A smart bird detection and deterrent device for substations

Abstract

This invention discloses an intelligent bird detection and deterrence device for substations, comprising a bird deterrence main body with a rotating gimbal mounted on top. The main body includes a main control unit, an ultrasonic generator, a strobe light, and a battery compartment. The rotating gimbal includes a gimbal base, a left rotating compartment, a right rotating compartment, and a ring-shaped microphone array positioned above the gimbal. The gimbal base is fixed to the top of the main body. An industrial camera is built into the left rotating compartment, and the industrial camera and the ring-shaped microphone array are communicatively connected to the main control unit. A laser emitter is built into the right rotating compartment. The rotation angle of the base and rotating compartments is controlled by a stepper motor. This invention addresses the bird detection and deterrence aspect of substation inspection tasks. Based on an improved YOLOv5 target detection algorithm and TDOA sound source localization algorithm, it perceives the pose information of bird targets from both visual and auditory perspectives, enabling precise location and tracking of bird flocks within the substation.

Description

Technical Field

[0001] This invention relates to the field of bird detection and deterrence equipment technology, specifically to an intelligent bird detection and deterrence device for substations. Background Technology

[0002] Substations are a crucial dispatching link in maintaining the stable operation of the power grid. In recent years, with economic development, my country's power grid has expanded rapidly, and the safety of the power system constantly affects social stability and people's daily lives. According to power grid operation data statistics, bird damage has become the leading cause of power equipment tripping accidents, posing a serious threat to the safe operation of substations.

[0003] Most existing bird deterrent devices only use single-function bird detection equipment, which has poor detection effect when faced with problems such as obstruction and long target distance. Furthermore, they lack detailed analysis of bird behavior and diversified bird deterrent strategies. As birds adapt to the bird deterrent devices, the bird deterrent efficiency will decrease significantly. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent bird detection and de-escalation device for substations, which solves the technical problems mentioned in the background section regarding the limited functionality of existing bird-repelling methods and the significant decrease in bird-repelling efficiency once birds adapt to the device.

[0005] To achieve this objective, the present invention adopts the following technical solution:

[0006] A smart bird detection and deflection device for substations includes a bird deterrent body with a rotating gimbal mounted on its top. The bird deterrent body includes a main control unit, an ultrasonic generator, a strobe light, and a battery compartment. The main control unit and battery compartment are fixed to the inner cavity of the bird deterrent body, and the battery compartment provides power to the entire device. The ultrasonic generator and strobe light are embedded in the outer shell of the bird deterrent body and electrically connected to the main control unit. The rotating gimbal includes a gimbal base, a left rotating compartment, a right rotating compartment, and a ring-shaped microphone array positioned above the gimbal. The gimbal base is fixed to the top of the bird deterrent body. The device comprises a rotating chamber on the left containing an industrial camera, which, along with a ring microphone array, is connected to a main control device. Based on target detection and sound source localization algorithms, it perceives bird location information in real time. A rotating chamber on the right contains a laser emitter equipped with a self-powered module, which is also connected to the main control device, which controls its operation. A rotating gimbal controls the rotation angle of the base and rotating chamber via a stepper motor, adjusting the direction of the industrial camera and laser emitter to achieve tracking-type bird detection and deterrence. A moving mechanism is mounted at the bottom of the main body of the bird deterrence device.

[0007] Furthermore, the moving mechanism includes a drive motor, a wheel set, and a support frame. The drive motor is located at the bottom of the moving mechanism, and the four wheels of the wheel set are located at the four corners of the support frame. The drive motor is connected to the wheel set, and the wheel set is driven by the motor to move. The main control device controls the operation of the drive motor.

[0008] Furthermore, the main control device is equipped with a wireless communication module and an information processing module. The wireless communication module is used for communication between the main control device and the ring microphone array, industrial camera, and laser emitter. The information processing module performs data analysis based on the auditory and visual information collected by the ring microphone array and industrial camera. The ring microphone array's built-in processor executes a sound source localization algorithm to initially perceive the bird's location and control the gimbal rotation. Then, a target detection algorithm is used to identify and locate the bird target in the video stream.

[0009] Furthermore, the ring microphone array collects, filters, and reduces the noise of bird sounds, and locates the sound source based on a sound source localization algorithm. The sound source localization algorithm is an improved Time Difference of Arrival (TDOA) method, which sets the time when the sound source arrives at the center of the microphone array as the standard time, thereby obtaining the time delay difference of the sound source arriving at each microphone, and deriving an overdetermined set of equations. Then, a particle swarm optimization algorithm is applied to calculate the relative coordinate position of the sound source. The relative coordinate position is used to calculate the distance and amplitude of the sound source, thereby controlling the rotating gimbal to perform directional rotation.

[0010] Furthermore, the particle swarm optimization algorithm is as follows: The sound source location coordinates are set to (x, y, z), and the location coordinates of the i-th and j-th microphone nodes are set to (x, y, z). i ,y i ,z i ) and (x j ,y j ,z j Then, based on the node position coordinates, the following relationship can be constructed:

[0011]

[0012] In the formula: i and j can take values ​​from 1 to 6, representing six microphone nodes, and i ≠ j; v is the speed of sound in the air; τ ij This represents the time difference in signal reception between microphone i and microphone j.

[0013] The problem model is transformed into a minimum value search model. Suppose there are N particles forming a community in a D-dimensional space, where the position vector and motion direction vector of the i-th particle are as follows:

[0014] X i =(x i1 ,xi2 ,…,x iD ), i = 1, 2, ..., N

[0015] V i =(v i1 ,v i2 ,…,v iD ), i = 1, 2, ..., N

[0016] In each iteration of the algorithm, the particle moves once. Therefore, the position of the i-th particle in the (t+1)-th iteration is represented as:

[0017] x ij (t+1)=x ij (t)+v ij (t+1)

[0018] The corresponding velocity vector is represented as:

[0019] v ij (t+1)=ωv ij (t)+c1r1(t)[P ij (t)-x ij (t)]+c2r2(t)[P gj (t)-x ij (t)]

[0020] In the formula: ω is the inertia coefficient; c1 and c2 are learning factors; r1(t) and r2(t) are random parameters; P ij (t), P gj (t) represents the optimal position perceived by the particle itself and the optimal position perceived by the group, respectively, in each iteration, ωv ij (t) is a memory term; c1r1(t)[P ij (t)-x ij [(t)] represents the self-cognition term, i.e., the optimal iteration direction of the particle's self-cognition; c2r2(t)[P gj (t)-x ij [(t)] represents the group cognition term, which is the indication vector sent by the particle at the global optimal position in this iteration;

[0021] The iteration termination condition is that the optimal position of the particle swarm search satisfies a predetermined minimum fitness threshold, let f i =T 1i -τ 1i (i = 1, 2, ..., 6), T 1i Let τ be the estimated time delay difference, representing the time delay difference between the i-th microphone and the reference microphone in acquiring sound information, while τ is... 1i The time delay difference is obtained based on the generalized cross-correlation algorithm. Therefore, the model can be transformed into a minimization problem and solved using the particle swarm optimization algorithm, as shown below:

[0022]

[0023] F(x,y,z) represents the estimation error function for the sound source location coordinates, characterized by the matching property of two time delay differences. The sound source distance D and argument θ are calculated based on the obtained location coordinates.

[0024]

[0025]

[0026] Furthermore, the industrial camera acquires video streams in real time and uses the collected visual information to execute the YOLOv5n target detection algorithm for bird identification via the built-in processor of the main control device. The network architecture of the YOLOv5n target detection algorithm consists of three parts: a backbone network, a Neck, and a Head. The backbone network extracts features from the input image and uses the EfficientFormerV2 lightweight network. The Neck is the connection part, which achieves multi-scale feature fusion through an FPN+PAN structure. The Head is the detection head part, which is responsible for generating the target's bounding box and class confidence, and making predictions on the acquired image.

[0027] The bounding box regression loss function in the Head section uses the Wise-IoUv3 function, as shown in the following formula:

[0028]

[0029]

[0030]

[0031] In the formula: IOU represents the intersection-union ratio. The normalization factor represents the mean loss. W is a monotonic focusing coefficient, and the parameter β is used to characterize the degree of anomaly in the anchor frame. g and H g The x and y values ​​represent the size of the smallest closed bounding box, and the x and y values ​​represent the position parameters of the predicted bounding box. gt and y gt This represents the position parameters of the true bounding box, where α and δ are hyperparameters that are set.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] This invention addresses the bird detection and control aspect of substation inspection tasks by using a sound source localization algorithm to assist cameras in real-time detection. This allows for the perception of bird target location information from both auditory and visual perspectives, enabling precise location and tracking of bird flocks within the substation.

[0034] This invention also employs a differentiated acoustic-optical hybrid bird-repelling module, which enhances the bird-repelling effect through the combined use of a laser emitter and an ultrasonic frequency conversion bird-repelling device. Different frequencies of ultrasonic waves can be emitted to target different bird species, achieving differentiated bird repelling. Furthermore, the bird-repelling strategy can be flexibly adjusted based on bird-repelling information obtained from remote devices, thereby reducing the birds' adaptability to the device. This device is capable of long-term and highly efficient bird-repelling operations. Attached Figure Description

[0035] Figure 1 This is a three-dimensional structural diagram of the present invention;

[0036] Figure 2 This is a frontal three-dimensional structural diagram of the rotating gimbal of the present invention;

[0037] Figure 3 This is a top-view perspective three-dimensional structural diagram of the bird repeller body equipped with the moving mechanism of the present invention;

[0038] Figure 4 This is the YOLOv5n basic model architecture used in this invention;

[0039] Figure 5 The EfficientFormerV2 lightweight network architecture used in this invention;

[0040] Figure 6 This is a functional block diagram of the bird detection and deterrence system of the present invention;

[0041] Figure 7 This is a flowchart illustrating the bird detection and deterrence device of the present invention.

[0042] The components include: bird deterrent main body-1, rotating gimbal-2, ring microphone array-3, left rotating chamber-4, industrial camera-5, right rotating chamber-6, laser emitter-7, moving mechanism-8, gimbal base-9, hex bolts-10, wheel set-11, support frame-12, drive motor-13, main control equipment-101, ultrasonic generator-102, strobe light-103, battery compartment-104, and stepper motor-201. Detailed Implementation

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the described embodiments are only some embodiments of the present invention, and not all embodiments. Specific implementation methods can be adjusted and changed according to actual needs.

[0044] Please see Figures 1-7The present invention provides a smart bird detection and deflection device for substations, including a bird deterrent body 1, a rotating gimbal 2 mounted on the top of the bird deterrent body 1, a ring microphone array 3 set on the top of the shell of the rotating gimbal 2, an industrial camera 5 built into the left rotating chamber 4 of the rotating gimbal 2, a laser emitter 7 built into the right rotating chamber 6, and a moving mechanism 8 assembled at the bottom of the bird deterrent body 1.

[0045] The bird repeller body 1 includes a main control device 101, an ultrasonic generator 102, a strobe light 103, and a battery compartment 104. The main control device 101 and the battery compartment 104 are fixed to the inner cavity side of the bird repeller body 1 by hot melt adhesive. The battery compartment 104 is used to power the entire device. The ultrasonic generator 102 and the strobe light 103 are embedded in the outer shell of the bird repeller body 1 and connected to the main control device 101.

[0046] The rotating gimbal 2 includes a gimbal base 9, a left rotating chamber 4, a right rotating chamber 6, and a ring microphone array 3 mounted above the gimbal. The gimbal base 9 is fixed to the top of the bird deterrent body 1 using a set of hexagonal bolts 10. The left rotating chamber 4 houses an industrial camera 5. The industrial camera 5 and the ring microphone array 3 communicate with the main control device 101 via a local area network, enabling real-time perception of bird location information based on target detection and sound source localization algorithms. The right rotating chamber 6 houses a laser emitter 7, which is equipped with a self-powered module and connects to the main control device 101 via Bluetooth. The main control device 101 can control its working status. The rotating gimbal 2 controls the rotation angle of the base and rotating chambers via a stepper motor 201, which is used to adjust the pointing of the industrial camera 5 and the laser emitter 7 to achieve tracking-type bird detection and deterrence.

[0047] The moving mechanism 8 includes a wheel set 11, a support frame 12 and a drive motor 13. The drive motor 13 is located at the bottom of the moving mechanism 8, and the wheel set 11 and the support frame 12 are located on both sides of the drive motor 13. The main control device 101 is connected to the wheel set 11 through the drive motor 13 and is driven by the motor to move.

[0048] The main control device 101 is equipped with a wireless communication module and an information processing module. The wireless communication module is used to connect the main control device 101 with devices such as the ring microphone array 3, industrial camera 5, and laser emitter 7 to realize the information closed loop within the device. The information processing module performs data analysis based on the auditory and visual information collected by the ring microphone array 3 and industrial camera 5. It uses the TDOA sound source localization algorithm built into the ring microphone array 3 to initially perceive the location of the bird and control the rotating gimbal 2 to perform corresponding movements. Subsequently, the YOLOv5 target detection algorithm is used to achieve accurate identification and localization of bird targets in the video stream.

[0049] The circular microphone array 3 contains six microphone nodes (the distance between the nodes is not specific). Its main function is to collect, filter, and reduce noise of bird sound data, and to use the time difference of arrival (TDOA) method based on particle swarm optimization to locate the sound source. This optimization algorithm can transform the TDOA localization problem model into a minimum value model, which has higher localization accuracy while ensuring computational efficiency.

[0050] The time it takes for the sound source to reach the center of the microphone array is defined as the standard time, thereby obtaining the time delay difference between the sound source and each microphone, and from this, an overdetermined set of equations is derived.

[0051] Let the coordinates of the sound source be (x, y, z), and the coordinates of the i-th and j-th microphone nodes be (x, y, z). i ,y i ,z i ) and (x j ,y j ,z j Then, based on the node position coordinates, the following relationship can be constructed:

[0052]

[0053] In the formula: i and j can take values ​​from 1 to 6, representing six microphone nodes, and i ≠ j; v is the speed of sound in the air; τ ij This represents the time difference in signal reception between microphone i and microphone j.

[0054] The problem model is transformed into a minimum-value-finding model using the particle swarm optimization algorithm. Suppose there are N particles forming a swarm in a D-dimensional space, where the position vector and motion direction vector of the i-th particle are shown in the following equation:

[0055] X i =(x i1 ,x i2 ,…,x iD ), i = 1, 2, ..., N

[0056] V i =(v i1 ,v i2 ,…,v iD ), i = 1, 2, ..., N

[0057] In each iteration of the algorithm, the particle moves once. Therefore, the position of the i-th particle in the (t+1)-th iteration can be represented as:

[0058] x ij (t+1)=x ij (t)+v ij (t+1)

[0059] The corresponding velocity vector can be expressed as:

[0060] v ij (t+1)=ωv ij (t)+c1r1(t)[P ij (t)-x ij (t)]+c2r2(t)[P gj (t)-x ij (t)]

[0061] In the formula: ω is the inertia coefficient; c1 and c2 are learning factors; r1(t) and r2(t) are random parameters; P ij (t), P gj (t) represents the optimal position traversed by the particle in each iteration, signifying the optimal position perceived by the particle itself and the optimal position perceived by the group, respectively. Because the initial position and velocity of each particle are random, there will be a difference between individual perception and group perception during actual calculation. ωv ij (t) is a memory term; c1r1(t)[P ij (t)-x ij [(t)] represents the self-cognition term, i.e., the optimal iteration direction of the particle's self-cognition; c2r2(t)[P gj (t)-x ij [(t)] represents the group cognition term, which is the indication vector sent by the particle at the global optimal position in this iteration.

[0062] The iteration terminates when the optimal position of the particle swarm search satisfies a predetermined minimum fitness threshold. Let f i =T 1i -τ 1i (i = 1, 2, ..., 6), T 1i Let τ be the estimated time delay difference, representing the time delay difference between the i-th microphone and the reference microphone in acquiring sound information, while τ is... 1i The time delay difference is obtained based on the generalized cross-correlation algorithm. Therefore, the model can be transformed into a minimization problem and solved using the particle swarm optimization algorithm, as shown below:

[0063]

[0064] F(x,y,z) represents the estimation error function for the sound source location coordinates, characterized by the matching property of the two time delay differences. The goal of this minimum model is to minimize the estimation error.

[0065] The distance D and the amplitude θ of the sound source can be calculated based on the obtained location coordinates.

[0066]

[0067]

[0068] Industrial camera 5 acquires video streams in real time and uses the improved YOLOv5n target detection algorithm built into main control device 101 to identify birds. The network architecture of the YOLOv5n target detection algorithm mainly consists of three parts: Backbone, Neck, and Head. Backbone is the backbone network that uses CSP structure to extract features from the input image. Neck is the connection part that uses FPN+PAN structure to achieve multi-scale feature fusion. Head is the detection head part, which is responsible for generating the bounding box and class confidence of the target and making predictions on the acquired image.

[0069] The improved YOLOv5n algorithm replaces the backbone part of the original network with the EfficientFormerV2 lightweight network and introduces a fine-grained joint search strategy, which can achieve lower latency and higher computational efficiency while ensuring detection accuracy. This enables efficient deployment on the main control device 101, meaning that the target detection algorithm has strong portability and good real-world detection performance.

[0070] The bounding box regression loss function, located in the head part, improves the matching degree between predicted and ground truth bounding boxes in object detection by iterating and modifying the gradient, thus making detection more accurate. The improved YOLOv5n algorithm replaces the original network's bounding box regression loss function with the Wise-IoUv3 function. The Wise-IoUv3 function incorporates a dynamic non-monotonic focusing mechanism to reduce the influence of harmful gradients, enabling the model to formulate the most suitable gradient gain allocation strategy for the current situation in real time, thereby improving the overall training effect. The specific formula is as follows:

[0071]

[0072]

[0073]

[0074] In the formula: IOU represents the intersection-union ratio. The normalization factor represents the mean loss. W is a monotonic focusing coefficient, and the parameter β is used to characterize the degree of anomaly in the anchor frame. g and H g The x and y values ​​represent the size of the smallest closed bounding box, and the x and y values ​​represent the position parameters of the predicted bounding box. gt and y gt The parameters represent the position of the true bounding box, and α and δ are hyperparameters that are set. The aforementioned dynamic non-monotonic focusing mechanism is reflected in the fact that Wise-IoU also allocates a small gradient gain r to focus the function on anchor boxes of normal quality, thereby avoiding gradient explosion.

[0075] Figure 4 and Figure 5 The diagram shows the overall architecture of the YOLOv5n network and the EfficientFormerV2 lightweight network used in this embodiment. EfficientFormerV2 replaces the original 3*3 average pooling layer with depthwise separable convolutions with the same kernel size, and injects a local information modeling layer into the feedforward network, which improves the model detection performance while avoiding latency overhead.

[0076] Figure 6 The diagram shown is a functional block diagram of the bird detection and de-icing system in this embodiment. This embodiment adopts a modular design for each function of the system, mainly divided into three parts: ① Audiovisual integrated perception module: mainly uses a binocular camera and microphone array to collect bird images and sound information in real time, identify the species and determine the location of bird targets, and rely on the sound source location to assist the camera in quick positioning, so as to realize the tracking bird de-icing module; ② Sound and light hybrid bird de-icing module: mainly includes a laser emitter and an ultrasonic bird de-icing device mounted on a gimbal. For different types of bird targets, it can emit ultrasonic waves of different frequencies and combine them with lasers to achieve differentiated bird de-icing; ③ Remote control module: mainly monitors bird activities through a mobile terminal to coordinate and control the bird de-icing robot to de-ic.

[0077] Figure 7 The diagram shows the workflow of the bird detection and deterrence device in this embodiment. After the system is powered on, it initializes the parameters of each module and performs a system self-check. If no problems are detected, the rotating gimbal enters the automatic patrol mode. The device uses a microphone array to initially locate birds and adjusts the camera angle based on the acquired location information to achieve precise positioning. Once a bird infestation is detected, the bird deterrence mode is triggered. The device selects an appropriate bird deterrence strategy based on the detected bird species and location, emitting ultrasonic waves of a specific frequency and simultaneously activating a laser to deter birds. The rotating gimbal can adjust the laser emission angle in real time to achieve tracking-style bird deterrence. The device can send relevant data on bird detection and deterrence to a remote control module in real time, including the time of bird infestation detection, the number and species of birds, the deterrence method, and the time of bird infestation resolution. The remote control module can organize and analyze the relevant information to optimize the bird deterrence strategy accordingly.

[0078] The above description is merely one specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A smart bird detection and deflection device for a substation, comprising a bird deterrent body, a rotating gimbal mounted on the top of the bird deterrent body, the bird deterrent body including a main control device, an ultrasonic generator, a strobe light, and a battery compartment, the main control device and the battery compartment being fixed to the inner cavity side of the bird deterrent body, the battery compartment being used to power the entire device, the ultrasonic generator and the strobe light being embedded in the outer shell of the bird deterrent body and electrically connected to the main control device; the rotating gimbal including a gimbal base, a left rotating compartment, a right rotating compartment, and a ring microphone array disposed above the gimbal, the gimbal base being fixed to the bird deterrent body. At the top, the left rotating chamber houses an industrial camera, which, along with a ring microphone array, communicates with the main control device. Based on target detection and sound source localization algorithms, it perceives bird location information in real time. The right rotating chamber houses a laser emitter equipped with a self-powered module, also communicating with the main control device, which controls its operation. The rotating gimbal uses a stepper motor to control the rotation angle of the base and rotating chamber, adjusting the pointing of the industrial camera and laser emitter to achieve tracking-type bird detection and deterrence. A moving mechanism is mounted at the bottom of the bird deterrent device. The moving mechanism includes a drive motor, a wheel set, and a support frame. The drive motor is located at the bottom of the moving mechanism, and the four wheels of the wheel set are located at the four corners of the support frame. The drive motor is connected to the wheel set, and the wheel set is driven by the motor to move. The main control device controls the operation of the drive motor. The main control device is equipped with a wireless communication module and an information processing module. The wireless communication module is used for communication between the main control device and the ring microphone array, industrial camera, and laser emitter. The information processing module performs data analysis based on the auditory and visual information collected by the ring microphone array and industrial camera. The ring microphone array's built-in processor executes a sound source localization algorithm to initially perceive the bird's location and control the gimbal rotation. Then, a target detection algorithm is used to identify and locate the bird target in the video stream. The circular microphone array collects, filters, and reduces noise from bird sounds, and locates the sound source based on a sound source localization algorithm. The sound source localization algorithm is an improved Time Difference of Arrival (TDOA) method, which sets the time when the sound source arrives at the center of the microphone array as the standard time, thereby obtaining the time delay difference between the sound source and each microphone, and deriving an overdetermined set of equations. Then, a particle swarm optimization algorithm is applied to calculate the relative coordinate position of the sound source. The relative coordinate position is used to calculate the distance and amplitude of the sound source, thereby controlling the rotating gimbal to perform directional rotation. Its features are, The particle swarm optimization algorithm is as follows: The sound source location coordinates are set to (x, y, z), and the location coordinates of the i-th and j-th microphone nodes are set to (x, y, z). i ,y i ,z i ) and (x j ,y j ,z j Then, based on the node position coordinates, the following relationship is constructed: In the formula: i and j take values ​​from 1 to 6, representing six microphone nodes, and i ≠ j; v is the speed of sound in the air; τ ij This represents the time difference in signal reception between microphone i and microphone j. The problem model is transformed into a minimum value search model. Suppose there are N particles forming a community in a D-dimensional space, where the position vector and motion direction vector of the i-th particle are as follows: X i =(x i1 ,x i2 ,…,x iD ),i=1,2,…,N V i =(v i1 ,v i2 ,…,v iD ),i=1,2,…,N In each iteration of the algorithm, the particle moves once. Therefore, the position of the i-th particle in the (t+1)-th iteration is represented as: x ij (t+1)=x ij (t)+v ij (t+1) In the formula: j is related to dimension D, representing the concrete growth of the dimension, and the corresponding velocity vector is expressed as: v ij (t+1)=ωv ij (t)+c1r1(t)[P ij (t)-x ij (t)]+c2r2(t)[P gj (t)-x ij (t)] In the formula: ω is the inertia coefficient; c1 and c2 are learning factors; r1(t) and r2(t) are random parameters; P ij (t), P gj (t) represents the optimal position perceived by the particle itself and the optimal position perceived by the group, respectively, in each iteration, ωv ij (t) is a memory term; c1r1(t)[P ij (t)-x ij [(t)] represents the self-cognition term, i.e., the optimal iteration direction of the particle's self-cognition; c2r2(t)[P gj (t)-x ij [(t)] represents the group cognition term, which is the indication vector sent by the particle at the global optimal position in this iteration; The iteration termination condition is that the optimal position of the particle swarm search satisfies a predetermined minimum fitness threshold, let f i =T 1i -τ 1i (i = 1, 2, ..., 6), where f i Let T be the time delay estimation error of node i. 1i Let τ be the reference value for the time delay difference, representing the time delay difference between the i-th microphone and the reference microphone in acquiring sound information, while τ... 1i The time delay difference is obtained based on the generalized cross-correlation algorithm. Therefore, the model is transformed into a minimization problem, which is solved using the particle swarm optimization algorithm, as follows: F(x,y,z) represents the estimation error function for the sound source location coordinates, characterized by the matching property of two time delay differences. i (x,y,z) represents the time delay estimation error of node i at different coordinate positions, and M represents the total number of microphone nodes participating in the iteration. The sound source distance D and the amplitude θ are calculated based on the obtained position coordinates.

2. The intelligent bird detection and deterrence device for substations as described in claim 1, characterized in that, The industrial camera acquires video streams in real time, and the acquired visual information is processed by the built-in processor of the main control device to execute the YOLOv5n target detection algorithm for bird identification. The network architecture of the YOLOv5n target detection algorithm consists of three parts: a backbone network, a Neck, and a Head. The backbone network extracts features from the input image and adopts the EfficientFormerV2 lightweight network. The Neck is the connection part, which realizes multi-scale feature fusion through an FPN+PAN structure. The Head is the detection head part, which is responsible for generating the bounding box and class confidence of the target and making predictions on the acquired image. The bounding box regression loss function in the Head section uses the Wise-IoUv3 function, as shown in the following formula: In the formula: IOU represents the intersection-union ratio. The normalization factor represents the mean loss. W is a monotonic focusing coefficient, and the parameter β is used to characterize the degree of anomaly in the anchor frame. g and H g The x and y values ​​represent the size of the smallest closed bounding box, and the x and y values ​​represent the position parameters of the predicted bounding box. gt and y gt The parameters represent the location of the true bounding box, α and δ are hyperparameters, and r is the gradient gain, which is controlled by hyperparameters α and δ, so that the bounding box regression focuses on the anchor box of normal quality.

Images