CN-122020489-A - Ammeter registration abnormity detection method and system

CN122020489ACN 122020489 ACN122020489 ACN 122020489ACN-122020489-A

Abstract

The invention discloses an ammeter registration anomaly detection method and system. The method comprises the steps of collecting current multi-source power service data of a user to be detected, preprocessing extraction features, generating detection effectiveness masks, constructing an address mapping index table according to multi-dimensional service attributes, executing physical rearrangement of memory data at a host end to construct a historical data container, calculating full-quantity self-adaptive threshold tensors in parallel, transmitting the current data to a GPU video memory to construct a feature tensor in batches, generating sparse calculation indexes based on the effectiveness masks, executing tensor aggregation and load fluctuation rate parallel calculation, comparing calculation results with the threshold tensors to generate an abnormal candidate list, and finally executing physical index inverse mapping, and filtering and updating based on global immune tensors. The invention solves the problems of access bottleneck and calculation efficiency of mass discrete data by utilizing heterogeneous calculation architecture and memory rearrangement technology, and realizes self-adaptive accurate and efficient detection of abnormal electric quantity.

Inventors

ZHOU HONGYONG
SONG WEIPING
LIU ZESAN
TIAN TIAN
SUN YUTING
REN CAIHONG
Xiao Wenyue
ZHANG XIAOWU
WANG HE
WANG DADI
LIN HONG

Assignees

国网江苏省电力有限公司营销服务中心
国网信息通信产业集团有限公司
国网江苏省电力有限公司
国家电网有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The method for detecting the abnormal ammeter reading is characterized by comprising the following steps: Preprocessing the current multi-source power business data, extracting power utilization characteristic parameters for calculating a load fluctuation rate index, and generating a detection effectiveness mask for marking the effective calculation state of the power utilization characteristic parameters; reading historical multi-source power business data of a user to be detected, aligning the calculation sequence of the historical multi-source power business data based on the address mapping index table, calculating historical load fluctuation rate indexes in parallel, and generating a full-quantity self-adaptive threshold tensor mapped with the rearranged data structure; constructing a to-be-detected electric quantity characteristic tensor based on the rearranged data structure and the detection effectiveness mask; Based on the sparse calculation index, carrying out tensor aggregation operation on the electric quantity characteristic tensor to be detected, calculating the load fluctuation rate index by utilizing aggregated data, carrying out element-by-element comparison on the load fluctuation rate index and the full-quantity self-adaptive threshold tensor, identifying abnormal data according to a comparison result, and generating an abnormal candidate list based on the abnormal data; And acquiring a preset global immune tensor, and filtering the abnormal candidate list based on the immune weights in the global immune tensor to generate a final diagnosis list.
2. The method for detecting abnormal meter reading according to claim 1, wherein the current multi-source power service data comprises daily granularity power consumption data, multi-dimensional service attribute parameters, full-scale expansion change state data and special period calendar data, The daily granularity electricity consumption data comprises an active electricity quantity metering indication with a time stamp; the multidimensional service attribute parameters comprise industry classification codes and platform region codes; The full-volume expansion change state data comprises new installation, capacity expansion, capacity reduction restoration, suspension and suspension restoration; the calendar data of the special period comprises legal holiday labels and user backup non-working day labels.
3. The method for detecting abnormal meter reading according to claim 2, wherein calculating the load fluctuation rate index specifically comprises: Based on the daily granularity electricity consumption data and the special period calendar data in the current-period multi-source power business data, the current-month total electric quantity, the current-month natural days and the current-month natural days, the current-month special period accumulated electric quantity, the current-month special period days and the current-month special period days are respectively extracted; Calculating the difference value between the total current month electric quantity and the accumulated electric quantity of the current month in the special period to obtain the current month pure working daily electric quantity; Calculating the difference value between the natural days of the current month and the days of the special period of the current month to obtain the pure working days of the current month; dividing the current month pure working day electric quantity by the current month pure working day number to obtain current month pure load intensity; dividing the previous month pure workday electric quantity by the previous month pure workday days to obtain previous month pure load intensity; And calculating the ratio of the current month pure load intensity to the previous month pure load intensity as the load fluctuation rate index.
4. The method for detecting abnormal electric meter reading according to claim 2, wherein constructing an address mapping index table, performing physical rearrangement on the current multi-source power service data based on the address mapping index table, and generating a continuously stored rearranged data structure, specifically comprises: Performing bit splicing on the industry classification code and the area code of the user to be detected to generate a user characteristic combination key; Performing sorting operation on the user characteristic combination keys of the user to be detected, and generating an address mapping index table pointing to a target physical index from an original logical index; opening up a continuous physical storage space in a memory of a host end; traversing the address mapping index table, and sequentially copying the current multisource power business data of the user to be detected, which are stored in a discrete manner, into the continuous physical storage space to obtain a continuously-stored rearranged data structure; In the rearranged data structure, data with the same user characteristic combination keys are arranged continuously in physical space to form a plurality of grouping data blocks.
5. The method for detecting abnormal meter reading according to claim 4, wherein the step of generating a full-scale adaptive threshold tensor by parallel calculation of historical load fluctuation rate indexes based on the calculation sequence of the historical multi-source power business data aligned by the address mapping index table comprises the following steps: Based on the historical multi-source power service data, calling a host-side multithread parallel instruction or a vectorization instruction, and calculating a plurality of historical load fluctuation rate indexes of each user to be detected in the rearranged data structure in batches in a preset historical period; Calculating arithmetic mean values and standard deviations of the historical load fluctuation rate indexes of all users to be detected in the clustered data blocks by using a statistical protocol algorithm for each clustered data block in the clustered data blocks, and calculating weighted sums of products of the arithmetic mean values and the standard deviations based on preset sensitivity coefficients to serve as adaptive abnormality judgment thresholds shared by the clustered data blocks; And giving the self-adaptive abnormal judgment threshold value to each user to be detected in the corresponding clustered data block by using a broadcasting mechanism, and packaging according to the physical sequence of the address mapping index table to generate the full-quantity self-adaptive threshold tensor.
6. The method for detecting an abnormality of an ammeter registration according to claim 1, wherein generating a sparse calculation index and performing a tensor aggregation operation based on the sparse calculation index, comprises: performing a parallel prefix and scan operation on the detection validity mask, calculating an accumulated count of valid identification values in the detection validity mask up to a current position, and determining the accumulated count as the sparse calculation index; and extracting data slices corresponding to the effective identification values in the detection effectiveness masks from the electric quantity characteristic tensor to be detected based on the sparse calculation index, and assembling the data slices into a compact calculation queue with continuous memory.
7. The method for detecting abnormal meter reading according to claim 1, wherein the filtering and updating are performed based on a global immune tensor, specifically comprising: According to the address mapping index table, reversely analyzing corresponding user identity information according to the physical storage index value in the abnormal candidate list; Reading corresponding immunity weights from the global immunity tensor according to the user identity information, wherein the immunity weights in the global immunity tensor are dynamically updated by executing multiplication attenuation operation based on a time attenuation factor according to a preset time period; if the immunity weight is greater than a preset false alarm shielding threshold, removing the corresponding user from the abnormal candidate list; and if the immunity weight is not greater than the preset false alarm shielding threshold value, reserving the corresponding user to generate the final diagnosis list.
8. Ammeter registration anomaly detection system, its characterized in that includes: the system comprises a data acquisition and characteristic preprocessing module, a detection effectiveness mask, a load fluctuation rate index calculation module and a load fluctuation rate index calculation module, wherein the data acquisition and characteristic preprocessing module is used for acquiring current multi-source power service data of a user to be detected; The memory rearrangement and threshold generation module is used for constructing an address mapping index table, performing physical rearrangement on the current multi-source power service data based on the address mapping index table, and generating a continuously stored rearranged data structure; reading historical multi-source power business data of a user to be detected, aligning the calculation sequence of the historical multi-source power business data based on the address mapping index table, and calculating historical load fluctuation rate indexes in parallel to generate a full-quantity self-adaptive threshold tensor mapped with the rearranged data structure; The heterogeneous data transmission and tensor construction module is used for constructing an electric quantity characteristic tensor to be detected based on the rearranged data structure and the detection effectiveness mask; Based on the sparse calculation index, performing tensor aggregation operation on the electric quantity characteristic tensor to be detected, calculating the load fluctuation rate index by using aggregated data, comparing the load fluctuation rate index with the full-quantity self-adaptive threshold tensor element by element, identifying abnormal data according to a comparison result, and generating an abnormal candidate list based on the abnormal data; The immune feedback and diagnosis output module is used for acquiring a preset global immune tensor, filtering the abnormal candidate list based on the immune weight in the global immune tensor, and generating a final diagnosis list.
9. A terminal comprises a processor and a storage medium, and is characterized in that: The storage medium is used for storing instructions; the processor being operative according to the instructions to perform the steps of the method as claimed in any one of claims 1 to 7.
10. Computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.

Description

Ammeter registration abnormity detection method and system Technical Field The invention relates to the technical field of electric power data processing, in particular to an electric meter indication abnormity detection method and system. Background The modern power enterprises have large numbers of users, and the marketing business such as cost calculation, market analysis and the like needs to perform anomaly detection on massive meter data before application. The electric energy meter data is accurately the basis of various electric power business development such as market transaction, marketing accounting, demand response, electricity stealing inspection and the like, but the electric energy meter data always has errors caused by various reasons, such as meter loading errors, disturbed data transmission, electric energy meter faults and the like, the errors are commonly called as anomalies before no field confirmation, and the anomalies data need to be found and processed separately in advance before the normal development of the business, so that the abnormal detection of the electric energy meter readings becomes an important work for guaranteeing the normal development of the business. Currently, for detecting the fluctuation abnormality of the electric quantity data, a fixed threshold rule judging method based on serial calculation of a Central Processing Unit (CPU) is adopted. The typical implementation flow of the method is that the system traverses the current month and last month electricity consumption data of each user in sequence, calculates the ratio of the current month electricity consumption to the last month electricity consumption, compares the calculation result with a preset global fixed threshold, and judges that the current period electricity consumption data of the user is abnormal if the fluctuation rate exceeds the threshold. However, the core problem faced by the detection of abnormal meter readings is the huge data size. Taking a provincial power grid company as an example, the power consumption clients managed by the provincial power grid company can reach thousands of households, and corresponding to electric energy meters of the same order of magnitude, the generated thousands of electric quantity data need to be automatically checked each month. At this data scale, existing methods expose a number of drawbacks. Firstly, in terms of computing architecture, the existing method mostly adopts a logic flow of computing and judging by a user-by-user cycle. This processing mode does not efficiently utilize the single instruction multiple data Stream (SIMD) instruction set of a modern Central Processing Unit (CPU) or the massively parallel computing capabilities of a Graphics Processor (GPU). When the user data of tens of millions of levels are processed, serial calculation is too long, a remarkable performance bottleneck is formed, and the timeliness requirement of batch rapid screening of mass data is difficult to meet. Secondly, the existing methods lack adaptability and refinement in terms of accuracy of judgment logic. On the one hand, most of the existing methods set uniform and static fluctuation rate abnormality determination thresholds for all types of users, and fail to consider inherent differences of electricity utilization behavior patterns of different users (such as residents, businesses and industries). The rule setting mode of 'one-cut' is difficult to accurately adapt to diversified real electricity utilization scenes, so that a large number of false alarms of normal situations are caused. Because of the huge user base, the absolute number of the false alarm results is larger, the burden of invalid manual auditing is increased, and the overall efficiency of electric charge distribution is restricted. On the other hand, the existing logic is only based on the original reading, and irregular disturbance of electricity consumption caused by special periods such as legal holidays is not considered and eliminated. For example, a normal increase in commercial power consumption during long-term fraud is easily determined as abnormal fluctuations, which further affect the accuracy and reliability of the detection results. Furthermore, the prior art lacks an optimization mechanism for sparse data in terms of computational resource utilization. The abnormal data of the power grid is sparse in nature in the whole data, namely most users are normal, but the existing general algorithm executes the whole-flow floating point calculation and logic judgment with the same precision on all users whether the users are obviously normal or in an immune period. This indiscriminate computing approach results in a large amount of computing power being wasted on the processing of invalid or low value data, failing to reduce system power consumption through sparsity optimization. Therefore, how to design a batch detection method for detecting abnormal power fluctuation, which can efficien