Multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP

Abstract

The present invention discloses a multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP algorithm, and relates to the technical field of tree-line distance prediction. The method includes: calculating the transmission line tower displacement data in real time based on the center frame difference method; calculating the transmission line tree-line distance based on monocular vision technology; obtaining multi-parameter data related to the tree-line distance; pre-processing the tower displacement data, the transmission line tree-line distance, and the multi-parameter data; constructing a PSO-BP multi-parameter fusion prediction model for tree-line distance prediction, and generating auxiliary decision-making information based on the predicted tree-line distance change trend to optimize the safety maintenance strategy of the power line. The present invention fuses the tower displacement, tree-line distance, and multi-parameter data into input, greatly improving the accuracy of the tree-line distance prediction, and helping to reduce the error caused by environmental changes; introducing the idea of ​​variation to prevent the algorithm from falling into local optimality, significantly improving the prediction accuracy and response speed.

Description

Technical Field

[0001] The present invention relates to the technical field of tree-line distance prediction, and in particular to a multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP. Background Art

[0002] The safe operation of power transmission lines is closely linked to the conditions of their surroundings, especially in complex terrain where trees too close to conductors can easily cause power outages. Traditional manual inspection methods are not only labor-intensive and resource-intensive, but also inefficient and time-consuming, making them unable to meet the real-time monitoring needs of high-risk areas.

[0003] With the expansion of smart grids, automated monitoring systems, such as lidar and ultrasonic rangefinders, have been developed in recent years to monitor the distance to tree barriers around power transmission lines. However, these systems are unable to cope with complex environmental conditions. For example, in complex terrains such as mountainous areas and forests, where tree growth rates and environmental conditions vary, existing monitoring equipment exhibits reduced accuracy in the face of these changing conditions, and the system's reliability is insufficient, making it difficult to meet the high-frequency, real-time monitoring needs.

[0004] Furthermore, traditional forecasting models are limited in scope, typically performing simple linear predictions based on a small number of parameters. They lack the ability to handle multi-parameter, nonlinear relationships, leading to large forecast errors and an inability to provide accurate early warnings of future risks. Most ranging systems rely on periodic data collection and processing, failing to provide timely warnings and forecasts, especially in situations with rapid tree growth or drastic weather changes. Summary of the Invention

[0005] In response to the problems of real-time performance, accuracy, and dynamic multi-parameter processing faced in existing transmission line tree-line distance monitoring and prediction, the present invention proposes a multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP algorithm, which comprehensively considers various environmental and line operation parameters and improves real-time response capabilities.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] In one aspect, the present invention provides a multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP algorithm, the method comprising:

[0008] Real-time calculation of transmission line tower displacement data based on the center frame difference method;

[0009] Calculate the distance between transmission lines and trees based on monocular vision technology;

[0010] Acquire multi-parameter data related to treeline distance, including historical and real-time data on span, height difference between two towers, conductor inclination, swing angle, and conductor operating temperature;

[0011] Preprocessing the tower displacement data, transmission line tree-line distance and multi-parameter data;

[0012] The pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data are fused and the fused data are used as comprehensive input features and input into the BP neural network;

[0013] The number of nodes in the input layer of the BP neural network is matched with the dimension of the multi-parameter data, and the hidden layer adopts a nonlinear activation function;

[0014] The BP neural network is optimized using the Mutation-PSO algorithm, and the optimal weights and thresholds obtained are assigned to the BP neural network to form an optimized PSO-BP multi-parameter fusion prediction model for real-time prediction of tree line distance.

[0015] As a preferred solution of the present invention, the real-time calculation of the transmission line tower displacement data based on the center frame difference method specifically includes:

[0016] Collecting tower displacement images and performing preprocessing, wherein the preprocessing includes distortion correction and color space conversion, and converting the images from RGB color space to HSV color space frame by frame;

[0017] The marker points in the image are clustered according to the DBSCAN clustering algorithm, and the center position of each marker point is identified by the centroid method, and the marker points are numbered according to their longitudinal coordinates;

[0018] The conversion factor is obtained through camera calibration, and the pixel displacement of the marker point is calculated based on the center frame difference method, which is the number of pixels occupied by the tower displacement, thereby obtaining the tower displacement data.

[0019] As a preferred solution of the present invention, the conversion factor is obtained by camera calibration, and the pixel displacement of the marker point is calculated based on the center frame difference method, specifically:

[0020] The conversion factor sf is obtained through the plane homography, the formula is:

[0021]

[0022] Where, f pix is the focal length of the camera lens, d is the distance from the camera optical center to the structural surface;

[0023] Or the conversion factor sf can be obtained by the proportional factor method, the formula is:

[0024]

[0025] Where P1Q1 is the pixel length corresponding to the marker point in the image; P s Q s is the physical distance of the same marker point in world coordinates;

[0026] The initial frame in the video sequence is used as the reference frame. For each frame, pixel-level differential analysis is performed with the reference frame to determine the pixel displacement between the current position of the tower marker and the reference position. The formula is:

[0027] △x=x current -x reference , △y=y current -y reference ;

[0028] Where △x and △y are the pixel displacements in the horizontal and vertical directions; x current 、y current is the position of the marker in the current frame; x reference 、y reference is the position of the same marker in the reference frame;

[0029] Combined with the conversion factor sf, the calculated pixel displacement is converted into the actual physical distance.

[0030] As a preferred solution of the present invention, the calculation of the transmission line tree-line distance based on monocular vision technology specifically includes:

[0031] After the monocular camera is installed and takes the first shot, use the first image as a sample reference image;

[0032] Analyze the tree features in the sample reference image and adjust the value range of the HSV color space to identify the trees in the image;

[0033] The HSV value range determined in the sample reference image is used as the HSV threshold value in subsequent monitoring images to automatically extract trees in the image;

[0034] A monocular camera is used to capture a wide-angle image of the power transmission line and its surroundings, and the wide-angle image is converted into an HSV color space; and an image processing technique is used to identify trees in the wide-angle image based on an HSV threshold.

[0035] The tree-line distance is calculated based on the camera imaging model and the spatial state parameters of the wires, including the vertical distance to the tree line, the horizontal distance to the tree line, and the tree-line clearance distance, thereby determining the three-dimensional position of the trees. Specifically, a monocular camera fixed to a tower monitors the area between two adjacent towers, and geometric imaging models are established for the towers and trees respectively. The calculation formula for the tree-line distance is derived based on the geometric imaging models.

[0036] The calculation formula for the vertical distance Ty from the tree line is:

[0037]

[0038] Where GH is the vertical pixel height of the tree in the image; GI is the pixel height of the tree crown projected on the screen to the image center I; Cy is the known camera installation height;

[0039] The calculation formula for the horizontal distance Tx from the tree line is:

[0040]

[0041] Where PQ and UV represent the known horizontal components and horizontal pixel distances on the tower, respectively; IG is the horizontal pixel distance from the tree crown G to the image center I; Pz is the known distance from the tower to the camera; Tz is the distance between the tree and the tower;

[0042]

[0043] Where Py is the known vertical height of the tower; CD is the pixel height of the image of the component Py of known height on the tower;

[0044] Calculate the tree line clearance distance Tn:

[0045]

[0046] Where h1 is the actual height of the tree; x is the distance from the intersection of the wire and the tree to the tower; l OA 、h OA They represent the horizontal distance and vertical distance from the camera center point O to the bottom point A of the tower respectively; M is the scale factor used for the conversion between pixel distance and actual distance; θ is the camera shooting angle.

[0047] As a preferred solution of the present invention, the tower displacement data, transmission line tree-line distance and multi-parameter data are pre-processed, including:

[0048] Data verification and multi-source data synchronization: Use timestamps and spatial distribution characteristics to cross-check tower displacement data, tree line distance and multi-parameter data, set timestamp error threshold △T, if the multi-source data at the same time meets the condition |T k -T j |<△T, it is considered as synchronous data, otherwise data compensation is performed; where T k 、T j Represent the timestamps of data source k and data source j respectively;

[0049] Abnormal data removal: Set the mean μ and standard deviation σ of each feature, and define the dynamic abnormal threshold as [μ-3σ, μ+3σ]. If the feature value of any data exceeds the dynamic abnormal threshold range, it will be regarded as abnormal data and removed;

[0050] Normalization and standardization: The maximum-minimum normalization combined with the standardization method is used to adjust the data to the range of [0,1]. In the standardization stage, the feature mean is adjusted to 0 and the standard deviation is adjusted to 1;

[0051] Feature enhancement: Feature extraction is performed based on principal component analysis. The original data matrix X is decomposed into a feature matrix X by principal component analysis. PCA , select the first m principal components with the highest contribution rate to form a simplified data set, and assign feature weights in combination with the entropy weight method;

[0052] Multi-layer wavelet denoising: Perform wavelet decomposition on the signal to obtain sub-signals in different frequency bands, retaining low-frequency signals to suppress high-frequency noise;

[0053] Missing data filling: The missing data are filled using the dynamic interpolation method based on Bayesian estimation. Let the observed data be X obs , the missing data is X miss , and the posterior probability distribution is obtained through Bayesian inference:

[0054] p(X miss |X obs )∝p(X obs |X miss )p(X miss );

[0055] Where, p(X miss |X obs ) is the missing data X miss In the observed data X obs The posterior probability under given conditions; p(X obs |X miss ) is the observed data X obs In the case of missing data X miss Likelihood function under given conditions; p(X miss ) is the missing data X miss The prior probability of

[0056] The missing values ​​are estimated by sampling from the posterior probability distribution to fill the missing data.

[0057] As a preferred solution of the present invention, the fusion of the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data specifically includes:

[0058] Perform independent component analysis on the preprocessed data to decompose each feature into independent signal components. This maximizes signal independence to achieve feature decoupling and eliminate redundant information between features.

[0059] Introducing a context-aware mechanism to enhance the relevance of current data by modeling the memory of historical data; assuming that the current feature Z t The context fusion output is C t , the calculation formula is:

[0060] C t =f(Z t ,Z t-1 ,Z t-2 ,…,Z t-b );

[0061] Where f represents the context dependency function, Z t-1 represents the characteristics at time t-1, Z t-2 represents the characteristics at time t-2, Z t-b Indicates the characteristics at time tb;

[0062] Build the context of the current feature through historical data at b moments to improve the model's ability to capture dynamic changes;

[0063] Fuzzy logic is introduced into the data fusion process, and the weight of each feature is dynamically adjusted through the membership function. Let feature Z s The weight adjustment function is:

[0064] w s =t(Z s )·w base +(1-t(Z s ))·w context ;

[0065] Where w s is the fused feature Z s The final weight of t(Z s ) represents feature Z s The fuzzy membership function has a range of [0,1]; w base is the benchmark weight; w context Represents the context weight.

[0066] As a preferred solution of the present invention, the BP neural network is optimized using the Mutation-PSO algorithm, specifically:

[0067] Initialize the particle group in the PSO algorithm, each particle represents the weight and threshold of the BP neural network, and randomly assign initial values ​​to the particle position and velocity;

[0068] According to the prediction error of the treeline distance, the fitness value of each particle is calculated using the Huber-Loss function. The smaller the fitness value, the higher the prediction accuracy.

[0069] In each iteration, the current fitness of each particle is compared with its historical best fitness. If the current fitness is better, the individual optimal value of the particle is updated. At the same time, the global optimal value of the entire particle swarm is updated to ensure that the particle swarm gradually approaches the optimal solution.

[0070] Randomly select particles from the particle swarm to perform mutation operations, randomly perturb the position and velocity of the particles to expand the search range and avoid falling into the local optimal solution. Recalculate the fitness of the particles after the mutation operation to determine whether the mutation has improved the prediction accuracy.

[0071] Adjust the position and velocity of all particles according to the following update formula:

[0072] v i (t+1)=ω·v i (t)+c1·rand1·(p best -x i (t))+c2·rand2·(g best -x i (t));

[0073] Where, v i (t+1) represents the velocity of particle i at the t+1th iteration, v i (t) represents the velocity of particle i at the tth iteration; ω is the inertia weight; c1 and c2 are both learning factors; rand1 and rand2 are both random factors; p best is the individual optimal position of the particle, g bes t is the global optimal position; x i (t) represents the current position of particle i at the tth iteration;

[0074] Repeat the above fitness calculation to position and speed adjustment steps until the fitness reaches the preset accuracy requirement or the number of iterations reaches the maximum limit, and finally obtain the optimal weights and thresholds of the optimized BP neural network.

[0075] As a preferred solution of the present invention, the fitness value of each particle is calculated using the Huber-Loss function, which is defined as follows:

[0076]

[0077] Where, is the Huber-Loss function value; y is the actual tree line distance, is the treeline distance predicted by BP neural network; δ is the adjustment parameter.

[0078] As a preferred solution of the present invention, the method further includes constructing an expert system, which is integrated into the PSO-BP multi-parameter fusion prediction model to analyze the deviation between the prediction results and the real data. The steps of constructing the expert system are as follows:

[0079] Receive output data from the PSO-BP multi-parameter fusion prediction model and real-time monitoring data;

[0080] Compare predicted output data with real-time data to identify deviations between the data;

[0081] Based on the deviation analysis results, correction suggestions are generated through an expert rule base, where the rule base is composed of historical data and prior knowledge;

[0082] Output correction suggestions to the PSO-BP multi-parameter fusion prediction model and adjust model parameters in real time.

[0083] On the other hand, the present invention also provides a multi-parameter transmission line tree-line distance prediction system based on an improved PSO-BP, which is applied to the multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP, and the system includes:

[0084] The tower displacement calculation module is used to calculate the tower displacement data of the transmission line in real time based on the center frame difference method;

[0085] The tree-line distance calculation module is used to calculate the tree-line distance of transmission lines based on monocular vision technology;

[0086] A multi-parameter data acquisition module is used to acquire multi-parameter data related to the tree-line distance, including historical and real-time data on the span, height difference between the two towers, conductor inclination, swing angle, and conductor operating temperature;

[0087] A data preprocessing module, used for preprocessing the tower displacement data, transmission line tree-line distance and multi-parameter data;

[0088] The prediction model building module is used to build a PSO-BP multi-parameter prediction model for treeline distance prediction, including a data fusion unit, a dimension matching unit, and an optimization unit;

[0089] The data fusion unit is used to fuse the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data, and input the fused data as comprehensive input features into the BP neural network;

[0090] The dimension matching unit is used to match the number of input layer nodes of the BP neural network with the dimension of the multi-parameter data, and the hidden layer adopts a nonlinear activation function;

[0091] The optimization unit is used to optimize the BP neural network using the Mutation-PSO algorithm, assign the optimized optimal weights and thresholds to the BP neural network, and form an optimized PSO-BP multi-parameter fusion prediction model;

[0092] The treeline distance prediction module is used to make real-time predictions of treeline distances using the optimized PSO-BP multi-parameter fusion prediction model.

[0093] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-parameter transmission line tree-line distance prediction method based on the improved PSO-BP.

[0094] The beneficial effects of the present invention are as follows: the central frame difference method and monocular vision technology are used to calculate the tower displacement and tree-line distance of the transmission line in real time, which can monitor the line status in real time and effectively reduce the safety hazards caused by tower displacement or conductor contact with trees, thereby improving the monitoring accuracy of the line; the preprocessing process of tower displacement data, tree-line distance and multi-parameter data reduces the misjudgment caused by data noise, ensuring the reliability of subsequent data fusion and prediction; the fusion of multiple environmental factors such as conductor inclination, swing angle and operating temperature improves the sensitivity to changes in the natural environment; the PSO-BP multi-parameter fusion prediction model is constructed, The model uses tower displacement, tree-line distance, and multi-parameter data as comprehensive input features. Compared with single-data prediction models, it significantly improves the accuracy of tree-line distance prediction and helps reduce errors caused by environmental changes. Optimization is performed on the basis of PSO-BP, introducing mutation ideas to prevent the algorithm from falling into local optimality, significantly improving the prediction accuracy and response speed. Auxiliary decision-making information is generated based on the real-time prediction results of tree-line distance, which can prompt line maintenance personnel to take measures such as pruning trees or adjusting lines in advance, thereby effectively optimizing the maintenance strategy of power lines and reducing the cost of daily inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0096] in:

[0097] Figure 1is a flow chart of the method of the present invention;

[0098] Figure 2 This is a flow chart of optimizing a BP neural network using the Mutation-PSO algorithm in an embodiment of the present invention;

[0099] Figure 3 This is a system modular structure diagram of the present invention. DETAILED DESCRIPTION

[0100] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0101] like Figure 1 、 Figure 2 FIG. 1 is an embodiment of the present invention, which provides a multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP, including the following steps:

[0102] S1: Real-time calculation of transmission line tower displacement data based on the center frame difference method, specifically including:

[0103] S11: Collect the tower displacement image and perform preprocessing, including but not limited to distortion correction and color space conversion, and convert the image from RGB (Red, Green, Blue) color space to HSV (Hue, Saturation, Value) color space frame by frame;

[0104] S12: Cluster the markers in the image according to the DBSCAN clustering algorithm, identify the center position of each marker using the centroid method, and number the markers according to their longitudinal coordinates;

[0105] S13: Obtain the conversion factor through camera calibration, and calculate the pixel displacement of the marker point based on the center frame difference method, that is, the number of pixels occupied by the tower displacement, thereby obtaining the tower displacement data.

[0106] Furthermore, step S13 includes:

[0107] The conversion factor sf is obtained through the plane homography, the formula is:

[0108]

[0109] Where, f pixis the focal length of the camera lens, d is the distance from the camera optical center to the structural surface;

[0110] Or the conversion factor sf can be obtained by the proportional factor method, the formula is:

[0111]

[0112] Where P1Q1 is the pixel length corresponding to the marker point in the image; P s Q s is the physical distance of the same marker point in world coordinates;

[0113] The initial frame in the video sequence is used as the reference frame. For each frame, pixel-level differential analysis is performed with the reference frame to determine the pixel displacement between the current position of the tower marker and the reference position. The formula is:

[0114] △x=x current -x reference , △y=y current -y reference ;

[0115] Where △x and △y are the pixel displacements in the horizontal and vertical directions; x current 、y current is the position of the marker in the current frame; x reference 、y reference is the position of the same marker in the reference frame;

[0116] Combined with the conversion factor sf, the calculated pixel displacement is converted into the actual physical distance.

[0117] S2: Calculate the distance between transmission lines and trees based on monocular vision technology, including:

[0118] S21: After the monocular camera is installed and takes the first shot, the first image is used as the sample reference image;

[0119] S22: analyzing tree features in the example reference image and adjusting the value range of the HSV color space to identify the trees in the image;

[0120] S23: Using the HSV value range determined in the example reference image as an HSV threshold value for use in subsequent monitoring images to automatically extract trees in the image;

[0121] S24: Capture wide-angle images of the power transmission line and its surroundings using a monocular camera and convert them into the HSV color space. Use image processing techniques based on HSV thresholds to identify trees in the wide-angle images.

[0122] S25: Calculate the tree-line distance based on the camera imaging model and the conductor spatial state parameters, including the vertical distance to the tree-line, the horizontal distance to the tree-line, and the tree-line clearance distance, thereby determining the three-dimensional position of the trees. Specifically, a monocular camera fixed to a tower monitors the area between two adjacent towers, establishes geometric imaging models for the towers and the trees respectively, and derives a calculation formula for the tree-line distance based on the geometric imaging model.

[0123] The calculation formula for the vertical distance Ty from the tree line is:

[0124]

[0125] Where GH is the vertical pixel height of the tree in the image; GI is the pixel height of the tree crown projected on the screen to the image center I; Cy is the known camera installation height;

[0126] The calculation formula for the horizontal distance Tx from the tree line is:

[0127]

[0128] Where PQ and UV represent the known horizontal components and horizontal pixel distances on the tower, respectively; IG is the horizontal pixel distance from the tree crown G to the image center I; Pz is the known distance from the tower to the camera; Tz is the distance between the tree and the tower;

[0129]

[0130] Where Py is the known vertical height of the tower; CD is the pixel height of the image of the component Py of known height on the tower;

[0131] Calculate the tree line clearance distance Tn:

[0132]

[0133] Where h1 is the actual height of the tree; x is the distance from the intersection of the wire and the tree to the tower; l OA 、h OA They represent the horizontal distance and vertical distance from the camera center point O to the bottom point A of the tower respectively; M is the scale factor used for the conversion between pixel distance and actual distance; θ is the camera shooting angle.

[0134] S3: Acquire multi-parameter data related to treeline distance, including but not limited to historical and real-time data on span, height difference between two towers, conductor inclination, swing angle, and conductor operating temperature;

[0135] Specifically, sensors distributed along the transmission lines collect multi-parameter data in real time and record them to form a historical database;

[0136] S4: Preprocess the tower displacement data, transmission line tree-line distance and multi-parameter data to eliminate data noise and dimensional differences and ensure the validity of the data input model;

[0137] In one specific embodiment, step S4 is implemented as follows:

[0138] S41: Data verification and multi-source data synchronization: Use timestamps and spatial distribution characteristics to cross-check tower displacement data, tree line distance and multi-parameter data, set timestamp error threshold △T, if the multi-source data at the same time meets the condition T k -T j |<△T, it is considered as synchronous data, otherwise data compensation is performed; where T k 、T j Represent the timestamps of data source k and data source j respectively;

[0139] S42: Abnormal data removal: Set the mean μ and standard deviation σ of each feature, and define the dynamic abnormal threshold as

[0140] [μ-3σ,μ+3σ], if the characteristic value of any data exceeds the dynamic abnormal threshold range, it will be regarded as abnormal data and removed;

[0141] S43: Normalization and standardization: The maximum-minimum normalization combined with the standardization method is used to adjust the data to the range of [0,1]. In the standardization stage, the feature mean is adjusted to 0 and the standard deviation is adjusted to 1;

[0142] S44: Feature enhancement: Feature extraction is performed based on principal component analysis. The original data matrix X is decomposed into a feature matrix X by principal component analysis. PCA , select the first m principal components with the highest contribution rate to form a simplified data set, and assign feature weights in combination with the entropy weight method;

[0143] S45: Multi-layer wavelet denoising: Perform wavelet decomposition on the signal to obtain sub-signals in different frequency bands, retaining low-frequency signals to suppress high-frequency noise;

[0144] S46: Missing data filling: The missing data are filled using the dynamic interpolation method based on Bayesian estimation. Suppose the observed data is X obs , the missing data is X miss , and the posterior probability distribution is obtained through Bayesian inference:

[0145] p(X miss |X obs )∝p(X obs |X miss )p(X miss );

[0146] Where, p(Xmiss |X obs ) is the missing data X miss In the observed data X obs The posterior probability under given conditions; p(X obs |X miss ) is the observed data X obs In the case of missing data X miss Likelihood function under given conditions; p(X miss ) is the missing data X miss The prior probability of

[0147] The missing values ​​are estimated by sampling from the posterior probability distribution to fill the missing data.

[0148] S5: Construct a PSO-BP (combining Particle Swarm Optimization (PSO) and Back Propagation (BP) neural network) multi-parameter fusion prediction model for treeline distance prediction, including:

[0149] S51: Fusing the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data, and using the fused data as comprehensive input features to input into a BP neural network (Back Propagation Neural Network);

[0150] Furthermore, step S51 is implemented by:

[0151] Independent Component Analysis (ICA) is performed on the preprocessed data to decompose each feature into independent signal components. This maximizes signal independence to achieve feature decoupling and eliminate redundant information between features.

[0152] Introducing a context-aware mechanism to enhance the relevance of current data by modeling the memory of historical data; assuming that the current feature Z t The context fusion output is C t , the calculation formula is:

[0153] C t =f(Z t ,Z t-1 ,Z t-2 ,…,Z t-b );

[0154] Where f represents the context dependency function, Z t-1 represents the characteristics at time t-1, Z t-2 represents the characteristics at time t-2, Zt-b Indicates the characteristics at time tb;

[0155] Build the context of the current feature through historical data at b moments to improve the model's ability to capture dynamic changes;

[0156] Fuzzy logic is introduced into the data fusion process, and the weight of each feature is dynamically adjusted through the membership function. Let feature Z s The weight adjustment function is:

[0157] w s =t(Z s )·w base +(1-t(Z s ))·w context ;

[0158] Where w s is the fused feature Z s The final weight of t(Z s ) represents feature Z s The fuzzy membership function has a range of [0,1]; w base is the benchmark weight, i.e., feature Z s The default weight when there is no context influence; w context Represents the context weight.

[0159] S52: Match the number of input layer nodes of the BP neural network with the dimension of the multi-parameter data, and use a nonlinear activation function in the hidden layer;

[0160] S53: Optimizing the BP neural network using the Mutation-PSO algorithm (mutation particle swarm optimization algorithm), assigning the optimal weights and thresholds obtained from the optimization to the BP neural network to form an optimized PSO-BP multi-parameter fusion prediction model. This model performs real-time prediction of the tree-line distance and generates auxiliary decision-making information based on the predicted tree-line distance change trend to optimize the safety maintenance strategy of the power line.

[0161] The Mutation-PSO algorithm is used to optimize the BP neural network, specifically:

[0162] Particle swarm initialization: Initialize the particle swarm in the PSO algorithm. Each particle represents the weight and threshold of the BP neural network, and randomly assigns initial values ​​to the particle position and velocity.

[0163] Fitness calculation: Based on the prediction error of the treeline distance, the Huber-Loss function is used to calculate the fitness value of each particle. The smaller the fitness value, the higher the prediction accuracy.

[0164] Update of individual optimal values ​​and global optimal values: In each iteration, the current fitness of each particle is compared with its historical best fitness. If the current fitness is better, the individual optimal value of the particle is updated. At the same time, the global optimal value of the entire particle swarm is updated to ensure that the particle swarm gradually approaches the optimal solution.

[0165] Mutation operation: Randomly select particles in the particle swarm to perform mutation operation, randomly perturb the position and velocity of the particles to expand the search range and avoid falling into the local optimal solution. The fitness of the particles after the mutation operation is recalculated to determine whether the mutation has improved the prediction accuracy.

[0166] Position and velocity adjustment: Adjust the position and velocity of all particles according to the following update formula:

[0167] v i (t+1)=ω·v i (t)+c1·rand1·(p best -x i (t))+c2·rand2·(g best -x i (t));

[0168] Where, v i (t+1) represents the velocity of particle i at the t+1th iteration, v i (t) represents the velocity of particle i at the tth iteration; ω is the inertia weight; c1 and c2 are both learning factors; rand1 and rand2 are both random factors; p best is the individual optimal position of the particle, g best is the global optimal position; x i (t) represents the current position of particle i at the tth iteration;

[0169] Iteration and convergence: Repeat the above steps of fitness calculation, individual optimal value and global optimal value update, mutation operation and position and speed adjustment until the fitness reaches the preset accuracy requirement or the number of iterations reaches the maximum limit, and finally obtain the optimal weights and thresholds of the optimized BP neural network.

[0170] Furthermore, the Huber-Loss function is used to calculate the fitness value of each particle, which is defined as follows:

[0171]

[0172] Where, is the Huber-Loss function value, which is used to measure the error between the predicted value and the actual value; y is the actual tree line distance, is the treeline distance predicted by the BP neural network; δ is an adjustment parameter used to control the switching between MSE and MAE; when the prediction error is small, Huber Loss is similar to MSE; when the error is large, Huber-Loss is similar to MAE, thereby reducing the sensitivity to outliers and improving the stability of the model. The smaller the fitness value, the higher the prediction accuracy.

[0173] In another specific embodiment, the method of the present invention further includes constructing an expert system, which is integrated into the PSO-BP multi-parameter fusion prediction model to analyze the deviation between the prediction results and the real data. The steps of constructing the expert system are as follows:

[0174] Receive output data from the PSO-BP multi-parameter fusion prediction model and real-time monitoring data;

[0175] Compare predicted output data with real-time data to identify deviations between the data;

[0176] Based on the deviation analysis results, correction suggestions are generated through an expert rule base, where the rule base is composed of historical data and prior knowledge;

[0177] Output correction suggestions to the PSO-BP multi-parameter fusion prediction model, and adjust the model parameters in real time to improve the accuracy of treeline distance prediction and the adaptability of the model.

[0178] like Figure 3 FIG. 1 is another embodiment of the present invention, which provides a multi-parameter transmission line tree-line distance prediction system based on an improved PSO-BP, which is applied to the multi-parameter transmission line tree-line distance prediction method based on an improved PSO-BP as described above. The system includes:

[0179] The tower displacement calculation module is used to calculate the tower displacement data of the transmission line in real time based on the center frame difference method;

[0180] The tree-line distance calculation module is used to calculate the tree-line distance of transmission lines based on monocular vision technology;

[0181] Multi-parameter data acquisition module, used to obtain multi-parameter data related to tree line distance, including historical and real-time data of span, height difference between two towers, conductor inclination, swing angle and conductor operating temperature;

[0182] Data preprocessing module, used to preprocess tower displacement data, transmission line tree-line distance and multi-parameter data;

[0183] The prediction model building module is used to build a PSO-BP multi-parameter prediction model for treeline distance prediction, including:

[0184] The data fusion unit is used to fuse the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data, and input the fused data as comprehensive input features into the BP neural network;

[0185] Dimension matching unit, used to match the number of input layer nodes of BP neural network with the dimension of multi-parameter data, and the hidden layer adopts nonlinear activation function;

[0186] The optimization unit is used to optimize the BP neural network using the Mutation-PSO algorithm, assign the optimal weights and thresholds obtained by optimization to the BP neural network, and form an optimized PSO-BP multi-parameter fusion prediction model;

[0187] The treeline distance prediction module is used to make real-time predictions of treeline distances using the optimized PSO-BP multi-parameter fusion prediction model.

[0188] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the steps of the multi-parameter transmission line tree-line distance prediction method based on the improved PSO-BP.

[0189] In summary, the present invention uses the central frame difference method and monocular vision technology to calculate the tower displacement and tree-line distance of the transmission line in real time, which can monitor the line status in real time and effectively reduce the safety hazards caused by tower displacement or wire contact with trees, thereby improving the monitoring accuracy of the line; the preprocessing process of tower displacement data, tree-line distance and multi-parameter data reduces the misjudgment caused by data noise, ensuring the reliability of subsequent data fusion and prediction; the fusion of multiple environmental factors such as wire inclination, swing angle and operating temperature improves the sensitivity to changes in the natural environment; constructs a PSO-BP multi-parameter fusion prediction model, Tower displacement, tree-line distance, and multi-parameter data are fed into the model as comprehensive input features. Compared with single-data prediction models, this significantly improves the accuracy of tree-line distance prediction and helps reduce errors caused by environmental changes. Optimization is performed on the basis of PSO-BP, introducing the idea of ​​mutation to prevent the algorithm from falling into local optimality, significantly improving the prediction accuracy and response speed. Auxiliary decision-making information is generated based on the real-time prediction results of tree-line distance, which can prompt line maintenance personnel to take measures such as pruning trees or adjusting lines in advance, thereby effectively optimizing the maintenance strategy of power lines and reducing the cost of daily inspections.

[0190] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP is characterized by: The method comprises: calculating the displacement data of the transmission line tower in real time based on the center frame difference method; Calculating the tree-line distance of transmission lines based on monocular vision technology involves calculating the tree-line distance based on a camera imaging model and conductor spatial state parameters, including vertical distance to the tree line, horizontal distance to the tree line, and tree-line clearance distance, thereby determining the three-dimensional position of trees. Specifically, a monocular camera is fixed to a tower to monitor the area between two adjacent towers, and geometric imaging models are established for the towers and trees respectively. A calculation formula for the tree-line distance is derived based on the geometric imaging model. The calculation formula for the vertical distance Ty from the tree line is: Where GH is the vertical pixel height of the tree in the image; GI is the pixel height of the tree crown projected on the screen to the image center I; Cy is the known camera installation height; The calculation formula for the horizontal distance Tx from the tree line is: Where PQ and UV represent the known horizontal components and horizontal pixel distances on the tower, respectively; IG is the horizontal pixel distance from the tree crown G to the image center I; Pz is the known distance from the tower to the camera; Tz is the distance between the tree and the tower; Where Py is the known vertical height of the tower; CD is the pixel height of the image of the component Py of known height on the tower; Calculate the tree line clearance distance Tn: Where h1 is the actual height of the tree; x is the distance from the intersection of the wire and the tree to the tower; l OA 、h OA They represent the horizontal distance and vertical distance from the camera center point O to the tower bottom point A respectively; M is the scale factor used for the conversion between pixel distance and actual distance; θ is the camera shooting angle; Acquire multi-parameter data related to treeline distance, including historical and real-time data on span, height difference between two towers, conductor inclination, swing angle, and conductor operating temperature; Preprocessing the tower displacement data, transmission line tree-line distance and multi-parameter data; The pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data are fused and the fused data are used as comprehensive input features and input into the BP neural network; The number of nodes in the input layer of the BP neural network is matched with the dimension of the multi-parameter data, and a nonlinear activation function is used in the hidden layer. The BP neural network is optimized using the Mutation-PSO algorithm, and the optimal weights and thresholds obtained by optimization are assigned to the BP neural network to form an optimized PSO-BP multi-parameter fusion prediction model for real-time prediction of treeline distance.

2. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 1 is characterized in that: The real-time calculation of the transmission line tower displacement data based on the center frame difference method specifically includes: Collecting tower displacement images and performing preprocessing, wherein the preprocessing includes distortion correction and color space conversion, and converting the images from RGB color space to HSV color space frame by frame; The marker points in the image are clustered according to the DBSCAN clustering algorithm, and the center position of each marker point is identified by the centroid method, and the marker points are numbered according to their longitudinal coordinates; The conversion factor is obtained through camera calibration, and the pixel displacement of the marker point is calculated based on the center frame difference method, which is the number of pixels occupied by the tower displacement, thereby obtaining the tower displacement data.

3. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 2 is characterized in that: The conversion factor is obtained by camera calibration, and the pixel displacement of the marker point is calculated based on the center frame difference method. Specifically, the conversion factor sf is obtained by plane homography, and the formula is: Where, f pix is the focal length of the camera lens, d is the distance from the camera optical center to the structural surface; Or the conversion factor sf can be obtained by the proportional factor method, the formula is: Where P1Q1 is the pixel length corresponding to the marker point in the image; P s Q s is the physical distance of the same marker point in the world coordinates; the initial frame in the video sequence is used as the reference frame. For each frame, pixel-level differential analysis is performed with the reference frame to determine the pixel displacement between the current position of the tower marker and the reference position. The formula is: Δx=x current -x reference ,Δy=y current -y reference ; Where Δx and Δy are the pixel displacements in the horizontal and vertical directions; x current 、y current is the position of the marker in the current frame; x reference 、y reference is the position of the same marker in the reference frame; Combined with the conversion factor sf, the calculated pixel displacement is converted into the actual physical distance.

4. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 2 is characterized in that: The method of calculating the distance between the transmission line and the tree line based on the monocular vision technology further includes: After the monocular camera is installed and takes the first shot, use the first image as a sample reference image; Analyze the tree features in the sample reference image and adjust the value range of the HSV color space to identify the trees in the image; use the HSV value range determined in the sample reference image as the HSV threshold and use it in subsequent monitoring images to automatically extract the trees in the image; A monocular camera is used to capture wide-angle images of the power transmission line and its surrounding environment, and the wide-angle images are converted into the HSV color space. Based on the HSV threshold, image processing technology is used to identify trees in the wide-angle images.

5. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 1 is characterized in that: Preprocessing the tower displacement data, transmission line tree-line distance and multi-parameter data includes: Data verification and multi-source data synchronization: Use timestamps and spatial distribution characteristics to cross-check tower displacement data, tree line distance and multi-parameter data, set timestamp error threshold ΔT, if the multi-source data at the same time meets the condition |T k -T j |<ΔT, it is considered as synchronous data, otherwise data compensation is performed; where T k 、T j Represent the timestamps of data source k and data source j respectively; Abnormal data removal: Set the mean μ and standard deviation σ of each feature, and define the dynamic abnormal threshold as [μ-3σ, μ+3σ]. If the feature value of any data exceeds the dynamic abnormal threshold range, it will be regarded as abnormal data and removed; Normalization and standardization: The maximum-minimum normalization combined with the standardization method is used to adjust the data to the range of [0,1]. In the standardization stage, the feature mean is adjusted to 0 and the standard deviation is adjusted to 1; Feature enhancement: Feature extraction is performed based on principal component analysis. The original data matrix X is decomposed into a feature matrix X by principal component analysis. PCA , select the first m principal components with the highest contribution rate to form a simplified data set, and assign feature weights in combination with the entropy weight method; Multi-layer wavelet denoising: Perform wavelet decomposition on the signal to obtain sub-signals in different frequency bands, retaining low-frequency signals to suppress high-frequency noise; Missing data filling: The missing data are filled using the dynamic interpolation method based on Bayesian estimation. Let the observed data be X obs , the missing data is X miss , and the posterior probability distribution is obtained through Bayesian inference: p(X miss |X obs )∝p(X obs |X miss )p(X miss ); Where, p(X miss |X obs ) is the missing data X miss When observing data X obs The posterior probability under given conditions; p(X obs |X miss ) is the observed data X obs In the case of missing data X miss Likelihood function under given conditions; p(X miss ) is the missing data X miss Prior probability; Sampling estimates of missing values ​​from the posterior probability distribution allows missing data to be completed.

6. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 1 is characterized in that: The fusion of the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data specifically includes: performing independent component analysis on the pre-processed data to decompose each feature into independent signal components, achieving feature decoupling by maximizing signal independence, and eliminating redundant information between features; Introducing a context-aware mechanism to enhance the relevance of current data by modeling the memory of historical data; assuming that the current feature Z t The context fusion output is C t , the calculation formula is: C t =f(Z t ,Z t-1 ,Z t-2 ,…,Z t-b ); Where f represents the context dependency function, Z t-1 represents the characteristics at time t-1, Z t-2 represents the characteristics at time t-2, Z t-b Indicates the characteristics at time tb; Build the context of the current feature through historical data at b moments to improve the model's ability to capture dynamic changes; Fuzzy logic is introduced into the data fusion process, and the weight of each feature is dynamically adjusted through the membership function. Let feature Z s The weight adjustment function is: In s =τ(Z s )·In base +(1-τ(Z s ))·In context ; Where w s is the fused feature Z s The final weight of τ(Z s ) represents feature Z s The fuzzy membership function has a range of [0,1]; w base is the benchmark weight; w context Represents the context weight.

7. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 1 is characterized in that: The Mutation-PSO algorithm is used to optimize the BP neural network, specifically: Initialize the particle group in the PSO algorithm, each particle represents the weight and threshold of the BP neural network, and randomly assign initial values ​​to the particle position and velocity; According to the prediction error of the treeline distance, the fitness value of each particle is calculated using the Huber-Loss function. The smaller the fitness value, the higher the prediction accuracy. In each iteration, the current fitness of each particle is compared with its historical best fitness. If the current fitness is better, the individual optimal value of the particle is updated. At the same time, the global optimal value of the entire particle swarm is updated to ensure that the particle swarm gradually approaches the optimal solution. Particles in the particle swarm are randomly selected to perform mutation operations, and the position and velocity of the particles are randomly perturbed to expand the search range and avoid falling into the local optimal solution. The fitness of the particles after the mutation operation is recalculated to determine whether the mutation has improved the prediction accuracy. Adjust the position and velocity of all particles according to the following update formula: v i (t+1)=ω·v i (t)+c1·rand1·(p best -x i (t))+c2erand2·(g best -x i (t)); Where, v i (t+1) represents the velocity of particle i at the t+1th iteration, v i (t) represents the velocity of particle i at the tth iteration; ω is the inertia weight; c1 and c2 are both learning factors; rand1 and rand2 are both random factors; p best is the individual optimal position of the particle, g best is the global optimal position; x i (t) represents the current position of particle i at the tth iteration; Repeat the above fitness calculation to position and speed adjustment steps until the fitness reaches the preset accuracy requirement or the number of iterations reaches the maximum limit, and finally obtain the optimal weights and thresholds of the optimized BP neural network.

8. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 7 is characterized in that: The fitness value of each particle is calculated using the Huber-Loss function, which is defined as follows: Where, is the Huber-Loss function value; y is the actual tree line distance, is the treeline distance predicted by BP neural network; δ is the adjustment parameter.

9. The multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP according to claim 1, characterized in that: The method further includes constructing an expert system, which is integrated into the PSO-BP multi-parameter fusion prediction model to analyze the deviation between the prediction results and the real data. The steps of constructing the expert system are as follows: Receive output data from the PSO-BP multi-parameter fusion prediction model and real-time monitoring data; Compare predicted output data with real-time data to identify deviations between the data; Based on the deviation analysis results, correction suggestions are generated through the expert rule base, which is composed of historical data and prior knowledge; the correction suggestions are output to the PSO-BP multi-parameter fusion prediction model to adjust the model parameters in real time.

10. A multi-parameter transmission line tree-line distance prediction system based on improved PSO-BP is characterized by: The method for predicting the tree-line distance of a multi-parameter transmission line based on an improved PSO-BP method according to any one of claims 1 to 9 comprises: The tower displacement calculation module is used to calculate the tower displacement data of the transmission line in real time based on the center frame difference method; The tree-line distance calculation module is used to calculate the tree-line distance of transmission lines based on monocular vision technology; A multi-parameter data acquisition module is used to acquire multi-parameter data related to the tree-line distance, including historical and real-time data on the span, height difference between the two towers, conductor inclination, swing angle, and conductor operating temperature; A data preprocessing module is used to preprocess the tower displacement data, transmission line tree-line distance and multi-parameter data; a prediction model construction module is used to construct a PSO-BP multi-parameter prediction model for tree-line distance prediction, including a data fusion unit, a dimension matching unit and an optimization unit; The data fusion unit is used to fuse the pre-processed tower displacement data, transmission line tree-line distance and multi-parameter data, and input the fused data as comprehensive input features into the BP neural network; The dimension matching unit is used to match the number of input layer nodes of the BP neural network with the dimension of the multi-parameter data, and the hidden layer adopts a nonlinear activation function; The optimization unit is used to optimize the BP neural network using the Mutation-PSO algorithm, assign the optimized optimal weights and thresholds to the BP neural network, and form an optimized PSO-BP multi-parameter fusion prediction model; The treeline distance prediction module is used to make real-time predictions of treeline distances using the optimized PSO-BP multi-parameter fusion prediction model.

11. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the multi-parameter transmission line tree-line distance prediction method based on improved PSO-BP are implemented as described in any one of claims 1 to 9.

Images