CN-121980143-A - Anomaly identification method, system, medium and product of electricity consumption data

CN121980143ACN 121980143 ACN121980143 ACN 121980143ACN-121980143-A

Abstract

The invention discloses an anomaly identification method, system, medium and product of electricity data, belonging to the field of anomaly identification of the electricity data, wherein the method comprises the steps of collecting current original electricity time sequence data of a plurality of loads in a target area, and obtaining corresponding historical electricity data and tide calculation data; the method comprises the steps of correcting original electricity time sequence data based on historical electricity data to obtain electricity time sequence data, extracting and predicting multi-scale time sequence characteristics of the historical electricity data by using a preset P-T imemi xer algorithm to generate predicted data with higher data density, carrying out bidirectional alignment fusion on the electricity time sequence data and the predicted data to obtain fused electricity data, calculating preset deviation indexes by combining the fused electricity data and power flow calculation data of a power distribution network, judging whether each data point in the fused data is abnormal, and outputting an abnormal identification result. By implementing the invention, the problem that the electricity consumption data is difficult to accurately identify under low-cost acquisition equipment in the prior art can be solved.

Inventors

CAI WENTING
XU JIAN
LIN CONG
Cao Zejiang
DENG HONGBIN
Liang tianying
LU XIN
Shen ruixuan

Assignees

中国南方电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251209

Claims (10)

1. An anomaly identification method for electricity consumption data, characterized by comprising the following steps: collecting current original power utilization time sequence data of a plurality of loads in a target area, and obtaining corresponding historical power utilization data and tide calculation data; correcting the original power utilization time sequence data based on the historical power utilization data to obtain power utilization time sequence data; Performing multi-scale time sequence feature extraction and prediction on the historical electricity utilization data according to a preset P-Timemixer algorithm to generate prediction data, wherein the data density of the prediction data is greater than that of the electricity utilization time sequence data; performing bidirectional alignment fusion on the power utilization time sequence data and the prediction data to obtain fusion power utilization data; And calculating a preset deviation index based on the fusion power consumption data and the power flow calculation data of the power distribution network, judging whether corresponding data points in the fusion power consumption data are abnormal data according to the deviation index, and outputting an abnormal recognition result.
2. The method for anomaly identification of electricity consumption data according to claim 1, wherein the correcting the original electricity consumption time series data based on the historical electricity consumption data to obtain the electricity consumption time series data comprises: In the original power utilization time sequence data, if a certain original data point is judged to have full-period offset, calculating an offset compensation amount according to a historical synchronous average value of the original data point and a synchronous actual measurement value of similar loads, and calculating a difference value between the original value of the original data point and the offset compensation amount to be used as an offset correction data point; In the original power utilization time sequence data, if a certain original data point is judged to have abnormal jump, interpolation is carried out according to the adjacent data of the original data point to obtain a jump correction value as a jump correction data point; In the original power utilization time sequence data, if a data missing point exists, the original power utilization time sequence data and a historical data value of the data missing point in the historical power utilization data are combined to conduct prediction to obtain a prediction result, and the prediction result is used as a missing correction data point; And integrating all corrected data points with uncorrected original data points to generate power utilization time sequence data.
3. The method for anomaly identification of electricity consumption data according to claim 1, wherein the performing multi-scale time sequence feature extraction and prediction on the historical electricity consumption data according to a preset P-Timemixer algorithm to generate prediction data comprises: Performing convolution operation step by step on the historical electricity utilization data to generate downsampled data of each time scale; applying a channel attention mechanism to each downsampled data respectively to generate corresponding multi-scale feature data weighted by each channel; Performing fast Fourier transform on each multi-scale characteristic data, selecting a plurality of frequency domain components with highest amplitude, performing zero filling expansion on a time sequence segment corresponding to each frequency domain component according to a period corresponding to each frequency domain component, and generating a corresponding expansion sequence of each frequency domain enhancement; performing imaging operation on the spreading sequences corresponding to the frequency domain components according to channel dimensions, generating corresponding multi-dimensional image data, and respectively extracting single-period features and multi-period features of each scale; Carrying out feature convolution fusion on each frequency domain component from a high scale to a low scale to obtain a single-period fusion feature, carrying out feature convolution fusion on each multi-period feature corresponding to each frequency domain component from a low scale to a high scale to obtain a multi-period fusion feature, and combining the single-period fusion feature and the multi-period fusion feature to obtain a fusion feature corresponding to the frequency domain component; And weighting each fusion feature based on the normalized amplitude weight of each frequency domain component to obtain a weighted fusion feature, and restoring the weighted fusion feature to the data dimension of the historical electricity utilization data to generate prediction data.
4. The method for anomaly identification of power consumption data according to claim 1, wherein the performing bidirectional alignment fusion on the power consumption time sequence data and the prediction data to obtain fused power consumption data comprises: And based on each data item channel of the power utilization time sequence data and the prediction data, respectively aligning and combining the power utilization time sequence data and the prediction data according to the channels to obtain the fusion power utilization data, wherein if the power utilization time sequence data and the prediction data simultaneously have data values at the same sampling time point, the data value of the power utilization time sequence data at the sampling time point is used as the data value of the fusion power utilization data.
5. The anomaly identification method for power consumption data according to claim 1, wherein the calculating a preset deviation index based on the combined power consumption data and the power flow calculation data of the power distribution network, and determining whether the corresponding data point in the combined power consumption data is anomaly data according to the deviation index, and outputting an anomaly identification result comprises: calculating the normalized deviation of the measured value of each data point in the fusion power consumption data and the power flow calculated value in the corresponding power flow calculated data of the power distribution network, and generating a power flow deviation index; And judging whether the corresponding data point is abnormal data or not based on each tide deviation index and a preset first deviation threshold value, and outputting an abnormal recognition result.
6. The anomaly identification method for power consumption data according to claim 1, wherein the calculating a preset deviation index based on the combined power consumption data and the power flow calculation data of the power distribution network, and determining whether the corresponding data point in the combined power consumption data is anomaly data according to the deviation index, and outputting an anomaly identification result, further comprises: calculating the normalized deviation of the measured value of each data point in the fusion power consumption data and the power flow calculated value in the corresponding power flow calculated data of the power distribution network, and generating a power flow deviation index; calculating a fluctuation index according to the difference value and the historical standard deviation of each data point in the fusion power consumption data and the historical synchronous value of each data point in the historical power consumption data; calculating the reciprocal of the average reachable distance between each data point in the fusion power consumption data and a plurality of nearest neighbor data points based on a local reachable density method, and generating a local density index; based on the current deviation index, the fluctuation index and the local density index of each data point in the fusion power consumption data, whether each data point is abnormal data or not is judged, and an abnormal recognition result is output.
7. The abnormal identification system for the electricity consumption data is characterized by comprising a data acquisition module, a correction module, a prediction module, an alignment fusion module and an output module; The data acquisition module is used for acquiring current original power utilization time sequence data of a plurality of loads in a target area and acquiring corresponding historical power utilization data and tide calculation data; the correction module is used for correcting the original power utilization time sequence data based on the historical power utilization data to obtain power utilization time sequence data; The prediction module is used for extracting and predicting the multi-scale time sequence characteristics of the historical power utilization data according to a preset P-Timemixer algorithm to generate prediction data, wherein the data density of the prediction data is larger than that of the power utilization time sequence data; the alignment fusion module is used for carrying out bidirectional alignment fusion on the power utilization time sequence data and the prediction data to obtain fusion power utilization data; the output module is used for calculating a preset deviation index based on the fusion power consumption data and the power flow calculation data of the power distribution network, judging whether corresponding data points in the fusion power consumption data are abnormal data or not according to the deviation index, and outputting an abnormal recognition result.
8. The anomaly identification system of electricity consumption data of claim 7, wherein the correction module comprises an offset correction unit, a jump correction unit, a missing supplementary unit and an integration unit; the offset correction unit is configured to calculate an offset compensation amount according to a historical synchronous average value of an original data point and a synchronous actual measurement value of a similar load if the original data point is determined to have full-period offset in the original power consumption time sequence data, and calculate a difference value between the original value of the original data point and the offset compensation amount as an offset correction data point; The jump correction unit is used for interpolating according to the adjacent data of a certain original data point to obtain a jump correction value as a jump correction data point if the abnormal jump exists in the original power utilization time sequence data; The missing supplementary unit is used for predicting a predicted result by combining the original power utilization time sequence data and a historical data value of the data missing point in the historical power utilization data if the data missing point exists in the original power utilization time sequence data, and taking the predicted result as a missing correction data point; the integration unit is used for integrating all corrected data points and uncorrected original data points to generate power utilization time sequence data.
9. A computer program product comprising a computer program or instructions which, when executed, implements a method of anomaly identification of electricity usage data according to any one of claims 1 to 6.
10. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements an anomaly identification method for electricity consumption data according to any one of claims 1 to 6.

Description

Anomaly identification method, system, medium and product of electricity consumption data Technical Field The invention belongs to the field of power data anomaly identification, and relates to an anomaly identification method, system, medium and product of power data. Background Under the trend of high-proportion access of new energy and intelligent development of an industrial park, the power distribution network has increasingly-increased fine monitoring demands on multi-source power consumption loads, and high-reliability management needs to be carried out on the time data such as voltage, current, power and the like so as to support regulation and control and state evaluation. The method has certain effectiveness on the premise of complete data sampling, stable transmission and higher equipment precision, but is difficult to effectively distinguish distortion of the data and abnormality of actual electricity consumption behaviors when facing to the composite quality problems of sampling sparsity, transmission delay, systematic offset and the like commonly existing in low-cost acquisition equipment, so that the robustness of an abnormality identification result is insufficient, and the severe requirement of monitoring the data quality of a high-reliability power distribution network is difficult to meet. Disclosure of Invention The application provides an anomaly identification method, system, medium and product of electricity consumption data, which can solve the problem that the anomaly of the electricity consumption data is difficult to accurately identify under low-cost acquisition equipment in the prior art. In order to achieve the above object, in a first aspect, the present invention provides an anomaly identification method for electricity consumption data, including: collecting current original power utilization time sequence data of a plurality of loads in a target area, and obtaining corresponding historical power utilization data and tide calculation data; correcting the original power utilization time sequence data based on the historical power utilization data to obtain power utilization time sequence data; Performing multi-scale time sequence feature extraction and prediction on the historical electricity utilization data according to a preset P-Timemixer algorithm to generate prediction data, wherein the data density of the prediction data is greater than that of the electricity utilization time sequence data; performing bidirectional alignment fusion on the power utilization time sequence data and the prediction data to obtain fusion power utilization data; And calculating a preset deviation index based on the fusion power consumption data and the power flow calculation data of the power distribution network, judging whether corresponding data points in the fusion power consumption data are abnormal data according to the deviation index, and outputting an abnormal recognition result. Compared with the prior art, the embodiment of the application has the beneficial effects that the current original power utilization time sequence data of a plurality of loads in a target area are collected, the historical power utilization data and the power flow calculation data are synchronously acquired, a multi-source information input foundation which covers time evolution, equipment association and a power grid physical model is constructed for abnormality identification, and a discrimination blind area caused by a single data source is avoided; the method comprises the steps of correcting original electricity time sequence data based on historical electricity data, restraining systematic deviation, local jump, transmission loss and other data distortion caused by low-cost acquisition equipment, improving the intrinsic consistency and reliability of input data, carrying out multi-scale feature extraction and prediction on the historical electricity data according to a preset time sequence prediction algorithm to generate predicted data with higher data density, compensating for the insufficient time resolution caused by sparse original sampling, enhancing time sequence continuity, carrying out bidirectional alignment fusion on the corrected electricity time sequence data and the high-density predicted data, compensating for the empty time granularity while keeping the high-confidence data, forming fused electricity data with high integrity and high time precision, calculating deviation indexes based on the fused electricity data and power flow calculation data of a power distribution network, judging abnormality according to the deviation indexes, taking physical constraints (such as power balance and voltage limit) of the power network operation as abnormal judgment basis, enabling abnormal recognition results to reflect whether the data violates the basic operation rule of the power network, forming a processing chain of 'trusted correction-high-density prediction-multi-source fusion-physical cons