to be tested and perform maintenance. First, data needs to be obtained.

[0043] The performance parameter time series data of each meter box in each historical sampling period in the historical database and the performance parameter time series data of the meter box to be tested in the sampling period to be tested are obtained from the meter box system; the performance parameter time series data includes the performance parameters at each sampling moment. Specifically, in an embodiment of the present invention, the current is used as the performance parameter, and the performance parameters of each meter box are synchronously sampled according to the preset sampling frequency to obtain the performance parameters of each meter box at each sampling moment; the preset time interval is used as a sampling period, and then the performance parameter time series data of the sampling period is obtained. The performance parameter time series data of the sampling period of each meter box stored in the historical database is used as the performance parameter time series data of each meter box in each historical sampling period; the performance parameter time series data of the sampling period of the meter box that needs to be tested the latest is used as the performance parameter time series data of the sampling period to be tested for the meter box to be tested. It should be noted that all the meter boxes targeted in the present invention are meter boxes in the same power supply area, and the same meter box refers to the meter box that detects the electricity consumption of the same user. Current is the basic performance parameter of the meter box, and monitoring it can timely discover the abnormal performance of the meter box. In addition, the performance parameter can also be voltage.

[0044] The research object of the present invention is in the same power supply area, that is, each meter box in the historical database is in the same power supply area, so as to facilitate the subsequent analysis of the impact of power supply anomalies on performance parameters in the same power supply area.

[0045] In one embodiment of the present invention, the preset sampling frequency is 10 minutes / time, and the preset time interval is from the beginning to the end of a day, which can be set by the implementer according to the implementation scenario. It should be noted that, in order to facilitate calculations, all indicator data involved in the calculations in the embodiments of the present invention are pre-processed to eliminate the dimension effect. The specific means of removing the dimension effect are technical means well known to those skilled in the art and are not limited here.

[0046] Step S2: In the historical database, the data anomaly of the performance parameter is obtained according to the deviation of the performance parameter from the preset performance standard value at the same sampling time; the fault possibility of the meter box is obtained according to the change of the data anomaly of all performance parameters of the meter box with time series; any performance parameter time series data is used as the data to be analyzed, and the training weight of the data to be analyzed is obtained according to the fault possibility of the meter box to which the data to be analyzed belongs, and the data anomaly of the performance parameters contained in the data to be analyzed and its adjacent performance parameter time series data in time series.

[0047] Considering that the unstable power supply to users and the failure of the meter box in the same power supply area will cause the performance parameters to fluctuate, the data anomaly degree is first used to preliminarily reflect the degree of abnormality of the performance parameters caused by the failure of the meter box. Considering that it is inaccurate to judge the abnormality based on a single performance parameter, and considering that the failure of the meter box will change more obviously over time, that is, in the early stage of the failure, a small number of components are in an abnormal working state. As the meter box works, the abnormal components affect the normal working components, causing the failure of the meter to change with time. The abnormal degree of the meter box operation becomes more obvious, and the probability of failure of the meter box is reflected by the failure possibility of the meter box. In order to improve the sensitivity of the neural network model to the performance parameter time series data caused by the failure of the meter box itself, the training weight of the data to be analyzed is obtained by combining the failure possibility of the meter box to which the data to be analyzed belongs, and the data anomaly degree of the performance parameters contained in the performance parameter time series data adjacent to the performance parameter time series data in the time series. The larger the training weight, the greater the degree of abnormality caused by the performance parameter time series data corresponding to the meter box itself. By constructing the training weight, the subsequent neural network model can pay more attention to the performance parameter time series data that is highly related to the failure during the training process, thereby improving the sensitivity of the trained neural network model to abnormal data.

[0048] Considering that the performance parameters are in line with the theoretical values ​​under normal circumstances, there are abnormal situations such as unstable power supply and meter box failures in the same power supply area, which cause the performance parameters to deviate from the theoretical values. The phenomenon that unstable power supply causes the performance parameters to deviate from the theoretical values ​​is caused by the change in the power supply demand of users at different times and time periods, which causes the performance parameters of the corresponding power supply system in the power supply area to generally deviate from the theoretical values. In order to analyze the impact of the performance parameters caused by meter failures deviating from the theoretical values, it is necessary to remove the impact of unstable power supply. Considering that the performance parameters caused by unstable power supply deviate from the theoretical values, it not only affects the meter box of a certain user, but also affects the meter boxes of users in the entire area, that is, at the moment of unstable power supply, the performance parameters of all meter boxes deviate from the theoretical values ​​in a consistent manner, but the performance parameters caused by meter box failures deviate from the theoretical values ​​in a specific manner. The abnormality of the performance parameters caused by the meter box failure is preliminarily reflected by constructing the data anomaly degree. Please refer to Figure 2 , which shows a flow chart of a method for obtaining data abnormality in one embodiment of the present invention. Preferably, in one embodiment of the present invention, the method for obtaining data abnormality includes:

[0049] Step S201: at any sampling moment, according to the difference between each performance parameter and a preset performance standard value, obtain a local deviation parameter of each performance parameter.

[0050] Considering that unstable power supply and meter box failure in the same power supply area may cause the performance parameters to deviate from the preset performance standard values, the local deviation parameter of each performance parameter is obtained. The larger the local deviation parameter, the greater the degree to which the performance parameter deviates from the preset performance standard value.

[0051] In one embodiment of the present invention, at any sampling moment, the absolute value of the difference between each performance parameter and the preset performance standard value is calculated to obtain the local deviation parameter of each performance parameter. In one embodiment of the present invention, the preset performance standard value is 30, and the implementer can set it according to the implementation scenario.

[0052] Step S202: Obtaining data abnormality of the performance parameter according to the prominence of the local deviation parameter of the performance parameter compared with the overall deviation parameter.

[0053] In order to distinguish the general deviation caused by unstable power supply and the specific deviation caused by meter box failure, the data anomaly degree of performance parameters is constructed. The data anomaly degree measures the prominence of a certain performance parameter relative to the overall deviation degree. The larger the data anomaly degree, the greater the specific deviation degree caused by the meter box failure, and the more likely the performance parameter is abnormal data caused by the meter box failure.

[0054] In one embodiment of the present invention, for any sampling moment, the mean of the local deviation parameters of the performance parameters corresponding to all performance parameters at the sampling moment is calculated to obtain the overall deviation parameter at the sampling moment; for any performance parameter at the sampling moment, the absolute value of the difference between the local deviation parameter of the performance parameter and the overall deviation parameter corresponding to the sampling moment is calculated and normalized to obtain the data abnormality of the performance parameter. It should be noted that the normalization method used is: normalization is performed using the norm normalization function to limit the value range to between 0 and 1. Normalization is a technical means well known to those skilled in the art, and the selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0055] For the above steps, the overall deviation parameter reflects the general impact of unstable power supply on the performance parameters of the power supply system in the entire region, and the data anomaly reflects the data anomaly of the performance parameter. The larger the data anomaly, the more prominent the deviation of the performance parameter is relative to the overall deviation, which means that the performance parameter is more affected by the specific factors of the meter box failure, and the more likely the performance parameter is caused by the meter box failure.

[0056] Considering that it is inaccurate to judge abnormalities based on only a single performance parameter, for example, considering that the user's power load will also cause the performance parameters of the meter box to deviate from the theoretical value, it is inaccurate to judge abnormalities based on only a single performance parameter. Considering that the failure of the meter box will become more obvious over time, that is, in the early stage of the failure, a small number of components are in abnormal working state. As the meter box works, the abnormal components affect the normal working components, causing the failure of the meter to change with time. The degree of abnormality of the meter box becomes more obvious. The probability of failure of the meter box is reflected by constructing the failure probability of the meter box. Preferably, in one embodiment of the present invention, the method for obtaining the failure probability includes:

[0057] For any meter box in the historical database, the period anomaly of each historical sampling period is obtained according to the overall distribution of the data anomaly corresponding to all performance parameters of each historical sampling period; the first abnormal parameter of the meter box is obtained according to the difference between each historical sampling period of the meter box and the corresponding period anomaly of the previous historical sampling period in the time series dimension;

[0058] Obtaining a second abnormal parameter of the meter box according to the overall distribution of the period abnormality of all historical sampling periods of the meter box;

[0059] The first abnormal parameter and the second abnormal parameter of the meter box are forwardly integrated to obtain the fault possibility of the meter box.

[0060] For the above steps, specifically, for any meter box in the historical database, calculate the mean of the data anomaly corresponding to all performance parameter values ​​in each historical sampling period to obtain the period anomaly of each historical sampling period; calculate the difference between the period anomaly corresponding to each historical sampling period and the previous historical sampling period in the time series dimension and perform normalization processing to obtain the period abnormal rising trend of the historical sampling period; take the mean of the period abnormal rising trend of all historical sampling periods of the meter box as the first abnormal parameter of the meter box; calculate the mean of the period anomaly of all historical sampling periods of the meter box to obtain the second abnormal parameter of the meter box; calculate the mean of the first abnormal parameter and the second abnormal parameter of the meter box to obtain the possibility of failure of the meter box.

[0061] For the above steps, the period abnormality reflects the overall abnormality degree of all performance parameters in the historical sampling period. In order to capture the trend of abnormal changes, the period abnormal rising trend of the historical sampling period is constructed. The larger the period abnormal rising trend, the more obvious the abnormal rising trend of the historical sampling period is compared with the previous historical sampling period. The first abnormal parameter reflects the overall situation of the period abnormal rising trend of the meter box in the historical sampling period. The second abnormal parameter reflects the overall situation of the period abnormality of the meter box in all historical sampling periods. The first abnormal parameter and the second abnormal parameter are forward fused to obtain the failure possibility of the meter box. The failure possibility analyzes the abnormality and abnormal rising trend of the performance parameters of the meter box in the historical sampling period, so as to more accurately evaluate the probability of failure of the meter box. This method is more comprehensive and accurate than abnormal judgment based on a single performance parameter, and can more effectively reflect the probability of failure of the meter box.

[0062] In order to improve the sensitivity of the neural network model to the performance parameter time series data caused by the meter box's own fault, considering that the corresponding performance parameter time series data is more important when the meter box has an abnormal fault, the training weight of the data to be analyzed is obtained by combining the fault possibility of the meter box to which the data to be analyzed belongs, and the data abnormality of the performance parameters contained in the performance parameter time series data adjacent to the performance parameter time series data in the time series. The larger the training weight, the greater the abnormality caused by the meter box's own fault corresponding to the performance parameter time series data. Preferably, in one embodiment of the present invention, the method for obtaining the training weight includes:

[0063] The mean of the data anomaly degrees corresponding to all performance parameters in the performance parameter time series data is used as the data segment anomaly index of the performance parameter time series data; the difference between the data to be analyzed and the data segment anomaly index of the previous performance parameter time series data in the time series is calculated and normalized to obtain the abnormal rise index of the data to be analyzed;

[0064] The data segment abnormality index, abnormal rise index and the fault probability of the meter box to which the data to be analyzed belong are forwardly integrated to obtain the training weight of the data to be analyzed. Among them, the specific method for obtaining the training weight of the data to be analyzed is: calculate the sum of the data segment abnormality index, abnormal rise index and the fault probability of the meter box to which the data to be analyzed belongs and perform normalization processing to obtain the training weight of the data to be analyzed. It should be noted that the normalization method is: normalize using the norm normalization function to limit the value range to between 0 and 1. Among them, normalization is a technical means well known to those skilled in the art, and the selection of the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here. It should be noted that the previous performance parameter time series data of the data to be analyzed in the time series is, among all the performance parameter time series data of the meter box corresponding to the data to be analyzed, the performance parameter time series data corresponding to the previous historical sampling period of the historical sampling period corresponding to the data to be analyzed.

[0065] For the above steps, the data segment abnormality index is constructed to reflect the degree of deviation of the performance parameter time series data from the normal state. The abnormal rising index of the data to be analyzed is used to reflect the aggravation of the abnormal situation corresponding to the data to be analyzed. Considering that the larger the abnormal rising index is, the more obvious the abnormal rising trend of the data to be analyzed is compared with the previous historical sampling period, and the data to be analyzed is more likely to be abnormal data caused by a fault; the larger the data segment abnormality index is, the more the deviation of the data to be analyzed from the normal state is, and the more likely the data to be analyzed is abnormal data caused by a fault; the greater the probability of the meter box failure is, the greater the probability of the meter box failure is; the training weight of the data to be analyzed is constructed by comprehensively considering the data segment abnormality index, the abnormal rising index and the meter box failure probability. The larger the training weight is, the greater the degree of abnormality caused by the performance parameter time series data corresponding to the meter box itself. By constructing the training weight for the subsequent neural network model to pay more attention to the data closely related to the meter box itself failure during the training process, the model can improve the sensitivity and recognition ability of the model to the data corresponding to the meter failure.

[0066] Step S3: Train the neural network model based on all performance parameter time series data in the historical database and their training weights to obtain a trained neural network model; input the performance parameter time series data of the sampling period to be tested into the trained neural network model to perform fault detection and maintenance on the metering box to be tested.

[0067] Considering that the training weight is the weight of the training of the performance parameter time series data during the model training process, the training weight is used to measure the adequacy of the model training. The larger the training weight of the performance parameter time series data, the more sufficient the training of the neural network model for the performance parameter time series data, and the higher the recognition accuracy of the trained neural network model for this type of performance parameter time series data, and the higher the recognition efficiency; the smaller the training weight of the performance parameter time series data, the less sufficient the training of the neural network model for the performance parameter time series data, and the lower the recognition accuracy of the trained neural network model for this type of performance parameter time series data, and the lower the training efficiency. The present invention assigns different training weights to the performance parameter time series data of different abnormal situations to improve the detection ability of the trained neural network model for the performance parameter time series data caused by the meter box's own fault, thereby improving the detection ability of the neural network model for the meter box fault and the maintenance accuracy.

[0068] In order to obtain a neural network model that is sensitive to the performance parameter time series data caused by the meter box's own fault, the neural network model is trained based on all the performance parameter time series data in the historical database and its training weights to obtain a trained neural network model.

[0069] Specifically, after determining the training weights of all performance parameter time series data, the performance parameter time series data are input into the neural network model to train the neural network model. During the training process, the neural network model will adjust its internal parameters according to the provided training weights and loss function until the loss function converges. The neural network model corresponding to the convergence of the loss function is determined as the trained neural network model.

[0070] The neural network model can select a recurrent neural network model for sequence data processing. The specific neural network model training process is a technical means well known to those skilled in the art and will not be described in detail here. In other embodiments of the present invention, since the deep neural network can learn the complex relationships in time series data and accurately predict and analyze it, the deep neural network model can also be sampled, which is not limited here.

[0071] In order to accurately detect the abnormality of the meter box, the performance parameter time series data of the sampling period to be tested is input into the trained neural network model to perform fault detection on the meter box to be tested.

[0072] In one embodiment of the present invention, the performance parameter time series data of the sampling period to be tested can be input into the trained neural network model through the trained neural network model, and the fault detection result of the meter box to be tested in the sampling period to be tested can be output. For example, the fault detection result is: the meter is qualified and the meter is unqualified. Specifically, when the fault detection result of the meter box to be tested in the sampling period to be tested is that the meter is qualified, there is no need to perform fault maintenance. When the fault detection result of the meter box to be tested in the sampling period to be tested is that the meter is unqualified, a special inspection and maintenance personnel is dispatched to perform on-site inspection and maintenance.

[0073] In summary, an embodiment of the present invention provides a maintenance and detection method for a meter box. First. In the embodiment of the present invention, the training weight of the data to be analyzed is obtained by the possibility of failure of the meter box to which the data to be analyzed belongs, and the data abnormality of the performance parameters contained in the data to be analyzed and the performance parameter time series data adjacent in time series; the neural network model is trained based on all the performance parameter time series data in the historical database and their training weights to obtain a trained neural network model; the performance parameter time series data of the sampling period to be tested is input into the trained neural network model to perform fault detection and maintenance on the meter box to be tested. The present invention assigns different training weights to the performance parameter time series data of different abnormal situations to improve the detection ability of the trained neural network model for the performance parameter time series data caused by the meter box failure, thereby improving the detection ability of the neural network model for the meter box failure and the maintenance accuracy.

[0074] The present invention also proposes a maintenance and detection system for a meter box, see Figure 3 , which shows a structural diagram of a maintenance and detection system for a meter box provided by an embodiment of the present invention. The system includes: a data acquisition module 101, a training weight analysis module 102 and a fault detection module 103.

[0075] The data acquisition module 101 is used to obtain the performance parameter time series data of each meter box in the historical database in each historical sampling period and the performance parameter time series data of the meter box to be tested in the sampling period to be tested; the performance parameter time series data includes the performance parameters at each sampling moment.

[0076] The training weight analysis module 102 is used to obtain the data anomaly of the performance parameters in the historical database according to the deviation of the performance parameters from the preset performance standard values ​​at the same sampling time; obtain the failure possibility of the meter box according to the change of the data anomaly of all performance parameters of the meter box with time series; take any performance parameter time series data as the data to be analyzed, and obtain the training weight of the data to be analyzed according to the failure possibility of the meter box to which the data to be analyzed belongs, and the data anomaly of the performance parameters contained in the data to be analyzed and its adjacent performance parameter time series data in time series.

[0077] The fault detection module 103 is used to train the neural network model based on all performance parameter time series data in the historical database and their training weights to obtain a trained neural network model; input the performance parameter time series data of the sampling period to be tested into the trained neural network model to perform fault detection and maintenance on the metering box to be tested.

[0078] It should be noted that: the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer equipment is divided into different functional modules to complete all or part of the functions described above. In addition, the maintenance and detection system for meter boxes provided in the above embodiment and the maintenance and detection method embodiment for meter boxes belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0079] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0080] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A maintenance and detection method for a metering box, characterized in that: Methods include: Obtain the performance parameter time series data of each meter box in each historical sampling period in the historical database and the performance parameter time series data of the meter box to be tested in the sampling period to be tested; The performance parameter time series data includes the performance parameters at each sampling moment; In the historical database, the data abnormality of the performance parameter is obtained according to the deviation of the performance parameter from the preset performance standard value at the same sampling time; the fault possibility of the meter box is obtained according to the change of the data abnormality of all the performance parameters of the meter box with time series; any performance parameter time series data is used as the data to be analyzed, and the training weight of the data to be analyzed is obtained according to the fault possibility of the meter box to which the data to be analyzed belongs, and the data abnormality of the performance parameter contained in the performance parameter time series data adjacent to the data to be analyzed in time series; Based on all the performance parameter time series data in the historical database and the training weights thereof, the neural network model is trained to obtain a trained neural network model; the performance parameter time series data of the sampling period to be tested is input into the trained neural network model, and fault detection and maintenance are performed on the meter box to be tested; The method for obtaining the training weights includes: The mean value of the data anomaly degree corresponding to all performance parameters in the performance parameter time series data is used as the data segment anomaly indicator of the performance parameter time series data; Calculate the difference between the data segment abnormality index of the data to be analyzed and the previous performance parameter time series data in time series and normalize them to obtain the abnormal rise index of the data to be analyzed; The data segment abnormality index of the data to be analyzed, the abnormal rise index and the fault possibility of the meter box to which the data to be analyzed belongs are forwardly integrated to obtain the training weight of the data to be analyzed.

2. The maintenance and detection method for a metering box according to claim 1, characterized in that: The method for obtaining the data abnormality degree includes: At any sampling moment, according to the difference between each performance parameter and the preset performance standard value, a local deviation parameter of each performance parameter is obtained; The data abnormality of the performance parameter is obtained according to the prominence of the local deviation parameter of the performance parameter compared with the overall deviation parameter.

3. A maintenance and detection method for a metering box according to claim 2, characterized in that: The method for obtaining the data abnormality of the performance parameter according to the prominence of the local deviation parameter of the performance parameter compared with the overall deviation parameter includes: Calculate the mean of the local deviation parameters of all performance parameters corresponding to the performance parameters at the sampling time to obtain the overall deviation parameter at the sampling time; For any performance parameter at a sampling moment, the absolute value of the difference between the local deviation parameter of the performance parameter and the overall deviation parameter corresponding to the sampling moment is calculated and normalized to obtain the data anomaly degree of the performance parameter.

4. A maintenance and detection method for a metering box according to claim 1, characterized in that: The method for obtaining the fault probability comprises: For any meter box in the historical database, the period anomaly of each historical sampling period is obtained according to the overall distribution of the data anomaly corresponding to all the performance parameters of each historical sampling period; the first abnormal parameter of the meter box is obtained according to the difference between each historical sampling period of the meter box and the period anomaly corresponding to the previous historical sampling period in the time series dimension; Obtaining a second abnormal parameter of the meter box according to the overall distribution of the abnormality of the time period in all historical sampling periods of the meter box; The first abnormal parameter and the second abnormal parameter of the meter box are forwardly integrated to obtain the fault possibility of the meter box.

5. A maintenance and detection method for a metering box according to claim 4, characterized in that: The method for obtaining the first abnormal parameter includes: The difference between the abnormality of each historical sampling period and the corresponding period in the previous historical sampling period in the time series dimension is calculated and normalized to obtain the abnormal rising trend of the historical sampling period; the average of the abnormal rising trends of all historical sampling periods of the metering box is used as the first abnormal parameter of the metering box.

6. A maintenance and detection method for a metering box according to claim 4, characterized in that: The method of forward fusing the first abnormal parameter and the second abnormal parameter of the meter box to obtain the fault possibility of the meter box includes: The mean values ​​of the first abnormal parameter and the second abnormal parameter of the metering box are calculated to obtain the fault possibility of the metering box.

7. A maintenance and detection method for a metering box according to claim 1, characterized in that: The method of forward fusing the data segment abnormality index, the abnormal rise index and the fault possibility of the meter box to which the data to be analyzed belongs to obtain the training weight of the data to be analyzed includes: The data segment abnormality index, the abnormal rise index and the sum of the fault probability of the meter box to which the data to be analyzed belongs are calculated and normalized to obtain the training weight of the data to be analyzed.

8. A maintenance and detection method for a meter box according to claim 1, characterized in that: The method for performing fault detection and repair on the meter box to be tested comprises: The performance parameter time series data of the sampling period to be tested is used as the input of the trained neural network to obtain the fault detection result of the meter box to be tested; and the meter box to be tested is repaired based on the fault detection result of the meter box to be tested.

9. A maintenance and detection method for a metering box according to claim 1, characterized in that: The preset performance standard value is 30.

Images