CN-122000857-A - Reactive power metering method suitable for nonlinear load

CN122000857ACN 122000857 ACN122000857 ACN 122000857ACN-122000857-A

Abstract

The invention discloses a reactive power metering method suitable for nonlinear loads, which belongs to the technical field of power grid metering and comprises the steps of obtaining all-condition electrical data, obtaining a load harmonic characteristic vector data set based on the all-condition electrical data, building a load classification model based on the load harmonic characteristic vector data set, obtaining a real-time characteristic vector based on current operation data, obtaining a load type based on the real-time characteristic vector and the load classification model, inputting the real-time characteristic vector into a dedicated load correction model to obtain equivalent circuit parameters, obtaining a virtual optimal circuit structure based on the equivalent circuit parameters, obtaining a high-precision reactive power value, obtaining a general reactive power value based on the real-time characteristic vector and the general reactive power correction model, and obtaining a nonlinear load reactive power calculation result based on the high-precision reactive power value and the general reactive power value. The reactive power metering method suitable for the nonlinear load disclosed by the invention realizes the accuracy and stability of the reactive power metering of the nonlinear load.

Inventors

ZHANG BENSONG
LI SHIJIE
HU JU
DU JINYANG

Assignees

中国南方电网有限责任公司

Dates

Publication Date: 20260508
Application Date: 20251202

Claims (10)

1. A reactive metering method suitable for nonlinear loads, comprising: Acquiring all-condition electrical data, and acquiring a load harmonic characteristic vector data set based on the all-condition electrical data and a preset harmonic characteristic extraction algorithm; Constructing a load classification model based on the load harmonic feature vector data set; Acquiring current operation data, and acquiring a real-time feature vector based on the current operation data and a preset load harmonic feature extraction algorithm; Acquiring a load type based on the real-time feature vector and a load classification model; calling a dedicated load correction model based on the load type and a preset load correction library, and inputting the real-time feature vector into the dedicated load correction model to obtain equivalent circuit parameters; acquiring a virtual optimal circuit structure based on the equivalent circuit parameters, and acquiring a high-precision reactive power value based on the virtual optimal circuit structure and a real-time feature vector; Acquiring a general reactive value based on the real-time feature vector and a general reactive correction model; And acquiring a nonlinear load reactive power calculation result based on the high-precision reactive power value and the universal reactive power value.
2. The reactive power metering method for nonlinear loads according to claim 1, wherein the acquiring all-condition electrical data, acquiring a load harmonic feature vector dataset based on the all-condition electrical data and a preset harmonic feature extraction algorithm, comprises: Acquiring all-condition electrical data based on a preset all-condition test scheme; the full-working-condition electrical data comprise a test load type, an original working-condition waveform and a true reactive power; acquiring an original characteristic vector based on the original working condition waveform; Performing quality evaluation based on the original feature vector and a preset isolated forest algorithm to obtain an anomaly score; Performing abnormal elimination based on the abnormal score and a preset abnormal threshold value to obtain a preprocessing working condition waveform; Zero crossing point detection is carried out based on the preprocessing working condition waveform, and a discrete period sample is obtained; acquiring an original material library based on the discrete period samples, the test load types and the true reactive power; and acquiring a load harmonic feature vector data set based on the original material library and a preset harmonic feature extraction algorithm.
3. The reactive power metering method suitable for nonlinear load according to claim 2, wherein the obtaining a load harmonic feature vector data set based on the original material library and a preset harmonic feature extraction algorithm comprises: acquiring extracted voltage and current waveform data, a load type tag and a true reactive power tag based on the original material library; Performing Fourier transformation based on the voltage and current waveform data to obtain fundamental wave characteristics and harmonic wave characteristics; Acquiring a power energy characteristic, a time domain waveform characteristic and a time domain analysis characteristic based on the voltage and current waveform data; Acquiring an initial load harmonic characteristic vector based on the fundamental wave characteristic, the harmonic characteristic, the power energy characteristic, the time domain waveform characteristic and the time domain analysis characteristic; carrying out principal component analysis and structure standardization based on the initial load harmonic characteristic vector to obtain a load harmonic characteristic vector; And acquiring a load harmonic characteristic vector data set based on the load harmonic characteristic vector, the load type tag and the true reactive power tag.
4. A reactive metering method for nonlinear loads according to claim 1, wherein said constructing a load classification model based on said load harmonic feature vector dataset comprises: constructing a training set and a verification set based on the load harmonic feature vector data set; constructing an original load classification model comprising a plurality of parallel load identification paths based on a preset target load type and a training set; acquiring an activation score based on a plurality of the parallel load identification paths and the verification set; judging whether the preset path identification condition is met or not based on the activation score and a preset consistency checking mechanism, and acquiring an identification judgment result; Performing super-parameter optimization on a plurality of parallel load identification paths based on the identification judgment result to acquire a current load identification path; and acquiring a load classification model based on the current load identification path.
5. A reactive metering method for nonlinear loads according to claim 1, wherein said obtaining load types based on said real-time eigenvectors and load classification model comprises: acquiring a path activation score based on the real-time feature vector and a load classification model; Acquiring load type probability distribution based on the activation score, and acquiring a current load classification result based on the load type probability distribution; Carrying out dynamic confidence assessment based on the load type probability distribution to obtain a confidence result; and when the preset high confidence condition is met based on the confidence result and a preset confidence threshold value, taking the current load classification result as a load type.
6. A reactive metering method for nonlinear loads according to claim 5, wherein said obtaining load types based on said real-time eigenvectors and load classification model comprises: when the confidence result and the preset confidence threshold value are judged to meet the preset low confidence condition, triggering a low confidence processing mechanism: acquiring a historical classification result of the real-time feature vector; performing time sequence voting based on the historical classification result to obtain a voting classification result; when the voting classification result is judged to meet a preset stable condition based on the voting classification result, the voting classification result is used as the load type; And when the voting classification result is judged to be not in accordance with a preset stable condition, marking the load type as an unknown type.
7. A reactive metering method for nonlinear loads according to claim 5, wherein said obtaining load types based on said real-time eigenvectors and load classification model comprises: When the confidence coefficient result and the preset confidence coefficient threshold value are judged to meet the preset secondary verification condition, triggering a secondary verification mechanism: acquiring load behavior change data based on a preset micro-excitation action; calculating the matching degree based on the load behavior change data and a reference physical fingerprint library corresponding to the current load classification result, and obtaining a matching result; And when the matching result is judged to be in accordance with a preset matching condition, taking the current load classification result as a load type, and updating the reference physical fingerprint library based on the real-time feature vector to perform reactive power metering next time.
8. A reactive metering method for nonlinear loads as claimed in claim 7, comprising: When the matching result is judged to be not in accordance with the preset matching condition, triggering a cross-category comparison flow to obtain a cross-category comparison result; When the cross-class comparison result is judged to meet a preset novel judgment condition, the load type is a novel load suspected sample, the load classification model is optimized and updated based on the load behavior change data, and an optimized load classification model is obtained so as to perform next reactive power metering based on the optimized load classification model.
9. A reactive power metering method for nonlinear loads according to claim 1, wherein obtaining nonlinear load reactive power calculation results based on the high-precision reactive power value and the general reactive power value comprises: acquiring a relative difference value based on the high-precision reactive power value and the universal reactive power value; when the relative difference value and the preset dynamic threshold value are judged to be not in accordance with the preset correction condition, the high-precision reactive power value marked with high confidence coefficient is used as a nonlinear load reactive power calculation result; When the relative difference value and the preset dynamic threshold value are judged to meet the preset correction condition, triggering a result credibility analysis mechanism: Acquiring adjacent period metering data, taking the general reactive value subjected to state marking to be checked as a nonlinear load reactive calculation result when the adjacent period metering data are judged to be not in accordance with a preset adoption condition, and carrying out parameter updating on the exclusive load correction model based on the general reactive value; And when the adjacent period metering data is judged to meet the preset adoption condition, the high-precision reactive power value subjected to the medium confidence labeling is used as a nonlinear load reactive power calculation result.
10. A reactive metering method for nonlinear loads as recited in claim 1, further comprising: Acquiring reactive power metering stable data in real time; acquiring classification confidence coefficient based on the nonlinear load reactive power calculation result; Acquiring an abnormal diagnosis result based on a preset abnormal database, the reactive power metering stable data and the classification confidence; Triggering an optimization alarm mechanism when the abnormal diagnosis result is judged to meet a preset optimization condition: semi-automatic labeling is carried out based on the current operation data, the real-time feature vector and a preset physical rule base, and a high-quality optimized data set is obtained; performing physical constraint incremental learning on the load classification model based on the high-quality optimized data set to obtain a target load classification model so as to perform reactive power metering next time based on the target load classification model; and carrying out dynamic network expansion on the exclusive load correction model based on the high-quality optimized data set to obtain a target load correction model so as to carry out next reactive power metering based on the target load correction model.

Description

Reactive power metering method suitable for nonlinear load Technical Field The invention relates to the technical field of power grid metering, in particular to a reactive power metering method suitable for nonlinear loads. Background With the wide application of nonlinear loads such as power electronic equipment in a power system, the generated harmonic waves and distortion currents form a serious challenge for the traditional reactive power metering method. When dealing with such nonlinear loads, the traditional metering theory based on sine wave assumption and the fixed compensation algorithm depending on Fourier transformation generally cannot accurately separate reactive components of fundamental waves and subharmonics, and the unique nonlinear characteristics of different loads can be ignored, so that the technical problem of remarkable reduction of metering precision is caused. Although some existing intelligent metering methods attempt to adopt a machine learning model for correction, most of the intelligent metering methods rely on a general 'black box' model, lack of consideration on a load physical mechanism, have insufficient generalization capability and poor interpretation performance when facing unknown or complex-characteristic loads, and cannot form a continuously optimized closed loop. Therefore, developing a high-precision reactive power metering method capable of accurately identifying load types, adaptively correcting errors and having continuous learning capability has become a key requirement for smart grid fine metering and management. Disclosure of Invention The invention provides a reactive power metering method suitable for nonlinear load, which can solve the technical problem that metering accuracy is obviously reduced when the load with unknown or complex characteristics is faced in the prior art, and realize the accuracy and stability of nonlinear load reactive power metering. The invention provides a reactive power metering method suitable for nonlinear load, comprising the following steps: Acquiring all-condition electrical data, and acquiring a load harmonic characteristic vector data set based on the all-condition electrical data and a preset harmonic characteristic extraction algorithm; Constructing a load classification model based on the load harmonic feature vector data set; Acquiring current operation data, and acquiring a real-time feature vector based on the current operation data and a preset load harmonic feature extraction algorithm; Acquiring a load type based on the real-time feature vector and a load classification model; calling a dedicated load correction model based on the load type and a preset load correction library, and inputting the real-time feature vector into the dedicated load correction model to obtain equivalent circuit parameters; acquiring a virtual optimal circuit structure based on the equivalent circuit parameters, and acquiring a high-precision reactive power value based on the virtual optimal circuit structure and a real-time feature vector; Acquiring a general reactive value based on the real-time feature vector and a general reactive correction model; And acquiring a nonlinear load reactive power calculation result based on the high-precision reactive power value and the universal reactive power value. The reactive power metering method suitable for the nonlinear load provided by the invention is characterized in that harmonic characteristic vectors are extracted through acquiring all-condition electrical data, a load classification model is constructed to identify the current load type, and then an exclusive load correction model is called to obtain equivalent circuit parameters so as to calculate high-precision reactive power and combine a general reactive power value to obtain a final result. Further, the acquiring the all-condition electrical data, and acquiring the load harmonic feature vector data set based on the all-condition electrical data and a preset harmonic feature extraction algorithm includes: Acquiring all-condition electrical data based on a preset all-condition test scheme; the full-working-condition electrical data comprise a test load type, an original working-condition waveform and a true reactive power; acquiring an original characteristic vector based on the original working condition waveform; Performing quality evaluation based on the original feature vector and a preset isolated forest algorithm to obtain an anomaly score; Performing abnormal elimination based on the abnormal score and a preset abnormal threshold value to obtain a preprocessing working condition waveform; Zero crossing point detection is carried out based on the preprocessing working condition waveform, and a discrete period sample is obtained; acquiring an original material library based on the discrete period samples, the test load types and the true reactive power; and acquiring a load harmonic feature vector data set based on the original ma