Hybrid three-dimensional underwater positioning method based on convolutional neural network optimization, medium and equipment

Abstract

The invention provides a hybrid three-dimensional underwater positioning method based on convolutional neural network optimization, a medium and equipment, and the method comprises the steps: designing a three-dimensional underwater positioning algorithm to position an unknown node, and enabling each group of anchor points used for positioning the unknown node to pass robustness verification in the positioning process; a CNN model is constructed, initialization and training are carried out, the input of the CNN model is the water flow information and the anchor point position, and the output of the CNN model is the predicted node position; iteratively and alternately using the three-dimensional underwater positioning algorithm and the CNN model until the maximum number of iterations is reached or all unknown nodes are positioned; in each iteration, node positions obtained through positioning of the three-dimensional underwater positioning algorithm are used as anchor point positions to be input into the CNN model, and then the CNN model is adopted to predict the actual positions of the nodes in the moving water flow. The method has stronger robustness and adaptability, the error of the positioning result influenced by environmental factors is smaller, and the influence of the mobility of the underwater node on the positioning accuracy can be effectively avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean exploration and development, underwater rescue, ocean current detection and other tasks that require underwater node positioning, and specifically relates to a hybrid three-dimensional underwater positioning method, medium and equipment based on convolutional neural network optimization. Background Art

[0002] In the exploration of underwater resources, underwater positioning technology plays a vital role. An accurate and low-energy underwater positioning algorithm can greatly improve the efficiency of underwater work, thereby creating greater benefits. The development of underwater positioning technology is inseparable from the precise positioning of mobile nodes. In marine environmental monitoring, the detection area is usually large and dynamically changing. The use of mobile nodes can help track marine biological resources and use their random mobility to monitor the entire area. In addition, the optimization of underwater paths and the realization of a series of node data transmission processes also require the precise positioning of mobile nodes. Therefore, the precise positioning of underwater mobile nodes has become the focus of current research.

[0003] At present, underwater node positioning technology usually uses anchors and beacons to assist communication. Because underwater nodes are mobile at any time and difficult to maintain, and time synchronization cannot be guaranteed, researchers usually set up surface buoys and underwater beacon nodes (or anchor nodes) to assist positioning. The surface buoy is equipped with a GPS device to achieve absolute position positioning, while the beacon node has strong computing power and can communicate directly with the surface. Due to the particularity of the underwater environment, underwater wireless network positioning faces some challenges. First, radio signals cannot be transmitted stably for a long time underwater. In the absence of long antennas and high transmission power, only ultra-low frequency radio signals can propagate over long distances underwater. Secondly, sensor nodes in the water will move with the water flow and cannot remain unchanged, which will affect the communication range of the nodes. Therefore, underwater positioning usually uses acoustic signal transmission. However, due to the different pressure, temperature and medium in the ocean, the actual sound speed in the water is not easy to obtain, and it is difficult for nodes in the water to maintain time synchronization. Therefore, underwater wireless sensor network positioning cannot directly use the algorithm of land network positioning.

[0004] The traditional three-dimensional underwater positioning (3DUL) algorithm is of great significance in the field of underwater positioning. It replaces the previous positioning mode that requires a large number of anchor points. It can achieve comprehensive and stable network positioning through iterative traversal with only three anchor points. Its model is as follows Figure 1As shown in Figure 2. In the ranging phase, the three buoys iterate along the 3D topology, estimate the distance to the neighboring nodes, and obtain their depth information. After the buoys sense the three neighboring nodes, they project to all sensor nodes to check whether a robust virtual anchor plane is formed. Once established, the position of the sensor node is located through three-coordinate measurement and marked as a reference node to continue iterative communication with other nodes.

[0005] The traditional 3DUL algorithm improves the shortcomings of requiring a large number of underwater anchor points in the past, and optimizes the energy consumption and efficiency of positioning to a certain extent. However, the disadvantages are long positioning time, low accuracy, large data volume, and easy data changes, which is more suitable for positioning in calm waters. The 3DUL algorithm uses three surface anchor points as the coordinate base, and traverses the entire underwater node network in a three-dimensional topology iteration to achieve positioning. Unknown nodes can only be used as anchor points for other unknown nodes to help locate after they are localized. The positioning time of the entire network is proportional to the number of nodes, and the basic positioning time is too long. In the field of underwater positioning, the positioning time is directly linked to the positioning accuracy. The movement of water flow causes the underwater unknown nodes to keep moving at all times, and the location information is constantly changing. In extreme cases, the water flow speed is too high and the positioning time is too long. When the unknown nodes in the deep water area are located, the positions of the unknown nodes that have been successfully located in the shallow water area change and need to be relocated. The coordinates of the unknown nodes located with the shallow water node as the anchor point will have a large error, and the final positioning accuracy is low. At the same time, the method of relying on information exchange to locate unknown nodes alone cannot obtain accurate node location information. The characteristic of nodes moving all the time will introduce errors when finally obtaining coordinate information, resulting in reduced accuracy. In addition, when the traditional 3DUL algorithm is positioned multiple times, the iterative convergence speed is prone to change, and the network coverage fluctuates, causing the problem of unstable positioning. This is because the unknown underwater nodes are random, and robustness judgment is required to select the appropriate anchor point when exchanging node information. Therefore, as the number of iterations increases, the iteration direction will change randomly, and sometimes the entire network cannot be located. Finally, underwater nodes are mobile, and unknown nodes that have been localized and transformed into anchor points need to be relocated after a certain period of time before they can continue to serve as anchor points. Therefore, the traditional 3DUL algorithm needs to be repeatedly applied many times, which also directly affects the positioning accuracy.

[0006] Convolutional neural networks have the characteristics of processing large-scale data, having little effect on parameter changes, and more accurate prediction results. Using CNN for node localization is a powerful technique that can improve positioning accuracy in a variety of situations. Using CNN, researchers have achieved significant improvements in indoor positioning systems. CNN can effectively process time series data such as received signal strength indicator (RSSI) values, thereby reducing noise and improving positioning accuracy. Following the steps of designing and building a complete CNN network, optimizing and regularizing it, and performing pooling and striding, positioning with regression estimation of node coordinates is achieved. In addition, using CNN on spatiotemporal angle of arrival (AOA) images shows good results in dealing with noise and errors in AOA estimation for mobile node positioning. These methods demonstrate the versatility and effectiveness of CNN, which can improve positioning accuracy by handling noise, enhancing robustness, and optimizing available data, thereby achieving accurate node positioning in a variety of environments.

[0007] Although CNN has contributed and helped in indoor positioning, underwater node positioning algorithms have not been improved by CNN. For underwater node positioning, since the transmitted signal is an acoustic signal, most distance-based methods such as time of arrival (TOA) and time difference of arrival (TDOA) are used. These methods measure the distance and angle information from the unknown node to the anchor node, and use the three-coordinate method and triangulation method to calculate the location information of the unknown node, which has high accuracy. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention provides a hybrid three-dimensional underwater positioning method, medium and equipment based on convolutional neural network optimization.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] In a first aspect, the present invention provides a hybrid three-dimensional underwater positioning method based on convolutional neural network optimization, characterized in that it comprises the following steps:

[0011] Design a 3D underwater positioning algorithm to locate unknown nodes. During the positioning process, each set of anchor points used to locate unknown nodes must pass robustness verification;

[0012] Constructing a CNN model and performing initialization and training, wherein the input of the CNN model is water flow information and anchor point positions, and the output of the CNN model is the predicted node position;

[0013] The 3D underwater positioning algorithm and the CNN model are used alternately and iteratively until the maximum number of iterations is reached or all unknown nodes are located. In each iteration, the node position located by the 3D underwater positioning algorithm is input into the CNN model as the anchor position, and then the CNN model is used to predict the actual position of the node in the moving water flow.

[0014] Optionally, the process of locating the unknown node by the three-dimensional underwater positioning algorithm includes a ranging stage and a projection and positioning stage;

[0015] In the ranging phase, the unknown node estimates its position to the neighboring anchor points by broadcasting and receiving data packets. When the unknown node determines that it is within the communication range of the three anchor points, it enters the projection and positioning phase;

[0016] In the projection and positioning stage, three anchor points are projected onto a horizontal plane at the same depth as the unknown node to perform three-sided positioning, and the accuracy of the unknown node positioning is determined through robustness verification.

[0017] Optionally, the specific process of the ranging stage is:

[0018] After positioning is started, the anchor point broadcasts anchor point data packets to the surrounding area, trying to get in touch with the unknown nodes within its communication range; when the sensor carried by the unknown node receives three anchor point data packets from different anchor points, the unknown node sends a ranging data packet to the three anchor points at t=t1, requesting information exchange with the anchor points; the anchor point receives the ranging data packet at t=t2, and at t=t3, the anchor point broadcasts a confirmation packet to the unknown node; the unknown node receives the confirmation packet sent by the anchor point at t=t4, which contains the values ​​of t2 and t3 as well as the coordinates and depth information of the anchor point, which is used for distance measurement between the unknown node and the anchor point.

[0019] Optionally, the distance measurement is achieved by multiplying the average one-way propagation time by the estimated underwater sound speed, as follows:

[0020] d = c*[(t2-t1)+(t4-t3)] / 2;

[0021] Where d represents the measured distance, c represents the speed of sound, and the time information of t1, t4 and t2, t3 are measured by the local clocks of the unknown node and the anchor point, respectively.

[0022] Optionally, the specific process of the projection and positioning stage is:

[0023] When the unknown node receives confirmation packets from three different anchor points, the node uses the coordinates and depth information in the confirmation packets to project the anchor points onto the horizontal plane at the same depth as the unknown node;

[0024] Robustness verification is performed on all sub-triangles in a quadrilateral consisting of three anchor points and one unknown node: when all sub-triangles meet the verification conditions, the quadrilateral is considered to be a robust quadrilateral. After verification, the anchor points send data packets to the unknown nodes for trilateral measurement to locate the unknown nodes; when a group of anchor points fails the verification, other anchor points are reselected for verification until the unknown nodes can be successfully located.

[0025] Optionally, the robustness verification condition is:

[0026] a*sin 2 δ>d min ;

[0027] In the formula, a is the shortest side length of each sub-triangle, δ is the minimum angle of each sub-triangle, and d min is the experimental threshold after noise detection and data measurement.

[0028] Optionally, the CNN model defines environmental parameters that match the actual positioning environment, including the size of the positioning area, the total number of nodes, the number of anchor points and the water flow speed; the CNN model also defines model parameters, including the number of input features, the number of samples for each training, the number of training rounds and the initial learning rate of the model; the CNN model adopts a learning rate decay strategy during training, the optimizer selects stochastic gradient descent, and sets training options including batch size and training cycle.

[0029] Optionally, in the process of iteratively alternating the use of the three-dimensional underwater positioning algorithm and the CNN model, for each node whose state is 0 and has sufficient remaining energy, the three-dimensional underwater positioning algorithm is first used to locate the node, and its state is updated to 1, and then its position information and water flow information are input into the CNN model to predict the actual position of the node, and the predicted actual position of the node is used for the three-dimensional underwater positioning algorithm in the next iteration; wherein, a state of 0 indicates not located, and a state of 1 indicates located.

[0030] In a second aspect, the present invention provides a computer-readable storage medium storing a computer program, characterized in that the computer program enables a computer to execute the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in the first aspect.

[0031] In a third aspect, the present invention provides an electronic device, characterized in that it comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in the first aspect.

[0032] The beneficial effects of the present invention are as follows: the improved 3DUL algorithm proposed by the present invention fully considers time synchronization, adds a time synchronization adjustment function, and further reduces errors; the anchor point does not need relative position information of the unknown node, and the anchor point can successfully assist any unknown node within its communication range to achieve positioning through robust judgment, and has high flexibility and adaptability. The present invention combines the 3DUL algorithm with a convolutional neural network to predict node movement, and inputs water flow information and anchor point coordinates for training, assists in predicting the position of the node after movement, improves the adaptability of the model, and has the ability to predict the position of underwater mobile nodes, avoiding the complex operation of repeatedly locating unknown nodes that have already moved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of the traditional 3DUL algorithm model.

[0034] Figure 2 It is a schematic diagram of RMSE and Loss in the CNN model training proposed in the present invention.

[0035] Figure 3 It is a performance diagram of the DUL_CNN algorithm proposed in the present invention.

[0036] Figure 4 This is a performance diagram of the traditional 3DUL algorithm. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0038] In one embodiment, the present invention proposes a hybrid three-dimensional underwater positioning method based on convolutional neural network optimization, which is an optimized hybrid algorithm obtained by combining the improved 3DUL algorithm with the CNN model. Through the training process of the neural network, the accuracy and efficiency of the positioning algorithm are improved. The optimized algorithm avoids the problem of requiring multiple high-energy beacon nodes for positioning in the past. Only three surface anchor nodes with known positions are required for positioning, and the CNN model is combined in the iterative process for learning, convergence and positioning prediction.

[0039] The 3DUL algorithm follows the traditional 3DUL algorithm mode, and the positioning process is divided into a ranging phase and a dynamic positioning phase. In the ranging phase, the node estimates the distance from itself to the neighboring nodes by broadcasting and receiving data packets, and obtains depth information through the pressure sensor it carries. After that, when the node determines that it is within the communication range of the three anchor points, it enters the second phase, namely the projection and dynamic positioning phase. The three anchor nodes are projected onto the horizontal plane at the same depth as the unknown node, and three-sided positioning is performed, and the exact position of the unknown node is determined by robustness verification. After successful localization, the node becomes an anchor point, republishes its new state to the network, and helps other nodes to localize. The entire iterative behavior will continue to repeat until the entire network is dynamically positioned. The traditional 3DUL algorithm does not consider time synchronization. This embodiment adds a time synchronization adjustment function to further reduce errors.

[0040] The experimental conditions of this embodiment are: three anchor points with GPS positioning, unknown nodes with information exchange capability and pressure sensors (to obtain depth information), the underwater test area is set to 300*300*50 by default, and the water flow velocity is set to {-0.1, 0.1} by default. The specific implementation process includes the following stages.

[0041] A. Ranging phase

[0042] After positioning is started, the anchor point broadcasts anchor_packet to the surrounding area, trying to get in touch with the unknown node within its communication range. When the unknown node sensor receives three data packets from different anchor points, the unknown node will send ranging_packet to the three anchor points, requesting information exchange with them. After the anchor point is authenticated, two-way information exchange is carried out to achieve distance measurement.

[0043] A pair of unknown nodes and anchors in the ranging process. The unknown node sends a ranging_packet to the anchor at t=t1, and the anchor receives the data packet at t=t2. When t=t3, the anchor broadcasts a confirmation packet to the unknown node, and the unknown node receives the message at t=t4, which contains the values ​​of t2, t3, and the coordinates and depth of the anchor. The time information t1-t4 is measured by the local clocks of the unknown node and the anchor respectively. The distance measurement is achieved by multiplying the average one-way propagation time by the estimated speed of sound underwater.

[0044] d = c*[(t2-t1)+(t4-t3)] / 2;

[0045] Where d represents the measuring distance and c represents the speed of sound.

[0046] After the unknown node obtains the acknowledgment_packet from three different anchor points, the distances to the three anchor points are measured. This phase ends with the projection and dynamic measurement phase.

[0047] B. Projection and Positioning Phase

[0048] When the unknown node receives a confirmation packet from three different anchor points, the node uses the depth and coordinate information in the packet to project the anchor point onto a horizontal plane at the same depth as the unknown node. The judgment of a robust quadrilateral is based on the conditional analysis of all sub-triangles consisting of three anchor points and an unknown node. When all sub-triangles meet the conditions, the quadrilateral is considered to be a robust quadrilateral. This definition effectively avoids trigger ambiguity errors and achieves effective robustness judgment. The conditional formula that needs to be met is: a*sin 2 δ>d min . Where a is the shortest side length of each sub-triangle, δ is the minimum angle of each sub-triangle, and d min is the experimental threshold after noise detection and data measurement, which is set to 10 in this embodiment. After many experiments and comparisons, when the threshold is too small, the density of the anchor point has a greater impact on the positioning result, and too much coordinate information that is not suitable as an anchor point is introduced, which increases the calculation time and positioning difficulty. If the threshold is too large, the positioning accuracy of the model will be significantly reduced. In order to make the output experimental results and images more intuitive and clear, the threshold of this embodiment is set to 10.

[0049] After robustness judgment and projection processing, the anchor points send data packets to the unknown nodes for dynamic trilateral measurement to achieve the positioning of the unknown nodes. When a group of anchor points cannot pass the robustness condition judgment, other anchor points are reselected to perform the same operation until the unknown node can be successfully positioned. Repeat this process so that all locatable nodes in the positioning area network are successfully positioned, and expand to all locations through an iterative method. The algorithm does not require the relative orientation information of the anchor points to the unknown nodes. The anchor points can successfully assist any unknown node within their communication range to achieve positioning through robust judgment, and have high flexibility and adaptability.

[0050] C. CNN establishment training phase

[0051] In order to enhance the stability and accuracy of positioning, this embodiment combines the 3DUL algorithm and the convolutional neural network to predict the node movement, and inputs the water flow information and anchor point coordinates for training to assist in predicting the position of the node after movement and improve the adaptability of the model.

[0052] In order to fully achieve the intended purpose, it is necessary to build a suitable CNN model structure according to the situation and set a reasonable training method to help the model run successfully. In the construction stage, the dropout layer, batch homogenization layer, activation function layer, multiple fully connected layers and a convolutional layer are defined to build the CNN model. The activation function uses the Relu function, combined with the dropout layer to improve the generalization ability of the model.

[0053] In the preparation stage, we first define the environmental parameters that match the actual positioning environment, such as the size of the positioning area, the total number of nodes, the number of anchor points, the water flow speed, etc. In addition to the environmental parameters, we also define the CNN model parameters. The model parameters include the number of input features, the number of samples for each training, the number of training rounds, and the initial learning rate of the model. Figure 2 As shown, this embodiment sets the initial learning rate to 0.01, and adopts the learning rate decay strategy during the training process to improve the generalization ability and convergence speed of the model. The optimizer selects stochastic gradient descent (SGD) and sets training options such as batch size and training cycle. The randomly generated training data includes node position, water flow velocity and target node. Extract input features and record training samples, and scale and normalize the input features and target values ​​between 0 and 1. The normalization operation that cannot be omitted reduces the value range of the input features to a similar scale, avoiding the training difficulties caused by the large value differences between different features. This helps the optimization algorithm to find the optimal solution faster and speed up the convergence speed of the model. Reducing the correlation between features helps the model better learn the relationship between features and improve the generalization ability and prediction performance of the model. After starting training, in the loop, for each training sample, the position and water flow velocity of the node nodes_pos are first randomly generated. Then a node is randomly selected as the target node target_node, and the input feature input_feature is extracted, which includes the position and water flow velocity information of the target node. Then the extracted input_feature is recorded in train_features (training features), and the position of the target node is recorded in train_positions (training positions). The model is trained using the data recorded in both, so that the model can predict the mobile nodes in the location environment. Finally, the trained model, normalization parameters, etc. are saved in the file cnn_model for later use. It is worth noting that the number and shape of samples input to the model input layer need to match the structure of the CNN input layer to avoid dimensional anomalies.

[0054] D. Algorithm combination

[0055] The trained CNN model has the initial ability to predict the position of underwater mobile nodes. Using the CNN model, the positioning algorithm avoids the complex operation of repeatedly positioning unknown nodes that have already moved.

[0056] With the help of this network model, this embodiment implements a hybrid underwater node positioning algorithm. In the stage of calculating the specific position of the node, each iteration alternately uses trilateral measurement and CNN prediction to continuously update and optimize the node's position estimation.

[0057] In each iteration, for nodes with a state of 0 and sufficient remaining energy, the code selects the three nearest anchor nodes, calculates the projection distance from the node to the anchor node, and if the robustness condition is met, uses trilateral measurement to calculate the exact position of the node and updates its state to 1. Repeat these two methods until the maximum number of iterations is reached or all nodes are located. When polygon positioning is performed on an unknown node, its calculated positioning position can be obtained. Its position information is input as an anchor point into the CNN model for prediction to obtain the predicted position.

[0058] like Figure 3 and Figure 4 As shown in the figure, compared with the traditional calculation method, this method has stronger robustness and adaptability, and the positioning result is less affected by environmental factors. The alternating use of these two methods can update the position of the node used as the anchor point at any time, so that it has higher accuracy in the subsequent auxiliary positioning of other nodes.

[0059] The node positions obtained by the traditional single three-sided positioning algorithm are time-sensitive. Consider one or more groups of underwater anchor points that remain relatively stationary in an area with relatively stable water flow speed, thus ignoring their inherent mobility. After the unknown node positioning is converted to a new anchor point, the coordinate information contained in the confirmation packet sent by the new anchor point to the unknown node it assisted in positioning does not correspond to the actual position at the time of sending. When the water flow speed approaches 0, this error is very small. The error increases with the increase of water flow speed. The algorithm is processed through the CNN model to reduce the difference between the actual position of the new anchor point located subsequently and the self-positioning position information contained in the data packet, further improving the positioning accuracy and efficiency.

[0060] The actual application stages of the algorithm are arranged in time sequence as ranging stage, CNN establishment and training stage, three-dimensional projection and robustness judgment stage, dynamic trilateral measurement stage and CNN prediction and positioning stage (the two stages are mixed and carried out simultaneously).

[0061] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in Example 1.

[0062] In another embodiment, the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in Example 1 is implemented.

[0063] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0064] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0065] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A hybrid three-dimensional underwater positioning method based on convolutional neural network optimization, characterized in that: The steps include: Design a 3D underwater positioning algorithm to locate unknown nodes. During the positioning process, each set of anchor points used to locate unknown nodes must pass robustness verification; Constructing a CNN model and performing initialization and training, wherein the input of the CNN model is water flow information and anchor point positions, and the output of the CNN model is the predicted node position; The 3D underwater positioning algorithm and the CNN model are used alternately and iteratively until the maximum number of iterations is reached or all unknown nodes are located. In each iteration, the node position located by the 3D underwater positioning algorithm is input into the CNN model as the anchor position, and then the CNN model is used to predict the actual position of the node in the moving water flow.

2. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization according to claim 1, characterized in that: The process of locating unknown nodes by the three-dimensional underwater positioning algorithm includes a ranging phase and a projection and positioning phase; In the ranging phase, the unknown node estimates its position to the neighboring anchor points by broadcasting and receiving data packets. When the unknown node determines that it is within the communication range of the three anchor points, it enters the projection and positioning phase; In the projection and positioning stage, three anchor points are projected onto a horizontal plane at the same depth as the unknown node to perform three-sided positioning, and the accuracy of the unknown node positioning is determined through robustness verification.

3. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as claimed in claim 2, characterized in that: The specific process of the ranging stage is: After positioning is started, the anchor point broadcasts anchor point data packets to the surrounding area, trying to get in touch with the unknown nodes within its communication range; when the sensor carried by the unknown node receives three anchor point data packets from different anchor points, the unknown node sends a ranging data packet to the three anchor points at t=t1, requesting information exchange with the anchor points; the anchor point receives the ranging data packet at t=t2, and at t=t3, the anchor point broadcasts a confirmation packet to the unknown node; the unknown node receives the confirmation packet sent by the anchor point at t=t4, which contains the values ​​of t2 and t3 as well as the coordinates and depth information of the anchor point, which is used for distance measurement between the unknown node and the anchor point.

4. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as claimed in claim 3, characterized in that: The distance measurement is achieved by multiplying the average one-way propagation time by the estimated speed of sound underwater, as follows: d = c*[(t2-t1)+(t4-t3)] / 2; Where d represents the measured distance, c represents the speed of sound, and the time information of t1, t4 and t2, t3 are measured by the local clocks of the unknown node and the anchor point, respectively.

5. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as claimed in claim 2, characterized in that: The specific process of the projection and positioning stage is as follows: When the unknown node receives confirmation packets from three different anchor points, the node uses the coordinates and depth information in the confirmation packets to project the anchor points onto the horizontal plane at the same depth as the unknown node; Robustness verification is performed on all sub-triangles in a quadrilateral consisting of three anchor points and one unknown node: when all sub-triangles meet the verification conditions, the quadrilateral is considered to be a robust quadrilateral. After verification, the anchor points send data packets to the unknown nodes for trilateral measurement to locate the unknown nodes; when a group of anchor points fails the verification, other anchor points are reselected for verification until the unknown nodes can be successfully located.

6. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization according to claim 5, characterized in that: The robustness verification conditions are: a*sin 2 δ>d min ; In the formula, a is the shortest side length of each sub-triangle, δ is the minimum angle of each sub-triangle, and d min is the experimental threshold after noise detection and data measurement.

7. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization according to claim 1, characterized in that: The CNN model defines environmental parameters that match the actual positioning environment, including the size of the positioning area, the total number of nodes, the number of anchor points and the water flow speed; the CNN model also defines model parameters, including the number of input features, the number of samples for each training, the number of training rounds and the initial learning rate of the model; the CNN model adopts a learning rate decay strategy during training, the optimizer selects stochastic gradient descent, and sets training options including batch size and training cycle.

8. The hybrid three-dimensional underwater positioning method based on convolutional neural network optimization according to claim 1, characterized in that: In the process of iteratively alternating the use of the three-dimensional underwater positioning algorithm and the CNN model, for each node whose state is 0 and whose remaining energy is sufficient, the three-dimensional underwater positioning algorithm is first used to locate the node, and its state is updated to 1, and then its position information and water flow information are input into the CNN model to predict the actual position of the node, and the predicted actual position of the node is used for the three-dimensional underwater positioning algorithm in the next iteration; wherein, a state of 0 indicates not located, and a state of 1 indicates located.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in any one of claims 1-8.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the hybrid three-dimensional underwater positioning method based on convolutional neural network optimization as described in any one of claims 1 to 8 is implemented.

Images