Method and system for identifying burnt electric energy meter, and medium
By obtaining the characteristic data of the power meter and inputting it into the machine learning model, and combining the sliding window model for quantity balance processing, the problem of low accuracy of the power meter burn loss identification in the existing technology is solved, and the accurate identification of the risk of burn loss of the power meter is achieved. , reducing security risks.
Patent Information
- Application Number
- CN202410308166.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-18
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing online monitoring methods for burn damage of electricity meter are relatively low in accuracy, making it difficult to effectively identify the risk of burn damage of electricity meter, and poses safety risks.
By obtaining the characteristic data of the electricity meter, including electrical characteristic data, product characteristic data and environmental characteristic data, and inputting it into the pre-trained machine learning model, we will establish a mapping relationship of identification results of whether the electricity meter has a risk of burning loss, and use a sliding window The model is subjected to quantitative balance processing to improve identification accuracy.
The accuracy of burn-out identification of electricity meter is improved, and the operation data analysis and calculation of the electricity meter can be accurately identified and the safety risks are reduced.
Smart Images

Figure CN119001586A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an online monitoring technology for electric energy meters, and in particular to an identification method, system and medium for a burned-out electric energy meter. Background Art
[0002] There are many factors that cause the burning of electric energy meters. In addition to fire, there are also factors such as untightened terminals, direct aluminum wires into households, too thin wire diameters, common neutral wires, etc. The burning of electric energy meters is a process phenomenon, which usually goes through several processes such as melting of terminals, burning of single terminals, burning of multiple terminals, and complete burning of electric energy meters. The whole process can take several days to several months, and only when the electric energy meter is completely burned out will it affect electricity consumption. The power company will only start replacing the meter after the user's power outage. However, there are great safety risks in the entire burning process, which is very likely to cause the burning of adjacent meters, meter boxes, and even houses. Therefore, it is imperative to realize online monitoring of electric energy meter burning.
[0003] In order to identify burned-out electric energy meters, a Chinese patent application document with publication number CN 114089257 A discloses an online monitoring method, system and medium for burned-out electric energy meters. The online monitoring method for burned-out electric energy meters of the present invention comprises: 1) obtaining characteristic data of the electric energy meter, including electrical characteristic data, product characteristic data and environmental characteristic data; 2) inputting the characteristic data into a pre-trained machine learning model to obtain a probability assessment result of the electric energy meter burning in the future output by the machine learning model. However, this method still has the problem of low accuracy. How to further improve the recognition accuracy of burned-out electric energy meters has become a key technical problem that needs to be solved urgently. Summary of the invention
[0004] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method, system and medium for identifying a burned-out electric energy meter are provided. The present invention aims to improve the recognition accuracy of a burned-out electric energy meter, and can be used to accurately identify electric energy meters with a risk of burning out by analyzing and calculating the operating data of the electric energy meters.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for identifying a burned-out electric energy meter, comprising: Step S101, obtaining characteristic data of the electric energy meter, the characteristic data including electrical characteristic data, product characteristic data and environmental characteristic data, the electrical characteristic data including part or all of the daily power consumption Q, daily maximum voltage U, daily maximum current I and meter running word terminal code A, zero-live current ratio fluctuation rate, daily power fluctuation rate, area line loss fluctuation rate and household meter voltage abnormality rate of the electric energy meter; the product characteristic data including part or all of the manufacturer B, the city or state C, the operation time D, the installation year E, the number of burned meters F in the area in the past two years and the area line loss G; the environmental characteristic data including part or all of the ambient temperature T and ambient humidity H of the electric energy meter; Step S102, inputting the characteristic data of the electric energy meter into a pre-trained machine learning model to obtain an identification result of whether the electric energy meter is at risk of being burned out, wherein the machine learning model is pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned out.
[0006] Optionally, before step S102, the step further includes training a machine learning model: Step S201, screening out a set of burned out electric energy meters and a set of non-burned electric energy meters from all replaced electric energy meters in a certain area; Step S202, extracting feature data from historical data of burned electric energy meters in the burned electric energy meter set to obtain burned electric energy meter data set A, and extracting feature data from historical data of burned electric energy meters in the non-burned electric energy meter set to obtain non-burned electric energy meter data set B; Step S203, using a sliding window model to perform quantity balance processing on the burned-out electric energy meter data set A; Step S204, training a machine learning model for identifying burned electricity meters with the burned electricity meter dataset A and the non-burned electricity meter dataset B after quantity balance processing, so as to establish a mapping relationship between the characteristic data of the electricity meter and the identification result of whether the electricity meter is at risk of being burned.
[0007] Optionally, when screening out a burnt electric energy meter set and a non-burnt electric energy meter set from all replaced electric energy meters in a certain area in step S201, the burnt electric energy meter set is screened out from all replaced electric energy meters in a certain area, including retaining electric energy meters with "burning phenomenon" as "neutral wire terminal burnt", "live wire terminal burnt" or "wiring terminal burnt", and excluding electric energy meters with "burning phenomenon" as "all burnt" or "burned meter does not include terminal burnt", thereby obtaining a burnt electric energy meter set consisting of electric energy meters with "burning phenomenon" as "neutral wire terminal burnt", "live wire terminal burnt" or "wiring terminal burnt".
[0008] Optionally, the sliding window model in step S202 includes characteristic data of an observation period and a performance period, wherein the observation period is from date t-2n to date tn, and the performance period is from date tn to date t, wherein date t is the table change date, and n is the sliding time window.
[0009] Optionally, step S203 includes: Step S301, assigning the burned out electric energy meter data set A to the burned out electric energy meter basic data set; assigning the non-burned electric energy meter data set B to the non-burned electric energy meter basic data set, and initializing the value of the loop variable i to 0; Step S302: take all a samples in the basic data set of burned-out electric energy meters as negative samples, randomly select a samples from the basic data set of non-burned electric energy meters as positive samples, and train the machine learning model for identifying burned-out electric energy meters with the 2a balanced positive and negative sample data consisting of negative samples and positive samples to obtain the specified performance evaluation index f of the machine learning model. 1i ; Step S303, increasing the window size of the sliding window model by a specified step length, extracting feature data from the historical data of the burned electric energy meters in the burned electric energy meter set and using the new sliding window model to obtain the burned electric energy meter data set A1, merging all a samples in the burned electric energy meter data set A1 and all a samples in the burned electric energy meter basic data set as a new burned electric energy meter basic data set containing 2a negative samples, randomly selecting 2a samples from the non-burned electric energy meter basic data set as positive samples to construct a new non-burned electric energy meter basic data set, training a machine learning model for identifying burned electric energy meters with 4a balanced positive and negative sample data consisting of negative samples of the new burned electric energy meter basic data set and positive samples of the new non-burned electric energy meter basic data set, and obtaining a specified performance evaluation index f of the machine learning model. 2i ; Step S304, determining the specified performance evaluation index f of the machine learning model 2i Is it less than a preset threshold? If yes, jump to step S305; otherwise, jump to step S302; Step S305: taking the final burned-out electric energy meter basic data set as the burned-out electric energy meter data set A after quantity balance processing.
[0010] Optionally, the machine learning model is a machine learning model based on XGBoost.
[0011] Optionally, before step S101, the step further includes grouping all low-voltage areas in the designated area to set calculation priorities and calculation frequencies, and performing steps S101 and S102 with the electric energy meters in the grouped low-voltage areas according to the priorities and calculation frequencies of the corresponding groups, wherein grouping all low-voltage areas in the designated area to calculate priorities and calculation frequencies includes: Step S401, the historical meter burnout data of all low-voltage areas in the specified area in the past m days are counted to form a dynamic dictionary C, and the key-value pair structure in the dynamic dictionary C is "area number: number of electric energy meter burnouts"; Step S402, based on the set threshold value n and the number of groups p, all low-voltage areas in the dynamic dictionary C are sorted and grouped according to the number of burnouts, and the grouping method uses equal-frequency grouping or equal-distance grouping; Step S403, for the grouped low-voltage substations, calculation priority and calculation frequency are set according to the sorting, so that for the electric energy meters in each low-voltage substation, steps S101 and S102 are executed according to the corresponding set calculation priority and calculation frequency to obtain the identification result of whether the corresponding electric energy meter has the risk of burning.
[0012] In addition, the present invention also provides a burnt-out electric energy meter identification system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the burnt-out electric energy meter identification method.
[0013] In addition, the present invention also provides a computer-readable storage medium, in which a computer program / instruction is stored. The computer program / instruction is programmed or configured to execute the method for identifying a burned-out electric energy meter through a processor.
[0014] In addition, the present invention also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the method for identifying a burned-out electric energy meter through a processor.
[0015] Compared with the prior art, the present invention mainly has the following advantages: the present invention includes obtaining characteristic data of an electric energy meter, including the fluctuation rate of the zero-to-live current ratio, the daily electricity consumption fluctuation rate, the line loss fluctuation rate in the substation area and the abnormal voltage rate of the household meter; the characteristic data of the electric energy meter is input into a pre-trained machine learning model to obtain an identification result of whether the electric energy meter is at risk of being burned out, and the machine learning model is pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned out. The present invention can improve the identification accuracy of a burned out electric energy meter, and can be used to accurately identify an electric energy meter at risk of being burned out by analyzing and calculating the operating data of the electric energy meter. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.
[0017] Figure 2 It is a sliding window model established in the embodiment of the present invention.
[0018] Figure 3These are the top 5 feature weights of the first generation model used for comparison in the embodiment of the present invention.
[0019] Figure 4 are the feature weights of the top 5 in the second generation model in the embodiment of the present invention. DETAILED DESCRIPTION
[0020] like Figure 1 As shown, this embodiment provides a method for identifying a burned-out electric energy meter, comprising: Step S101, obtaining characteristic data of the electric energy meter, the characteristic data including electrical characteristic data, product characteristic data and environmental characteristic data, the electrical characteristic data including part or all of the daily power consumption Q, daily maximum voltage U, daily maximum current I and meter running word terminal code A, zero-live current ratio fluctuation rate, daily power fluctuation rate, area line loss fluctuation rate and household meter voltage abnormality rate of the electric energy meter; the product characteristic data including part or all of the manufacturer B, the city or state C, the operation time D, the installation year E, the number of burned meters F in the area in the past two years and the area line loss G; the environmental characteristic data including part or all of the ambient temperature T and ambient humidity H of the electric energy meter; Step S102, inputting the characteristic data of the electric energy meter into a pre-trained machine learning model to obtain an identification result of whether the electric energy meter is at risk of being burned out, wherein the machine learning model is pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned out.
[0021] According to the actual situation on site and other data analysis, the factors that affect the burning of electric energy meters include weather, the quality of the electric energy meter itself, user load, and the installation quality of the electric energy meter. However, many factors have certain limitations, such as meteorological conditions. Because all the electric energy meters included in the data set are basically from the same province or city and county, under normal conditions, their meteorological conditions will not differ greatly, and it is problematic to use them as core features for analysis. In addition, the installation quality of the electric energy meter is considered to be the main cause of the burning of the electric energy meter on site, but its basic data is difficult to obtain, and it is impossible to evaluate the installation quality of each electric energy meter and include it in the data set. In order to optimize the model effect, the characteristic data selected in this embodiment include electrical characteristic data, product characteristic data and environmental characteristic data. The electrical characteristic data include the daily power consumption Q, daily maximum voltage U, daily maximum current I and meter running code A of the electric energy meter, the fluctuation rate of the zero-live current ratio, the daily power fluctuation rate, the line loss fluctuation rate of the substation and the abnormal voltage rate of the household meter; the product characteristic data include the manufacturer B of the electric energy meter, the city or state C to which it belongs, the operating time D, the installation year E, the number of burned meters F in the substation in the past two years and the line loss G in the substation; the environmental characteristic data include the ambient temperature T and ambient humidity H of the electric energy meter; among them, the daily power consumption Q, the daily maximum voltage U, the daily maximum current I and the meter running code A, the manufacturer B of the electric energy meter, the city or state C to which it belongs, the operating time D, the installation year E, the number of burned meters F in the past two years in the substation and the line loss G in the substation, the ambient temperature T and ambient humidity H of the electric energy meter are all known data, see the record in the Chinese patent application document with publication number CN114089257 A for details. On this basis, this embodiment adds the fluctuation rate of zero-live current ratio, daily power fluctuation rate, area line loss fluctuation rate and household meter voltage abnormality rate as characteristic data. The reasons for selection are as follows: Fluctuation rate of zero-live current ratio: In theory, the zero-live current of a single-phase energy meter will remain consistent at any time, but the zero-live current of an energy meter in the process of burning will be different. The reason is that the live current of a single-phase energy meter is generally sampled using a manganese copper resistor, and the zero-live current is generally sampled using an induction coil. Both the manganese copper resistor and the induction coil are close to their respective terminals. Once a terminal begins to melt or slightly burn, it is easy to affect the current sampling circuit inside it, causing the sampled zero-live current to differ.
[0022] The calculation method of zero-live current ratio fluctuation rate is as follows: Step 1: First, calculate the daily zero-live current ratio. Assuming that the acquisition system can collect zero-live currents at q points every day, it can be expressed as: ; Step 2: Set the threshold Q and remove all points with missing values or current values less than the threshold Q. To avoid the interference of the minimum value or zero value on the subsequent calculation, the zero-live current ratio at a certain moment is set to: ; Step 3, calculate the average value of the daily zero-fire current ratio as: , The rate of change between two consecutive days is: , In the above formula, is the average value of the daily zero-fire current ratio of the previous day, is the average value of the daily zero-fire current ratio on the following day; The fluctuation rate of the zero-live current ratio within a certain period of time is defined as:
[0023] In the above formula, n is the total number of moments in this period of time, is the zero-live current ratio at the ith moment.
[0024] Daily electricity consumption fluctuation rate / substation line loss fluctuation rate: Consistent with the principle of "zero-live current ratio", single-phase electricity meters generally use live current as the measuring current. The integral sum of the product of the measuring voltage, the measuring current and the power factor per unit time is the electricity. If the live current sampling element is inaccurate due to slight burning of the live terminal, it will cause abnormal electricity consumption, which will cause abnormal electricity fluctuation of the electricity meter, and the line loss of the substation where the electricity meter is located will also be abnormal. The calculation method of electricity consumption fluctuation rate and substation line loss fluctuation rate is consistent with the method in step 3 of "zero-live current ratio fluctuation rate".
[0025] Abnormal rate of household meter voltage: In actual situations, the total meter in the substation is usually a three-phase meter, and the household meter is a single-phase meter. The household meter is connected to a phase of the total meter. In theory, the total meter voltage will be consistent with the household meter voltage at any time, but due to the influence of the resistance on the line, the total meter voltage will be slightly higher than the household meter voltage, and there will be a voltage difference between the two. The electric energy meter in the process of burning will produce oxides or carbides at the terminal, which will increase the line resistance. When the same current flows through the line, the voltage difference between the total meter and the household meter will increase, resulting in a low voltage for the user, which is lower than the normal range in the substation. For a certain substation, based on its historical voltage and current data, its physical topological relationship is sorted out, and the specific steps are as follows: (1) Obtain the voltage and current data of the substation at each time within k consecutive days; (2) Eliminate invalid data points, including missing values and outliers. Outliers are mainly for abnormal voltage data. Traverse the voltage and current data of all times. When the voltage of a user at a certain time is greater than the total meter voltage of the substation, the voltage of the user at that time is considered invalid data and is removed; (3) Traverse each user, and for each user, combine its current data and sort them according to the voltage value. Finally, record its sorting percentage according to the total number of users in the substation. The sorting principle is: 1) At the same time, when the current magnitude is approximately the same, the user with a higher voltage is ranked higher; 2) If the current magnitude of two users Y1 and Y2 is not approximately the same, they can be ranked higher according to their respective differences with the third user. The order of each user Y3 is determined by the relative size of each user; (4) According to the voltage value of each user at each moment, the highest ranking percentage and the lowest ranking percentage within the substation within k days are recorded; (5) After completing the historical data processing, the voltage ranking of all users in their substation at each moment is recorded. When the ranking at a certain moment is lower than its historical lowest ranking, it is recorded as an abnormal point. The abnormal rate of the household meter voltage every day is (total number of abnormal points) / (total number of points on that day-number of invalid points). The calculation method of the invalid points is the same as the calculation method of the invalid data points in step (2).
[0026] In addition, before step S102, this embodiment also includes training a machine learning model: Step S201, screening out a set of burned out electric energy meters and a set of non-burned electric energy meters from all replaced electric energy meters in a certain area; Step S202, extracting feature data from historical data of burned electric energy meters in the burned electric energy meter set to obtain burned electric energy meter data set A, and extracting feature data from historical data of burned electric energy meters in the non-burned electric energy meter set to obtain non-burned electric energy meter data set B; Step S203, using a sliding window model to perform quantity balancing processing on the burned-out electric energy meter data set A (because the amount of data on the non-burned meters is much larger than that on the burned-out meters, only the burned-out electric energy meter data set needs to be balanced to avoid unnecessary quantity balancing processing operations); Step S204, training a machine learning model for identifying burned electricity meters with the burned electricity meter dataset A and the non-burned electricity meter dataset B after quantity balance processing, so as to establish a mapping relationship between the characteristic data of the electricity meter and the identification result of whether the electricity meter is at risk of being burned.
[0027] When selecting the burned out electric energy meter set and the non-burned electric energy meter set from all replaced electric energy meters in a certain area in step S201 of this embodiment, the burned out electric energy meter set is selected from all replaced electric energy meters in a certain area, including electric energy meters whose "burning phenomenon" is "neutral terminal burned", "live terminal burned" or "wiring terminal burned", and electric energy meters whose "burning phenomenon" is "all burned" or "burned meter does not include terminal burned", so as to obtain the burned out electric energy meter set consisting of electric energy meters whose "burning phenomenon" is "neutral terminal burned", "live terminal burned" or "wiring terminal burned". Specifically, in this embodiment, according to the management method for dismantling and sorting electric energy meters, all dismantled and replaced electric energy meters within a certain area are uniformly recovered and sorted, and all electric energy meters are classified. For electric energy meters with "burned" phenomenon, detailed classification is performed. Since the burning of some electric energy meters is caused by external fires, if this part of the data is included in the data set, it may affect the overall distribution of data features. Therefore, it is necessary to screen out electric energy meters with "burning phenomenon" of "neutral terminal burnt", "live terminal burnt" or "connection terminal burnt", and eliminate electric energy meters with "burning phenomenon" of "all burnt" or "burned meter excluding terminal burnt". The burned electric energy meter data set is defined as A.
[0028] After sorting the dismantled energy meters, it is necessary to trace the historical data of the burned energy meters. However, in practice, only the replacement time of the burned energy meters can be confirmed, and the starting date of the burning cannot be determined. Therefore, a sliding window model is established for the historical data of the energy meters, such as Figure 2 As shown, the sliding window model in step S202 of this embodiment includes characteristic data of the observation period and the performance period, wherein the observation period is from date t-2n to date tn, and the performance period is from date tn to date t, wherein date t is the table replacement date, and n is the sliding time window. Figure 2 It can be seen that the date of table change is recorded as t, all data from t-2n to t are taken, and the data from t-2n to tn are regarded as observation period data, and the data from tn to t are performance period data.
[0029] Artificial intelligence algorithm models are essentially dependent on statistical algorithms, which generally require that the amount of data in the training set is as large as possible, and that the positive and negative samples in the data set are relatively uniform, so that the generalization effect of the model can be optimized. However, according to historical statistical data, electric energy meter burnout failures only account for about 10% of the total number of failures. In other words, in actual operation, the number of samples in data sets A and B may be extremely uneven, resulting in poor model effects. To solve this problem, step S203 of this embodiment includes: Step S301, assigning the burned out electric energy meter data set A to the burned out electric energy meter basic data set; assigning the non-burned electric energy meter data set B to the non-burned electric energy meter basic data set, and initializing the value of the loop variable i to 0; Step S302: take all a samples in the basic data set of burned-out electric energy meters as negative samples, randomly select a samples from the basic data set of non-burned electric energy meters as positive samples, and train the machine learning model for identifying burned-out electric energy meters with the 2a balanced positive and negative sample data consisting of negative samples and positive samples to obtain the specified performance evaluation index f of the machine learning model. 1i ; Step S303, increasing the window size of the sliding window model by a specified step length, extracting feature data from the historical data of the burned electric energy meters in the burned electric energy meter set and using the new sliding window model to obtain the burned electric energy meter data set A1, merging all a samples in the burned electric energy meter data set A1 and all a samples in the burned electric energy meter basic data set as a new burned electric energy meter basic data set containing 2a negative samples, randomly selecting 2a samples from the non-burned electric energy meter basic data set as positive samples to construct a new non-burned electric energy meter basic data set, training a machine learning model for identifying burned electric energy meters with 4a balanced positive and negative sample data consisting of negative samples of the new burned electric energy meter basic data set and positive samples of the new non-burned electric energy meter basic data set, and obtaining a specified performance evaluation index f of the machine learning model. 2i ; Step S304, determining the specified performance evaluation index f of the machine learning model 2i Is it less than a preset threshold? If yes, jump to step S305; otherwise, jump to step S302; Step S305: taking the final burned-out electric energy meter basic data set as the burned-out electric energy meter data set A after quantity balance processing.
[0030] In this embodiment, A is used as the basic sample, and its sample number is a. A number of samples are randomly selected from B to complete the establishment of the positive and negative sample data set. The preset model parameters are used to train and test the model effect, and the model evaluation index (F1-score) is obtained as f 10On the basis of the sliding data window model, for each sample data in A, adjust the size of n, the step value is 1, and obtain a new sample A1; take the union of A and A1, reassign it to A, the number of samples is theoretically 2a, randomly take 2a samples in B, complete the establishment of the positive and negative sample data set, establish a new model in the same way and with the same parameters, and calculate the new evaluation index f 11 . Repeat the above steps. When the evaluation index f 1x When it is less than the threshold, the test is stopped. A and B set by the previous step value are taken as the final positive and negative sample data sets, so as to solve the problem that the number of samples in the data sets A and B in the prior art is extremely uneven, resulting in poor model effect.
[0031] It should be noted that the machine learning model is used to be pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned, and the required machine learning model can be used as needed. For example, as an optional implementation, the machine learning model used in this embodiment is a machine learning model based on XGBoost.
[0032] After completing the modeling and training of the machine learning model, it is necessary to use the machine learning model to calculate the characteristics of all the electric energy meters in the area and output the size of the probability of each electric energy meter burning out. This will consume a lot of computing resources in some time periods every day, which will put great pressure on the system platform. According to business needs, the system will collect the frozen data of all electric energy meters in the province the previous day at zero o'clock every day, complete the calculation of the probability of electric energy meter burning out before 8 o'clock every morning, and send the details of all electric energy meters with probabilities exceeding the threshold to the operation and maintenance personnel, who will complete the work order feedback within the specified time limit. According to this business model, the computing resources of the system platform are quite scarce in the early morning of each day, and there will be a large waste of resources in the subsequent time period. In order to reduce the daily computing pressure of the system platform, before step S101 of this embodiment, it also includes grouping all low-voltage areas in the specified area to set the calculation priority and calculation frequency, and the electric energy meters in the grouped low-voltage areas perform steps S101 and S102 according to the priority and calculation frequency of the corresponding group. The grouping of all low-voltage areas in the specified area to calculate the priority and calculation frequency includes: Step S401, the historical meter burnout data of all low-voltage areas in the specified area in the past m days are counted to form a dynamic dictionary C, and the key-value pair structure in the dynamic dictionary C is "area number: number of electric energy meter burnouts"; Step S402, based on the set threshold value n and the number of groups p, all low-voltage areas in the dynamic dictionary C are sorted and grouped according to the number of burnouts, and the grouping method uses equal-frequency grouping or equal-distance grouping; Step S403, for the grouped low-voltage areas, calculation priorities and calculation frequencies are set according to the sorting, so that for the electric energy meters in each low-voltage area, steps S101 and S102 are respectively executed according to the corresponding set calculation priorities and calculation frequencies to obtain the identification result of whether the corresponding electric energy meter has a risk of being burned. For example, in this embodiment, the results obtained by setting the calculation priorities and calculation frequencies for the grouped low-voltage areas according to the sorting are shown in Table 1.
[0033] Table 1 Data table of calculation priority and calculation frequency for low voltage area settings
[0034] In Table 1, t1 to t3 are the periods corresponding to different calculation frequencies. As an optional implementation, in this embodiment, the dynamic dictionary C is updated in real time according to the sorting data every day, and the data in the above table is updated. The system completes the feature processing and model calculation of all the electric energy meters in the area that should be calculated every day. Through this method, the system can complete the tasks that must be calculated on the day from 0 to 8 o'clock every day, and perform feature processing on other data that need to be calculated after 8 o'clock, and store them, reducing the subsequent calculation pressure and ensuring the balanced use of system computing resources.
[0035] In order to verify the identification method of the burned-out electric energy meter of this embodiment, the identification method of the burned-out electric energy meter of this embodiment and the model disclosed in the Chinese patent application document with publication number CN114089257A were used to process and compare the burned-out electric energy meter data of a certain province. A total of 317,640 burned-out electric energy meters were accumulated, of which 256,049 met the requirements. Through the expansion of the data set, a total of 1,024,196 burned-out electric energy meter data samples were obtained. Through the above-mentioned method for obtaining the non-burned data table, a total of 1,024,196 non-burned electric energy meter data samples were obtained. The model training and evaluation were completed according to the model training method disclosed in the Chinese patent application document with publication number CN114089257A. The precision rate P, recall rate R and F1 score (F1-Measure) were used as evaluation indicators. The evaluation indicators P=96.7%, R=81.3%, and F1=88.33% were finally calculated. The evaluation indexes corresponding to the model disclosed in the Chinese patent application document with publication number CN114089257A are P=88.3%, R=67.7%, and F1=76.6%. It can be seen that the identification method of the burned-out electric energy meter in this embodiment has been greatly improved in various evaluation indicators compared with the previous generation model (the model disclosed in the Chinese patent application document with publication number CN114089257A).
[0036] Figure 3 The weights of the top five features of the first generation model (the model disclosed in the Chinese patent application document with publication number CN114089257A) used for comparison, Figure 4 is the weight of the top 5 features of the second generation model (method of this embodiment) in the embodiment of the present invention. Figure 3 and Figure 4 It can be seen from the two figures that by adding new features, the feature weight ranking of the model has changed significantly, so the model effect can be greatly improved.
[0037] In step S203 of this embodiment, the burned-out electric energy meter dataset A is processed for quantity balance using a sliding window model. As the number of samples in the burned-out electric energy meter dataset increases from 256049 to 1024196, the improvement of the model evaluation index is shown in Table 2.
[0038] Table 2: Improvement of model evaluation indicators with different sample sizes by resampling in sliding time windows Sample size Evaluation index (F1) 256049 81.25% 512098 83.43% 768147 87.25% 1024196 88.33% As shown in Table 2, in step S203 of this embodiment, the burned-out electric energy meter data set A is processed for quantity balance using a sliding window model, thereby achieving resampling through a sliding time window, thereby significantly improving the training samples of the model.
[0039] To summarize, the method for identifying a burned-out electric energy meter in the present embodiment includes obtaining characteristic data of the electric energy meter, including the fluctuation rate of the zero-to-live current ratio, the daily electricity consumption fluctuation rate, the line loss fluctuation rate in the substation area, and the abnormal voltage rate of the household meter; inputting the characteristic data of the electric energy meter into a pre-trained machine learning model to obtain an identification result of whether the electric energy meter is at risk of being burned out. The machine learning model is pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned out. The present invention can improve the identification accuracy of the burned-out electric energy meter, and can be used to accurately identify the electric energy meter with a risk of being burned out by analyzing and calculating the operating data of the electric energy meter.
[0040] In addition, this embodiment also provides a burnt-out electric energy meter identification system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the burnt-out electric energy meter identification method.
[0041] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program / instruction is stored. The computer program / instruction is programmed or configured to execute the method for identifying a burned-out electric energy meter through a processor.
[0042] In addition, this embodiment also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the method for identifying a burned-out electric energy meter through a processor.
[0043] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0044] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
Claims
1. A method for identifying a burned-out electric energy meter, characterized in that: include: Step S101, obtaining characteristic data of the electric energy meter, the characteristic data including electrical characteristic data, product characteristic data and environmental characteristic data, the electrical characteristic data including part or all of the daily power consumption Q, daily maximum voltage U, daily maximum current I and meter running word terminal code A, zero-live current ratio fluctuation rate, daily power fluctuation rate, area line loss fluctuation rate and household meter voltage abnormality rate of the electric energy meter; the product characteristic data including part or all of the manufacturer B, the city or state C, the operation time D, the installation year E, the number of burned meters F in the area in the past two years and the area line loss G; the environmental characteristic data including part or all of the ambient temperature T and ambient humidity H of the electric energy meter; Step S102, inputting the characteristic data of the electric energy meter into a pre-trained machine learning model to obtain an identification result of whether the electric energy meter is at risk of being burned out, wherein the machine learning model is pre-trained to establish a mapping relationship between the characteristic data of the electric energy meter and the identification result of whether the electric energy meter is at risk of being burned out.
2. The method for identifying a burned-out electric energy meter according to claim 1, characterized in that: Before step S102, the process also includes training a machine learning model: Step S201, screening out a set of burned out electric energy meters and a set of non-burned electric energy meters from all replaced electric energy meters in a certain area; Step S202, extracting feature data from historical data of burned electric energy meters in the burned electric energy meter set to obtain burned electric energy meter data set A, and extracting feature data from historical data of burned electric energy meters in the non-burned electric energy meter set to obtain non-burned electric energy meter data set B; Step S203, using a sliding window model to perform quantity balance processing on the burned-out electric energy meter data set A; Step S204, training a machine learning model for identifying burned electricity meters with the burned electricity meter dataset A and the non-burned electricity meter dataset B after quantity balance processing, so as to establish a mapping relationship between the characteristic data of the electricity meter and the identification result of whether the electricity meter is at risk of being burned.
3. The method for identifying a burned-out electric energy meter according to claim 2, characterized in that: When filtering out the burned out electric energy meter set and the non-burned electric energy meter set from all the replaced electric energy meters in a certain area in step S201, the burned out electric energy meter set is filtered out from all the replaced electric energy meters in a certain area, including retaining the electric energy meters with "burning phenomenon" of "neutral terminal burned", "live terminal burned" or "wiring terminal burned", and excluding the electric energy meters with "burning phenomenon" of "all burned" or "burned meter does not include terminal burned", so as to obtain the burned out electric energy meter set consisting of the electric energy meters with "burning phenomenon" of "neutral terminal burned", "live terminal burned" or "wiring terminal burned".
4. The method for identifying a burned-out electric energy meter according to claim 2, characterized in that: The sliding window model in step S202 includes characteristic data of an observation period and a performance period, wherein the observation period is from date t-2n to date tn, and the performance period is from date tn to date t, wherein date t is the table change date, and n is the sliding time window.
5. The method for identifying a burned-out electric energy meter according to claim 4, characterized in that: Step S203 includes: Step S301, assigning the burned out electric energy meter data set A to the burned out electric energy meter basic data set; assigning the non-burned electric energy meter data set B to the non-burned electric energy meter basic data set, and initializing the value of the loop variable i to 0; Step S302: take all a samples in the basic data set of burned-out electric energy meters as negative samples, randomly select a samples from the basic data set of non-burned electric energy meters as positive samples, and train the machine learning model for identifying burned-out electric energy meters with the 2a balanced positive and negative sample data consisting of negative samples and positive samples to obtain the specified performance evaluation index f of the machine learning model. 1i ; Step S303, increasing the window size of the sliding window model by a specified step length, extracting feature data from the historical data of the burned electric energy meters in the burned electric energy meter set and using the new sliding window model to obtain the burned electric energy meter data set A1, merging all a samples in the burned electric energy meter data set A1 and all a samples in the burned electric energy meter basic data set as a new burned electric energy meter basic data set containing 2a negative samples, randomly selecting 2a samples from the non-burned electric energy meter basic data set as positive samples to construct a new non-burned electric energy meter basic data set, training a machine learning model for identifying burned electric energy meters with 4a balanced positive and negative sample data consisting of negative samples of the new burned electric energy meter basic data set and positive samples of the new non-burned electric energy meter basic data set, and obtaining a specified performance evaluation index f of the machine learning model. 2i ; Step S304, determining the specified performance evaluation index f of the machine learning model 2i Is it less than a preset threshold? If yes, jump to step S305; otherwise, jump to step S302; Step S305: taking the final burned-out electric energy meter basic data set as the burned-out electric energy meter data set A after quantity balance processing.
6. The method for identifying a burned-out electric energy meter according to claim 1, characterized in that: The machine learning model is a machine learning model based on XGBoost.
7. The method for identifying a burned-out electric energy meter according to claim 4, characterized in that: Before step S101, the method further includes grouping all low-voltage areas in the designated area to set calculation priorities and calculation frequencies, and executing steps S101 and S102 according to the priorities and calculation frequencies of the corresponding groups with the electric energy meters in the grouped low-voltage areas. The method further includes grouping all low-voltage areas in the designated area to calculate priorities and calculation frequencies, including: Step S401, statistically analyze the historical meter burnout data of all low-voltage areas in the specified area within the past m days to form a dynamic dictionary C, and the key-value pair structure in the dynamic dictionary C is "area number: number of electric energy meter burnouts"; Step S402, based on the set threshold value n and the number of groups p, all low-voltage areas in the dynamic dictionary C are sorted and grouped according to the number of burnouts, and the grouping method uses equal-frequency grouping or equal-distance grouping; Step S403, for the grouped low-voltage substations, calculation priority and calculation frequency are set according to the sorting, so that for the electric energy meters in each low-voltage substation, steps S101 and S102 are executed according to the corresponding set calculation priority and calculation frequency to obtain the identification result of whether the corresponding electric energy meter has the risk of burning.
8. A burnt-out electric energy meter identification system, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the method for identifying a burned-out electric energy meter as described in any one of claims 1 to 7.
9. A computer-readable storage medium having a computer program / instruction stored therein, characterized in that: The computer program / instructions are programmed or configured to execute the method for identifying a burned-out electric energy meter as described in any one of claims 1 to 7 through a processor.
10. A computer program product comprising a computer program / instructions, characterized in that The computer program / instructions are programmed or configured to execute the method for identifying a burned-out electric energy meter as described in any one of claims 1 to 7 through a processor.
Citation Information
Patent Citations
Machine learning model training method and system for power load identification
CN111242391A
User-transformer relation identification method based on electric quantity and voltage information fusion
CN114021430A
Electric energy meter burning online monitoring method and system and medium
CN114089257A
Two-stage monitoring analysis method and system for transformer area line loss
CN115940393A
Voltage sag estimation model establishment method, medium and system
CN117578481A
Cited By
Abnormal power consumption behavior real-time monitoring and positioning method and system
CN119397313A
Electric energy meter burning loss online monitoring method and system
CN120087230A
