Power distribution network abnormal data identification method, system, equipment and medium

Abstract

The invention discloses a power distribution network abnormal data identification method, system and device, and a medium. The method comprises the steps of determining first data corresponding to power distribution data based on a power distribution area corresponding to the power distribution data; screening out second data corresponding to the power distribution data from the first data according to a power distribution date corresponding to the power distribution data; adjusting the second data based on the environment information corresponding to the second data and the environment information corresponding to the power distribution data to obtain prediction data corresponding to the power distribution data; determining a comprehensive loss function according to the prediction data and the power distribution data; constructing an anomaly recognition model based on the comprehensive loss function; and comparing the prediction data with the power distribution data based on the anomaly identification model, and screening to obtain abnormal data in the power distribution data.

Description

Technical Field

[0001] The present invention relates to the technical field of data identification, and more specifically to a method, system, device and medium for identifying abnormal data in a distribution network. Background Art

[0002] With the rapid development of smart grids, the scale and complexity of distribution networks are increasing, and various operating data directly affect the normal operation of distribution networks. Therefore, higher requirements are placed on the identification of abnormal data in distribution networks. In order to improve the stability and security of distribution networks, it is necessary to accurately identify abnormal data in distribution networks and eliminate the interference factors caused by abnormal data on the normal operation of distribution networks.

[0003] The publication number is CN114116832A, and the name is a data-driven distribution network anomaly identification method, which includes the following steps: establishing a high-dimensional matrix model composed of voltage measurement data of each node, selecting the average spectrum radius as the linear eigenvalue statistic, and determining the time when the fault occurs; applying augmented matrix theory to perform data correlation analysis, constructing the average spectrum radius deviation rate as a quantitative indicator, and measuring the influence of each node voltage on the system operation state.

[0004] The publication number is CN111651503A, and the name is a distribution network data anomaly identification method, system and terminal equipment. Sample data, monitoring data, a first trend curve and a first trend deviation degree are obtained through the first processing and the second processing of the distribution network operation data. Reliable data is obtained based on whether the first trend deviation degree meets the first preset condition. Based on the second trend curve of the sample data and the monitoring data, the reliability of the second trend curve, and whether the reliability of the second trend curve meets the second preset condition, the operation data of the distribution network corresponding to the second trend deviation degree is identified as unreliable data, and then it is identified whether the unreliable data meets the third preset condition, and the operation data of the distribution network corresponding to the unreliable data is obtained as abnormal data.

[0005] In the existing distribution data identification methods, due to the many factors affecting distribution data and the inconsistent comparison standards for distribution data in different distribution areas, problems such as difficulty in identifying abnormal data frequently occur.

[0006] Therefore, how to provide a method for identifying abnormal data in a distribution network to solve the difficulties in identifying abnormal data in distribution data of different distribution areas is an urgent problem that technicians in this field need to solve. Summary of the invention

[0007] In view of this, the present invention provides a method, system, device, and medium for identifying abnormal data in a distribution network, aiming to solve the technical problem of difficulties in identifying abnormal data in distribution data of different distribution areas.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] A method for identifying abnormal data in a distribution network, comprising:

[0010] determining first data corresponding to the power distribution data based on a power distribution area corresponding to the power distribution data;

[0011] Filtering out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data;

[0012] Adjusting the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain prediction data corresponding to the power distribution data;

[0013] Determining a comprehensive loss function based on the prediction data and the power distribution data;

[0014] Constructing an anomaly recognition model based on the comprehensive loss function;

[0015] The predicted data and the power distribution data are compared based on the abnormality identification model to filter out abnormal data in the power distribution data.

[0016] Further, determining a comprehensive loss function according to the prediction data and the power distribution data includes:

[0017] Calculating a deviation between the predicted data and the power distribution data as a first loss parameter;

[0018] Calculating a mean absolute error between the prediction data and the power distribution data as a second loss function;

[0019] The comprehensive loss function is determined based on the first loss parameter and the second loss parameter.

[0020] Furthermore, the calculation formula of the comprehensive loss function is:

[0021]

[0022] Among them, Loss is the loss function, MAE is the mean absolute error, Bias is the deviation, N is the number of samples, y true,i is the power distribution data of the ith sample, y pred,i is the predicted data of the i-th sample.

[0023] Furthermore, before comparing the forecast data with the power distribution data, the method further includes:

[0024] Determine a business expansion index corresponding to the power distribution data based on business expansion information of the power distribution area corresponding to the power distribution data; the business expansion index is used to characterize the degree of business expansion;

[0025] The business expansion index is greater than a preset index, and the forecast data is adjusted based on the business expansion index.

[0026] Further, the business expansion information includes the number of newly added users, increased capacity, and reduced capacity; and determining the business expansion index corresponding to the power distribution data based on the business expansion information of the power distribution area corresponding to the power distribution data includes:

[0027] Calculate the business expansion index corresponding to the power distribution data by using a first formula;

[0028] The first formula is:

[0029] S=α∑k i q i +(1-α)(q 增 -q 减 )

[0030] In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q 增 To increase capacity, q 减 To reduce capacity.

[0031] Further, the adjusting the second historical data based on the environmental information corresponding to the second historical data and the environmental information corresponding to the power distribution data to obtain the prediction data corresponding to the power distribution data includes:

[0032] Inputting the environmental information corresponding to the second data into a pre-trained deep learning model to obtain a first environmental impact index corresponding to the second data;

[0033] Inputting environmental information corresponding to the power distribution data into the pre-trained deep learning model to obtain a second environmental impact index corresponding to the power distribution data;

[0034] The second data is adjusted based on the ratio of the second environmental impact index to the first environmental impact index to obtain prediction data corresponding to the power distribution data.

[0035] Furthermore, it also includes:

[0036] Setting reference power distribution data, and determining the environmental impact index of the reference power distribution data as a first index;

[0037] Adjusting the first index based on the ratio of each historical power distribution data to the reference power distribution data to obtain an environmental impact index corresponding to each historical power distribution data;

[0038] Each historical power distribution data and the environmental impact index corresponding to each historical power distribution data are used as a training set to train a deep learning model.

[0039] A distribution network abnormal data identification system, comprising:

[0040] A first data screening module, configured to determine first data corresponding to the power distribution data based on a power distribution area corresponding to the power distribution data;

[0041] A second data screening module, configured to screen out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data;

[0042] a data adjustment module, configured to adjust the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, so as to obtain prediction data corresponding to the power distribution data;

[0043] A loss function determination module, used to determine a comprehensive loss function based on the prediction data and the power distribution data;

[0044] A model building module, used to build an anomaly recognition model based on the comprehensive loss function;

[0045] A data identification module is used to compare the predicted data with the power distribution data based on the abnormality identification model, and screen out abnormal data in the power distribution data.

[0046] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method as claimed in any one of claims 1 to 7 when executing the computer program.

[0047] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

[0048] The above technical solution includes at least the following technical effects:

[0049] It can be seen from the above technical solutions that the present invention discloses a method, system, device, and medium for identifying abnormal data in a distribution network. Compared with the prior art, the present invention can take into account that the distribution data corresponding to different distribution areas and distribution dates have different characteristics. The method for identifying abnormal data in a distribution network provided in this embodiment screens the second data according to different distribution areas and distribution dates to avoid the situation where the comparison standards of distribution data in different distribution areas and distribution dates are inconsistent, resulting in inconsistent identification standards for abnormal data, thereby reducing the identification accuracy.

[0050] Taking the influence of variable environmental information on power distribution data into consideration, the second data is adjusted according to the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, so as to obtain corresponding prediction data. Based on the prediction data, the normal data variation range of the power distribution data can be predicted, thereby reducing the number of data comparisons. At the same time, based on the comparison results of the power distribution data and the prediction data, abnormal data can be effectively screened, thereby improving the efficiency of identifying abnormal data. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0052] Figure 1 The present invention is a flow chart of a method for identifying abnormal data in a distribution network.

[0053] Figure 2 The present invention is a schematic diagram of the structure of a device for identifying abnormal data in a power distribution network.

[0054] Figure 3 The present invention is a schematic diagram of the structure of an electronic device for identifying abnormal data in a power distribution network. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See attached Figure 1-3 The embodiment of the present invention discloses a method for identifying abnormal data in a distribution network, comprising:

[0057] determining first data corresponding to the power distribution data based on the power distribution area corresponding to the power distribution data;

[0058] Filtering second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data;

[0059] Adjusting the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain prediction data corresponding to the power distribution data;

[0060] Determine the comprehensive loss function based on the forecast data and power distribution data;

[0061] Build an anomaly recognition model based on the comprehensive loss function;

[0062] The predicted data and distribution data are compared based on the anomaly recognition model to filter out the abnormal data in the distribution data.

[0063] In order to further optimize the above technical solution, a comprehensive loss function is determined based on the forecast data and power distribution data, including:

[0064] Calculate the deviation between the predicted data and the power distribution data as the first loss parameter;

[0065] Calculate the mean absolute error between the prediction data and the power distribution data as the second loss function;

[0066] Based on the first loss parameter and the second loss parameter, a comprehensive loss function is determined.

[0067] In order to further optimize the above technical solution, the calculation formula of the comprehensive loss function is:

[0068]

[0069] Among them, Loss is the loss function, MAE is the mean absolute error, Bias is the deviation, N is the number of samples, y true,i is the power distribution data of the ith sample, y pred,i is the predicted data of the i-th sample.

[0070] To further optimize the above technical solution, before comparing the forecast data with the power distribution data, it also includes:

[0071] Determine the business expansion index corresponding to the power distribution data based on the business expansion information of the power distribution area corresponding to the power distribution data; the business expansion index is used to characterize the degree of business expansion;

[0072] The business expansion index is greater than the preset index, and the forecast data is adjusted based on the business expansion index.

[0073] To further optimize the above technical solution, the business expansion information includes the number of new users, increased capacity, and reduced capacity; the business expansion index corresponding to the power distribution data is determined based on the business expansion information of the power distribution area corresponding to the power distribution data, including:

[0074] Calculate the business expansion index corresponding to the power distribution data by using the first formula;

[0075] The first formula is:

[0076] S=α∑k i q i +(1-α)(q 增 -q 减 )

[0077] In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q 增 To increase capacity, q 减 To reduce capacity.

[0078] To further optimize the above technical solution, the second historical data is adjusted based on the environmental information corresponding to the second historical data and the environmental information corresponding to the power distribution data to obtain the prediction data corresponding to the power distribution data, including:

[0079] Inputting environmental information corresponding to the second data into a pre-trained deep learning model to obtain a first environmental impact index corresponding to the second data;

[0080] Inputting environmental information corresponding to the power distribution data into a pre-trained deep learning model to obtain a second environmental impact index corresponding to the power distribution data;

[0081] The second data is adjusted based on the ratio of the second environmental impact index to the first environmental impact index to obtain prediction data corresponding to the power distribution data.

[0082] To further optimize the above technical solution, it also includes:

[0083] Setting reference power distribution data, and determining the environmental impact index of the reference power distribution data as a first index;

[0084] Adjusting the first index based on the ratio of each historical power distribution data to the reference power distribution data to obtain an environmental impact index corresponding to each historical power distribution data;

[0085] Each historical power distribution data and the environmental impact index corresponding to each historical power distribution data are used as training sets to train a deep learning model.

[0086] A distribution network abnormal data identification system, comprising:

[0087] A first data screening module, configured to determine first data corresponding to the power distribution data based on a power distribution area corresponding to the power distribution data;

[0088] A second data screening module, used for screening out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data;

[0089] A data adjustment module, configured to adjust the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, so as to obtain prediction data corresponding to the power distribution data;

[0090] A loss function determination module, used to determine a comprehensive loss function based on the prediction data and the power distribution data;

[0091] A model building module, used to build an anomaly recognition model based on a comprehensive loss function;

[0092] The data identification module is used to compare the predicted data and the power distribution data based on the anomaly identification model, and screen out the abnormal data in the power distribution data.

[0093] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods of claims 1 to 7 when executing the computer program.

[0094] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the methods of claims 1 to 7 are implemented.

[0095] refer to Figure 1 , Figure 1 A flow chart of a method for identifying abnormal data in a distribution network provided by an embodiment of the present disclosure, the method comprising:

[0096] S101: Determine first data corresponding to power distribution data based on a power distribution area corresponding to the power distribution data.

[0097] In this embodiment, the power distribution data refers to the power consumption data and load data generated during the operation of the distribution network. The power distribution area refers to the types of different regions where different distribution networks provide power supply. The corresponding first data is filtered according to the different distribution areas corresponding to the power distribution data, and the first data refers to the historical data belonging to the same distribution area, that is, the first data can be the first historical data, and the first historical data can be the historical data belonging to the same distribution area.

[0098] Determine the regional attribute of the power distribution area corresponding to the power distribution data; and determine the historical data of the power distribution area belonging to the same regional attribute as the historical data.

[0099] The regional attribute refers to the regional category to which different power distribution areas belong. The first data corresponding to the power distribution data is determined based on the power distribution area corresponding to the power distribution data, which can be described in detail as follows:

[0100] The historical data of each power distribution area is clustered to obtain multiple data classes, and a regional attribute is added to each data class. The regional attributes to which multiple different power distribution areas belong are divided, so as to determine the regional attributes of the power distribution areas corresponding to the power distribution data, and the historical data of the power distribution areas belonging to the same regional attribute are determined as the first data.

[0101] This embodiment provides a reference example, such as setting multiple regional attributes as residential areas, commercial areas, and industrial areas, dividing the corresponding regional attributes for multiple different power distribution areas, determining the regional attribute of the power distribution area corresponding to a certain power distribution data as: commercial area, and determining the historical data of all power distribution areas with regional attributes as commercial areas as first data, that is, the first data is the historical data corresponding to all commercial areas.

[0102] This embodiment takes into account that power distribution data may be affected by different power distribution areas. In order to screen historical data of power distribution areas with the same attributes, regional differences are eliminated by determining regional attributes to ensure the scientific nature of the distribution network abnormal data identification process.

[0103] S102: Filter out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data.

[0104] In this embodiment, the power distribution date refers to the time information corresponding to the power distribution data. The power distribution data is divided into time periods according to the power consumption or load size. The power distribution data of different time periods corresponds to a date classification, and the date classification includes different power distribution dates. According to the power distribution date corresponding to the power distribution data, the corresponding second data is filtered out from the first data, and the second data is the historical power distribution data corresponding to the specific power distribution area and the power distribution date, that is, the second data can be the second historical data, and the second historical data can be the historical power distribution data corresponding to the specific power distribution area and the power distribution date.

[0105] This embodiment provides a reference example, for example, setting the distribution area corresponding to a certain distribution data to a residential area, determining the corresponding first historical data (first data) according to the distribution area, and the first historical data (first data) is the electricity consumption data of the residential area in the past period of time. Setting the distribution date corresponding to the distribution data to the summer vacation period, filtering the corresponding second historical data (second data) from the first historical data (first data) according to the distribution date, and obtaining the second historical data (second data) is the electricity consumption data of the residential area during the summer vacation period.

[0106] Determine the date category to which the power distribution date corresponding to the power distribution data belongs;

[0107] The data with similar dates in the first data are determined as the second data corresponding to the power distribution data. The similar dates are dates in the corresponding date category.

[0108] Specifically, since the power distribution data has not only spatial differences but also temporal differences, in order to better determine the second data corresponding to the power distribution data, the first data can be further screened according to the power distribution date. The power distribution date refers to the time period corresponding to the power distribution data, the date classification refers to the classification category obtained by classifying different time periods, and the similar date refers to the date in the date classification to which the power distribution date corresponding to the power distribution data belongs. The second data corresponding to the power distribution data can be screened out from the first data according to the power distribution date corresponding to the power distribution data, which can be described in detail as follows:

[0109] The first data corresponding to each power distribution area is clustered to obtain multiple data classes, and a time attribute, i.e., a date classification, is added to each data class. The date classification to which the power distribution data belongs is determined according to the power distribution date corresponding to the power distribution data, and the data corresponding to a similar date related to the date classification in the first data is determined as the second data corresponding to the power distribution data.

[0110] This embodiment provides a reference example, such as setting the first data as historical data corresponding to all commercial areas, the power distribution date corresponding to the power distribution data is the weekend, determining that the date category to which the power distribution date corresponding to the power distribution data belongs is the weekend, and determining the data of the date corresponding to the weekend in the first data as the second data corresponding to the power distribution data, that is, the second data is the data corresponding to the commercial area on the weekend.

[0111] This embodiment can simultaneously overcome the temporal and spatial differences in power distribution data and improve the accuracy of identifying abnormal data in the distribution network.

[0112] S103: Adjust the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain predicted data corresponding to the power distribution data.

[0113] In this embodiment, environmental information refers to certain meteorological factors existing in the surrounding environment, such as temperature, humidity, wind speed, rainfall, etc. The aforementioned environmental information has an important impact on the operation of the distribution network, and may cause the distribution data to have different data values ​​in the same distribution area and distribution date. For example, high temperature will lead to frequent use of air-conditioning equipment, thereby increasing the demand for electricity consumption; rainy and snowy weather may affect the normal operation of distribution network equipment due to low temperature, thereby reducing electricity consumption. The second data is adjusted according to the environmental information corresponding to the second data and the environmental information corresponding to the distribution data, and the predicted data corresponding to the distribution data can be obtained. The predicted data is used to predict the standard value corresponding to the distribution data based on the change trend of the second data.

[0114] S104: Determine a comprehensive loss function based on the prediction data and the power distribution data.

[0115] In this embodiment, the comprehensive loss function is used to measure the difference between the predicted data and the power distribution data. Different loss function calculation methods can be used, such as mean square error, mean absolute error, etc. By determining the comprehensive loss function, the degree of deviation between the predicted value and the actual value can be quantified, providing a basis for the subsequent construction of anomaly recognition model.

[0116] S105: Construct an anomaly recognition model based on the comprehensive loss function.

[0117] In this embodiment, an abnormality recognition model can be constructed using a comprehensive loss function. The model can be trained based on a machine learning algorithm, and its purpose is to accurately identify abnormal data by learning the characteristics and rules of normal data. In the process of building the model, the parameters of the model can be adjusted to minimize the comprehensive loss function and improve the accuracy and reliability of the model.

[0118] S106: Compare the predicted data and the power distribution data based on the abnormality identification model, and filter out abnormal data in the power distribution data.

[0119] In this embodiment, the constructed abnormality recognition model is used to compare the prediction data and the power distribution data.

[0120] The prediction data and the power distribution data are compared, and the abnormal data in the power distribution data are screened out through the comparison results, thereby realizing the identification process of abnormal data in the distribution network.

[0121] This embodiment provides another reference example, for example, setting the electricity consumption data of a residential area during this summer vacation in a certain power distribution data to 1,500,000 kWh, and the second data corresponding to the data is the electricity consumption data of the residential area during last summer vacation, which is approximately 1,560,000 kWh. The data is adjusted according to the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, and the predicted data corresponding to the power distribution data is 1,580,000 kWh. The predicted data is compared with the power distribution data, and a result that the two are inconsistent is obtained. Therefore, it can be obtained that the "electricity consumption data of the residential area during this summer vacation" is an identification result of abnormal data.

[0122] From the above, it can be concluded that this embodiment can obtain the second data corresponding to the same power distribution area and power distribution date by screening, and combine the influence of variable environmental information to achieve true prediction of power distribution data, and screen abnormal data by comparing the predicted data and the power distribution data. This embodiment can overcome the influence of environmental information on power distribution data, effectively identify dynamically changing power distribution data, and improve the efficiency of abnormal data identification.

[0123] In one embodiment of the present disclosure, determining a comprehensive loss function based on the prediction data and the power distribution data includes:

[0124] Calculate the deviation between the predicted data and the power distribution data as the first loss parameter;

[0125] Calculate the mean absolute error between the prediction data and the power distribution data as the second loss function;

[0126] Based on the first loss parameter and the second loss parameter, a comprehensive loss function is determined.

[0127] In this embodiment, the predicted data is a reasonable estimate of the current power distribution data obtained by analyzing and adjusting the historical data. The predicted data is compared with the power distribution data, and the deviation between them is calculated. The deviation can reflect the degree of difference between the predicted value and the actual value. For example, the deviation value can be obtained by subtracting the power distribution data from the predicted data. This deviation value is used as the first loss parameter, which measures the accuracy of the model prediction from one perspective. If the deviation is large, it means that the prediction result of the model is far from the actual situation.

[0128] Mean absolute error is a commonly used indicator to measure the accuracy of predictions. It calculates the average of the absolute differences between the predicted data and the distribution data. Specifically, for each data point, the absolute difference between its predicted value and the actual value is calculated, and then the absolute differences of all data points are summed and divided by the total number of data points to get the mean absolute error. This mean absolute error, as the second loss function, reflects the overall error level of the model prediction from another perspective. Compared with the deviation, the mean absolute error takes into account the situation of all data points and can more comprehensively reflect the performance of the model.

[0129] The comprehensive loss function is determined by combining the first loss parameter (bias) and the second loss parameter (mean absolute error). These two parameters can be combined in different ways, such as simple addition, weighted summation, etc.

[0130] It can be concluded from the above that by comprehensively considering the two loss parameters, the prediction performance of the model can be more comprehensively measured. The comprehensive loss function determined in this way can simultaneously reflect the degree of deviation between the predicted value and the actual value and the overall average error level, thereby providing a basis for building a more accurate anomaly recognition model. In the process of building an anomaly recognition model, the comprehensive loss function can be minimized by adjusting the parameters of the model to improve the accuracy and reliability of the model.

[0131] In one embodiment of the present disclosure, the calculation formula of the comprehensive loss function is:

[0132]

[0133] Among them, Loss is the loss function, MAE is the mean absolute error, Bias is the deviation, N is the number of samples, y true,i is the power distribution data of the ith sample, y pred,i is the predicted data of the i-th sample.

[0134] In one embodiment of the present disclosure, before comparing the prediction data with the power distribution data, the method for identifying abnormal data of the power distribution network further includes:

[0135] The business expansion index corresponding to the power distribution data is determined based on the business expansion information of the power distribution area corresponding to the power distribution data. The business expansion index is used to characterize the degree of business expansion.

[0136] In response to the business expansion index being greater than a preset index, the forecast data is adjusted based on the business expansion index.

[0137] In this embodiment, business expansion refers to the expansion and transformation of the distribution network. Business expansion activities will affect the structure and scale of the distribution network, and then affect the operating performance and reliability of the distribution network. As business expansion activities proceed, business expansion information will also have a certain impact on the distribution data. In order to eliminate the recognition process of abnormal data of the distribution network data by business expansion information, before comparing the predicted data with the distribution data, the business expansion index is used to determine whether the predicted data needs to be adjusted. The business expansion index is used to characterize the degree of business expansion. The business expansion index corresponding to the distribution data can be determined by the business expansion information of the distribution area corresponding to the distribution data, and the business expansion index is judged. When it is greater than the preset index, the predicted data is adjusted according to the business expansion index. If the business expansion index is less than or equal to the preset index, there is no need to adjust the predicted data, and the next step of comparing the predicted data and the distribution data can be carried out.

[0138] From the above, it can be concluded that this embodiment takes into account the influence of business expansion information on the identification of abnormal data of the distribution network, and determines whether the predicted data needs to be adjusted by setting the business expansion index, so as to avoid misjudgment of the predicted data due to the influence of business expansion information, thereby misidentifying the abnormal data of the distribution network. This embodiment can reduce human subjective judgment and increase the objectivity and authenticity of the method for identifying abnormal data of the distribution network by calculating the business expansion index.

[0139] In one embodiment of the present disclosure, the business expansion information includes the number of newly added users, increased capacity, and decreased capacity. Determining the business expansion index corresponding to the power distribution data based on the business expansion information of the power distribution area corresponding to the power distribution data includes:

[0140] The business expansion index corresponding to the power distribution data is calculated by the first formula.

[0141] The first formula is:

[0142] S=α∑k i q i +(1-α)(q增 -q 减 )

[0143] In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q 增 To increase capacity, q 减 To reduce capacity.

[0144] In this embodiment, the business expansion information includes information such as the number of new users, capacity increase, and capacity reduction, among which capacity increase and capacity reduction refer to two common operations in the power system in which users increase or decrease the original power capacity according to changes in their own power demand. Capacity increase means that the capacity of the user's original power equipment is insufficient, and new equipment needs to be added or replaced with higher-capacity equipment to meet the power demand in production, life, etc. Capacity increase may involve the expansion and transformation of the power grid to ensure that the distribution network can meet the user's increased power demand. Capacity reduction means that the capacity of the user's original power equipment is excessive, and some equipment needs to be reduced or the operating capacity of the equipment needs to be reduced to reduce electricity costs. Capacity reduction may involve the adjustment and optimization of the distribution network to ensure that the distribution network operates more efficiently and reliably while meeting the power demand of users. Capacity increase and capacity reduction refer to the increase and decrease of power capacity on the basis of the original power capacity during capacity increase and capacity reduction operations. The business expansion index corresponding to the distribution data is determined based on the business expansion information of the distribution area corresponding to the distribution data, which can be detailed as follows:

[0145] The business expansion index corresponding to the power distribution data is calculated by the first formula, which can be expressed as:

[0146] S=α∑k i q i +(1-α)(q 增 -q 减 )

[0147] In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q 增 To increase capacity, q 减 To reduce capacity.

[0148] This embodiment provides a reference example. For instance, set the preset index to 1,000,000. Suppose a power distribution area corresponding to certain power distribution data can be divided into three categories according to its regional attributes, namely: residential area 1, commercial area 2, and industrial area 3. Among them, the number of users in residential area 1 is 2,000, and the electricity consumption capacity is 200 degrees per month; the number of users in commercial area 2 is 400, and the electricity consumption capacity is 5,000 degrees per month; the number of users in industrial area 3 is 100, and the electricity consumption capacity is 100,000 degrees per month; the increased capacity is 500,000 degrees per month; the decreased capacity is 300,000 degrees per month; the weight coefficient is 0.6. Calculate the service expansion index corresponding to the power distribution data through the first formula as follows:

[0149] S = α∑k i q i +(1 - α)(q 增 - q 减 ) == 0.6

[0150] * (2000 * 200 + 400 * 5000 + 100 * 100000)+(1 - 0.6)(

[0151] 500000 - 300000)= <7520000> 1000000

[0152] Therefore, it is necessary to adjust the prediction data based on the service expansion index.

[0153] It can be concluded from the above that this embodiment can calculate the service expansion index according to various service expansion information of the power distribution area corresponding to the power distribution data, and judge whether it is necessary to further adjust the prediction data according to the calculation result, effectively ensuring the accuracy of the prediction data.

[0154] In an embodiment of the present disclosure, the second data is adjusted based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain the prediction data corresponding to the power distribution data, including:

[0155] Input the environmental information corresponding to the second data into the deep learning model to obtain the first environmental impact index corresponding to the second data.

[0156] Input the environmental information corresponding to the power distribution data into the deep learning model to obtain the second environmental impact index corresponding to the power distribution data.

[0157] Adjust the second data based on the ratio of the second environmental impact index and the first environmental impact index to obtain the prediction data corresponding to the power distribution data.

[0158] In this embodiment, since environmental information has a high impact on power distribution data, in order to eliminate the interference of environmental information in the process of obtaining predicted data, an impact index of environmental information on power distribution data can be generated through a deep learning model, and the power distribution data can be adjusted using the environmental impact index to obtain predicted data. Among them, the deep learning model is a classification model that is trained based on historical data and can output environmental impact indexes corresponding to different environmental information. The environmental impact index represents the degree of influence of environmental information on power distribution data. The second data is adjusted based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain predicted data corresponding to the power distribution data, which can be described in detail as follows:

[0159] The environmental information corresponding to the second data is input into the deep learning model, and the deep learning model is used to identify and output the corresponding first environmental impact index. The environmental information corresponding to the power distribution data is input into the deep learning model, and the deep learning model is used to classify and identify the second environmental impact index corresponding to the power distribution data. The ratio of the second environmental impact index to the first environmental impact index is calculated, and the second data is adjusted according to the ratio to obtain the predicted data corresponding to the power distribution data.

[0160] This embodiment provides a reference example, such as setting the first environmental impact index to 70 and the second environmental impact index to 90, and calculating the ratio of the second environmental impact index to the first environmental impact index to be 9 / 7, adjusting the second data according to the ratio, and multiplying the second data by the ratio to obtain the predicted data corresponding to the power distribution data.

[0161] From the above, it can be concluded that this embodiment takes into account the impact of dynamic environmental information on power distribution data. In order to more accurately predict the predicted data corresponding to the power distribution data, the second data is adjusted by calculating the environmental impact index to obtain the predicted data. This embodiment can eliminate the impact of accidental factors on power distribution data and improve the applicability of abnormal data identification.

[0162] In one embodiment of the present disclosure, the method for identifying abnormal data in a distribution network further includes:

[0163] Reference power distribution data is set, and an environmental impact index of the reference power distribution data is determined as a first index.

[0164] The first index is adjusted based on the ratio of each historical power distribution data to the reference power distribution data to obtain an environmental impact index corresponding to each historical power distribution data.

[0165] Each historical power distribution data and the environmental impact index corresponding to each historical power distribution data are trained to obtain a deep learning model.

[0166] In this embodiment, in order to accurately obtain the first environmental impact index and the second environmental impact index, it is necessary to train a deep learning model to output the environmental impact index for environmental information corresponding to different power distribution data. The training process is:

[0167] Reference power distribution data is set, and the reference power distribution data can be any historical power distribution data. The environmental impact index of the reference power distribution data is determined as the first index. The ratio of each historical power distribution data to the power distribution data is calculated, and the first index is adjusted according to the obtained multiple ratios to obtain the environmental impact index corresponding to each historical power distribution data. Each historical power distribution data and its corresponding environmental impact index are trained based on the classification algorithm to obtain a deep learning model.

[0168] It can be concluded from the above that this embodiment can train a deep learning model based on a classification algorithm to improve the accuracy and speed of data classification, thereby providing data support for the abnormal data identification process of the distribution network.

[0169] Corresponding to the method for identifying abnormal data in the distribution network in the above embodiment, Figure 2 This is a structural block diagram of a device for identifying abnormal data in a distribution network provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 2 The distribution network abnormal data identification device 20 includes: a first data screening module 21, a second data screening module 22, a data adjustment module 23, a loss function determination module 24, a model construction module 25 and a data identification module 26.

[0170] Wherein, the first data screening module 21 is used to determine the first data corresponding to the power distribution data based on the power distribution area corresponding to the power distribution data;

[0171] A second data screening module 22, configured to screen out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data;

[0172] A data adjustment module 23, configured to adjust the second data based on the environment information corresponding to the second data and the environment information corresponding to the power distribution data, to obtain prediction data corresponding to the power distribution data;

[0173] A loss function determination module 24, used to determine a comprehensive loss function based on the prediction data and the power distribution data;

[0174] A model building module 25, used to build an anomaly recognition model based on a comprehensive loss function;

[0175] The data identification module 26 is used to compare the predicted data and the power distribution data based on the abnormality identification model, and screen out abnormal data in the power distribution data.

[0176] In one embodiment of the present disclosure, the loss function determination module 24 is specifically used to:

[0177] Calculate the deviation between the predicted data and the power distribution data as the first loss parameter;

[0178] Calculate the mean absolute error between the prediction data and the power distribution data as the second loss function;

[0179] Based on the first loss parameter and the second loss parameter, a comprehensive loss function is determined.

[0180] In one embodiment of the present disclosure, the loss function determination module 24 is further configured to:

[0181] The calculation formula of the comprehensive loss function is:

[0182]

[0183] Among them, Loss is the loss function, MAE is the mean absolute error, Bias is the deviation, N is the number of samples, y true,i is the power distribution data of the ith sample, y pred,i is the predicted data of the i-th sample.

[0184] In one embodiment of the present disclosure, the data identification module 26 is specifically used to:

[0185] The business expansion index corresponding to the power distribution data is determined based on the business expansion information of the power distribution area corresponding to the power distribution data.

[0186] The business expansion index is used to characterize the degree of business expansion;

[0187] In response to the business expansion index being greater than a preset index, the forecast data is adjusted based on the business expansion index.

[0188] In one embodiment of the present disclosure, the business expansion information includes the amount of newly added users, increased capacity, and decreased capacity. The data identification module 24 is specifically used to:

[0189] Calculate the business expansion index corresponding to the power distribution data by using the first formula;

[0190] The first formula is:

[0191] S=α∑k i q i +(1-α)(q 增 -q 减 )

[0192] In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q增 To increase capacity, q 减 To reduce capacity.

[0193] In one embodiment of the present disclosure, the data adjustment module 23 is specifically used to:

[0194] Inputting environmental information corresponding to the second data into a deep learning model to obtain a first environmental impact index corresponding to the second data;

[0195] Inputting environmental information corresponding to the power distribution data into a deep learning model to obtain a second environmental impact index corresponding to the power distribution data;

[0196] The second data is adjusted based on the ratio of the second environmental impact index to the first environmental impact index to obtain prediction data corresponding to the power distribution data.

[0197] In one embodiment of the present disclosure, the data adjustment module 23 is specifically used to:

[0198] Setting reference power distribution data, and determining the environmental impact index of the reference power distribution data as a first index;

[0199] Adjusting the first index based on the ratio of each historical power distribution data to the reference power distribution data to obtain an environmental impact index corresponding to each historical power distribution data;

[0200] Each historical power distribution data and the environmental impact index corresponding to each historical power distribution data are trained to obtain a deep learning model.

[0201] See also Figure 3 , Figure 3 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 3 The electronic device 300 in the embodiment shown may include: a processor 301, an input device 302, an output device 303 and a memory 304, wherein the processor 301, the input device 302, the output device 303 and the memory 304 are one or more. The communication between the above-mentioned processor 301, the input device 302, the output device 303 and the memory 304 can be completed through the communication bus 305. The function of the memory 304 is to store computer programs, and the computer program may include program instructions, and the program instructions may include the reading and calculation of data. The execution subject of the program instructions stored in the memory 304 is the processor 301. Among them, the processor 301 can be used to call the program instructions stored in the memory 304 to execute the functions of the first data screening module 21, the second data screening module 22, the data adjustment module 23, the loss function determination module 24, the model construction module 25 and the data identification module 26 in the above-mentioned distribution network abnormal data identification device 20 embodiment, such as Figure 2 shown.

[0202] The beneficial effects of the abnormal data identification method for distribution network provided by the embodiment of the present disclosure are:

[0203] First, considering that the distribution data corresponding to different distribution areas and distribution dates have different characteristics, the distribution network abnormal data identification method provided in this embodiment screens the second data according to different distribution areas and distribution dates to avoid the situation where the distribution data of different distribution areas and distribution dates have inconsistent comparison standards, resulting in inconsistent identification standards for abnormal data, thereby reducing the identification accuracy.

[0204] Second, this embodiment takes the influence of variable environmental information on power distribution data into consideration, and adjusts the second data according to the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, so as to obtain corresponding prediction data. Based on the prediction data, the normal data variation range of the power distribution data can be predicted, thereby reducing the number of data comparisons. At the same time, based on the comparison results between the power distribution data and the prediction data, abnormal data can be effectively screened, thereby improving the efficiency of identifying abnormal data.

[0205] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0206] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying abnormal data in a distribution network, characterized in that: include: determining first data corresponding to the power distribution data based on a power distribution area corresponding to the power distribution data; Filtering out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data; Adjusting the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data to obtain prediction data corresponding to the power distribution data; Determining a comprehensive loss function based on the prediction data and the power distribution data; Constructing an anomaly recognition model based on the comprehensive loss function; The predicted data and the power distribution data are compared based on the abnormality identification model to filter out abnormal data in the power distribution data.

2. A method for identifying abnormal data in a distribution network according to claim 1, characterized in that: Determining a comprehensive loss function according to the prediction data and the power distribution data includes: Calculating a deviation between the predicted data and the power distribution data as a first loss parameter; Calculating a mean absolute error between the prediction data and the power distribution data as a second loss function; The comprehensive loss function is determined based on the first loss parameter and the second loss parameter.

3. A method for identifying abnormal data in a distribution network according to claim 2, characterized in that: The calculation formula of the comprehensive loss function is: Among them, Loss is the loss function, MAE is the mean absolute error, Bias is the deviation, N is the number of samples, y true,i is the power distribution data of the ith sample, y pred,i is the predicted data of the i-th sample.

4. A method for identifying abnormal data in a distribution network according to claim 1, characterized in that: Before comparing the forecast data with the power distribution data, the method further includes: Determine a business expansion index corresponding to the power distribution data based on business expansion information of the power distribution area corresponding to the power distribution data; the business expansion index is used to characterize the degree of business expansion; The business expansion index is greater than a preset index, and the forecast data is adjusted based on the business expansion index.

5. A method for identifying abnormal data in a distribution network according to claim 4, characterized in that: The business expansion information includes the number of newly added users, capacity increase, and capacity reduction; The determining the business expansion index corresponding to the power distribution data based on the business expansion information of the power distribution area corresponding to the power distribution data includes: Calculate the business expansion index corresponding to the power distribution data by using a first formula; The first formula is: S=α∑k i q i +(1-a)(q 增 -q 减 ) In the formula, S is the industry expansion index, α is the weight coefficient, i is the regional attribute number of the distribution area corresponding to the distribution data, and k i is the number of users in the th area, q i is the power consumption capacity of a single user in the area of ​​the first type, q 增 To increase capacity, q 减 To reduce capacity.

6. A method for identifying abnormal data in a distribution network according to claim 1, characterized in that: The adjusting the second historical data based on the environmental information corresponding to the second historical data and the environmental information corresponding to the power distribution data to obtain the prediction data corresponding to the power distribution data includes: Inputting the environmental information corresponding to the second data into a pre-trained deep learning model to obtain a first environmental impact index corresponding to the second data; Inputting environmental information corresponding to the power distribution data into the pre-trained deep learning model to obtain a second environmental impact index corresponding to the power distribution data; The second data is adjusted based on the ratio of the second environmental impact index to the first environmental impact index to obtain prediction data corresponding to the power distribution data.

7. A method for identifying abnormal data in a distribution network according to claim 6, characterized in that: Also includes: Setting reference power distribution data, and determining the environmental impact index of the reference power distribution data as a first index; Adjusting the first index based on the ratio of each historical power distribution data to the reference power distribution data to obtain an environmental impact index corresponding to each historical power distribution data; Each historical power distribution data and the environmental impact index corresponding to each historical power distribution data are used as a training set to train a deep learning model.

8. A distribution network abnormal data identification system, applicable to any distribution network abnormal data identification method described in claims 1-7, characterized in that: include: A first data screening module, configured to determine first data corresponding to the power distribution data based on a power distribution area corresponding to the power distribution data; A second data screening module, configured to screen out second data corresponding to the power distribution data from the first data according to the power distribution date corresponding to the power distribution data; a data adjustment module, configured to adjust the second data based on the environmental information corresponding to the second data and the environmental information corresponding to the power distribution data, so as to obtain prediction data corresponding to the power distribution data; A loss function determination module, used to determine a comprehensive loss function based on the prediction data and the power distribution data; A model building module, used to build an anomaly recognition model based on the comprehensive loss function; A data identification module is used to compare the predicted data with the power distribution data based on the abnormality identification model, and screen out abnormal data in the power distribution data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images