CN-121999440-A - Power transmission and transformation project construction state early warning method and system

CN121999440ACN 121999440 ACN121999440 ACN 121999440ACN-121999440-A

Abstract

The invention discloses a power transmission and transformation project construction state early warning method which comprises the steps of obtaining image data of a power transmission and transformation project construction state, preprocessing and classifying to construct a training data set, constructing a power transmission and transformation project construction state early warning initial model based on a wide residual error network, a normalization layer and a full connection layer, training to obtain a power transmission and transformation project construction state early warning model, and carrying out state early warning in the actual power transmission and transformation project construction process by adopting the obtained power transmission and transformation project construction state early warning model. The invention also discloses a system for realizing the power transmission and transformation project construction state early warning method. The invention not only realizes the early warning of the construction state of the power transmission and transformation project, but also has higher reliability and better accuracy.

Inventors

XU BO
ZHENG BINYU
PAN MENGZHEN
YAO YING
CHEN JIAN
HE SHUYUN
LIU JUN
LIU WENJUN
ZHANG HUA
LIU YONGFENG

Assignees

国网湖南省电力有限公司经济技术研究院
国网湖南省电力有限公司
国家电网有限公司

Dates

Publication Date: 20260508
Application Date: 20260212

Claims (10)

1. A power transmission and transformation project construction state early warning method comprises the following steps: s1, acquiring image data of a construction state of a power transmission and transformation project; s2, preprocessing and classifying the image data obtained in the step S1 to construct a training data set; S3, constructing an early warning initial model of the construction state of the power transmission and transformation project based on the wide residual error network, the normalization layer and the full connection layer; S4, training the power transmission and transformation project construction state early warning initial model constructed in the step S3 by adopting the training data set constructed in the step S2 to obtain a power transmission and transformation project construction state early warning model; s5, adopting the power transmission and transformation project construction state early warning model obtained in the step S4 to perform state early warning in the actual power transmission and transformation project construction process.
2. The power transmission and transformation project construction state early warning method according to claim 1, characterized in that the step S1 of acquiring the image data of the power transmission and transformation project construction state specifically comprises the following steps: acquiring image data of a construction state of a power transmission and transformation project; The image data comprises image data under various construction scenes, wherein the construction scenes comprise mountain crossing construction scenes, tower resistance construction scenes and wire erection construction scenes; the image data comprises tagged image data and untagged image data; the label comprises a normal construction state and an abnormal construction state, wherein the abnormal construction state comprises an abnormal state with an abnormal frequency higher than or equal to a set value and an abnormal state with an abnormal frequency lower than the set value, the abnormal state with the abnormal frequency higher than or equal to the set value comprises an unbuckled safety belt, an unbuckled helmet and loosening bolts, and the abnormal state with the abnormal frequency lower than the set value comprises live line crossing operation and scaffold displacement.
3. The power transmission and transformation project construction state early warning method according to claim 2, characterized in that the step S2 of preprocessing and classifying the image data obtained in the step S1 to construct a training data set, specifically comprises the following steps: for tagged images: Dividing the image data with the tag into a head type region, a middle type region and a tail type region, wherein the head type region corresponds to the image data with the tag in a normal construction state, the middle type region corresponds to the image data with the tag in an abnormal state with the abnormal frequency higher than or equal to a set value, and the tail type region corresponds to the image data with the tag in an abnormal state with the abnormal frequency lower than the set value; The classified image data with the labels is adjusted to be an image with uniform pixel size, and data enhancement is carried out by adopting a RandAugment method; For unlabeled images: Adjusting the unlabeled image data into an image with uniform pixel size; Respectively carrying out weak enhancement processing and strong enhancement processing on the adjusted unlabeled image data to obtain corresponding weak enhancement unlabeled data And strongly enhanced unlabeled data The weak enhancement processing is enhancement processing by adopting a lightweight data enhancement mode, wherein the lightweight data enhancement mode comprises random rotation, random horizontal overturn and random vertical overturn, and the strong enhancement processing is data enhancement by adopting a RandAugment method.
4. The power transmission and transformation engineering construction state early warning method according to claim 3, wherein the step S3 is based on a wide residual network, a normalization layer and a full connection layer, and the method specifically comprises the following steps: The constructed power transmission and transformation engineering construction state early warning initial model comprises a feature extraction module, a multi-expert calibration module, a multi-expert classification module, a self-adaptive aggregation module and a grading early warning module which are sequentially connected in series; The feature extraction module is used for extracting original features of the input image; the multi-expert calibration module is used for independently calibrating and decoupling different types of areas according to the obtained original characteristics; the multi-expert classification module is used for calculating the distribution probability of each category according to the characteristics output by the multi-expert calibration module and correcting the distribution probability to obtain category prediction probability distribution; The self-adaptive aggregation module is used for predicting probability distribution results according to each category and calculating to obtain final probability distribution; And the grading early warning module is used for carrying out grading early warning according to the final probability distribution result.
5. The power transmission and transformation project construction state early warning method according to claim 4, characterized in that the feature extraction module is specifically a WRN-28-2 network in a wide residual error network; the multi-expert calibration module specifically comprises the following schemes: The multi-expert calibration module comprises a head calibration layer, a middle calibration layer and a tail calibration layer, wherein the head calibration layer, the middle calibration layer and the tail calibration layer are mutually independent; The head calibration layer is built by adopting a similar batch normalization layer, the middle calibration layer is built by adopting a similar batch normalization layer, and the tail calibration layer is built by adopting a cumulative batch normalization layer; The system comprises a head calibration layer, a middle calibration layer, a tail calibration layer, a middle calibration layer and a tail calibration layer, wherein the head calibration layer is only used for processing non-label data and labeled image data divided into head type areas; Inputting a preprocessed standardized image sample, wherein the standardized image sample comprises marked data And nonstandard data Acquiring original characteristic output The original features output construction scene key information for comprehensively describing the sample; Original sample features to be attributed to head class regions Input header alignment layer, original sample features to be attributed to middle class region Inputting the original sample characteristics belonging to tail class area into middle calibration layer Inputting the tail calibration layer, obtaining the corresponding calibrated characteristics , ; The processing procedure of the class batch normalization layer is expressed as follows: In the middle of Is an input original sample feature; An index moving average value of sample characteristics in a region corresponding to the calibration layer corresponding to k; The index moving average variance of the sample characteristics in the area corresponding to the calibration layer corresponding to k; Is a set minimum number and is used for preventing denominator from being 0; scaling parameters of the calibration layer corresponding to k; the offset parameter of the calibration layer corresponding to k; And Iteratively updating by adopting an index moving average strategy; the head calibration layer, the middle calibration layer and the tail calibration layer are independent of each other.
6. The power transmission and transformation engineering construction state early warning method according to claim 5 is characterized in that the multi-expert classification module specifically comprises the following scheme: the multi-expert classification module comprises a head classification layer, a middle classification layer and a tail classification layer, wherein the head classification layer, the middle classification layer and the tail classification layer are mutually independent; the head classifying layer is built by adopting a full-connection layer, the middle classifying layer is built by adopting the full-connection layer, and the tail classifying layer is built by adopting the full-connection layer; The head classifying layer only processes the data output by the head calibrating layer, the middle classifying layer only processes the data output by the middle calibrating layer, and the tail classifying layer only processes the data output by the tail calibrating layer; the head classification layer outputs the initial head classification predictive probability distribution of the full connection layer Predicting probability distribution as final head class ; Initial middle category predictive probability distribution output by the middle classification layer to the full connection layer And correcting by adopting the following formula to obtain the middle category prediction probability distribution: In the middle of Predicting probability distribution for the middle category; for the data set, the unbalance ratio is a priori, and , The number of samples included in the category having the largest number of samples, The number of samples included in the category having the smallest number of samples; The tail classification layer predicts probability distribution of initial tail category output by the full connection layer And correcting by adopting the following formula to obtain the tail category prediction probability distribution: In the middle of A probability distribution is predicted for the tail class.
7. The power transmission and transformation engineering construction state early warning method according to claim 6, characterized in that the self-adaptive aggregation module comprises the following steps: for the input image, respectively acquiring corresponding head category prediction probability distribution Mid-class predictive probability distribution And tail class predictive probability distribution ; The corresponding shannon entropy is calculated by the following formula: In the middle of C is the total category number in the training data set; The probability distribution of the input data belonging to the c-th category corresponding to the probability distribution k; The value of k is H, M and T, the probability distribution is the head type prediction probability distribution when the value of k is H, the probability distribution is the middle type prediction probability distribution when the value of k is M, and the probability distribution is the tail type prediction probability distribution when the value of k is T; the weight corresponding to the probability distribution k is calculated by adopting the following formula : Wherein m is the code number of the probability distribution, Representation of Take the value of , Representation of Take the value of , Representation of Take the value of ; Calculating to obtain final probability distribution Is that ; Based on the probability distribution obtained And selecting the category with the highest probability value as the final prediction category.
8. The power transmission and transformation engineering construction state early warning method according to claim 7, characterized in that the grading early warning module comprises the following steps: the head category prediction probability distribution is calculated by adopting the following formula Mid-class predictive probability distribution And tail class predictive probability distribution A divergence parameter between any two predictive probability distributions: In the middle of To predict probability distribution And predicting probability distribution The parameters of the divergence between the two, The range of the values is as follows 、 And , ; Is Kullback-Leibler divergence; To predict probability distribution And predicting probability distribution Is an average distribution of (3); the comprehensive uncertainty measurement value is calculated by adopting the following formula : In the middle of Is a risk sensitive factor, and ; As a tail indication function, if either i or j corresponds to a tail Otherwise ; Hierarchical early warning is carried out by adopting the following rules: If it is And the final prediction type is in a normal construction state, and early warning is not carried out; If it is Or if the final prediction category is an abnormal state corresponding to the middle category region, normal early warning is carried out; If it is Or the final prediction category is an abnormal state corresponding to the tail category area, and then emergency early warning is carried out.
9. The power transmission and transformation engineering construction state early warning method according to claim 8, characterized by training in step S4, specifically comprising the following steps: the training process only aims at a feature extraction module, a multi-expert calibration module and a multi-expert classification module in the model to train; for tagged images: The label loss is calculated by the following formula : In the middle of Is the total number of tagged images; A predictive tag for sample i; A real label of the sample i; Is a cross entropy loss function; For unlabeled images: Weak enhancement of untag data Processing through the constructed model to obtain a corresponding classification result, and taking the classification result as a pseudo label of the label-free image ; Enhancing strongly untagged data Processing through the constructed model to obtain corresponding prediction probability distribution ; Judging if it is If the maximum value of the set value is larger than the set value, marking the corresponding sample as an effective sample, otherwise marking the corresponding sample as an ineffective sample; for all the effective samples, the label loss is calculated by adopting the following formula : In the middle of Is the total number of unlabeled images; A predictive probability distribution for an effective sample j; a pseudo tag that is a valid sample j; During training, the following total loss function is adopted Training the model: In the middle of A first weight is set; is a set second weight.
10. A system for realizing the construction state early warning method of the power transmission and transformation project is characterized by comprising a data acquisition module, a data processing module, a model construction module, a model training module and a state early warning module, wherein the data acquisition module, the data processing module, the model construction module, the model training module and the state early warning module are sequentially connected in series, the data acquisition module is used for acquiring image data of the construction state of the power transmission and transformation project and uploading data information to the data processing module, the data processing module is used for preprocessing and classifying the acquired image data according to the received data information to construct a training data set and uploading the data information to the model construction module, the model construction module is used for constructing an initial model of the construction state early warning of the power transmission and transformation project based on a wide residual error network, a normalization layer and a full connection layer according to the received data information, the model training module is used for training the constructed initial model of the construction state early warning of the power transmission and transformation project according to the received data information to acquire an actual construction state early warning module of the power transmission and transformation project.

Description

Power transmission and transformation project construction state early warning method and system Technical Field The invention belongs to the field of electric automation, and particularly relates to a power transmission and transformation project construction state early warning method and system. Background Along with the development of economic technology and the improvement of living standard of people, electric energy becomes an indispensable secondary energy source in the production and living of people, and brings endless convenience to the production and living of people. Therefore, ensuring stable and reliable supply of electric energy becomes one of the most important tasks of the electric power system. The power transmission and transformation project is an important component of the power system, and the safety monitoring of the construction process is directly related to the safe and stable operation of the power system and the safety of constructors. Currently, an intelligent monitoring scheme based on a computer vision scheme is widely applied to the safety monitoring process of power transmission and transformation engineering. The key of the scheme is that visual data such as pictures of a monitoring site are analyzed, and abnormal states such as normal construction states and abnormal construction states in a construction scene are automatically distinguished, so that early warning and accurate management and control of safety risks are realized. At present, the monitoring scheme based on computer vision often adopts a semi-supervised learning method to relieve the defect of insufficient labeling data in training data. However, in the construction process of power transmission and transformation engineering, the data sample of the normal construction state is extremely high in proportion, the data of abnormal states such as illegal construction and construction hidden danger are extremely low in proportion, and the sample has characteristic distribution showing dispersion characteristics and obvious intra-class morphological difference due to factors such as various operation environments and changeable illumination conditions. Although the existing scheme adopts a semi-supervised learning method to alleviate the defect of insufficient labeling data in training data, the existing scheme has the problems of unbalanced characteristic space, selective deviation of pseudo labels, positive feedback amplification of noise and the like in the specific application process, so that the recognition rate of abnormal states is extremely low, and the accuracy requirement of on-site safety control cannot be met. Disclosure of Invention The invention aims to provide a power transmission and transformation project construction state early warning method with high reliability and good accuracy. The second purpose of the invention is to provide a system for realizing the power transmission and transformation project construction state early warning method. The invention provides a construction state early warning method for power transmission and transformation engineering, which comprises the following steps: s1, acquiring image data of a construction state of a power transmission and transformation project; s2, preprocessing and classifying the image data obtained in the step S1 to construct a training data set; S3, constructing an early warning initial model of the construction state of the power transmission and transformation project based on the wide residual error network, the normalization layer and the full connection layer; S4, training the power transmission and transformation project construction state early warning initial model constructed in the step S3 by adopting the training data set constructed in the step S2 to obtain a power transmission and transformation project construction state early warning model; s5, adopting the power transmission and transformation project construction state early warning model obtained in the step S4 to perform state early warning in the actual power transmission and transformation project construction process. The step S1 of acquiring the image data of the construction state of the power transmission and transformation project specifically comprises the following steps: acquiring image data of a construction state of a power transmission and transformation project; The image data comprises image data under various construction scenes, wherein the construction scenes comprise mountain crossing construction scenes, tower resistance construction scenes and wire erection construction scenes; the image data comprises tagged image data and untagged image data; The label comprises a normal construction state and an abnormal construction state, wherein the abnormal construction state comprises an abnormal state with an abnormal frequency higher than or equal to a set value and an abnormal state with an abnormal frequency lower than the set value, the abnormal state w