CN-121997721-A - Pilot bifurcation prediction model construction method and application

CN121997721ACN 121997721 ACN121997721 ACN 121997721ACN-121997721-A

Abstract

The invention belongs to the technical field of pilot bifurcation prediction, and discloses a pilot bifurcation prediction model construction method and application, wherein the method comprises the steps of obtaining multi-scale discharge data in a pilot discharge process; dividing a pilot development path into a series of space-time fragments along the axial movement direction of the pilot development path based on multi-scale discharge data, acquiring multi-dimensional physical characteristics of each space-time fragment including pilot head injection charge quantity, pilot head background electric field intensity, pilot head deflection angle and voltage change rate to construct a training sample set, training a neural network model by using the training sample set, merging physical rules as trend constraints into training, and taking the physical rules as a pilot bifurcation probability prediction model when the model converges. The invention can overcome the limitation that a single threshold criterion in the traditional simulation depends on experience setting and cannot characterize a multidimensional physical quantity coupling driving mechanism, and provides a bifurcation judging mechanism which combines physical certainty and randomness for pilot discharge simulation.

Inventors

GUO LILI
HE JUNJIA
ZHAO XIANGEN
WANG XIANKANG
LIU XIAOPENG
YE ZHI
MAO SIBO

Assignees

华中科技大学

Dates

Publication Date: 20260508
Application Date: 20260105

Claims (10)

1. The method for constructing the pilot bifurcation prediction model is characterized by comprising the following steps of: Acquiring multi-scale discharge data in a pilot discharge process, wherein the multi-scale discharge data comprises space-time evolution image data, discharge current waveforms and voltage waveforms of pilot channel development; Dividing a pilot development path synthesized by the space-time evolution image data into a series of space-time fragments along the axial movement direction of the space-time fragments, and acquiring multidimensional physical characteristics of each space-time fragment, including pilot head injection charge quantity, pilot head background electric field intensity, pilot head deflection angle and voltage change rate, based on space-time coordinates of each space-time fragment, the discharge current and the voltage; And training the neural network model by using the training sample set, wherein when the neural network model converges, the neural network model is used as a pilot bifurcation probability prediction model.
2. The method of claim 1, wherein the neural network model comprises: the feature embedding layer is used for mapping the input multidimensional physical features into a feature token sequence containing physical semantic intensity; the feature attention module is used for capturing nonlinear coupling relations of the fractional contributions among different physical features in the feature token sequence by adopting a multi-head attention mechanism, and extracting nonlinear feature vectors h feat containing high-order semantic information through aggregation operation; The linear mapping module is used for carrying out linear mapping on the input multidimensional physical characteristics so as to extract linear characteristic vectors h res among different physical characteristics; The decision output module is used for splicing and fusing the nonlinear feature vector h feat and the linear feature vector h res and outputting predicted leading bifurcation probability through an activation function 。
3. The method for constructing the leading bifurcation predicting model according to claim 2, wherein the feature embedding layer maps the input multidimensional physical feature into a feature token sequence containing physical semantic strength, comprising: Injecting the pilot header into the charge Strength of background electric field of leading head Deflection angle of pilot head Rate of change of voltage Respectively mapped to physical semantic intensity values in the intervals of [0,1] 、、、 ; Will physical semantic intensity value 、、、 And splicing in the channel dimension, and performing linear projection to obtain the characteristic token sequence.
4. A method of constructing a model of a pilot bifurcation prediction according to claim 3, wherein the loss function used in training the neural network model comprises the pilot bifurcation probability predicted by the neural network model Predictive loss with tags and physical coherence constraint loss function The physical consistency constraint loss function The method comprises the following steps: Wherein, the Representing the calculation of the pearson correlation coefficient.
5. The method for constructing a pilot bifurcation prediction model according to any one of claims 1-4, wherein the space-time coordinates of each space-time segment are calculated by: Based on the discharge current, identifying a lead start time t 0 and mapping it to a start spatio-temporal coordinate (y 1 , t 1 ) of the first spatio-temporal segment, wherein t 1 = t 0 ,y 1 =0,t 1 is the start time of the first spatio-temporal segment and y 1 is the start position of the first spatio-temporal segment; Extracting the exposure time of each frame of image and the position of the corresponding leading head in the axial movement direction in the space-time evolution image data to form a discrete development track point set { (Y i , T i ) }, wherein Y i is the position of the corresponding leading head in the axial movement direction of the ith frame of image, and T i is the exposure time of the ith frame of image; Interpolating the set of development trajectory points { (Y i , T i ) } to form a fitted development trajectory curve; Based on the development trajectory curve, calculating the space-time coordinates (y j , t j ）、（y j+1 , t j+1 ) of the leading head crossing each space-time segment boundary, wherein j=2, 3..N, N is the total number of space-time segments, y j and y j+1 are the start position and the end position of the j-th space-time segment, and t j and t j+1 are the start time and the end time of the j-th space-time segment.
6. The method for constructing a pilot bifurcation prediction model according to any one of claims 1 to 4, wherein the acquiring the multidimensional physical characteristics of each space-time segment including the pilot head injection charge amount, the pilot head background electric field strength, the pilot head deflection angle and the voltage change rate includes: And calculating linear correlation coefficients among the candidate features by adopting a Pearson correlation analysis method, and identifying and eliminating instantaneous voltage with strong correlation with the background electric field intensity of the pilot head to obtain the multidimensional physical features, wherein each candidate feature comprises the injection charge quantity of the pilot head, the background electric field intensity of the pilot head, the deflection angle of the pilot head and the voltage change rate.
7. A pilot bifurcation prediction method, comprising: Inputting multidimensional physical characteristics corresponding to the current stepping node and comprising pilot head injection charge quantity, pilot head background electric field intensity, pilot head deflection angle and voltage change rate into a pilot bifurcation probability prediction model to obtain predicted pilot bifurcation probability ; If the multidimensional physical characteristics are all higher than the preset minimum physical threshold value, the predicted pilot bifurcation probability is obtained The probability as the final prediction, P final , otherwise, P final = 0; Wherein the pilot bifurcation probability prediction model is constructed by the pilot bifurcation prediction model construction method according to any one of claims 1-6; Or/and, further comprising: Generating a compliance 0 1, Uniformly distributed random numbers r; If P final is more than or equal to r, judging that the current stepping node is bifurcated, otherwise, judging that the current stepping node is not bifurcated.
8. A pilot bifurcation predicting method applied to long-gap discharge simulation calculation, characterized in that the pilot bifurcation predicting method is the pilot bifurcation predicting method according to claim 7.
9. An electronic device comprising a computer readable storage medium and a processor; The computer-readable storage medium is for storing executable instructions; The processor is configured to read executable instructions stored in the computer readable storage medium, execute the pilot bifurcation prediction model construction method according to any one of claims 1-6, execute the pilot bifurcation prediction method according to claim 7, or/and execute the pilot bifurcation prediction method according to claim 8, which is applied to long-gap discharge simulation calculation.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the pilot bifurcation prediction model construction method according to any one of claims 1-6, implements the pilot bifurcation prediction method according to claim 7, or/and implements the pilot bifurcation prediction method according to claim 8 for use in long-gap discharge simulation calculation.

Description

Pilot bifurcation prediction model construction method and application Technical Field The invention belongs to the technical field of pilot bifurcation prediction, and particularly relates to a pilot bifurcation prediction model construction method and application. Background Long air gap discharge is a key physical process in lightning protection, power system insulation design, and high voltage equipment tolerance assessment. The development behavior of the lead is used as a leading channel structure before breakdown, and the geometric evolution and final breakdown characteristics of the discharge channel are directly determined. Among these, bifurcation is one of the most typical nonlinear behaviors in the lead development, which significantly changes channel morphology, local electric field distribution and energy injection characteristics, and has a decisive influence on breakdown paths and breakdown mechanisms. Therefore, accurately identifying and predicting the triggering condition of the pilot furcation is an important scientific problem of long-term concern in high-pressure engineering. The existing pilot bifurcation simulation research mainly forms two main stream modeling paths, but the existing pilot bifurcation simulation research has the limitations on mechanism and applicability: 1. morphology generation method based on fractal geometry: Such methods (e.g., NPW models and improved algorithms thereof) generally rely on fractal growth and random growth theory, and determine the direction of expansion of the channel by maximizing the local electric field or potential energy gain, so as to macroscopically reproduce a multi-branched tree structure. However, this type of approach is essentially a "morphogenic strategy", whose growth rules are mainly based on probabilistic descriptions of phenomenology, rather than the true physical driving force derived from the lead head. Therefore, it cannot explain mechanically why the bifurcation is what is and what is exactly the trigger condition, lacks physical causality, and hardly reflects the substantial influence of the physical parameters such as voltage waveform, electrode structure, etc. on the bifurcation. 2. Criteria method based on single physical threshold: Such methods typically set a certain threshold value of "critical electric field strength" or "critical potential drop" in the simulation model, and when the parameters of the pilot header exceed this empirical threshold value, it is determined that a branch is generated. Although the method makes up the defect that the geometric model lacks a trigger mechanism, the method has the obvious defect that the multi-dimensional coupling mechanism is absent, and the driving mechanism of the pilot fork coupled by the multi-dimensional physical quantity cannot be described. And the single threshold method breaks the organic connection among the physical quantities, and cannot capture the competition mechanism when the high-energy promotion bifurcation and the low-electric-field inhibition bifurcation coexist. And secondly, "threshold universality is poor and difficult to determine". The relevant threshold value is seriously dependent on the back-stepping experience under specific experimental conditions, and lacks a unified physical derivation basis. When facing different gap scales (such as deduction from meter level to ten meter level) or complex voltage waveforms, the original experience threshold value often fails, and an accurate threshold value adapting to a new working condition is difficult to find, so that the model generalization capability is extremely weak. In summary, the prior art mainly stays at the stage of approximating bifurcation behaviors by using a simple criterion or a phenomenological probability, the geometric preferential model (fractal) and the single threshold triggering method can not reflect the multidimensional physical coupling mechanism behind the leading bifurcation, and lack definite physical causality and universality, the existing model is difficult to adapt to the changes of working conditions of different gap scales (such as from meter level to ten meter level), different electrode structures and the like, and the model has poor interpretability, multi-working condition generalization capability and consistency. Disclosure of Invention Aiming at the defects or improvement demands of the prior art, the invention provides a pilot bifurcation prediction model construction method and application, and aims to improve the accuracy of pilot bifurcation prediction. In order to achieve the above object, the present invention provides a method for constructing a pilot bifurcation prediction model, comprising: Acquiring multi-scale discharge data in a pilot discharge process, wherein the multi-scale discharge data comprises space-time evolution image data, discharge current waveforms and voltage waveforms of pilot channel development; Dividing a pilot development path synthe