CN-122022001-A - New energy power prediction method based on meteorological big model

CN122022001ACN 122022001 ACN122022001 ACN 122022001ACN-122022001-A

Abstract

The invention discloses a new energy power prediction method based on a meteorological large model, which particularly relates to the field of new energy power prediction and edge network data processing, and comprises the steps of carrying out time difference on continuous data comprising wind speed, irradiance, temperature, humidity, air pressure, electric load and equipment state reported by a plurality of stations to generate time change fragments, counting sampling frequency changes to form sampling rhythm fragments, identifying spatial association relation of the sampling rhythm fragments to obtain spatial association fragments and folding the spatial association fragments into an original structure sequence.

Inventors

CHEN ZHENGJIAN
SHI LEI
HE YIN
DENG ZIXIAO
LIU FULING
MA JUNLONG
LI QIU
Yu Shuran
WANG YAXIN
XUE ZHIXIN
Xiong Jinzhong
Chen Haiao
XU CHENBO
Liang Shanheng

Assignees

深能智慧能源科技有限公司
深能北方能源控股有限公司
深能南京能源控股有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (8)

1. A new energy power prediction method based on a meteorological large model is characterized by comprising the following steps: S1, generating time variation fragments by carrying out time difference on continuous data comprising wind speed, irradiance, temperature, humidity, air pressure, electric load and equipment state, which are reported by a plurality of stations, and then counting sampling frequency variation on the time variation fragments to form sampling rhythm fragments; S2, comparing the original structure sequence with an uploading time sequence generated by an edge node when uploading the original structure sequence to identify a time hopping fragment, detecting a sampling breakpoint of the time hopping fragment to form a sampling breakpoint fragment, and fitting a modal change to the sampling breakpoint fragment to generate a modal change fragment; s3, resolving an offset correction amount from the offset track sequence, and performing reverse convolution calculation on the offset correction amount, the time-varying segment, the sampling rhythm segment and the space-related segment to obtain a deconvolution segment; S4, inputting the correction structure sequence into a meteorological large model to solve characteristic response so as to generate a model response segment, comparing the model response segment with the correction structure sequence so as to identify a consistency deviation segment, and then executing structure recombination calculation on the consistency deviation segment so as to form a consistency sequence; s5, solving a power prediction result based on the consistency sequence, then carrying out waveform coupling on the power prediction result and the correction structure sequence to solve a power offset segment, and finally solving a risk quantity on the power offset segment to output a risk constraint quantity.
2. The new energy power prediction method based on the meteorological large model of claim 1, wherein in S1, the new energy power prediction method further comprises the steps of obtaining wind speed, irradiance, temperature, humidity, air pressure, electric load and equipment state continuously reported by a plurality of stations, performing difference calculation on data records of adjacent moments, and performing sequence splicing operation on all the difference values according to time sequence to form a time variation segment; Performing adjacent comparison calculation on original time marks corresponding to the data difference values forming the time variation segment, and counting time intervals between the adjacent time marks to form a time interval sequence; performing corresponding comparison calculation on the sampling rhythm segments and the space coordinates of all stations, and calculating the space association score by calculating the distance between stations, the similarity of the direction difference value and the rhythm variation sequence and performing weighted combination calculation; And performing time line alignment operation on the spatial correlation score and the rhythm variation sequence recorded in the spatial correlation fragment and the time sequence, performing multidimensional folding calculation according to a preset folding rule, and compressing and merging the aligned structures segment by segment to construct an original structure sequence.
3. The new energy power prediction method based on the meteorological large model of claim 2, wherein in S2, the method further comprises the steps of performing point-by-point comparison calculation on a time mark of each structural point in an original structural sequence and an uploading time sequence formed by uploading time synchronously recorded by an edge node when uploading the original structural sequence according to the sequence of the structural points, and obtaining a time difference variation sequence by calculating time differences of adjacent structural points and performing adjacent difference variation calculation on the time differences; And executing threshold comparison operation on the time difference variation sequence to judge whether the time difference variation exceeds a variation threshold value: when any time difference variation exceeds the variation threshold, the structural points of which the time difference variation exceeds the variation threshold are formed into time jump candidate fragments according to the original sequence; respectively taking the time hopping candidate fragments and the non-hopping processing sequences as sequences to be detected, executing sequence comparison calculation of adjacent structural points on each sequence to be detected, and determining continuity of the sequence by judging whether the time marks of the adjacent structural points are in reverse sequence and hop number or not: And when the reverse sequence or the jump number does not occur, outputting the whole time jump candidate fragments or the jump-free processing sequences as continuous fragments.
4. The method for predicting new energy power based on meteorological large model of claim 3, wherein in S2, time jump segments and continuous segments are used as segments to be detected respectively, time difference comparison calculation of adjacent structural points is executed for each segment to be detected, and whether sampling break points are formed is determined by judging whether the time difference of the adjacent structural points exceeds the sampling rhythm range or not: When the time difference is beyond the sampling rhythm range, the exceeding points form sampling breakpoint fragments, and when all the time differences are not beyond the sampling rhythm range, the continuous fragments form candidate sampling breakpoint fragments; and performing modal change fitting operation on the sampling breakpoint fragments and the candidate sampling breakpoint fragments, generating modal change fragments by fitting the trend of the structural points changing along with time, performing discretization operation on the modal change fragments according to the sequence of the structural points, and combining the discrete structural points according to the time sequence to form an offset track sequence.
5. The method of claim 4, wherein in S3, the method further comprises performing adjacent structure point differential operation on structure points in the offset track sequence, and solving an offset correction amount by performing accumulated deduction on the offset of the differential result according to the sequence of the structure points; and respectively executing structural point corresponding comparison operation on the offset correction quantity and the time variation segment, the sampling rhythm segment and the space correlation segment, and determining a reverse convolution path by judging whether the difference value of the offset correction quantity at the structural points corresponding to the time variation segment, the sampling rhythm segment and the space correlation segment is within the upper limit and the lower limit range of the consistency threshold value: When any difference value is not in the upper limit range and the lower limit range of the consistency threshold, split compensation operation is carried out on the offset correction amount according to the type of the structural fragment, and segmented reverse convolution calculation is carried out on the offset correction amount after compensation, the time variation fragment, the sampling rhythm fragment and the space correlation fragment respectively, so that segmented reverse convolution fragments are generated; The synchronous deconvolution fragments and the segmented deconvolution fragments are recombined into deconvolution fragments according to the sequence of structural points, structural point position alignment operation is carried out on the deconvolution fragments, difference value calculation is carried out on deconvolution values positioned at the same structural point position in the deconvolution fragments, deconvolution values with difference values being at the lower limit of a threshold value are selected, and cross fusion operation is carried out, so that deconvolution fusion fragments are generated; And performing structural reconstruction operation on the deconvolution fusion segment, generating a reconstruction segment by performing continuity correction and rhythm correction on the structural points, performing back projection correction operation on the reconstruction segment and the offset correction amount, and generating a corrected structural sequence by performing compensation operation on the offset of the structural points.
6. The new energy power prediction method based on the meteorological large model according to claim 5, wherein in S4, the new energy power prediction method based on the meteorological large model further comprises the steps of constructing an input tensor according to the sequence of structural points by performing embedded mapping and sequence derivation on the structural points of the input tensor, inputting the mapped correction structural sequence into the meteorological large model to perform feature solving operation, constructing a response mapping sequence, and generating model response fragments according to the sequence of structural points by the response mapping sequence; Performing structural point corresponding comparison calculation on the model response fragment and the correction structural sequence, constructing a deviation mapping sequence by calculating a corresponding structural point differential sequence, and judging whether any differential in the deviation mapping sequence exceeds the upper limit and the lower limit of a deviation threshold value or not: Executing breakpoint segmentation and difference foldback processing when any difference exceeds the upper limit and the lower limit of a deviation threshold, namely executing differential comparison of adjacent structural points on the deviation mapping sequence according to the sequence of the structural points, taking the structural points exceeding the upper limit and the lower limit of the deviation threshold as segmentation points, dividing the deviation mapping sequence according to the segmentation points in a front-back sequence to form a plurality of deviation subsequences, executing threshold foldback operation on the difference exceeding the upper limit and the lower limit of the deviation threshold in the deviation subsequences, and reversely subtracting the excess quantity from the excess quantity to enable the foldback difference to be in the upper limit and the lower limit of the deviation threshold; And when all the differences are within the upper and lower limits of the deviation threshold, performing continuous segment extraction processing, namely sequentially combining the structural points continuously within the upper and lower limits of the deviation threshold to form continuous deviation segments by sequentially performing differential inspection on the deviation mapping sequence according to the structural points.
7. The method of claim 6, wherein S4 further comprises reorganizing the folded sub-sequence of deviations and the continuous deviation segments into deviation processing segments according to the sequence of structural points, performing structural point reconstruction operation on the deviation processing segments, and performing accumulation operation on the difference between the derived position of the previous structural point and the current structural point to form a sequence of position values of the current structural point; And then, executing stability judgment on the reconstructed sequence, and determining whether the structural points meet the stability condition by judging whether the difference between the difference of the structural points and the difference of the adjacent structural points is within the upper limit and the lower limit of a stability threshold value or not: Locking the position of the structural point when the structural point meets the stability condition, and performing position value adjustment operation on the deduced position value of the structural point when the structural point does not meet the stability condition to form a deviation recombination segment; the deviation recombination fragments and the correction structure sequence are sequentially rearranged and continuously connected according to the structure points to generate a consistency sequence, secondary consistency check operation is carried out on the consistency sequence, and whether the differences corresponding to the structure points are all within the upper limit range and the lower limit range of a consistency threshold value is judged: And when any difference is not in the upper limit and the lower limit of the consistency threshold, performing backtracking compensation operation on the structure points with the deviation to generate a final consistency sequence.
8. The method of claim 7, wherein in S5, the method further comprises performing power derivation operations on the consistency sequence in order of structure points, generating a base power value by extracting a numerical component for each structure point and performing power mapping calculations, and performing differential expansion operations on the base power values of adjacent structure points, and constructing an expanded power amount by overlapping the differential to the current base power value; performing cumulative recursion operation on the extended power quantity according to the sequence of the structural points to form a power push chain, generating a power prediction result by the power push chain according to the sequence of the structural points, performing waveform coupling operation on the power prediction result and the correction structure sequence according to the sequence of the structural points, constructing a coupling differential chain by performing subtraction operation on the power prediction result of the corresponding structural points and a reference power value and combining according to the sequence of the structural points, and judging whether the coupling differential exceeds the upper limit and the lower limit of an offset threshold value or not: When the coupling difference exceeds the upper limit and the lower limit of the offset threshold, the corresponding structure points are subjected to offset point calibration operation and form an offset calibration segment; combining the offset calibration segment and the continuous coupling segment according to the sequence of the structural points to form a power offset segment, executing fluctuation difference calculation on the power offset segment according to the sequence of the structural points to form a fluctuation segment, and judging whether the difference between a fluctuation value and an adjacent fluctuation value exceeds the upper limit and the lower limit of a fluctuation threshold value or not: When the fluctuation difference value exceeds the upper limit and the lower limit of the fluctuation threshold, carrying out fluctuation turn-back operation on the corresponding structural points and carrying out position value correction operation on the deduced position values of the corresponding structural points; combining the structural points subjected to foldback or recursion treatment according to the sequence of the structural points to generate a risk deduction input segment; and performing risk deduction calculation on the risk deduction input fragments according to the sequence of the structural points, solving the risk amount by performing fluctuation recursion and offset accumulation operation on the structural points, and performing continuity correction on the risk amount according to the sequence of the structural points to output a risk constraint amount.

Description

New energy power prediction method based on meteorological big model Technical Field The invention relates to the technical field of new energy power prediction and edge network data processing, in particular to a new energy power prediction method based on a meteorological large model. Background In the current new energy power prediction flow, various meteorological and operation data enter an edge node at a station side, are sent to a cloud end through an edge network, are uniformly inferred by a meteorological large model, and finally provide a risk assessment basis for a dispatching system; However, the real running environment is far more complicated than the surface flow, because different stations, different manufacturer devices, different sampling periods and a plurality of network links are mutually independent, the data such as wind speed, irradiation, temperature, humidity and pressure often generate inconspicuous structural deformation in the edge network transmission process, the time stamp can be rearranged, the sampling rhythm can be disordered, the network queuing leads to the delayed arrival of partial data, the multi-mode weather data are nominally synchronous and actually misplaced due to inconsistent updating frequency, the deformation can not lead single data to seem abnormal, but after converging to the cloud, the original physical sequence, causal association and changing rhythm of the single data are destroyed, thus forming a mixed weather input which looks intact but structurally distorted, the cloud weather large model is difficult to recover the real barely evolved path from the sequence after distortion even if the structure is advanced, only can give a power prediction of strong reasonable but intrinsic distortion, the original evaluation system on the dispatching side directly depends on the predictions to judge the key indexes such as power fluctuation risk, safety margin and the like, and the structure of the edge of the power grid is further amplified and judged as the risk of the distortion in the network is further amplified; Therefore, the contradiction in the current system is not the insufficient precision of the data itself, but the data is silently disturbed when the data is transmitted by the edge network, so that the large model sees the data which is not the original real data, and the situation that the structure is distorted becomes the most urgent problem to be solved in the current new energy power prediction. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides a new energy power prediction method based on a meteorological large model, which is used for restoring the meteorological and operation data entering the meteorological large model into a consistent structure in three dimensions of time, rhythm and space by constructing a multi-stage structure processing chain from uploading offset identification and structural consistency reconstruction to model response calibration and risk constraint derivation of an edge network, and outputting risk constraint quantity capable of being used for scheduling side risk assessment according to the weather and operation data, so as to solve the problem that data structure dislocation caused by the edge network in the background technology cannot be identified and repaired by the model. In order to achieve the purpose, the invention provides the following technical scheme that the new energy power prediction method based on the meteorological big model comprises the following steps: S1, generating time variation fragments by carrying out time difference on continuous data comprising wind speed, irradiance, temperature, humidity, air pressure, electric load and equipment state, which are reported by a plurality of stations, and then counting sampling frequency variation on the time variation fragments to form sampling rhythm fragments; S2, comparing the original structure sequence with an uploading time sequence generated by an edge node when uploading the original structure sequence to identify a time hopping fragment, detecting a sampling breakpoint of the time hopping fragment to form a sampling breakpoint fragment, and fitting a modal change to the sampling breakpoint fragment to generate a modal change fragment; s3, resolving an offset correction amount from the offset track sequence, and performing reverse convolution calculation on the offset correction amount, the time-varying segment, the sampling rhythm segment and the space-related segment to obtain a deconvolution segment; S4, inputting the correction structure sequence into a meteorological large model to solve characteristic response so as to generate a model response segment, comparing the model response segment with the correction structure sequence so as to identify a consistency deviation segment, and then executing structure recombination calculation on the consistency deviation segment so a