Power transformation facility anti-bird measure parameter dynamic adjustment system and method based on big data

Abstract

The invention discloses a power transformation facility anti-bird measure parameter dynamic adjustment system and method based on big data, and relates to the technical field of power transformation facility anti-bird big data processing, and the system comprises a data collection module which integrates various types of equipment to collect bird, environment and anti-bird device data; the big data processing module is used for preprocessing data and constructing a standardized data set; the bird species recognition sub-module recognizes bird species and marks the infringement level; the bird situation analysis module outputs a risk level; the environment adaptation sub-module corrects the risk level; the parameter adjustment decision module solves an optimal parameter; the execution control module drives equipment to adjust; the feedback monitoring module collects data to form a closed loop. According to the method, the data quality and the bird species recognition accuracy are improved, the bird infringement risk is accurately pre-judged, the anti-bird parameter adaptation scene is optimized, the device energy consumption and the operation and maintenance cost are reduced, and the intelligent operation and maintenance requirements of a power system are met.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology for bird prevention in substation facilities, specifically to a system and method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data. Background Technology

[0002] As the core node of power transmission and distribution in the power system, the operational safety of substation facilities directly determines the stability of power supply, and bird activity is a significant contributing factor to substation facility failures. Fallen branches, feathers, and other foreign objects from birds nesting and roosting can easily cause short circuits in substation equipment. Direct contact between birds and live parts can also lead to insulation flashover and grounding faults. Such faults can cause localized power outages or, in severe cases, regional power grid fluctuations, resulting in significant economic losses. Current bird control measures for substation facilities mainly rely on traditional fixed solutions, including the installation of bird spikes at preset intervals, the operation of fixed-frequency acoustic bird deterrents, and fixed-angle laser bird deterrents. The parameter settings for these measures depend entirely on the on-site experience of maintenance personnel, lacking the ability to dynamically adapt to bird activity patterns and failing to cope with complex and ever-changing bird behavior scenarios.

[0003] Different bird species exhibit significant differences in their sensitivity to bird deterrence methods. Large predatory birds such as magpies and crows respond weakly to conventional sound frequencies, making it difficult for fixed-parameter sound deterrents to effectively drive them away. Smaller predatory birds such as sparrows and pigeons easily adapt to fixed-parameter bird deterrence environments, resulting in a significant decrease in deterrence effectiveness over long-term operation and an inability to provide sustained protection. Some existing bird control systems attempt to incorporate sensors to collect bird activity data, but the data processing stage has significant shortcomings. They can only perform simple filtering of single-dimensional data, failing to integrate historical bird activity data, environmental factors, and other multi-dimensional data for in-depth analysis. This prevents the construction of a correlation model between bird activity and environmental conditions, leading to low accuracy in predicting bird attack risks and difficulty in adjusting bird control strategies in advance, leaving the system in a reactive "failure-to-respond" state.

[0004] Traditional image recognition algorithms used in bird species identification also have limitations. Their sample databases cover a limited range of scenarios and fail to fully consider the impact of different lighting conditions, shooting angles, and distances on identification results. This makes them unable to accurately distinguish between invasive and non-invasive birds, easily leading to over-repelling or under-identification of invasive birds. Over-repelling increases equipment energy consumption and maintenance costs; for example, repelling non-invasive birds like swallows and woodpeckers not only wastes energy but may also disrupt the ecological balance. Under-identification directly increases the risk of facility malfunctions; for instance, failing to identify highly invasive birds like magpies allows them to nest and cause short circuits. Furthermore, existing bird control systems lack a complete closed-loop control mechanism. After adjusting bird control measures, the effectiveness is assessed solely through manual inspections, failing to quantify key indicators such as changes in bird activity and facility operating status in real time. The assessment results are not fed back to the parameter decision-making process, resulting in a lack of iterative optimization basis for parameter adjustments. This approach remains reactive and fails to meet the demands of intelligent and lean operation and maintenance of power systems for accurate, dynamic, and economical bird control measures. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a system and method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data, with the aim of solving the problems in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data, comprising: Data acquisition module: integrates video acquisition equipment and multiple sensors to synchronously acquire images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices; Big Data Processing Module: Preprocesses the data collected by the data acquisition module; constructs a standardized substation bird monitoring dataset based on the preprocessed data; Bird species identification submodule: It has a preset sample library covering multiple types of invasive and non-invasive birds; it is used to match birds in real-time captured images of substation facilities with bird samples in the sample library to determine the bird species and mark the level of invasiveness. Bird Situation Analysis Module: Used to build a deep learning model; correlates the standardized substation bird situation dataset and the real-time bird situation identification results output by the bird species identification submodule in terms of spatiotemporal, bird species, and environmental dimensions to form a complete substation bird situation dataset; extracts a model training dataset with actual damage labels from the standardized substation bird situation dataset to train the deep learning model; extracts a risk prediction dataset from the complete substation bird situation dataset, inputs the risk prediction dataset into the trained deep learning model, and outputs the initial risk level of bird damage to the substation area at different future time periods; Environment Adaptation Submodule: Constructs a correlation model based on real-time environmental parameters and outputs a correction coefficient for bird activity levels; The comprehensive risk correction module for bird infestation is used to correct the initial risk level of bird infestation in substation areas at different future time periods based on preset multi-dimensional weighting coefficients and bird activity activity correction coefficients. Parameter adjustment decision module: With the optimization objectives of reducing the risk level after the comprehensive risk correction module to a preset threshold, reducing the energy consumption of bird-proofing devices, and minimizing the interference with the operation of substation facilities, a multi-objective optimization function is constructed, and an intelligent algorithm is used to solve for the optimal combination of bird-proofing measures parameters. Execution control module: Includes drive circuit and actuator, which adjusts the bird-proofing device based on the optimal combination of bird-proofing measure parameters obtained from the solution.

[0007] Furthermore, in the big data processing module, the standardized substation bird monitoring dataset includes: Historical bird activity data: historical bird species types and severity of infestation; Corresponding historical environmental parameters; Historical bird activity data in time and space: Time dimension: observation period and season; Spatial dimension: activity area coordinates, records of stay at high-risk locations, and nesting locations.

[0008] Furthermore, the bird species identification submodule specifically includes: The system receives real-time monitoring images of the substation facility area collected by the data acquisition module, performs component decomposition on each pixel in the real-time monitoring image of the substation facility area, and separates the reflection component that reflects the inherent outline and texture of the bird, as well as the illumination component that reflects the ambient light interference. Adaptive brightness equalization processing is performed on the separated light components, and brightness is increased in low-light areas of the real-time monitoring screen of the substation area, while brightness is suppressed in overexposed highlight areas. Based on the illumination and reflection components after adaptive brightness equalization processing, adaptive illumination enhancement calculation is performed to output a standardized image that eliminates illumination interference. A dual-branch parallel multi-scale feature fusion convolutional neural network was constructed. This network consists of three core units: a shared shallow feature layer, a main classification branch, and a small target bird detection branch. After a standardized image is input into the network, it first enters the shared shallow feature layer. This layer extracts common basic features of all birds in the standardized image through convolutional layers, including bird edges, contours, and basic textures. The output of these common basic features is simultaneously transmitted to the main classification branch and the small target bird detection branch. The main classification branch further extracts semantic information from the common basic features through deep convolutional layers, obtaining deep semantic features of large and medium-sized predatory birds. The small target bird detection branch retains the common basic features output by the shared shallow feature layer and simultaneously sets up dedicated anchor boxes adapted for small predatory birds to extract high-resolution detail features. A feature pyramid network is then constructed to perform cross-scale fusion of the deep semantic features of large and medium-sized predatory birds output by the main classification branch and the high-resolution detail features of small predatory birds output by the small target bird detection branch, outputting standardized bird species matching features that simultaneously cover large, medium, and small predatory birds. A bird sample bank covering all scenarios of substation facilities was constructed. The sample bank contains samples of various types of invasive and non-invasive birds, covering different lighting conditions, shooting angles, shooting distances, and scenarios that frequently occur in substations, such as backlighting, nighttime infrared, cloudy and rainy conditions, and strong wind shaking. It also includes bird samples from different habitat locations. For each type of bird sample in the bird sample bank, complete multi-dimensional basic annotation, including bird body size parameters, typical behavioral characteristics, and risk coefficients of different habitat locations; The standardized bird species matching features are matched with the features of bird samples in the bird sample database using cosine similarity. The bird species type corresponding to the bird sample with the highest similarity is taken as the final identification result. After bird species identification is completed, scores are calculated for the identified bird species' body size parameters, typical behavioral characteristics, and risk coefficients at different habitat locations. Based on these scores, a weighted average is used to calculate the final base score for bird species infestation. Based on the final base score for bird species infringement. The corresponding level of infringement is marked.

[0009] Furthermore, in the bird situation analysis module, the model training dataset built based on the standardized substation bird situation dataset includes: historical bird situation data, corresponding historical environmental parameters, historical bird spatiotemporal activity data, and actual damage result labels; The process of training a deep learning model involves simultaneously extracting two types of features from the data in the model training dataset: Temporal feature extraction: Temporal features in the model training dataset are extracted using the sliding window method; Spatial distribution feature extraction: Spatial distribution features in the model training dataset are extracted through spatial convolution; By fusing temporal features with spatial distribution features, spatiotemporal features are obtained; After fusing spatiotemporal features with historical bird species types, attack levels, and environmental parameters, the data is input into a deep learning model for training, using the Adam optimizer; the final trained deep learning model is obtained. The risk prediction dataset extracted from the complete substation bird situation dataset includes real-time bird species types and attack levels output by the bird species identification submodule, real-time environmental parameters after preprocessing by the big data processing module, and historical data extracted from the standardized substation bird situation dataset that corresponds to the current bird situation data, environmental parameters, and bird spatiotemporal activity data. The risk prediction dataset is input into the trained deep learning model to output the initial risk level of bird infestation in the substation facility area at different future time periods. This includes: extracting real-time temporal features from the data in the risk prediction dataset using the sliding window method, extracting real-time spatial distribution features using spatial convolution, and fusing the real-time temporal features with the real-time spatial distribution features to obtain real-time spatiotemporal features; and then inputting the real-time spatiotemporal features, along with real-time bird species type, infestation level, and environmental parameters, into the trained deep learning model to output the initial risk level of bird infestation in the substation facility area at different future time periods.

[0010] Furthermore, the big data processing module also includes a data standardization submodule, which is used to perform normalization processing on the constructed standardized substation bird situation dataset using standardization methods, mapping all data to the 0-1 range; at the same time, it constructs data quality assessment indicators, and scores the normalized standardized substation bird situation dataset through three dimensions: data integrity, accuracy, and consistency. If the score meets the preset score, it is determined to be a qualified dataset and used in the bird situation analysis module; if the score does not meet the preset score, it is determined to be an unqualified dataset and returned to the data preprocessing stage for reprocessing.

[0011] Furthermore, the comprehensive infringement risk correction module specifically includes: Time adaptation scores are set based on bird activity and inactivity periods. Bird density correction scores are set based on the number of invasive birds in the same area. The feedback score for historical bird infestation was calculated based on the frequency coefficient of historical bird infestation and the interval coefficient of most recent infestation. ; A correction dimension indicator system was constructed, which includes bird species attack dimension, environmental adaptation dimension, time dimension, bird flock density dimension, and historical attack feedback dimension. The subjective weights of each correction dimension are calculated using the analytic hierarchy process (AHP); the objective weights of each correction dimension are calculated using the entropy weight method; based on the principle of minimum discriminative information, the subjective and objective weights of the correction dimensions are integrated to obtain the dynamic adaptive weight coefficients of each correction dimension; based on the dynamic adaptive weight coefficients of each correction dimension, a comprehensive risk level correction formula is constructed. ; In the formula, This indicates the revised risk level for bird attacks; Represents the dynamic adaptive weighting coefficients for the bird species attack dimension; Represents the dynamic adaptive weight coefficients for the environment adaptation dimension; Represents the dynamic adaptive weighting coefficients in the time dimension; Represents the dynamic adaptive weighting coefficients for the bird flock density dimension; Represents the dynamic adaptive weighting coefficients for the historical infringement feedback dimension; This represents the environmental adaptation correction score, which is derived from the bird activity level correction coefficient output by the environmental adaptation submodule.

[0012] Furthermore, the parameter adjustment decision module uses an improved particle swarm optimization algorithm to solve for the optimal combination of bird deterrence parameters. The optimal combination of bird deterrence parameters obtained includes the frequency and volume of the sonic bird deterrent, the scanning angle and frequency of the laser bird deterrent, and the deployment spacing and height of the bird deterrent spikes. The improved particle swarm optimization algorithm is obtained by introducing an adaptive inertial weight mechanism and a mutation operator on the basis of the standard particle swarm optimization algorithm.

[0013] A method for dynamically adjusting parameters of bird-proofing measures for substation facilities based on big data includes the following steps: Start the data acquisition module and various devices to obtain images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices; The big data processing module preprocesses the data collected by the data acquisition module, and a standardized substation bird monitoring dataset is constructed based on the preprocessed data. The bird species identification submodule matches birds in real-time captured images of the substation facility area with bird samples in a preset sample library to determine the bird species and label the level of harm. The bird situation analysis module, based on the outputs of the bird species identification submodule and the big data processing module, generates the initial risk level of bird infestation in the substation area at different future time periods. The environment adaptation submodule constructs a correlation model based on real-time environmental parameters and outputs a correction coefficient for bird activity levels. By using the parameter adjustment decision module, with the optimization objectives of reducing the risk level after the infringement risk comprehensive correction module to a preset threshold, reducing the energy consumption of bird-proofing devices, and minimizing the interference with the operation of substation facilities, a multi-objective optimization function is constructed, and an intelligent algorithm is used to solve for the optimal combination of bird-proofing measures parameters. The bird-proofing device is adjusted by executing the control module based on the optimal combination of bird-proofing parameters obtained from the solution.

[0014] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data.

[0015] A non-volatile computer storage medium storing computer-executable instructions that execute a method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data.

[0016] Compared with existing technologies, the present invention has the following advantages:

[0017] (1) This invention, through the coordinated cooperation of a multi-dimensional, multi-source data acquisition and standardized processing system, a multi-scale feature fusion-based bird species identification and dynamic attack classification mechanism, a spatiotemporal feature fusion-based deep learning model, a multi-dimensional dynamic adaptive weight risk correction system, a multi-objective intelligent optimization-based dynamic adjustment strategy for bird prevention parameters, and a full-link closed-loop feedback optimization architecture, achieves a comprehensive transformation of bird prevention measures for substation facilities from a traditional, experience-driven, fixed protection mode to a big data-driven, intelligent, dynamic, and precise protection mode. This ensures the reliability and consistency of subsequent analysis and decision-making from the data source, and achieves accurate differentiation between harmful and non-harmful birds and the determination of attack levels from the bird species identification stage. Dynamic quantitative grading represents a core upgrade from reactive response after a fault occurs to proactive pre-emptive risk prediction from a risk management perspective. From a parameter optimization standpoint, it achieves a multi-objective balance between protection effectiveness, energy consumption control, and minimizing operational interference with substation facilities. Furthermore, a full-cycle closed-loop iteration mechanism enables continuous optimization of system performance. This effectively addresses the core shortcomings of existing technologies, such as poor adaptability to bird species identification scenarios, insufficient risk prediction accuracy, fixed bird prevention parameters that cannot adapt to complex and changing bird conditions, and the lack of a complete closed-loop control system. It comprehensively enhances the intelligence level and operational reliability of the bird prevention system for substation facilities, providing complete and efficient technical support for the safe and stable operation of substation facilities.

[0018] (2) This invention effectively eliminates the interference of complex environments such as strong light, backlight, and low light in substations on the identification results by performing adaptive illumination robustness enhancement processing in the bird species identification process, combined with a dual-branch parallel multi-scale feature fusion convolutional neural network and a bird sample library with multi-dimensional annotation of the whole scene. It solves the industry pain point of easy loss of small infestation bird features and high false negative rate. At the same time, based on the multi-dimensional weighted calculation mechanism of bird body size parameters, real-time behavioral features, and habitat risk coefficient, it realizes dynamic and accurate annotation of bird infestation level, providing accurate species dimension basis for subsequent risk assessment. On this basis, through the deep learning model that integrates time and space dimension features, it effectively captures the temporal pattern and spatial distribution preference of bird activities. Combined with the multi-dimensional dynamic adaptive weight risk correction system covering bird species, environment, time, bird density, and historical infestation records, it greatly improves the accuracy and scene adaptability of bird infestation risk prediction. It can output time-segmented risk prediction results that fit the actual bird situation at the substation site, providing a reliable decision-making basis for the early adjustment and precise implementation of bird prevention measures.

[0019] (3) This invention constructs a multi-objective optimization function with the optimization objectives of reducing the risk level of bird infestation to a preset threshold, reducing the energy consumption of bird protection devices, and minimizing the interference of substation operation. By introducing an improved intelligent algorithm with adaptive inertial weights and mutation operators, the optimal combination of bird protection measures parameters is solved. It can output suitable bird protection device operation parameters for different bird species and risk level scenarios, effectively avoiding the bird adaptability problem and ineffective energy consumption caused by long-term operation of fixed parameters. At the same time, the dual-path redundant drive design and action verification mechanism ensure the accurate and reliable execution of parameter adjustment commands. The supporting full-cycle feedback monitoring and hierarchical optimization strategy can complete the continuous iterative optimization of system parameters based on the actual protection effect on site. While ensuring the bird protection effect of substation facilities, it significantly reduces the operating energy consumption of bird protection devices and the on-site operation and maintenance labor costs. It takes into account the operation safety, operation and maintenance economy and ecological friendliness of substation facilities. It can be adapted to a variety of complex application scenarios such as urban substations and outdoor wind power substation facilities, and has strong engineering application value and promotion prospects. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system structure of the present invention.

[0021] Figure 2 This is a flowchart of the method of the present invention.

[0022] Figure 3 A comparison chart showing the error in predicting bird attack risks at different times.

[0023] Figure 4 A graph showing the daily energy consumption of bird-proofing devices as a function of operating time. Detailed Implementation

[0024] Example 1

[0025] like Figure 1 As shown, the present invention provides a technical solution: a dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data, comprising: Data acquisition module: Integrates high-definition video acquisition equipment and multiple sensors (infrared sensors, environmental sensors, and bird deterrent device status sensors, etc.) to simultaneously acquire images of the substation area, surrounding environmental parameters (temperature, humidity, wind speed, air pressure, etc.), and operating status parameters of bird deterrent devices (sonic bird repellers, laser bird repellers, bird spikes, etc.). The acquired data is transmitted to the big data processing module through a 5G encrypted communication link with a transmission latency of ≤10ms, ensuring the security and real-time performance of data transmission.

[0026] The high-definition video acquisition equipment has a frame rate of 25fps and a resolution of 1920×1080; the infrared sensor has a detection range of 0-100m and a response time of ≤50ms.

[0027] The big data processing module employs a Hadoop distributed storage architecture (to meet the storage requirements of massive amounts of collected data) to preprocess the data collected by the data acquisition module. Preprocessing includes sequentially performing missing value imputation, duplicate data removal, and outlier identification. Missing value imputation uses the K-nearest neighbor algorithm with a neighborhood size of 10, and outlier identification is based on the 3σ criterion. The preprocessed data is then analyzed using principal component analysis to extract core feature dimensions. Based on these extracted core feature dimensions, a standardized substation bird monitoring dataset is constructed. The extracted core feature dimensions include: Historical bird activity data: historical bird species types and severity of infestation; Corresponding historical environmental parameters: temperature, humidity, wind speed, air pressure, etc.; Historical bird activity data in time and space: Time dimension: observation period, season; Spatial dimension: activity area coordinates, records of stay at high-risk locations, nesting locations, etc.

[0028] The outlier identification process of the 3σ criterion is as follows: First, calculate the mean and standard deviation of the collected data, and determine the outlier judgment interval as [mean - 3 × standard deviation, mean + 3 × standard deviation]. Data outside this interval are marked as outliers. For outliers in environmental parameters, interpolation is used to replace them. During principal component analysis, core feature dimensions with a variance contribution rate of ≥80% are retained to reduce data redundancy and improve the efficiency of subsequent model calculations.

[0029] The big data processing module also includes a data standardization submodule, which performs normalization processing on the constructed standardized substation bird situation dataset using the min-max standardization method, mapping all data to the 0-1 range. At the same time, it constructs data quality assessment indicators, scoring the normalized standardized substation bird situation dataset through three dimensions: data integrity, accuracy, and consistency. A score ≥ 85 is judged as a qualified dataset and used for bird situation analysis (bird situation analysis module), while a score < 85 is judged as an unqualified dataset and returned to the data preprocessing stage for reprocessing to maintain the data quality of the input model.

[0030] Traditional convolutional neural networks (CNNs) typically consist of simple layer combinations, usually containing 2-3 convolutional layers and 1-2 pooling layers. Some simplified models even have only one convolutional layer and one pooling layer. Convolutional and pooling designs often use relatively large kernel sizes such as 5×5 or 7×7, with a stride of 2. Pooling layers typically use 2×2 kernels with a stride of 2, resulting in insufficient balance between feature extraction accuracy and redundancy compression. Activation functions are mainly Sigmoid or Tanh functions, without specific optimization. Output layers only support basic classification output, typically only outputting bird species probabilities, without additional labeling capabilities.

[0031] Bird species identification submodule execution: Enhanced image adaptive lighting robustness processing: The system receives real-time monitoring images of the substation area transmitted by the data acquisition module, performs component decomposition on each pixel in the real-time monitoring images of the substation area, separates the reflection component that reflects the inherent outline and texture of birds, and the illumination component that reflects ambient light interference, and eliminates the obstruction of image details by strong light, backlight, low light at night, and rainy weather. Adaptive brightness equalization is applied to the separated light components, and brightness is increased in low-light areas of the real-time monitoring image of the substation area. Brightness is suppressed in overexposed highlight areas, while retaining the core identification features of birds such as edges and feather textures, avoiding feature blurring caused by traditional brightness adjustment. Based on the illumination and reflection components after adaptive brightness equalization, adaptive illumination enhancement calculation is performed to output a standardized image that eliminates illumination interference. The calculation formula is as follows: ; In the formula, Indicates the number of pixels in the enhanced image. The final brightness value; Represents the pixels in the original input image. The original brightness value; Represents pixels The reflection component corresponds to the inherent texture features of birds; Represents pixels The light component corresponds to ambient light interference information; This represents the illumination compensation coefficient.

[0032] Dual-branch feature extraction: A dual-branch parallel multi-scale feature fusion convolutional neural network was constructed. The multi-scale feature fusion convolutional neural network is divided into three core units: a shared shallow feature layer, a main classification branch, and a small target bird detection branch. After the standardized image is input into the network, it first enters the shared shallow feature layer. The shared shallow feature layer extracts the common basic features of all birds in the standardized image through two convolutional layers with 3×3 convolutional kernels, including bird edges, outlines, and basic textures. The output common basic features are synchronously transmitted to the main classification branch and the small target bird detection branch to ensure that the basic features of the two branches are of the same origin. The main classification branch uses three deep convolutional layers to further extract semantics from general basic features, obtaining high-dimensional classification features (deep semantic features) of large and medium-sized invasive birds (magpies, crows, etc.) for subsequent accurate matching of bird species; The small target bird detection branch retains the common basic features output from the shared shallow feature layer, avoiding the loss of small attacking bird features caused by deep convolution. At the same time, it sets up dedicated anchor boxes adapted to small attacking birds (such as sparrows), with the anchor box size matching the pixel proportion of small attacking birds in the image, to extract the fine texture and contour features (high-resolution detail features) of small attacking birds. In addition, it sets an additional feature enhancement coefficient for small attacking birds to increase the weight of small attacking bird features in the fusion process and solve the problem of missed detection of small attacking birds. A feature pyramid network is constructed to fuse the deep semantic features of large and medium-sized invasive birds output by the main classification branch with the high-resolution detail features of small invasive birds output by the small target bird detection branch across scales. The output is a standardized bird species matching feature that covers large, medium and small invasive birds, ensuring consistent recognition accuracy for birds of different sizes.

[0033] Bird sample bank construction and species matching: A bird sample bank covering all scenarios of substation facilities was constructed. The sample bank includes 12 types of invasive birds and 8 types of non-invasive birds, with no less than 8,000 samples for each type of bird. It covers different lighting conditions, shooting angles, shooting distances, and scenarios that frequently occur in substations, such as backlighting, nighttime infrared, blurry images in rainy weather, and shaking in strong winds. It also includes bird samples from different habitats such as insulator strings, transformer tops, and tower crossarms. For each type of bird sample in the bird sample bank, multi-dimensional basic annotations were completed, including bird body size parameters (wingspan, weight range), typical behavioral characteristics (nesting and picking branches, staying on electrified parts, short-term flight, etc.), and risk coefficients of different habitat locations, to provide a standardized benchmark for subsequent damage level annotation; The standardized bird species matching features are matched with the bird sample features in the bird sample database (the bird sample features are obtained by performing the same steps as the standardized bird species matching features on the bird samples). The bird species type corresponding to the bird sample with the highest similarity is taken as the final identification result. If the matching similarity threshold is ≥95%, it is considered a valid identification, and the identified bird species type is output.

[0034] Infringement level and basic score for bird species: After bird species identification is completed, calculate the scores for the identified bird species' body size parameters, typical behavioral characteristics, and risk coefficients for different habitat locations: Body shape dimension score The values ​​were assigned based on the size parameters of the bird species: 8.0-10.0 for large invasive birds (magpies, crows), 4.0-7.9 for medium invasive birds (pigeons), 1.0-3.9 for small invasive birds (sparrows), and 0 for non-invasive birds. Behavioral dimension score The values ​​are assigned based on the typical behavioral characteristics of the bird species: nest building behavior with twigs is assigned 8.0-10.0, prolonged stay near electrified parts is assigned 4.0-7.9, short-term risk-free flight is assigned 1.0-3.9, and no harmful behavior is assigned 0. Habitat risk score The risk coefficient of the habitat location is assigned based on the identification of bird species: the straight-line distance between the habitat location and the live parts of the substation or high-risk area is less than 0.5m and assigned 8.0-10.0; 0.5m-2m and assigned 4.0-7.9; 2m-5m and assigned 1.0-3.9; non-high-risk open areas are assigned 0. Based on the scores of the identified bird species' body size parameters, typical behavioral characteristics, and risk coefficients at different habitat locations, a weighted average is used to calculate the final base score for bird species infestation. , is represented as: ; In the formula, This represents the weighting coefficient for body shape dimensions; This represents the weight coefficient for the behavioral dimension; This represents the weighting coefficient of the habitat location dimension, satisfying... The weighting coefficients can be adaptively adjusted based on the historical attack records of the corresponding bird species. For bird species with multiple historical attack records, the weighting of body size and behavior dimensions will be increased simultaneously. Based on the final base score for bird species damage The corresponding infringement levels are marked as follows: 8.0-10.0 is severe infringement, 4.0-7.9 is moderate infringement, 1.0-3.9 is mild infringement, and 0 is no infringement.

[0035] Bird situation analysis module: The core is to build a deep learning model that integrates spatiotemporal features, and the specific implementation is as follows: 1. Data Association and Complete Dataset Construction: The system receives two types of input data: first, a standardized substation bird activity dataset (qualified dataset) output from the big data processing module, containing historical bird activity data for the past three years, corresponding historical environmental parameters, and historical spatiotemporal activity data of birds; second, real-time bird identification results output from the bird species identification submodule, including real-time bird species identified in real-time footage of the substation area and their labeled threat levels. These two types of input data are then correlated across spatiotemporal, bird species, and environmental dimensions to form a complete substation bird activity dataset, ensuring the consistency between historical data and real-time scenarios.

[0036] 2. Model Training and Iterative Updates: A model training dataset is constructed based on a standardized substation bird monitoring dataset (qualified dataset). The data composition of the model training dataset includes: Historical bird activity data for the past three years: historical bird species types and corresponding threat levels; Corresponding historical environmental parameters: environmental monitoring data such as temperature, humidity, wind speed, and air pressure that are matched with historical bird activity data; Historical bird activity data in time and space: time dimension (observation period, season, migration period, etc.) and spatial dimension (coordinates of activity area, records of stay at high-risk locations, nesting locations, etc.); Actual damage result label: The bird damage result corresponding to the historical scenario (such as "caused equipment short circuit", "no substantial impact", etc.); The model training dataset is divided into training set, validation set, and test set in a 7:2:1 ratio. Training set: used for model body fitting and parameter learning; Validation set: used for hyperparameter tuning and overfitting control during training; Test set: Used to verify the final risk prediction accuracy of the model after training is completed.

[0037] The specific process for model training is as follows: two types of feature extraction are performed simultaneously on the data in the training set to ensure uniform feature dimensions. Temporal feature extraction: Temporal features (daily cycle patterns, seasonal cycle patterns, migration trends, etc. of bird activity) are extracted using the sliding window method (window size 24 hours, time segments divided by hours). Spatial distribution feature extraction: Spatial distribution features (birds' preference for high-risk areas, nesting site habits, activity space distribution, etc.) are extracted through spatial convolution, i.e., three two-dimensional convolutional layers (with 32, 64, and 128 kernels). By fusing temporal and spatial distribution features, a spatiotemporal feature model is obtained that simultaneously considers the temporal patterns and spatial distribution attributes of bird activity. This spatiotemporal feature model is then fused with historical bird species types, attack severity levels, and environmental parameters, and input into a deep learning model (containing an input layer, four convolutional layers, two pooling layers, one fully connected layer, and an output layer). The model is trained using the Adam optimizer (initial learning rate 1e-3, 200 training epochs, validation every 10 epochs) until the risk prediction accuracy on the test set stabilizes at ≥90%. When a full month of historical data is added to the standardized substation bird monitoring dataset (qualified dataset), the model is automatically iterated and updated, incorporating the new data into the training dataset to optimize parameters.

[0038] 3. Risk Prediction and Result Output: A risk prediction dataset is extracted based on the complete substation bird monitoring dataset. The risk prediction dataset includes: The bird species identification submodule outputs real-time bird species types and attack levels; Real-time environmental parameters after preprocessing by the big data processing module; Historical data from the past three years that are similar to current bird data, environmental parameters, and spatiotemporal bird activity data, extracted from standardized substation bird datasets (qualified datasets); Following the feature engineering logic of the model training and iterative update stages, real-time temporal features are extracted from the risk prediction dataset using the sliding window method, and real-time spatial distribution features are extracted using spatial convolution. The real-time temporal features and real-time spatial distribution features are then fused to obtain real-time spatiotemporal features. These real-time spatiotemporal features are then fused with real-time bird species type, attack level, and environmental parameters and input into the trained deep learning model. The model outputs the initial risk level of bird attacks on substation facilities in different time periods within the next 24 hours (range 0.0-10.0, corresponding to non-attack to severe attack). This result is then transmitted to the attack risk comprehensive correction module for subsequent correction.

[0039] Among them, the sliding window method is used to extract the time dimension features of the periodic patterns and trends of bird activities. For example, the model will learn: daily cycle pattern: bird attack scores usually peak in the early morning and evening; seasonal pattern: the average activity frequency of birds in spring is significantly higher than in winter; weather influence: the historical activity frequency will decrease when the wind speed is greater than 5 m / s. Spatial distribution features of bird activity can be extracted through spatial convolution. For example, the model can learn that birds tend to stay near the insulator strings of towers (high-risk areas) rather than in open areas; and that some bird species (such as magpies) have the habit of nesting above certain transformers. The sliding window method uses a 24-hour window size to extract the time dimension features of the periodic patterns and trends of bird activities by dividing the time into hourly segments. The spatial convolution uses three two-dimensional convolutional layers. The model training process uses the Adam optimizer with an initial learning rate of 1e-3 and 200 training rounds. A validation set evaluation is performed every 10 rounds, and the model parameters are adjusted based on the evaluation results to keep the risk level prediction accuracy above 90%.

[0040] Environment Adaptation Submodule: Constructs a correlation model based on real-time environmental parameters, outputs a correction coefficient for bird activity, with the correction coefficient ranging from 0.7 to 1.3. It also has a built-in seasonal feature library to adjust the sensitivity of environmental parameters according to the season.

[0041] The comprehensive risk correction module for bird intrusion, based on preset multi-dimensional weighting coefficients and bird activity level correction coefficients, corrects the initial risk level of bird intrusion in the substation area at different times within the next 24 hours output by the bird activity analysis module. Specifically: Calculate the time adaptation score The bird activity period was 2.0, and the other periods (inactive periods) were 0.5. Calculate the bird flock density correction score The system assigns values ​​based on the number and density of invasive birds in the same area as identified in real time. For invasive birds with ≥5 birds, the value is 8.0-10.0; for 3-4 birds, the value is 4.0-7.9; for 1-2 birds, the value is 1.0-3.9; and for no invasive birds, the value is 0. Calculate the historical bird damage feedback score , is represented as: ; In the formula, The historical attack frequency coefficient is represented by a value of 1.0 for a region and a bird species that has been attacked ≥3 times in the past year, 0.5 for 1-2 times, and 0 for no attack record. This represents the most recent attack interval coefficient. For the corresponding region and bird species, the interval between the most recent attack event is less than 30, with a talent value of 10.0; 30-90, with a talent value of 5.0; and greater than 90, with a talent value of 1.0. If there is no attack record, a value of 0 is assigned. A revised indicator system was constructed, comprising bird species infestation, environmental adaptability, time, bird flock density, and historical infestation feedback dimensions, with corresponding weight coefficients: [Weight coefficients for bird species infestation dimension are listed below]. Weighting coefficients for the environment adaptation dimension Weighting coefficients for the time dimension Weighting coefficients for bird flock density dimension Weighting coefficients for the historical infringement feedback dimension ,satisfy ; The subjective weights of each correction dimension are calculated using the analytic hierarchy process: Based on the field experience of substation facility operation and maintenance experts, an importance judgment matrix of 5 correction dimensions is constructed. The maximum eigenvalue and corresponding eigenvector of the importance judgment matrix are calculated. After completing the consistency verification, the subjective weights of the 5 correction dimensions are obtained to ensure that the weights meet the actual protection needs of on-site operation and maintenance. The objective weights of each correction dimension are calculated using the entropy weight method: Based on more than 3 years of historical data from the standardized substation bird situation dataset, the information entropy of each correction dimension index is calculated. The smaller the information entropy, the greater the effective information of that dimension and the stronger its impact on the risk outcome, and the higher the corresponding weight. Finally, the objective weights of each dimension are obtained to ensure that the weights fit the actual patterns of bird situation on site. Based on the principle of minimum discriminative information, the subjective and objective weights of the correction dimensions are integrated to obtain the dynamic adaptive weight coefficients of each correction dimension. The calculation formula is as follows: ; In the formula, Indicates the first Dynamic adaptive weight coefficients for each corrected dimension; Indicates the first Subjective weights for each correction dimension; Indicates the first The objective weights of each corrected dimension; These correspond to the bird species invasion dimension, environmental adaptation dimension, time dimension, bird flock density dimension, and historical invasion feedback dimension, respectively. Based on the dynamic adaptive weight coefficients of each correction dimension, a comprehensive risk level correction formula is constructed: ; In the formula, This indicates the revised risk level for bird infestations, with a value ranging from 0.0 to 10.0. Represents the dynamic adaptive weighting coefficients for the bird species attack dimension; Represents the dynamic adaptive weight coefficients for the environment adaptation dimension; Represents the dynamic adaptive weighting coefficients in the time dimension; Represents the dynamic adaptive weighting coefficients for the bird flock density dimension; Represents the dynamic adaptive weighting coefficients for the historical infringement feedback dimension; This represents the environmental adaptation correction score, which is derived from the bird activity level correction coefficient output by the environmental adaptation submodule.

[0042] Parameter adjustment decision module: With the optimization objectives of reducing the bird attack risk level after correction by the comprehensive attack risk correction module to a preset threshold, reducing the energy consumption of bird prevention devices, and minimizing the interference of substation operation, a multi-objective optimization function is constructed, and an improved intelligent algorithm is used to solve for the optimal combination of bird prevention measures parameters.

[0043] In the parameter adjustment decision module, an improved particle swarm optimization algorithm is used to solve for the optimal combination of bird deterrence parameters. The improved particle swarm optimization algorithm introduces adaptive inertia weight and mutation operator on the basis of the original particle swarm optimization algorithm. The inertia weight decreases linearly from 0.9 to 0.4 with the number of iterations, and the mutation probability is set to 0.05. A global optimal solution verification is performed every 50 iterations to deviate from the local optimal solution. The optimal combination of bird deterrence parameters obtained includes the frequency and volume of the sonic bird deterrent, the scanning angle and frequency of the laser bird deterrent, and the deployment spacing and height of the bird deterrent spikes.

[0044] The original particle swarm optimization algorithm includes updating the velocity and position of particles. In this invention, the adaptive inertial weight is applied before the velocity update to dynamically calculate the current weight value and substitute it into the velocity formula. The Gaussian mutation operator is executed randomly with a set probability after the position update, or it is forcibly executed on all particles except the optimal particle when the global optimal stagnation is detected every 50 generations.

[0045] The adaptive inertia weights employ a linear decreasing strategy, expressed as follows: ; In the formula, Indicates the first The inertia weight value at the next iteration; This represents the maximum value of the inertia weight; This represents the minimum value of the inertia weight. This indicates the preset maximum number of iterations; Using the Gaussian mutation operator, the mutation operation is calculated using the following formula: ; In the formula, Indicates the position of the mutated particle; Indicates the particle position before mutation; This represents a random number that follows a standard normal distribution with a mean of 0 and a variance of 1.

[0046] The execution control module includes a drive circuit and an actuator. The drive circuit supports dual modes of 485 bus and wireless communication, with an output voltage range of 0-24V and an output current of ≤5A. Based on the optimal combination of bird deterrent parameters, it drives the adjustment mechanism of the bird deterrent device (sonic bird deterrent, laser bird deterrent, bird spike) to complete parameter adjustment.

[0047] The execution control module also includes an action verification submodule. After the parameter adjustment command is issued, the action verification submodule detects the actual action state of the actuator through position sensors and pressure sensors, compares it with the preset state of the command, and determines that the action meets the command requirements if the deviation is ≤5%. If the deviation is >5%, the command is reissued and the backup execution channel is started. The dual-redundancy design of the drive circuit enables the backup channel to be switched on and activated within 100ms when a single channel fails, maintaining the reliability of parameter adjustment.

[0048] Feedback monitoring module: Real-time acquisition of images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices after parameter adjustment by the execution control module. These are transmitted to the big data processing module to form a closed-loop optimization. The sampling period is 30 seconds.

[0049] The feedback monitoring module also includes an effect evaluation submodule, which constructs an evaluation index system for bird control effectiveness. This system includes four core indicators: the rate of reduction in bird activity frequency, the rate of repelling invasive birds, the rate of substation facility failure, and the rate of reduction in energy consumption of bird control devices. Each indicator has a weight of 0.25. A comprehensive evaluation score is calculated based on the four core indicators. A comprehensive evaluation score of ≥80 indicates that the parameter adjustment meets the expected effect, a score of 60-79 indicates that the parameter adjustment needs further optimization, and a comprehensive evaluation score of <60 indicates that the parameter adjustment has not met the expectations. The evaluation results are simultaneously fed back to the parameter adjustment decision module as the basis for the next round of parameter optimization.

[0050] Specifically, when the comprehensive evaluation score is less than 60 points, an emergency adjustment process is initiated. The parameter adjustment decision module prioritizes risk level reduction as the core objective, quickly solves for the parameter combination of emergency bird prevention measures, and issues it for execution. When the score is between 60 and 79 points, the weight allocation of the multi-objective optimization function is adjusted based on feedback data, increasing the weight of energy consumption optimization. When the score is ≥80 points, the current optimal bird prevention measure parameter combination is stored in the parameter library for direct use in subsequent similar scenarios, shortening the decision-making time. The hierarchical adjustment strategy enhances the flexibility and scenario-specificity of the system response.

[0051] The second embodiment of the present invention also provides a method for dynamically adjusting the parameters of bird prevention measures for substation facilities based on big data, such as... Figure 2 As shown, it includes the following steps: 1. Data acquisition startup steps: Start all types of equipment in the data acquisition module to obtain images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices; High-definition video acquisition equipment captures real-time images of the substation facility area, infrared sensors monitor the approach of birds, environmental sensors continuously collect data on temperature, humidity, wind speed, and air pressure, and bird-proof device status sensors record the operating status parameters of the bird-proof device. 2. Data preprocessing steps: The data collected by the data acquisition module is preprocessed by the big data processing module, and a standardized substation bird monitoring dataset is constructed based on the preprocessed data; 3. Bird species identification and classification steps: The bird species identification submodule matches the birds in the real-time captured images of the substation area with bird samples in the preset sample library to determine the bird species and mark the level of harm. 4. Bird Situation Risk Prediction Steps: Through the bird situation analysis module, based on the output of the bird species identification submodule and the big data processing module, the initial risk level of bird infestation in the substation area at different future time periods is generated.

[0052] 5. Environmental Adaptation Correction Steps: The environmental adaptation submodule constructs a correlation model based on real-time environmental parameters and outputs a correction coefficient for bird activity levels; specifically: The environmental adaptation submodule calculates the bird activity level correction coefficient based on real-time environmental data. When the temperature exceeds the 15℃-25℃ range, the correction coefficient decreases linearly with the temperature difference. When the wind speed is ≥5m / s, the correction coefficient drops to 0.7. When the relative humidity is ≥80%, the correction coefficient rises to 1.2. The weights are adjusted in conjunction with the seasonal feature library, and the correction coefficient is converted into an environmental adaptation correction score. 6. Parameter Decision Solving Steps: Through the parameter adjustment decision module, with the optimization objectives of reducing the risk level after the infringement risk comprehensive correction module to a preset threshold, reducing the energy consumption of bird deterrent devices, and minimizing the interference of substation operation, a multi-objective optimization function is constructed. An improved particle swarm optimization algorithm is used to solve for the optimal combination of bird deterrent measures parameters. During the algorithm iteration process, the solution space is optimized through adaptive inertial weights and mutation operators, and parameters such as the frequency of the acoustic bird deterrent (20-20000Hz), the scanning angle of the laser bird deterrent (0-180°), and the deployment spacing of the bird deterrent spikes are output. 7. Execution control steps: The drive circuit of the execution control module receives the optimal bird deterrence parameter combination command, drives the actuator through 485 bus or wireless communication mode, the sonic bird deterrent works at the set frequency and volume, the laser bird deterrent scans at the set angle and frequency, the bird deterrent spike adjustment mechanism adjusts the deployment spacing and height, and the action verification submodule verifies the action status in real time. 8. Feedback closed-loop optimization steps: The feedback monitoring module collects data after parameter adjustment at a sampling period of 30 seconds. The effect evaluation submodule calculates the comprehensive evaluation score and transmits the evaluation results to the big data processing module as the basis for the next round of data preprocessing and parameter decision-making, forming a closed-loop iterative optimization.

[0053] A third embodiment of the present invention also provides an electronic device, including a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data.

[0054] The fourth embodiment of the present invention also provides a non-volatile computer storage medium storing computer-executable instructions that execute a method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data.

[0055] The following two examples further illustrate specific embodiments of the present invention:

[0056] Example 5: Bird prevention scenario in a 220kV urban substation

[0057] This embodiment is applied to a 220kV urban substation, which covers an area of ​​approximately 15,000㎡ and includes core equipment such as main transformers, circuit breakers, and disconnect switches. Historical data shows that there are an average of 2-3 faults caused by bird activity per year. The main birds that cause damage are magpies (severe), sparrows (moderate), and pigeons (moderate). The system needs to cope with the complex bird situation during the summer high temperature (up to 40℃), winter low temperature (down to -10℃), and spring and autumn bird migration seasons. After the system is deployed, it needs to achieve a bird attack risk prediction accuracy rate of ≥90% and reduce the energy consumption of bird prevention devices by more than 30%.

[0058] I. System Module Implementation

[0059] 1. Deployment of the data acquisition module

[0060] High-definition video acquisition equipment was installed at 12 key locations around the substation equipment area, with a frame rate of 25fps and a resolution of 1920×1080. The lenses were pointed towards the top of the equipment and its supports, areas where birds are likely to nest. Eight infrared sensors were installed near equipment such as the main transformer and circuit breakers, with a detection range of 0-100m and a response time of 45ms, to monitor the approaching trajectories of birds in real time. Six sets of temperature and humidity sensors and four sets of wind speed sensors were installed along the substation perimeter wall, with a data acquisition frequency of 1 time / second, to record environmental parameters synchronously. Status sensors were installed on three sonic bird deterrents, two laser bird deterrents, and ten bird spikes to record the sonic frequency, laser scanning angle, and bird spike deployment spacing. The current sonic frequency was 2000Hz, the current laser scanning angle was 60°, and the current bird spike deployment spacing was 15cm. All devices transmitted data through a 5G encrypted communication link, with a measured transmission latency of 8ms, meeting the ≤10ms requirement.

[0061] 2. Big data processing module in operation

[0062] The system employs a Hadoop distributed storage architecture, comprising 3 data nodes and 1 master node, with a storage capacity of 10TB. After the collected data enters the module, missing value imputation is performed first: for example, if temperature and humidity sensor data is missing for a certain period, the K-nearest neighbor algorithm is used, setting the neighborhood to 10, selecting temperature and humidity data from 10 adjacent time periods, calculating the mean (temperature 26℃, humidity 65%), and imputing the missing value; then, duplicate data is removed, such as retaining one valid data point if the wind speed is 3m / s for three consecutive collections; outliers are identified based on the 3σ criterion: taking July temperature data as an example, the mean is calculated to be 28℃, and the standard deviation is 3℃. The outlier range is [28-3×3, 28+3×3]=[19, 37]℃. The high temperature data of 40℃ is marked as an outlier, and a linear interpolation method is used, referencing the previous temperature of 29℃ and the subsequent temperature of 30℃, to replace it with 29.5℃. After processing, principal component analysis was used to extract four core feature dimensions: bird activity frequency, environmental temperature and humidity, wind speed, and equipment operating status. The variance contribution rate of the four core feature dimensions was 82%, and a standardized bird situation dataset was constructed.

[0063] 3. Bird species identification and environmental adaptation implementation

[0064] The bird species identification submodule is equipped with a dual-branch parallel multi-scale feature fusion convolutional neural network model. The model has a pre-mounted illumination robustness feature enhancement unit, and the input layer receives high-definition video frames (1920×1080 pixels). It performs component decomposition on the input real-time monitoring footage of the substation, separating the reflection components (inherent bird contours and texture features) and illumination components (ambient light interference information) of the image pixels. It enhances brightness in low-light areas and suppresses brightness in overexposed highlight areas, performing adaptive illumination enhancement calculations to eliminate the occlusion of image details by strong light, backlight, and rainy scenes in the substation, outputting a standardized image free of illumination interference. After the standardized image is input into the network, it first enters a shared shallow feature layer, where two 3×3 convolutional kernel layers extract common basic features of birds (edges, contours, basic textures). The output features are simultaneously transmitted to the main classification branch and the small target bird detection branch. The main classification... The first branch extracts high-dimensional semantic classification features of large and medium-sized invasive birds such as magpies and crows through three deep convolutional layers. The second branch, which detects small birds, retains high-resolution feature maps and extracts fine texture and contour features of small invasive birds such as sparrows through exclusive anchor boxes adapted for small birds. The features output by the two branches are fused across scales through a feature pyramid network to output standardized bird species matching features. The model is equipped with a full-scene bird sample library, covering 12 types of invasive and 8 types of non-invasive birds. The magpie sample includes 1200 images in strong light scenes, 1000 images in backlight scenes, and 800 images in nighttime infrared scenes. The sparrow sample includes 1500 images at a distance of 5m, 1300 images at a distance of 10m, and 1200 images at a distance of 15m. Each type of sample is labeled with bird size parameters, typical behavioral characteristics, and risk coefficients for different habitat locations.

[0065] On a certain day, a single magpie was detected. After the model extracted features, it performed cosine similarity matching with the sample database, outputting a magpie matching probability of 98%, which meets the effective identification threshold of ≥95%, thus confirming the bird species as a magpie. A base score for bird species infringement was calculated based on multi-dimensional weighted averages, using the following formula: ; In this scenario, the magpie is a large invasive bird, and its size score is [not specified]. =9.0; Real-time monitoring showed that it stayed near the insulator string, less than 0.5m away from live parts, and the risk score of the habitat location was 9.0. =9.5; No high-risk behaviors such as nest building and twig carrying; behavioral dimension score. =8.5; the weighting coefficients are α=0.4, β=0.3, and γ=0.3, which satisfy α+β+γ=1. Substituting them into the calculation, we get S=0.4×9.0+0.3×8.5+0.3×9.5=9.0, and the infringement level is marked as severe. The environmental adaptation submodule collects real-time temperature data at 30℃, which exceeds the suitable activity range for birds (15-25℃). It also collects data on wind speed (2 m / s) and humidity (70%). Correction coefficients are calculated as follows: for a temperature difference of 5℃, the correction coefficient linearly decreases to 0.9; for wind speed ≤ 5 m / s, the correction coefficient remains at 0.9; for humidity < 80%, the correction coefficient remains at 0.9. The overall output correction coefficient is 0.9. The base environmental score is set to 5, which is converted to an environmental adaptation correction score E = 0.9 × 5 = 4.5.

[0066] 4. Risk prediction and parameter decision-making

[0067] The bird activity analysis module loads historical and real-time environmental parameters from the past 3 years. The sliding window method is set to a window size of 24 hours, and time segments are divided by hour. Features of the 6:00-8:00 period are extracted, which is the active period for birds: the historical average activity frequency during this period is 12 times / hour, the real-time frequency is 10 times / hour, and the initial risk level is output as 5.2.

[0068] The comprehensive risk mitigation module for infringement is based on the bird behavior patterns at the substation site. It uses a combination of the analytic hierarchy process (AHP) and the entropy weight method to solve for dynamic adaptive weight coefficients. Combined with the monitoring results of a single magpie in this scenario, the weights for each dimension are determined: =0.4 (Bird Species Infestation Dimension) =0.2 (Environmental Adaptability Dimension) =0.2 (time dimension) =0.1 (bird flock density dimension) =0.1 (historical infringement feedback dimension), satisfies ; Determine the scores for each correction dimension: Time Adaptation Score =2.0 (6:00-8:00 is the peak time for bird activity); Bird density correction score =1.0 (One invasive bird was detected; the baseline value is taken from the range of 1.0-3.9). Historical infringement feedback score Magpies have attacked this substation three times in the past year. =1.0, with the most recent incident occurring 60 days ago. =5.0, therefore H=H freq ×H interval =1.0 × 5.0 = 5.0; Calculation based on the comprehensive risk level correction formula: ; Substitute the values: R=0.4×9.0+0.2×4.5+0.2×2.0+0.1×1.0+0.1×5.0=3.6+0.9+0.4+0.1+0.5=6.5; The high-risk level range is 4.1-6.0. The revised risk level of 6.5 has risen above the high-risk range, and bird protection parameters need to be adjusted immediately.

[0069] The parameter adjustment decision module constructs a multi-objective optimization function to achieve the desired result. With the goals of reducing to a preset threshold, lowering energy consumption, and minimizing interference, an improved particle swarm optimization algorithm is used to solve the problem: the inertia weight of the algorithm decreases linearly from 0.9 to 0.4 with the number of iterations, the mutation probability is set to 0.05, and the global optimal solution is verified after 50 iterations. The optimal parameter combination is output as follows: the sound wave bird repeller has a frequency of 5000Hz and a volume of 80dB, the laser bird repeller has a scanning angle of 90° and a frequency of 3 times / minute, and the anti-bird spikes have an deployment spacing of 20cm.

[0070] 5. Execution and Feedback Closed Loop

[0071] The drive circuit of the execution control module receives parameter commands via a 485 bus, driving the sonic bird repeller to adjust its frequency to 5000Hz (its original frequency was 2000Hz) and simultaneously adjusting the volume to 80dB. The laser bird repeller motor rotates, expanding the scanning angle from 60° to 90° and increasing the scanning frequency to 3 times / minute. The bird spike adjustment mechanism, pushed by a hydraulic rod, increases the deployment spacing from 15cm to 20cm (its original deployment spacing was 15cm). The action verification submodule detects the actual angle of the laser bird repeller as 89.5° using a position sensor, with a deviation of 0.5% from the commanded angle. This deviation is ≤5%, indicating the action is valid. The actual spacing of the bird spikes is detected as 19.8cm, with a deviation of 1% from the commanded spacing. This deviation is ≤5%, indicating the action meets the requirements.

[0072] The feedback monitoring module collects data at a 30-second sampling cycle: the frequency of magpie activity decreased from 5 times to 1 time within 1 hour, and the frequency of sparrow activity decreased from 8 times to 2 times; the energy consumption of the sonic bird repeller decreased from 150W to 120W, and the energy consumption of the laser bird repeller decreased from 200W to 180W; the substation facilities are operating normally, with no abnormal signs such as short circuits or flashovers. The effect evaluation submodule calculates the comprehensive score: bird activity frequency reduction rate 80%, invasive bird deterrence rate 90%, fault occurrence rate 0%, energy consumption reduction rate 25%, comprehensive score = (0.8+0.9+1.0+0.75)×0.25×100=86.25 points. This score ≥80 points indicates that the parameter adjustment is effective, and the evaluation results are transmitted to the big data processing module as the basis for the next round of optimization.

[0073] II. Data Representation and Interpretation

[0074] Table 1: Comparison of the effectiveness of different bird prevention schemes in 220kV substations (after 3 months of operation)

[0075] Table 1 shows that the traditional fixed-parameter solution, unable to adapt to changes in bird activity and the environment, still experienced two bird-related incidents within three months, with an average daily energy consumption of 8.6 kWh, a bird deterrence rate of only 65%, and a monthly manual inspection cost of 3,000 yuan. The system of this invention, through dynamic parameter adjustment, achieves zero incidents, reduces energy consumption by 30.2%, increases the bird deterrence rate to 92%, and reduces maintenance costs by 60% due to closed-loop optimization that reduces manual intervention. The data demonstrates that this invention significantly outperforms traditional solutions in terms of protection effectiveness, energy consumption control, and operational economics, and is suitable for complex bird-related scenarios in urban substations.

[0076] Example 6: Bird prevention scenario for 1.5MW wind power substation facilities

[0077] This embodiment is applied to a 1.5MW wind power substation. The facility is located in an outdoor wind farm, surrounded by farmland and forests. The main birds that are pests are crows (severe), pigeons (moderate), and swallows (mild). The wind speed is relatively high throughout the year (average 3-6 m / s), the lowest temperature in winter is -15℃, and the highest temperature in summer is 38℃. It is necessary to deal with the impact of electromagnetic interference during the start-up and shutdown of wind power equipment on the bird protection device. The system needs to meet the requirements of fault handling response time ≤100ms and continuous operation without failure time ≥3000 hours.

[0078] I. System Module Implementation

[0079] 1. Deployment of the data acquisition module

[0080] Four high-definition video acquisition devices were installed on top of the transformer substation and next to the cable trays of the wind power substation facility. The lenses were fitted with windproof covers, with a frame rate of 25fps and a resolution of 1920×1080. Six infrared sensors were installed around the transformer substation, with a detection range of 0-100m and a response time of 48ms, avoiding the area where the wind turbine blades rotate. Four sets of temperature and humidity sensors with an IP65 protection rating were installed, along with four sets of wind speed sensors with a wind resistance rating of level 12. These sensors were fixed in the wind-sheltered areas of the wind turbine tower, collecting data at a frequency of 1 time / second. Status sensors were installed on two acoustic bird deterrents, two laser bird deterrents, and eight bird spikes to record equipment operating parameters. The current frequency of the acoustic bird deterrents was 3000Hz, and the current scanning frequency of the laser bird deterrents was 2 scans / minute. The collected data was transmitted via a 5G encrypted link, with a measured transmission latency of 9ms.

[0081] 2. Big Data Processing and Bird Species Identification

[0082] The big data processing module adopts a Hadoop architecture, containing two data nodes and one master node. Missing value imputation uses the K-nearest neighbor algorithm, with a neighborhood size of 10. For example, if wind speed sensor data is missing at a certain moment, wind speed data from 10 time periods before and after the missing value are selected. These data range from 2-5 m / s, and the calculated mean is 3.5 m / s. This mean is used to impute the missing value. Duplicate data is then removed. Outliers are identified based on the 3σ criterion: taking wind speed data as an example, with a mean of 4 m / s and a standard deviation of 1.5 m / s, the outlier interval is calculated as [4-3×1.5, 4+3×1.5], resulting in [-0.5, 8.5] m / s. The instantaneous strong wind data of 9 m / s is removed. After processing, principal component analysis is used to extract core features: "bird activity trajectory, wind speed, temperature, and equipment electromagnetic interference value." The variance contribution rate of these core features is 81%, and a dataset is constructed.

[0083] In the sample library of the bird species identification submodule, the crow samples include 1,000 images in windy scenes and 800 images in low-temperature scenes, while the pigeon samples include 1,200 images at a distance of 15-20m. The model analyzed bird images collected on a certain day and found that the extracted beak and feather features matched the crow samples with a 97% match rate. The level of damage was marked as severe, and the corresponding basic score for bird species damage was S=8.5.

[0084] 3. Environmental Adaptation and Risk Prediction

[0085] The environmental adaptation submodule collects real-time data: temperature 10℃, which falls within the suitable range for bird activity (15-25℃); wind speed 6m / s (≥5m / s); and humidity 60%. Correction factors are calculated: the correction factor drops to 0.7 when wind speed is ≥5m / s, and is 1.0 when temperature and humidity are normal, resulting in a comprehensive correction factor of 0.7. The base environmental score is set to 5, which is then converted to an environmental adaptation correction score. .

[0086] The bird activity analysis module loads historical data from the past three years. This data shows that crows are frequently active in the wind farm during spring and autumn. The 24-hour features are extracted using the sliding window method: historical activity frequency of 8 times / hour during the 16:00-17:00 period and real-time frequency of 7 times / hour. The initial risk level is output as 5.0.

[0087] The comprehensive risk mitigation module for infringement is based on the patterns of bird activity near wind power facilities. It uses a combination of the analytic hierarchy process (AHP) and the entropy weight method to solve for dynamic adaptive weight coefficients. Combined with the monitoring results of a single crow in this scenario, the weights for each dimension are determined: =0.4 (Bird Species Infestation Dimension) =0.2 (Environmental Adaptability Dimension) =0.2 (time dimension) =0.1 (bird flock density dimension) =0.1 (historical infringement feedback dimension), satisfies ; Determine the scores for each correction dimension: Time Adaptation Score =2.0 (16:00-17:00 is the peak bird season). Bird density correction score =1.0 (One invasive bird was detected; the baseline value is taken from the range of 1.0-3.9). Historical infringement feedback score The wind turbine has been targeted by crows four times in the past year. =1.0, with the most recent incident occurring 45 days ago. =5.0, therefore H=H freq ×H interval =1.0 × 5.0 = 5.0;

[0088] Calculation based on the comprehensive risk level correction formula: ; Substitute the values: R=0.4×8.5+0.2×3.5+0.2×2.0+0.1×1.0+0.1×5.0=3.4+0.7+0.4+0.1+0.5=6.1; The revised risk level of 6.1 is high-risk, and bird protection parameters need to be adjusted immediately.

[0089] 4. Parameter Decision-Making and Execution Control

[0090] The parameter adjustment decision module employs an improved particle swarm optimization algorithm, iterating 60 times before outputting optimal parameters: 6000Hz frequency and 85dB volume for the sonic bird repeller, 120° scanning angle and 4 times / minute for the laser bird repeller, and 22cm spacing for the bird spikes. The execution control module's drive circuit receives commands, and the main channel sends control signals via the 485 bus. The sonic bird repeller frequency increases from 3000Hz to 6000Hz (its original frequency was 3000Hz), and the laser bird repeller scanning angle expands to 120°. The action verification submodule detects that the actual laser bird repeller angle is 119°, with a deviation of 0.8% from the commanded angle. This deviation is ≤5%, indicating the action is effective. After 10 minutes of operation, if the main channel experiences a sudden failure, the backup channel switches on within 95ms, maintaining parameter output via wireless communication to ensure uninterrupted operation.

[0091] 5. Feedback loop and fault handling

[0092] The feedback monitoring module collected data at a 30-second sampling cycle: the frequency of crow activity decreased from 7 times to 1 time within 1 hour, and the frequency of pigeon activity decreased from 5 times to 1 time; the energy consumption of the bird-repelling device decreased from 2.5 kWh / hour to 2.1 kWh / hour; the wind power substation facilities operated normally, with electromagnetic interference values ​​≤50 dB, and no equipment malfunctions were observed. The overall score of the effectiveness evaluation submodule = (bird activity reduction rate 85% + bird deterrence rate 93% + failure rate 0 + energy consumption reduction rate 16%) × 0.25 × 100 = 88.5 points, indicating that the parameter adjustment was effective.

[0093] One day, the system detected an abnormal volume in one of the sonic bird repellers. The actual volume was 60dB, lower than the required 85dB. The feedback monitoring module identified the anomaly within 30 seconds and transmitted it to the parameter adjustment decision module. The decision module initiated an emergency adjustment, temporarily increasing the frequency of the laser bird repeller to 5 scans per minute (previously 4 scans per minute). At the same time, a maintenance command was issued. During the maintenance period, the bird deterrent effect did not decrease significantly, and the rate of repelling intrusive birds remained above 88%, demonstrating the system's reliability.

[0094] II. Data Representation and Interpretation

[0095] Table 2: Comparison of Fault Handling Effects of Bird Prevention Systems in 1.5MW Wind Power Substations

[0096] Table 2 data illustrates that traditional bird-proofing systems rely on manual intervention for fault handling, with processing times ranging from 500 to 1200 ms. During a fault, the bird-proofing effectiveness drops significantly, with a retention rate between 60% and 75%, and recovery takes 20 to 30 seconds. The system of this invention, through real-time feedback and automatic decision-making, controls the fault handling time within 85-98 ms, achieving a 90% effectiveness retention rate even when the drive channel fails, with a minimum recovery time of 5 seconds. The data demonstrates that this invention maintains high reliability and rapid response capabilities even in the harsh environments of outdoor wind power substations, meeting continuous operation requirements and preventing wind power equipment failures due to bird-proofing ineffectiveness.

[0097] Reference Figure 3 This figure highlights the accuracy of the risk prediction in this invention: traditional methods, lacking integration of spatiotemporal features and environmental data, suffer from error rates exceeding 8% and reaching as high as 12% during critical periods such as early morning, evening, and migration, easily leading to delays in bird control parameter adjustments. This invention, through the sliding window method to extract temporal features and spatial convolution to capture regional patterns, combined with environmental adaptation correction, achieves an error rate below 2.5% for all time periods, even keeping the high-risk migration period below 2.5%. This low-error risk prediction ensures that parameter adjustment decisions are more aligned with actual bird behavior.

[0098] Reference Figure 4The figure illustrates the energy-saving advantages of this invention: Traditional fixed-parameter systems, due to the inability to dynamically adjust parameters, maintain high-power operation even in the absence of bird activity, with daily energy consumption remaining stable at 8.4-8.7 kWh. This invention, through dynamic parameter optimization, reduces the power of the bird deterrent device and decreases the scanning frequency during bird-free periods, while precisely matching parameters during bird-infested periods, controlling daily energy consumption to 4.8-5.2 kWh, a reduction of approximately 40% compared to traditional systems. Over long-term operation, this energy-saving effect can significantly reduce the operation and maintenance costs of substation facilities, meeting the needs of lean operation and maintenance in power systems, while also reducing energy waste and improving economic efficiency.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic adjustment system for bird prevention measures parameters in substation facilities based on big data, characterized in that, include: Data acquisition module: integrates video acquisition equipment and multiple sensors to synchronously acquire images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices; Big Data Processing Module: Preprocesses the data collected by the data acquisition module; A standardized substation bird monitoring dataset was constructed based on the preprocessed data. Bird species identification submodule: It has a preset sample library covering multiple types of invasive and non-invasive birds; it is used to match birds in real-time captured images of substation facilities with bird samples in the sample library to determine the bird species and mark the level of invasiveness. Bird Situation Analysis Module: Used to build deep learning models; correlates the standardized substation bird situation dataset and the real-time bird situation identification results output by the bird species identification submodule in terms of time, space, bird species, and environment to form a complete substation bird situation dataset; extracts a model training dataset with actual damage results labels from the standardized substation bird situation dataset to train the deep learning model; Risk prediction datasets are extracted from the complete bird situation dataset of substation facilities. The risk prediction datasets are then input into the trained deep learning model to output the initial risk level of bird infestation in the substation facility area at different future time periods. Environment Adaptation Submodule: Constructs a correlation model based on real-time environmental parameters and outputs a correction coefficient for bird activity levels; The comprehensive risk correction module for bird infestation is used to correct the initial risk level of bird infestation in substation areas at different future time periods based on preset multi-dimensional weighting coefficients and bird activity activity correction coefficients. Parameter adjustment decision module: With the optimization objectives of reducing the risk level after the comprehensive risk correction module to a preset threshold, reducing the energy consumption of bird-proofing devices, and minimizing the interference with the operation of substation facilities, a multi-objective optimization function is constructed, and an intelligent algorithm is used to solve for the optimal combination of bird-proofing measures parameters. Execution control module: Includes drive circuit and actuator, which adjusts the bird-proofing device based on the optimal combination of bird-proofing measure parameters obtained from the solution.

2. The system for dynamically adjusting parameters of bird-proofing measures for substation facilities based on big data, as described in claim 1, is characterized in that: In the big data processing module, the standardized substation bird monitoring dataset includes: Historical bird activity data: historical bird species types and severity of infestation; Corresponding historical environmental parameters; Historical bird activity data in time and space: Time dimension: observation period and season; Spatial dimension: activity area coordinates, records of stay at high-risk locations, and nesting locations.

3. The dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data according to claim 2, characterized in that: The bird species identification submodule specifically includes: The system receives real-time monitoring images of the substation facility area collected by the data acquisition module, performs component decomposition on each pixel in the real-time monitoring image of the substation facility area, and separates the reflection component that reflects the inherent outline and texture of the bird, as well as the illumination component that reflects the ambient light interference. Adaptive brightness equalization processing is performed on the separated light components, and brightness is increased in low-light areas of the real-time monitoring screen of the substation area, while brightness is suppressed in overexposed highlight areas. Based on the illumination and reflection components after adaptive brightness equalization processing, adaptive illumination enhancement calculation is performed to output a standardized image that eliminates illumination interference. A dual-branch parallel multi-scale feature fusion convolutional neural network was constructed. This network consists of three core units: a shared shallow feature layer, a main classification branch, and a small target bird detection branch. After a standardized image is input into the network, it first enters the shared shallow feature layer. This layer extracts common basic features of all birds in the standardized image through convolutional layers, including bird edges, contours, and basic textures. The output of these common basic features is simultaneously transmitted to the main classification branch and the small target bird detection branch. The main classification branch further extracts semantic information from the common basic features through deep convolutional layers, obtaining deep semantic features of large and medium-sized predatory birds. The small target bird detection branch retains the common basic features output by the shared shallow feature layer and simultaneously sets up dedicated anchor boxes adapted for small predatory birds to extract high-resolution detail features. A feature pyramid network is then constructed to perform cross-scale fusion of the deep semantic features of large and medium-sized predatory birds output by the main classification branch and the high-resolution detail features of small predatory birds output by the small target bird detection branch, outputting standardized bird species matching features that simultaneously cover large, medium, and small predatory birds. A bird sample bank covering all scenarios of substation facilities was constructed. The sample bank contains samples of various types of invasive and non-invasive birds, covering different lighting conditions, shooting angles, shooting distances, and scenarios that frequently occur in substations, such as backlighting, nighttime infrared, cloudy and rainy conditions, and strong wind shaking. It also includes bird samples from different habitat locations. For each type of bird sample in the bird sample bank, complete multi-dimensional basic annotation, including bird body size parameters, typical behavioral characteristics, and risk coefficients of different habitat locations; The standardized bird species matching features are matched with the features of bird samples in the bird sample database using cosine similarity. The bird species type corresponding to the bird sample with the highest similarity is taken as the final identification result. After bird species identification is completed, scores are calculated for the identified bird species' body size parameters, typical behavioral characteristics, and risk coefficients at different habitat locations. Based on these scores, a weighted average is used to calculate the final base score for bird species infestation. Based on the final base score for bird species infringement. The corresponding level of infringement is marked.

4. The dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data according to claim 3, characterized in that: In the bird situation analysis module, the model training dataset built based on the standardized substation bird situation dataset includes: historical bird situation data, corresponding historical environmental parameters, historical bird spatiotemporal activity data, and actual damage result labels; The process of training a deep learning model involves simultaneously extracting two types of features from the data in the model training dataset: Temporal feature extraction: Temporal features in the model training dataset are extracted using the sliding window method; Spatial distribution feature extraction: Spatial distribution features in the model training dataset are extracted through spatial convolution; By fusing temporal features with spatial distribution features, spatiotemporal features are obtained; After fusing spatiotemporal features with historical bird species types, attack levels, and environmental parameters, the data is input into a deep learning model for training, using the Adam optimizer; the final trained deep learning model is obtained. The risk prediction dataset extracted from the complete substation bird situation dataset includes real-time bird species types and attack levels output by the bird species identification submodule, real-time environmental parameters after preprocessing by the big data processing module, and historical data extracted from the standardized substation bird situation dataset that corresponds to the current bird situation data, environmental parameters, and bird spatiotemporal activity data. The risk prediction dataset is input into the trained deep learning model to output the initial risk level of bird infestation in the substation facility area at different future time periods. This includes: extracting real-time temporal features from the data in the risk prediction dataset using the sliding window method, extracting real-time spatial distribution features using spatial convolution, and fusing the real-time temporal features with the real-time spatial distribution features to obtain real-time spatiotemporal features; and then inputting the real-time spatiotemporal features, along with real-time bird species type, infestation level, and environmental parameters, into the trained deep learning model to output the initial risk level of bird infestation in the substation facility area at different future time periods.

5. A dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data, as described in claim 4, is characterized in that: The big data processing module also includes a data standardization submodule, which is used to perform normalization processing on the constructed standardized substation bird situation dataset using standardization methods, mapping all data to the 0-1 range; at the same time, it constructs data quality assessment indicators, and scores the normalized standardized substation bird situation dataset through three dimensions: data integrity, accuracy, and consistency. If the score meets the preset score, it is determined to be a qualified dataset and used in the bird situation analysis module; if the score does not meet the preset score, it is determined to be an unqualified dataset and returned to the data preprocessing stage for reprocessing.

6. The dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data according to claim 5, characterized in that: The comprehensive risk mitigation module specifically includes: Time adaptation scores are set based on bird activity and inactivity periods. Bird density correction scores are set based on the number of invasive birds in the same area. The feedback score for historical bird infestation was calculated based on the frequency coefficient of historical bird infestation and the interval coefficient of most recent infestation. ; A correction dimension indicator system was constructed, which includes bird species attack dimension, environmental adaptation dimension, time dimension, bird flock density dimension, and historical attack feedback dimension. The subjective weights of each correction dimension are calculated using the analytic hierarchy process (AHP); the objective weights of each correction dimension are calculated using the entropy weight method; based on the principle of minimum discriminative information, the subjective and objective weights of the correction dimensions are integrated to obtain the dynamic adaptive weight coefficients of each correction dimension; based on the dynamic adaptive weight coefficients of each correction dimension, a comprehensive risk level correction formula is constructed. ； In the formula, This indicates the revised risk level for bird attacks; Represents the dynamic adaptive weighting coefficients for the bird species attack dimension; Represents the dynamic adaptive weight coefficients for the environment adaptation dimension; Represents the dynamic adaptive weighting coefficients in the time dimension; Represents the dynamic adaptive weighting coefficients for the bird flock density dimension; Represents the dynamic adaptive weighting coefficients for the historical infringement feedback dimension; This represents the environmental adaptation correction score, which is derived from the bird activity level correction coefficient output by the environmental adaptation submodule.

7. A dynamic adjustment system for bird prevention measures parameters of substation facilities based on big data, as described in claim 6, is characterized in that: In the parameter adjustment decision module, the optimal combination of bird deterrence parameters obtained by solving includes the frequency and volume of the sonic bird deterrent, the scanning angle and frequency of the laser bird deterrent, and the deployment spacing and height of the bird deterrent spikes.

8. A method for dynamically adjusting parameters of bird-proofing measures for substation facilities based on big data, implemented based on the dynamic adjustment system for bird-proofing measures for substation facilities based on big data as described in any one of claims 1-7, characterized in that, Includes the following steps: Start the data acquisition module and various devices to obtain images of the substation area, surrounding environmental parameters, and operating status parameters of bird-proof devices; The big data processing module preprocesses the data collected by the data acquisition module, and a standardized substation bird monitoring dataset is constructed based on the preprocessed data. The bird species identification submodule matches birds in real-time captured images of the substation facility area with bird samples in a preset sample library to determine the bird species and label the level of harm. The bird situation analysis module, based on the outputs of the bird species identification submodule and the big data processing module, generates the initial risk level of bird infestation in the substation area at different future time periods. The environment adaptation submodule constructs a correlation model based on real-time environmental parameters and outputs a correction coefficient for bird activity levels. By using the parameter adjustment decision module, with the optimization objectives of reducing the risk level after the infringement risk comprehensive correction module to a preset threshold, reducing the energy consumption of bird-proofing devices, and minimizing the interference with the operation of substation facilities, a multi-objective optimization function is constructed, and an intelligent algorithm is used to solve for the optimal combination of bird-proofing measures parameters. The bird-proofing device is adjusted by executing the control module based on the optimal combination of bird-proofing parameters obtained from the solution.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a bus. The processor and the memory are connected via the bus. The memory stores a set of program code, and the processor calls the program code stored in the memory to execute the method for dynamically adjusting the parameters of bird prevention measures for substation facilities based on big data, as described in claim 8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer can execute instructions to perform the method for dynamically adjusting parameters of bird prevention measures for substation facilities based on big data, as described in claim 8.

Images