Power transmission line galloping prediction method integrating optical fiber monitoring and refined meteorological data
Through the combination of adaptive weighted data fusion, BP neural network and K-Mediods algorithm, the problems of poor fusion of multi-source data and poor timing feature capture in traditional methods are solved, and high-precision transmission line dance prediction is achieved.
Patent Information
- Application Number
- CN202411445115.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-16
- Publication Date
- 2025-03-21
AI Technical Summary
Traditional transmission line dance prediction methods lack deep fusion of multi-source data, difficult to capture timing characteristics, and fail to effectively consider vertical and horizontal clustering processing, resulting in low prediction accuracy.
Adaptive weighted primary data fusion and BP neural network secondary data fusion are used, and horizontal and vertical timing law mining is performed in combination with the K-Mediods algorithm, and prediction accuracy is improved through improved GRU gating structure and attention closed-loop mechanism.
It realizes deep fusion of multi-source data and accurate mining of timing laws, improving the accuracy and accuracy of transmission line dance prediction.
Smart Images
Figure CN119670927A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a transmission line galloping prediction method integrating optical fiber monitoring and refined meteorological data, and belongs to the technical field of electric power communication. Background Art
[0002] At present, transmission lines, as the largest component of the power system, bear the important task of electric energy transmission. In daily operation, transmission lines are easily affected by meteorological environments such as low temperature, strong winds, and heavy rains, causing vibrations, which can cause line tripping, disconnection, power outages, and other accidents, greatly affecting the normal operation of society and the daily lives of residents.
[0003] Transmission line ice-covered dancing prediction uses the close relationship between time-series data such as wind speed, humidity, and temperature and the dancing state of ice-covered lines. Through the vertical correlation of the same data and the horizontal correlation of different data, the line dancing risk level is obtained, and the level of dancing is predicted in advance. Further measures such as deploying spacers and artificial deicing are taken for the predicted lines to reduce the impact of ice-covered dancing on residents' daily lives and the power grid economy. However, traditional methods often have the following problems.
[0004] First, the traditional line dancing prediction method lacks deep integration of multi-source data, and it is difficult to capture the time series characteristics in the data. At the same time, it does not take into account the vertical and horizontal clustering of data of different time periods and types, and ignores the complex mutual influence and correlation of time series data, resulting in poor extraction of horizontal and vertical time series rules of multi-source data.
[0005] Second, the traditional method uses the clustering method based on Euclidean distance to fuse multi-source data, which leads to irregular changes in the time intervals of the data subsets after clustering, causing horizontal uneven timing between single data types and vertical uneven timing of overall clustering of multiple data types. At the same time, there is a lack of closed-loop feedback mechanism, and the network parameters are not updated in a timely manner. The model training process is prone to fall into local optimality, resulting in low accuracy of the final line dance prediction, making it difficult to achieve fast and accurate line dance prediction.
[0006] In view of the above-mentioned defects, the present invention aims to create a transmission line sway prediction method that integrates optical fiber monitoring and refined meteorological data to make it more valuable for industrial use. Summary of the invention
[0007] In order to solve the above technical problems, the purpose of the present invention is to provide a transmission line galloping prediction method integrating optical fiber monitoring and refined meteorological data.
[0008] The present invention provides a transmission line galloping prediction method integrating optical fiber monitoring and refined meteorological data, and the specific prediction steps are as follows:
[0009] The optical fiber real-time monitoring data and meteorological forecast data are fused based on adaptive weighting at the first level, and the potential time series patterns in the data are learned and captured based on the second level data fusion of BP neural network. At the same time, the horizontal and vertical time series regularity mining of the ice dancing fusion data of the transmission line based on K-Mediods is carried out to improve the depth and accuracy of the mining of horizontal and vertical time series regularity of the data.
[0010] The time series data is processed according to the horizontal and vertical time series rules to obtain the horizontal and vertical uneven time series perception adjustment parameters. The obtained parameters are used to improve the traditional GRU gating structure and reduce the prediction bias of the network. At the same time, the attention closed-loop mechanism is used to improve the accuracy of model training, and finally the transmission line dancing prediction that integrates optical fiber monitoring and refined meteorological data is realized.
[0011] Furthermore, the specific method of the first-level data fusion based on adaptive weighting is:
[0012] The types of fiber optic monitoring sensor data and meteorological forecast data related to ice dancing on transmission lines are defined as temperature, humidity, and wind speed. There are N temperature fiber optic monitoring sensors on a transmission line, and each fiber optic monitoring sensor collects the temperature of the transmission line at the same fixed time interval τ; it is defined that there are M ground meteorological stations that predict the temperature data of the line at the same time interval to obtain M groups of meteorological forecast data; the multi-source temperature data set of fiber optic monitoring and meteorological forecast at the t-th sampling time is:
[0013] X te (t) = {X te，1 (t),…,X te，i (t),…,X te，N (t),…,X te，N+M (t)} (1)
[0014] Where, X te，i (t), i = 1, 2, ..., N represents the temperature collected by the i-th temperature optical fiber monitoring sensor at the t-th sampling time, which is the true value of the temperature Unbiased judgment result; X te，i (t), i = N + 1, N + 2, ..., N + M is the temperature forecast data of the i-th ground weather station, which is the true value The biased judgment result of represents the variance of the true value error of the temperature data, then the total mean square error of the temperature data at time t is:
[0015]
[0016] The adaptive weighted first-level fusion result of each optical fiber sensor and meteorological forecast data at the t-th sampling time is expressed as:
[0017] Based on the above steps, the transmission line time series temperature fusion data set can be obtained. Similarly, the adaptive weighted first-level fusion time series corresponding to humidity and wind speed data can be obtained:
[0018] Furthermore, the specific method of the secondary data fusion based on BP neural network is:
[0019] (1) Initialize network weights and bias b;
[0020] (2) Input temperature time series based on adaptive weighted first-level fusion As training samples, network forward training is performed;
[0021] (3) Based on Calculate the loss function, where is the number of fused time series samples; is the fusion value; is the true value, which is obtained by field measurement; the loss function L Fu Used to measure the error of the current BP neural network secondary data fusion to facilitate subsequent network training;
[0022] (4) According to the loss function, back propagation is performed based on gradient descent to update the network parameters;
[0023] (5) The result obtained from the first-level fusion is input into the trained second-level data fusion BP neural network again to obtain the final second-level data fusion result.
[0024] Furthermore, the specific method for mining the horizontal and vertical time series rules of the K-Mediods-based fusion data of ice dancing on power transmission lines is as follows:
[0025] (1) Initialize the cluster center and randomly select K te The transmission line temperature fusion data samples are used as the initial longitudinal clustering center point set;
[0026] (2) Calculate the Euclidean distances from all sample points in the fused data set to the center points of each vertical cluster in turn, and place the sample points into the cluster represented by the nearest vertical cluster center point. The calculation formula of the Euclidean distance is expressed as k te is an initial random cluster center point, is the value of the temperature fusion data at the i-th sample point;
[0027] (3) After the vertical clustering is completed, each sample point in each cluster is selected as the vertical cluster center, the Euclidean distance is recalculated, and the sample point with the smallest total distance within the cluster is selected as the new center point, which is expressed as is the sample point in the cluster; k te is the vertical cluster center;
[0028] (4) When the set of vertical clustering center points remains unchanged, the algorithm terminates and the final vertical clustering result is output as in The vertical fusion data of temperature Otherwise, update the cluster center points and repeat S1.3.1 to S1.3.4 until the cluster center point set remains unchanged;
[0029] (5) Based on the above method, the vertical clustering results of humidity and wind speed are obtained in turn. Among them, K hu , K vc are the longitudinal clustering numbers of humidity fusion data and wind speed fusion data, respectively;
[0030] (6) The three types of fused data are mixed horizontally. The horizontal data format of a sample point of the fused mixed data is: ω te ,ω hu ,ω vc are the normalized weights of the corresponding fused data. The fused mixed data is clustered horizontally, and the results are expressed as K mix is the number of clusters for fusion mixed data.
[0031] Furthermore, the processing of the time series data according to the horizontal and vertical time series rules to obtain the horizontal and vertical staggered time series perception adjustment parameters is specifically:
[0032]
[0033]
[0034] In the formula, α te , β te , ε te represents the temperature perception parameter, which jointly determines the construction of the time-distance function and can adaptively adjust the time-distance function according to the data of different subclasses; d te (k) represents the time interval before adjustment of the kth subclass in the temperature clustering result; α mix , β mix , ε mix represents the lateral perception parameter; d mix ={d mix (1),d mix (2),…,d mix (k),…,d mix (K mix )} is the set of horizontally staggered time intervals; dmix (k) represents the time interval before adjustment of the kth subclass in the mixed data clustering results.
[0035] Furthermore, the specific method of using the obtained parameters to improve the traditional GRU gating structure is:
[0036] Combining the improved time-series perception function with the traditional GRU, the update methods of each gate structure are as follows
[0037]
[0038]
[0039]
[0040]
[0041] In the formula, Represents the current input unit after interaction update; is the hidden layer output after interactive update; It is an update unit that integrates time-distance information; It is a reset unit that integrates time-distance information; is the summary of the input and past hidden layer states; is the hidden layer state output; is the weight parameter; the GRU unit comprehensively considers the prediction data of all clustering results, and the final prediction output is expressed as
[0042] Furthermore, the specific method of using the attention closed-loop mechanism to improve the accuracy of model training is:
[0043] The loss function and attention closed-loop training method are expressed as:
[0044]
[0045]
[0046] In the formula, is the predicted network parameter; ρ is the learning rate;
[0047] Based on the prediction network training of the uneven time series perception data, the prediction data sets of temperature, humidity and wind speed are obtained respectively. Taking temperature as an example, it is expressed as Similarly, the predicted data set for humidity and wind speed is The transmission line galloping risk coefficient is expressed as
[0048]
[0049] In the formula, Respectively represent the predicted values of temperature, wind speed and humidity at time t; T ice is the duration of ice cover; is the dancing risk factor per unit time;
[0050] Further based on the mathematical statistics principle and the actual line ice galloping data, the galloping risk coefficient results of different galloping events are classified, and the transmission line ice galloping risk level characterized by the galloping risk coefficient is obtained, which is expressed as
[0051]
[0052] A transmission line galloping prediction system integrating optical fiber monitoring and refined meteorological data, comprising a meteorological prediction module, an optical fiber monitoring sensing module, a data fusion module, an uneven time series perception module, and a transmission line galloping prediction module;
[0053] The weather forecast module is responsible for forecasting various weather data such as temperature, humidity and wind speed, and transmitting the obtained data to the data fusion module;
[0054] The optical fiber monitoring sensor module is responsible for collecting optical fiber sensing data generated by Brillouin scattering, Raman scattering and Rayleigh scattering, and transmitting the obtained data to the data fusion module;
[0055] The data fusion module is responsible for fusing meteorological data and optical fiber monitoring sensor time series data, and transmitting the obtained data to the time series data clustering module for clustering;
[0056] The time series data clustering module is responsible for data clustering, mining the horizontal and vertical time series rules of the time series data, and transmitting the clustering data to the uneven time series perception module;
[0057] The staggered timing perception module is responsible for adaptively perceiving the longitudinal and transverse staggered timing, obtaining prediction network parameters, and transmitting the parameters to the prediction network construction module;
[0058] The prediction network construction module is responsible for constructing a time series prediction network using parameters, training the prediction model, and transmitting the trained model to the transmission line galloping prediction module;
[0059] The transmission line galloping prediction module is responsible for using the trained network and time series data to make predictions, obtain transmission line galloping prediction results and perform closed-loop feedback to strengthen training.
[0060] A transmission line galloping prediction device integrating optical fiber monitoring and refined meteorological data, comprising
[0061] Data fusion module: used to perform primary data fusion of optical fiber real-time monitoring data and meteorological forecast data based on adaptive weighting, and secondary data fusion learning and capturing of potential time series patterns in data based on BP neural network, and at the same time, to mine horizontal and vertical time series patterns of ice dancing fusion data of power transmission lines based on K-Mediods, so as to improve the depth and accuracy of mining horizontal and vertical time series patterns of data;
[0062] Prediction module: It is used to process time series data according to the horizontal and vertical time series rules, obtain the horizontal and vertical uneven time series perception adjustment parameters, use the obtained parameters to improve the traditional GRU gating structure, reduce the prediction deviation of the network, and use the attention closed-loop mechanism to improve the accuracy of model training, and finally realize the transmission line sway prediction that integrates optical fiber monitoring and refined meteorological data.
[0063] Furthermore, in the data fusion module, the specific method for mining the horizontal and vertical time series rules of the ice dancing fusion data of the transmission line based on K-Mediods is:
[0064] (1) Initialize the cluster center and randomly select K te The transmission line temperature fusion data samples are used as the initial longitudinal clustering center point set;
[0065] (2) Calculate the Euclidean distances from all sample points in the fused data set to the center points of each vertical cluster in turn, and place the sample points into the cluster represented by the nearest vertical cluster center point. The calculation formula of the Euclidean distance is expressed as k te is an initial random cluster center point, is the value of the temperature fusion data at the i-th sample point;
[0066] (3) After the vertical clustering is completed, each sample point in each cluster is selected as the vertical cluster center, the Euclidean distance is recalculated, and the sample point with the smallest total distance within the cluster is selected as the new center point, which is expressed as is the sample point in the cluster; k te is the vertical cluster center;
[0067] (4) When the set of vertical clustering center points remains unchanged, the algorithm terminates and the final vertical clustering result is output as in The vertical fusion data of temperature Otherwise, update the cluster center points and repeat S1.3.1 to S1.3.4 until the cluster center point set remains unchanged;
[0068] (5) Based on the above method, the vertical clustering results of humidity and wind speed are obtained in turn. Among them, K hu , Kvc are the longitudinal clustering numbers of humidity fusion data and wind speed fusion data, respectively;
[0069] (6) The three types of fused data are mixed horizontally. The horizontal data format of a sample point of the fused mixed data is: ω te ,ω hu ,ω vc are the normalized weights of the corresponding fused data. The fused mixed data is clustered horizontally, and the results are expressed as K mix is the number of clusters for fusion mixed data.
[0070] Furthermore, in the prediction module, the specific method of using the obtained parameters to improve the traditional GRU gating structure is:
[0071] Combining the improved time-series perception function with the traditional GRU, the update methods of each gate structure are as follows
[0072]
[0073]
[0074]
[0075]
[0076] In the formula, Represents the current input unit after interaction update; is the hidden layer output after interactive update; It is an update unit that integrates time-distance information; It is a reset unit that integrates time-distance information; is the summary of the input and past hidden layer states; is the hidden layer state output; is the weight parameter; the GRU unit comprehensively considers the prediction data of all clustering results, and the final prediction output is expressed as
[0077] By means of the above scheme, the present invention has at least the following advantages:
[0078] (1) Considering multi-source data and time series regularity mining based on vertical and horizontal clustering of data: The present invention proposes a method for mining horizontal and vertical time series regularities of multi-source data based on adaptive weighted learning, performs first-level adaptive weighting on optical fiber real-time monitoring data and meteorological forecast data, and uses a second-level BP neural network to learn and capture potential time series patterns in the data to achieve deep fusion of multi-source data; at the same time, the K-Mediods algorithm is used to mine horizontal and vertical time series regularities of the fused data, effectively improving the depth and accuracy of mining horizontal and vertical time series regularities of the data.
[0079] (2) Considering the perception of uneven timing and attention closed-loop feedback: The present invention proposes a transmission line sway prediction method based on the perception of uneven timing and attention closed-loop. The timing data is processed according to the problem of uneven timing in the horizontal and vertical directions, and the horizontal and vertical uneven timing perception adjustment parameters are obtained. The obtained parameters are used to improve the traditional GRU gating structure, reduce the prediction deviation of the network, and use the attention closed-loop mechanism to improve the accuracy of model training.
[0080] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0081] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show a certain embodiment of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0082] Figure 1 It is a schematic diagram of the structure of the power transmission line galloping prediction system integrating optical fiber monitoring and refined meteorological data of the present invention;
[0083] Figure 2 It is a working block diagram of the power transmission line galloping prediction method integrating optical fiber monitoring and refined meteorological data of the present invention. DETAILED DESCRIPTION
[0084] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0085] The present invention proposes a transmission line galloping prediction system integrating optical fiber monitoring and refined meteorological data. Figure 1 As shown, it includes weather forecast module, optical fiber monitoring sensor module, data fusion module, uneven time series perception module, and transmission line dancing prediction module. The modules are introduced as follows:
[0086] Meteorological forecast module: responsible for predicting various meteorological data such as temperature, humidity and wind speed, and transmitting the obtained data to the data fusion module.
[0087] Fiber optic monitoring sensor module: responsible for collecting fiber optic sensing data generated by Brillouin scattering, Raman scattering and Rayleigh scattering, and transmitting the obtained data to the data fusion module.
[0088] Data fusion module: responsible for fusing meteorological data and optical fiber monitoring sensor time series data, and transmitting the obtained data to the time series data clustering module for clustering.
[0089] Time series data clustering module: responsible for data clustering, mining the horizontal and vertical time series rules of time series data, and transmitting the clustered data to the irregular time series perception module.
[0090] Irregular timing perception module: responsible for the adaptive perception of longitudinal and lateral irregular timing, obtaining the prediction network parameters, and transmitting the parameters to the prediction network construction module.
[0091] Prediction network construction module: responsible for using parameters to build a time series prediction network, training the prediction model, and transmitting the trained model to the transmission line dancing prediction module.
[0092] Transmission line gallop prediction module: responsible for using the trained network and time series data to make predictions, obtain transmission line gallop prediction results and perform closed-loop feedback to strengthen training.
[0093] like Figure 2 As shown, the present invention proposes a transmission line galloping prediction method that integrates optical fiber monitoring and refined meteorological data, including two parts: a multi-source data horizontal and vertical time series law mining method based on adaptive weighted learning and a transmission line galloping prediction method based on uneven time series perception and attention closed loop.
[0094] A method for mining horizontal and vertical time series patterns from multi-source data based on adaptive weighted learning
[0095] The present invention proposes a method for mining horizontal and vertical time series rules of multi-source data based on adaptive weighted learning, including primary data fusion based on adaptive weighted learning, secondary data fusion based on BP neural network, and horizontal and vertical time series rule mining of ice dancing fusion data of power transmission lines based on K-Mediods. The specific steps are as follows:
[0096] S1.1: First-level data fusion based on adaptive weighting
[0097] The types of fiber optic monitoring sensor data and meteorological forecast data related to ice dancing on transmission lines are defined as temperature, humidity, and wind speed. Taking the temperature data of transmission lines as an example, it is assumed that there are N temperature fiber optic monitoring sensors on a transmission line, and each fiber optic monitoring sensor collects the temperature of the transmission line at the same fixed time interval τ. At the same time, it is assumed that there are M ground meteorological stations that predict the temperature data of the line at the same time interval, and M groups of meteorological forecast data can be obtained. The multi-source temperature data set of fiber optic monitoring and meteorological forecast at the t-th sampling time is expressed as:
[0098] X te (t) = {X te,1 (t),…,X te,i (t),…,X te,N (t),…,X te,N+M (t)} (1)
[0099] Where, X te,i (t), i = 1, 2, ..., N represents the temperature collected by the i-th temperature optical fiber monitoring sensor at the t-th sampling time, which is the true value of the temperature The unbiased judgment result of X is obtained. At the same time, there is a monitoring error caused by equipment aging, and each monitoring value is independent of each other; te,i (t), i = N + 1, N + 2, ..., N + M is the temperature forecast data of the i-th ground meteorological station, which is the true value The biased judgment results of each prediction value are independent of each other. represents the variance of the true value error of the temperature data, then the total mean square error of the temperature data at time t is expressed as
[0100]
[0101] The adaptive weighted multi-source time series data first-level fusion process takes into account the inherent correlation of data collected by different temperature sensors to the maximum extent, and takes the data fusion result under the condition of mean square error as the lower limit as the final multi-dimensional temperature data fusion result. The adaptive weighted first-level fusion result of each optical fiber sensor and meteorological forecast data at the t-th sampling time can be expressed as Based on the above steps, the transmission line time series temperature fusion data set can be obtained. Similarly, the adaptive weighted first-level fusion time series corresponding to humidity and wind speed data can be obtained:
[0102] S1.2: Secondary data fusion based on BP neural network
[0103] The secondary data fusion method based on BP neural network selects the temperature, humidity, wind speed and other transmission line dancing data after adaptive weighted learning fusion as the input of BP neural network, learns and captures the potential time series pattern in the data, and finally outputs the secondary fusion data of each sampling moment after learning, effectively reducing the noise and irregularity of the data. Taking the temperature first-level fusion time series as an example, the training set and the test set are divided according to the ratio of 3:1.
[0104] S1.2.1: Initialize network weights and bias b;
[0105] S1.2.2: Input temperature time series based on adaptive weighted first-level fusion As training samples, network forward training is performed;
[0106] S1.2.3: Based on Calculate the loss function, where is the number of fused time series samples; is the fusion value; is the true value, which is obtained by field measurement; the loss function L Fu It is used to measure the error of the current BP neural network secondary data fusion to facilitate subsequent network training.
[0107] S1.2.4: According to the loss function, backpropagation is performed based on gradient descent to update the network parameters.
[0108] S1.2.5: The result obtained by the first-level fusion is input into the trained second-level data fusion BP neural network again to obtain the final second-level data fusion result.
[0109] S1.3: Mining horizontal and vertical time series rules of ice-covered transmission line dancing fusion data based on K-Mediods
[0110] After obtaining the fused data, due to the complexity and strong nonlinearity of the optical fiber monitoring data and meteorological data, directly predicting the fused data will cause the neural network to fail to converge, thus affecting the data prediction accuracy. Therefore, the K-Mediods algorithm is used to cluster the fused high-dimensional data to effectively reduce the redundancy of the data and better mine the nonlinear characteristics of the ice-covered dancing data of the transmission line. Clustering includes longitudinal clustering analysis of the data itself and transverse clustering analysis of multiple fused data. The longitudinal clustering can identify the typical change rules of different time periods by analyzing the correlation of the fused data itself, thereby revealing the periodicity and anomalies of the data, improving the interpretability of the data and the accuracy of the prediction model; the transverse clustering can reveal the complex mutual influence and correlation between different data sequences by combining multiple fused data for overall analysis, helping to identify the interaction patterns and potential causal relationships of multidimensional data, so as to more comprehensively understand and predict the risk of ice-covered dancing of transmission lines.
[0111] S1.3.1 Initialize the cluster center and randomly select K te The transmission line temperature fusion data samples are used as the initial longitudinal clustering center point set.
[0112] S1.3.2 Calculate the Euclidean distance from all sample points in the fused data set to the center point of each vertical cluster in turn, and put the sample point into the cluster represented by the nearest vertical cluster center point. The calculation formula of the Euclidean distance is expressed as k te is an initial random cluster center point, is the value of the temperature fusion data at the i-th sample point.
[0113] S1.3.3 After the vertical clustering is completed, each sample point in each cluster is selected as the vertical cluster center, the Euclidean distance is recalculated, and the sample point with the smallest total distance within the cluster is selected as the new center point, which is expressed as is the sample point in the cluster; k te is the vertical cluster center.
[0114] S1.3.4 When the set of vertical clustering center points remains unchanged, the algorithm terminates and the final vertical clustering result is output as in The vertical fusion data of temperature Otherwise, update the cluster center points and repeat S1.3.1 to S1.3.4 until the cluster center point set remains unchanged.
[0115] S1.3.5 Based on the above method, the vertical clustering results of humidity and wind speed are obtained in turn Among them, K hu , Kvc They are the vertical clustering numbers of humidity fusion data and wind speed fusion data respectively.
[0116] S1.3.6 The three fused data are horizontally mixed. The horizontal data format of a sample point of the fused mixed data is ω te ,ω hu ,ω vc are the normalized weights of the corresponding fused data. The fused mixed data is clustered horizontally, and the results are expressed as K mix is the number of clusters for fusion mixed data.
[0117] S2: Transmission line galloping prediction method based on uneven time series perception and attention closed loop
[0118] The present invention constructs a transmission line dancing prediction method based on uneven timing perception and attention closed loop, processes timing data according to the horizontal and vertical timing unevenness problem and obtains horizontal and vertical uneven timing perception adjustment parameters to improve the traditional GRU prediction network and improve the prediction accuracy through the attention closed loop mechanism.
[0119] S2.1 Adaptive adjustment method based on perception of uneven timing
[0120] In order to solve the problem of longitudinal time series inconsistency of the same type of fused data and the problem of transverse time series inconsistency of different data sources, taking multi-source temperature data fusion as an example, the transverse and longitudinal inconsistency time series perception adjustment parameters are obtained, which are expressed as
[0121]
[0122]
[0123] In the formula, α te , β te , ε te represents the temperature perception parameter, which jointly determines the construction of the time-distance function and can adaptively adjust the time-distance function according to the data of different subclasses; d te (k) represents the time interval before adjustment of the kth subclass in the temperature clustering result. mix , β mix , ε mix represents the lateral perception parameter; d mix ={d mix (1),d mix (2),…,d mix (k),…,d mix (K mix )} is the set of horizontally staggered time intervals; d mix (k) represents the time interval before adjustment of the kth subclass in the mixed data clustering results.
[0124] The gating structure of the traditional GRU is improved by using the adjustment parameters of the irregular timing perception, giving the update gate and the reset gate the adaptive learning ability for irregular timing interval information. The improved timing perception function is combined with the traditional GRU to obtain the following update methods for each gating structure:
[0125]
[0126]
[0127]
[0128]
[0129] In the formula, Represents the current input unit after interaction update; is the hidden layer output after interactive update; It is an update unit that integrates time-distance information; It is a reset unit that integrates time-distance information; is the summary of the input and past hidden layer states; is the hidden layer state output; is the weight parameter. The GRU unit comprehensively considers the prediction data of all clustering results, and the final prediction output is expressed as
[0130] S2.2 Transmission line gallop prediction method based on attention closed loop
[0131] Considering minimizing the error between the predicted data and the real data, the prediction network of the time series perception data is trained by combining the mean absolute error and the root mean square error. The parameters of the model are updated by the gradient of the loss function, so that the loss function gradually decreases and the accuracy of the model prediction is gradually improved. The loss function and the attention closed-loop training method are expressed as
[0132]
[0133]
[0134] In the formula, is the prediction network parameter; ρ is the learning rate. The prediction method and network training method of wind speed and humidity data are the same as the prediction and training methods of the prediction network of the above-mentioned temperature uneven time series perception data.
[0135] Based on the prediction network training of the uneven time series perception data, the prediction data sets of temperature, humidity and wind speed can be obtained respectively. Taking temperature as an example, it is expressed as Similarly, the predicted data set for humidity and wind speed is The galloping of transmission line ice is a process in which energy accumulates continuously under the combined effects of ambient temperature, relative air humidity and wind speed, which eventually leads to line galloping. When analyzing the factors affecting the galloping, in addition to meteorological parameters such as temperature and humidity at a specific time, physical quantities representing the accumulation process, such as duration, should also be considered. Among them, the four parameters that have the greatest impact on the galloping of transmission lines are temperature, wind speed, humidity and ice duration that meet the conditions for ice galloping. The transmission line galloping risk coefficient is established and expressed as
[0136]
[0137] In the formula, Respectively represent the predicted values of temperature, wind speed and humidity at time t; T ice is the duration of ice cover; is the dancing risk factor per unit time.
[0138] Based on the mathematical statistics principle and the actual line ice galloping data, the galloping risk coefficient results of different galloping events are classified, and the transmission line ice galloping risk level characterized by the galloping risk coefficient is obtained, which is expressed as
[0139]
[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.
Claims
1. A method for predicting transmission line galloping by integrating optical fiber monitoring and refined meteorological data, characterized in that The specific prediction steps are: The optical fiber real-time monitoring data and meteorological forecast data are fused based on adaptive weighting at the first level, and the potential time series patterns in the data are learned and captured based on the second level data fusion of BP neural network. At the same time, the horizontal and vertical time series regularity mining of the ice dancing fusion data of the transmission line based on K-Mediods is carried out to improve the depth and accuracy of the mining of horizontal and vertical time series regularity of the data. The time series data is processed according to the horizontal and vertical time series rules to obtain the horizontal and vertical uneven time series perception adjustment parameters. The obtained parameters are used to improve the traditional GRU gating structure and reduce the prediction bias of the network. At the same time, the attention closed-loop mechanism is used to improve the accuracy of model training, and finally the transmission line dancing prediction that integrates optical fiber monitoring and refined meteorological data is realized.
2. A method for predicting power transmission line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of the first-level data fusion based on adaptive weighting is: The types of fiber optic monitoring sensor data and meteorological forecast data related to ice dancing on transmission lines are defined as temperature, humidity, and wind speed. There are N temperature fiber optic monitoring sensors on a transmission line, and each fiber optic monitoring sensor collects the temperature of the transmission line at the same fixed time interval τ; it is defined that there are M ground meteorological stations that predict the temperature data of the line at the same time interval to obtain M groups of meteorological forecast data; the multi-source temperature data set of fiber optic monitoring and meteorological forecast at the t-th sampling time is: X te (t)={X te,1 (t),…,X te,i (t),…,X te,N (t),…,X te,N+M (t)} (1) Where, X te,i (t), i = 1, 2, ..., N represents the temperature collected by the i-th temperature optical fiber monitoring sensor at the t-th sampling time, which is the true value of the temperature Unbiased judgment result; X te,i (t), i = N + 1, N + 2, ..., N + M is the temperature forecast data of the i-th ground weather station, which is the true value The biased judgment result of represents the variance of the true value error of the temperature data, then the total mean square error of the temperature data at time t is: The adaptive weighted first-level fusion result of each optical fiber sensor and meteorological forecast data at the t-th sampling time is expressed as: Based on the above steps, the transmission line time series temperature fusion data set can be obtained. Similarly, the adaptive weighted first-level fusion time series corresponding to humidity and wind speed data can be obtained:
3. The method for predicting power line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of the secondary data fusion based on BP neural network is: (1) Initialize network weights and bias b; (2) Input temperature time series based on adaptive weighted first-level fusion As training samples, network forward training is performed; (3) Based on Calculate the loss function, where is the number of fused time series samples; is the fusion value; is the true value, which is obtained by field measurement; the loss function L Fu Used to measure the error of the current BP neural network secondary data fusion to facilitate subsequent network training; (4) According to the loss function, back propagation is performed based on gradient descent to update the network parameters; (5) The result obtained from the first-level fusion is input into the trained second-level data fusion BP neural network again to obtain the final second-level data fusion result.
4. A method for predicting power line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of mining the horizontal and vertical time series rules of the K-Mediods-based fusion data of ice dancing on power transmission lines is as follows: (1) Initialize the cluster center and randomly select K te The transmission line temperature fusion data samples are used as the initial longitudinal clustering center point set; (2) Calculate the Euclidean distances from all sample points in the fused data set to the center points of each vertical cluster in turn, and place the sample points into the cluster represented by the nearest vertical cluster center point. The calculation formula of the Euclidean distance is expressed as k te is an initial random cluster center point, is the value of the temperature fusion data at the i-th sample point; (3) After the vertical clustering is completed, each sample point in each cluster is selected as the vertical cluster center, the Euclidean distance is recalculated, and the sample point with the smallest total distance within the cluster is selected as the new center point, which is expressed as is the sample point in the cluster; k te is the vertical cluster center; (4) When the set of vertical clustering center points remains unchanged, the algorithm terminates and the final vertical clustering result is output as in The vertical fusion data of temperature Otherwise, update the cluster center points and repeat S1.3.1 to S1.3.4 until the cluster center point set remains unchanged; (5) Based on the above method, the vertical clustering results of humidity and wind speed are obtained in turn. Among them, K hu , K vc are the longitudinal clustering numbers of humidity fusion data and wind speed fusion data, respectively; (6) The three types of fused data are mixed horizontally. The horizontal data format of a sample point of the fused mixed data is: ω te ,ω hu ,ω vc are the normalized weights of the corresponding fused data. The fused mixed data is clustered horizontally, and the results are expressed as K mix is the number of clusters for fusion mixed data.
5. The method for predicting power transmission line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of processing the time series data according to the horizontal and vertical time series rules to obtain the horizontal and vertical uneven time series perception adjustment parameters is: In the formula, α te , β te , ε te represents the temperature perception parameter, which jointly determines the construction of the time-distance function and can adaptively adjust the time-distance function according to the data of different subclasses; d te (k) represents the time interval before adjustment of the kth subclass in the temperature clustering result; α mix , β mix , ε mix represents the lateral perception parameter; d mix ={d mix (1),d mix (2),…,d mix (k),…,d mix (K mix )} is the set of horizontally staggered time intervals; d mix (k) represents the time interval before adjustment of the kth subclass in the mixed data clustering results.
6. The method for predicting power transmission line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of using the obtained parameters to improve the traditional GRU gating structure is: Combining the improved time-series perception function with the traditional GRU, the update methods of each gate structure are as follows In the formula, Represents the current input unit after interaction update; is the hidden layer output after interactive update; It is an update unit that integrates time-distance information; It is a reset unit that integrates time-distance information; is the summary of the input and past hidden layer states; is the hidden layer state output; is the weight parameter; the GRU unit comprehensively considers the prediction data of all clustering results, and the final prediction output is expressed as 7. The method for predicting power line galloping by integrating optical fiber monitoring and refined meteorological data according to claim 1, characterized in that: The specific method of using the attention closed-loop mechanism to improve the accuracy of model training is: The loss function and attention closed-loop training method are expressed as: In the formula, is the predicted network parameter; ρ is the learning rate; Based on the prediction network training of the uneven time series perception data, the prediction data sets of temperature, humidity and wind speed are obtained respectively. Taking temperature as an example, it is expressed as Similarly, the predicted data set for humidity and wind speed is The transmission line galloping risk coefficient is expressed as In the formula, Respectively represent the predicted values of temperature, wind speed and humidity at time t; T ice is the duration of ice cover; is the dancing risk factor per unit time; Further based on the mathematical statistics principle and the actual line ice galloping data, the galloping risk coefficient results of different galloping events are classified, and the transmission line ice galloping risk level characterized by the galloping risk coefficient is obtained, which is expressed as 8. A transmission line galloping prediction system integrating optical fiber monitoring and refined meteorological data, characterized by: It includes weather forecast module, optical fiber monitoring sensor module, data fusion module, uneven time series perception module, and transmission line wave prediction module; The weather forecast module is responsible for forecasting various weather data such as temperature, humidity and wind speed, and transmitting the obtained data to the data fusion module; The optical fiber monitoring sensor module is responsible for collecting optical fiber sensing data generated by Brillouin scattering, Raman scattering and Rayleigh scattering, and transmitting the obtained data to the data fusion module; The data fusion module is responsible for fusing meteorological data and optical fiber monitoring sensor time series data, and transmitting the obtained data to the time series data clustering module for clustering; The time series data clustering module is responsible for data clustering, mining the horizontal and vertical time series rules of the time series data, and transmitting the clustering data to the uneven time series perception module; The staggered timing perception module is responsible for adaptively perceiving the longitudinal and transverse staggered timing, obtaining prediction network parameters, and transmitting the parameters to the prediction network construction module; The prediction network construction module is responsible for constructing a time series prediction network using parameters, training the prediction model, and transmitting the trained model to the transmission line galloping prediction module; The transmission line galloping prediction module is responsible for using the trained network and time series data to make predictions, obtain transmission line galloping prediction results and perform closed-loop feedback to strengthen training.
9. A transmission line galloping prediction device integrating optical fiber monitoring and refined meteorological data, characterized in that: include Data fusion module: used to perform primary data fusion of optical fiber real-time monitoring data and meteorological forecast data based on adaptive weighting, and secondary data fusion learning and capturing of potential time series patterns in data based on BP neural network, and at the same time, to mine horizontal and vertical time series patterns of ice dancing fusion data of power transmission lines based on K-Mediods, so as to improve the depth and accuracy of mining horizontal and vertical time series patterns of data; Prediction module: It is used to process time series data according to the horizontal and vertical time series rules, obtain the horizontal and vertical uneven time series perception adjustment parameters, use the obtained parameters to improve the traditional GRU gating structure, reduce the prediction deviation of the network, and use the attention closed-loop mechanism to improve the accuracy of model training, and finally realize the transmission line sway prediction that integrates optical fiber monitoring and refined meteorological data.
10. The transmission line galloping prediction device integrating optical fiber monitoring and refined meteorological data according to claim 9, characterized in that: In the data fusion module, the specific method for mining the horizontal and vertical time series rules of the ice dancing fusion data of the transmission line based on K-Mediods is: (1) Initialize the cluster center and randomly select K te The transmission line temperature fusion data samples are used as the initial longitudinal clustering center point set; (2) Calculate the Euclidean distances from all sample points in the fused data set to the center points of each vertical cluster in turn, and place the sample points into the cluster represented by the nearest vertical cluster center point. The calculation formula of the Euclidean distance is expressed as k te is an initial random cluster center point, is the value of the temperature fusion data at the i-th sample point; (3) After the vertical clustering is completed, each sample point in each cluster is selected as the vertical cluster center, the Euclidean distance is recalculated, and the sample point with the smallest total distance within the cluster is selected as the new center point, which is expressed as is the sample point in the cluster; k te is the vertical cluster center; (4) When the set of vertical clustering center points remains unchanged, the algorithm terminates and the final vertical clustering result is output as in The vertical fusion data of temperature Otherwise, update the cluster center points and repeat S1.3.1 to S1.3.4 until the cluster center point set remains unchanged; (5) Based on the above method, the vertical clustering results of humidity and wind speed are obtained in turn. Among them, K hu , K vc are the longitudinal clustering numbers of humidity fusion data and wind speed fusion data, respectively; (6) The three types of fused data are mixed horizontally. The horizontal data format of a sample point of the fused mixed data is: ω te ,ω hu ,ω vc are the normalized weights of the corresponding fused data. The fused mixed data is clustered horizontally, and the results are expressed as K mix is the number of clusters for fusion mixed data.
11. The transmission line galloping prediction device integrating optical fiber monitoring and refined meteorological data according to claim 9, characterized in that: In the prediction module, the specific method of using the obtained parameters to improve the traditional GRU gating structure is: Combining the improved time-series perception function with the traditional GRU, the update methods of each gate structure are as follows In the formula, Represents the current input unit after interaction update; is the hidden layer output after interactive update; It is an update unit that integrates time-distance information; It is a reset unit that integrates time-distance information; is the summary of the input and past hidden layer states; is the hidden layer state output; is the weight parameter; the GRU unit comprehensively considers the prediction data of all clustering results, and the final prediction output is expressed as