A floating particle and a high-precision positioning method for multiple floating particles

Abstract

The present invention relates to a floating particle and a high-precision positioning method for multiple floating particles. The floating particle includes a signal acquisition module, an environmental monitoring module, a positioning module, a communication module, a particle signal forwarding module, and an energy management module, and all the foregoing modules are integrated inside a housing. By designing a multi-float collaborative positioning system, combining an environment adaptive calibration mechanism and low-power communication technology, efficient and reliable real-time monitoring of floating particles is achieved in a complex and dynamic marine environment, meeting the requirements of multiple fields such as environmental protection, scientific research, and disaster emergency, and realizing high-precision real-time positioning of floating particles in a complex marine environment.

Description

Technical Field

[0001] The present invention relates to a floating particle and a high-precision positioning method for multiple floating particles, belonging to the field of node positioning in marine wireless sensor networks. Background Art

[0002] With the continuous growth of the demand for marine ecological protection and environmental governance, the marine floating particle positioning technology has important applications in multiple fields, including plastic waste distribution tracking, oil spill pollution diffusion control, marine ecological monitoring, buoy observation data collection, fishery resource protection, and precise recovery of floating instruments. These scenarios pose higher requirements for the real-time and high-precision positioning of the distribution of floating objects, especially in long-term monitoring, dynamic tracking, and emergency response.

[0003] Currently, the positioning methods for floating particles mainly include the following: satellite-based marine monitoring and positioning technology, unmanned aerial vehicle (UAV) visual monitoring, monitoring technology for a single floating node, acoustic positioning technology, and radar monitoring technology. Regarding satellite-based marine monitoring and positioning technology, satellite monitoring can cover a wide sea area and is suitable for large-scale statistics of the distribution of floating objects. However, due to resolution limitations, it is difficult to accurately locate small-volume floating particles, and the real-time performance is poor, unable to meet the dynamic tracking requirements. Regarding UAV visual monitoring, it is carried out by a UAV equipped with a camera or sensor, so it can identify and track floating objects in a local area. However, limited by battery life and environmental conditions (such as wind speed and weather), it is difficult to operate for a long time and in a large range. The monitoring technology for a single floating node uses a single buoy to monitor floating objects through GPS or signal ranging. Due to the lack of multi-point cooperation, its measurement results are easily affected by multipath effects and sea wave interference, and the positioning accuracy is low. Acoustic positioning technology uses the characteristics of sound wave propagation to locate underwater or surface objects. In a complex marine environment, the propagation of sound waves is greatly affected by factors such as temperature, salinity, and water depth, which easily causes ranging errors. In addition, the power consumption of acoustic devices is relatively high, and it is not suitable for long-term operation. Radar monitoring technology uses a high-frequency radar to detect the position and distribution of floating objects through reflected signals. This technology has insufficient detection sensitivity for small-volume floating particles (such as plastic micro-particles), and the equipment cost is relatively high, which is not suitable for large-scale marine applications.

[0004] Therefore, it is necessary to provide a new positioning method for floating particles to overcome the problems in the prior art, such as "insufficient positioning accuracy, especially for small-volume floating particles", "unstable monitoring results affected by complex marine environments", "high cost and large power consumption, making it difficult to meet the long-term operation requirements", "large communication delay and poor real-time performance", and "lack of large-scale collaborative monitoring ability". Summary of the Invention

[0005] The present invention provides a floating particle and a high-precision positioning method for multiple floating particles. Through the collaborative work of multiple floats, environment adaptive calibration, and low-power communication technology, high-precision real-time positioning of floating particles in a complex marine environment is achieved.

[0006] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0007] A floating particle includes a signal acquisition module, an environment monitoring module, a positioning module, a communication module, a particle signal forwarding module, and an energy management module. All the foregoing modules are integrated inside a housing;

[0008] The signal acquisition module includes an optical sensor, an ultrasonic sensor, and an electromagnetic sensor. It is used to detect the signals of the floating particles and transmit the signals to a central processing unit deployed at a coastal station or on a ship;

[0009] The environment monitoring module includes an anemometer, a turbidity sensor, an acceleration sensor, and a temperature sensor. It is used to collect environmental data in real time to assist in correcting positioning errors;

[0010] The positioning module is used to measure the distance or positional relationship between floating particles;

[0011] The communication module is based on the LoRa communication protocol for long-distance data transmission;

[0012] The particle signal forwarding module expands the signal coverage range of the floating particles through a mutual forwarding mechanism;

[0013] The energy management module includes an integrated solar or wave energy power generation system;

[0014] A high-precision positioning method for multiple floating particles specifically includes the following steps:

[0015] Step S1, put a number of floating particles into the ocean;

[0016] Step S2, start the floating particles. The signal acquisition module sends the signals of the floating particles and the collected environmental data to the central processing unit through the communication module;

[0017] Step S3, construct a multi-target processing model, separate the characteristics of different floating particles from the signals of a number of floating particles, and at the same time assign an identifier to each floating particle, and output the predicted coordinates of a single floating particle;

[0018] Step S4, based on the predicted coordinates obtained in step S3, narrow down the target position range of the floating particles;

[0019] Step S5: Within the target position range, select a floating particle as the target floating node. Establish a signal propagation model for a single floating particle based on the time difference of the signals from the target floating node reaching different floating particles, and determine the position of the target floating node;

[0020] Step S6: Based on the environmental data collected in Step S2, use a neural network to dynamically adjust the propagation speed in the signal propagation model established in Step S5 to correct the error caused by environmental changes;

[0021] Step S7: Use the propagation speed corrected in Step S6 to obtain a corrected signal propagation model, and at the same time, combine the position of the target floating node determined in Step S5 to construct an optimized single floating particle trajectory prediction model;

[0022] Further, in Step S3, the specific steps of constructing a multi-target processing model to output the predicted coordinates of a single floating particle are as follows:

[0023] Step S31: Construct the model architecture of the multi-target processing model, including an input layer, a feature extraction layer, and a fusion classification layer;

[0024] Step S32: The input layer converts the original signal x ( t ) into a frequency spectrum through a fast Fourier transform X ( F ). The feature extraction layer extracts a frequency feature vector and an amplitude feature vector respectively through a dual-branch convolutional network. The fusion classification layer concatenates the dual-branch features and outputs a classification result through a fully connected layer. The classification result output formula is

[0025]

[0026] In the formula, y is the classification result, is the frequency feature vector, is the amplitude feature vector, W is the weight, b is the bias;

[0027] Step S33: Regard each floating particle as a target, defined as k , and generate a unique identifier for a single floating particle according to the classification result, that is

[0028]

[0029] In the formula, is the identifier, is the target k frequency, is the target k amplitude, as the target k the timestamp of the first occurrence, Hash the function ensures uniqueness;

[0030] Step S34, using a recurrent neural network to process the normalized frequency and amplitude to obtain the hidden time series data at the current moment. The hidden state update formula is

[0031]

[0032] In the formula, h t as the target k the hidden state at the current moment, h t-1 as the target k the hidden state at the previous moment, x t as the target k the input at the current moment, that is, the normalized frequency F t and amplitude A t , W h is the weight at the current moment, b h is the bias at the current moment, σ is the Sigmoid activation function;

[0033] Step S35, output the predicted coordinates of a single floating particle. The prediction formula is

[0034]

[0035] In the formula, is the output containing the predicted coordinates of the next moment ( x ′, y ′, z ′), W y is the weight containing the next moment, h t as the target k the hidden state at the current moment, b y is the bias containing the next moment;

[0036] Furthermore, in Step S5, the steps for constructing the signal propagation speed model are specifically as follows:

[0037] Step S511, select a floating particle from several floating particles and define it as the target float node. The formula for calculating the time difference between other several floating particles and the target float node is

[0038]

[0039] In the formula, is the time difference, is the time of other floating particles, is the time of the target float node, N is the total number of floating particles;

[0040] Step S512, the constructed signal propagation model is

[0041]

[0042] In the formula, is the distance difference corresponding to the time difference, is the propagation speed, is the time difference;

[0043] Furthermore, in step S5, the steps of determining the position of the target float node are specifically as follows:

[0044] Step S521, construct the hyperboloid equation corresponding to the time difference between every two floating particles, specifically

[0045]

[0046] In the formula, are all obtained through the predicted coordinates of a single floating particle output in step S35;

[0047] If N ≥4, N floating particles form multiple hyperboloid equations to establish an overdetermined system of equations:

[0048] ;

[0049] Step S522, optimize the results of the constructed hyperboloid equation to determine the position of the target float node. The optimization formula is:

[0050]

[0051] In the formula, , M is the total number of equations;

[0052] Furthermore, in step S6, the specific steps of dynamically adjusting the propagation speed in the signal propagation model are:

[0053] Step S61, construct the signal propagation speed model, and the correction formula is

[0054]

[0055] In the formula, is the propagation speed correction value, is the standard propagation speed, E is the environmental factor, and the environmental factor is temperature or humidity or wind speed, is the non - linear function for environmental correction;

[0056] Step S62, evaluate the accuracy of the prediction of the correction formula, adjust the parameters of the correction formula through the evaluation results, take the environmental factor as the input, and finally output the propagation speed correction value. Among them, the evaluation formula is

[0057]

[0058] In the formula, is the propagation speed correction value predicted by the correction formula, is the true propagation speed correction value obtained by offline data collection, and ζ is the loss function of the correction formula;

[0059] Further, in step S7, the specific steps to construct the optimized single floating particle trajectory prediction model are as follows:

[0060] Step S71, use a convolutional neural network to perform multi - dimensional feature extraction on the time - difference matrix and the signal matrix generated when determining the position of the target float node, and fuse the extracted features to output an accurate position estimate value. The fusion formula is

[0061]

[0062] In the formula, is the predicted position vector, T is the time - difference matrix, R is the signal matrix;

[0063] Step S72, evaluate the gap between the predicted position and the true position, adjust the parameters of the single floating particle trajectory prediction model, and finally accurately locate the position of the floating particle. Among them, the evaluation formula is

[0064]

[0065] In the formula, is the loss function of the evaluation formula, is the predicted position vector, is the known true position, is the Euclidean distance.

[0066] Through the above technical solutions, compared with the prior art, the present invention has the following beneficial effects:

[0067] 1. The floating particles provided by the present invention utilize low-power LoRa communication and solar power supply technologies, are easy to produce and deploy, can achieve large-scale coverage of a floating network, and realize the long-term monitoring ability of devices.

[0068] 2. The multi-floating particle high-precision positioning method provided by the present invention meets the real-time monitoring requirements of wide sea areas through a distributed network structure and real-time algorithms.

[0069] 3. The multi-floating particle high-precision positioning method provided by the present invention solves the signal interference problem in complex marine environments and realizes high-precision particle positioning through multi-float collaborative positioning and dynamic error correction.

[0070] 4. The multi-floating particle high-precision positioning method provided by the present invention is a positioning solution with low cost, strong adaptability, and high reliability, meeting the application requirements of multiple scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The present invention will be further described below in conjunction with the drawings and embodiments.

[0072] Figure 1 is a comparison example of the root mean square error between the multi-target processing method and the Kalman filtering method of the present invention;

[0073] Figure 2 is a comparison example of the false alarm rate between the multi-target processing method and the Kalman filtering method of the present invention;

[0074] Figure 3 is an example diagram of the distributed computing method provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] The present invention will now be described in further detail with reference to the drawings.

[0076] As described in the background art, due to the complex and dynamic characteristics of the marine environment (such as waves, wind speed, tides, etc.), existing positioning methods usually have difficulty ensuring reliability and accuracy under these conditions, and there are significant technical limitations. For example, satellite-based positioning technology has high costs, poor real-time performance, and its accuracy is greatly affected by weather. Monitoring floating objects visually by drones carried with cameras has a short endurance time and cannot achieve long-term monitoring. A single floating buoy monitors floating objects through GPS or signal ranging technology, lacking multi-point collaboration, having large positioning errors, and being easily interfered by the environment.

[0077] Therefore, the present application provides a high-precision positioning method for multiple floating particles, aiming to solve the technical problems of high-precision real-time positioning and monitoring of floating particles in the marine environment, and achieve high-precision real-time positioning of multiple floating particles in a complex marine environment. If high-precision real-time positioning of several floating particles is to be achieved, the structure of the floating particles is also crucial. The floating particles provided by the present application include a signal acquisition module, an environmental monitoring module, a positioning module, a communication module, a particle signal forwarding module, and an energy management module. All of the foregoing modules are integrated inside a housing. The signal acquisition module includes an optical sensor, an ultrasonic sensor, an electromagnetic sensor, or other applicable detection devices, which are used to detect the signals of the floating particles and transmit the signals to a central processing unit deployed at a coastal station or on a ship. The environmental monitoring module includes an anemometer, a turbidity sensor, an acceleration sensor, and a temperature sensor, which are used to collect environmental data in real time (such as wind speed, sea current speed, temperature, wave height, wave period, and seawater turbidity) to assist in positioning error correction. The positioning module supports GPS positioning and wireless signal ranging (such as RSSI, TOA) functions, and is used to measure the distance or positional relationship between floating particles. The communication module is based on the LoRa communication protocol and supports low-power, long-distance data transmission. The particle signal forwarding module enhances the communication ability between particles, expands the signal coverage range through an inter-forwarding mechanism, and improves the reliability of data transmission. The energy management module includes an integrated solar or wave energy power generation system to ensure long-term power supply capacity.

[0078] Regarding the central processing unit, it is responsible for receiving the data (actively transmitted radio signals) from the floating particles, performing a positioning algorithm, and outputting the real-time positions of the floating particles.

[0079] Next is a high-precision positioning method for multiple floating particles provided by the present application, which specifically includes the following steps:

[0080] Step S1, put several of the aforesaid floating particles into the ocean. Each floating particle is a basic unit in a distributed network, responsible for collecting signals, and processing and transmitting them.

[0081] Step S2, after starting the floating particles, perform self-positioning and establish a distributed LoRa communication network; the signal acquisition module transmits the signals of the floating particles and the collected environmental data to the central processing unit through the communication module.

[0082] Due to the large number of floating particles at sea, the signals are complex. Therefore, we need to separate the characteristics of different floating particles from the mixed signals to initially predict the position of a single floating particle, roughly narrow down the position of the floating particle to a certain range, avoid signal loss, improve the calculation speed, and facilitate the tracking of floating particles. This is step S3. The central processing unit receives the data of multiple floating particles, constructs a multi-target processing model, separates the characteristics of different floating particles from the signals of several floating particles, and at the same time assigns an identifier to each floating particle, and outputs the predicted coordinates of a single floating particle.

[0083] The specific steps are as follows:

[0084] Step S31, construct the model architecture of the multi-target processing model, including an input layer, a feature extraction layer, and a fusion classification layer;

[0085] Step S32, the input layer converts the original signal x(t) into a spectrum through fast Fourier transform X(F) . The feature extraction layer extracts the frequency feature vector and the amplitude feature vector respectively through a dual-branch convolutional network (CNN). Preferably, for the frequency branch: 1D convolutional kernel (size = 5, channels = 32) → extract local frequency domain features. For the amplitude branch: 1D convolutional kernel (size = 3, channels = 16) → extract amplitude change features. The fusion classification layer concatenates the dual-branch features and outputs the classification result through a fully connected layer. The classification result output formula is

[0086]

[0087] In the formula, y is the classification result, is the frequency feature vector, is the amplitude feature vector, W is the weight, b is the bias;

[0088] Step S33, regard each floating particle as a target, defined as k , generate a unique identifier for a single floating particle according to the classification result , that is

[0089]

[0090] In the formula, is the identifier, is the target k 's frequency, is the target k 's amplitude, is the target k 's timestamp of the first appearance, Hash The function ensures uniqueness;

[0091] Next, a recurrent neural network (RNN) is used to process the time-series signal features, track the dynamic changes of floating particles, and improve the accuracy of target discrimination and recognition. The process is as follows: In step S34, the normalized frequency and amplitude are processed using a recurrent neural network to obtain the hidden time-series data at the current moment. The hidden state update formula is

[0092]

[0093] In the formula, h t is the target k hidden state at the current moment, h t-1 is the target k hidden state at the previous moment, is the target k input at the current moment, that is, the normalized frequency and amplitude , is the weight at the current moment, is the bias at the current moment, is the Sigmoid activation function;

[0094] In step S35, the predicted coordinates of a single floating particle are output. The prediction formula is

[0095]

[0096] In the formula, is the predicted coordinate including the next moment output , is the weight including the next moment, h t is the target k hidden state at the current moment, b y is the bias including the next moment.

[0097] Regarding the benefits of using this method, it is compared with the commonly used Kalman filtering method here. The conditions are set as shown in Table 1.

[0098] Table 1

[0099]

[0100] From Figure 1 the root mean square error to Figure 2 the false alarm rate comparison, it can be seen that under different sea conditions, the performance obtained by the multi-target processing method adopted in this application is significantly higher than that obtained by the Kalman filtering method.

[0101] ​​​​In step S4, based on the predicted coordinates obtained in step S3, the target position range of the floating particles is narrowed down.

[0102] After narrowing down the target range, the TDOA algorithm is used to accurately locate the floating particles. The position of the floating particles is determined by measuring the time difference of the target buoy node signal arriving at different floating particles.

[0103] Specifically, in step S5, first, a signal propagation speed model is constructed. The specific steps are as follows:

[0104] In step S511, select a floating particle among several floating particles and define it as the target buoy node. The formula for calculating the time difference between other several floating particles and the target buoy node is

[0105]

[0106] In the formula, is the time difference, is the time of other floating particles, is the time of the target buoy node, N is the total number of floating particles;

[0107] The constructed signal propagation model in step S512 is

[0108]

[0109] In the formula, is the distance difference corresponding to the time difference, is the propagation speed, is the time difference.

[0110] Continue to determine the position of the target buoy node. The specific steps are as follows:

[0111] In step S521, the time difference between every two floating particles corresponds to a hyperboloid equation. Taking the target buoy node and a certain other floating particle as an example, the hyperboloid equation corresponding to the time difference is constructed as follows

[0112]

[0113] In the formula, are all obtained through the predicted coordinates of a single floating particle output in step S35;

[0114] For N floating particles (numbered from 1 to N), multiple double-sided curves can be formed. The above description is the general form. If N ≥4, N floating particles can form multiple hyperboloid equations. Therefore, an overdetermined system of equations is established:

[0115] ;

[0116] Step S522, process noise and environmental errors using the least squares method and neural network method, that is, optimize the results of the constructed hyperbolic equation to determine the target float node position. The optimization formula is:

[0117]

[0118] In the formula, , M is the total number of equations.

[0119] Next is the more critical technical detail of this application. To enhance the positioning accuracy and reliability, an optimization strategy needs to be added to the provided TDOA algorithm, that is, step S6. Use the environmental data such as temperature, humidity, wind speed, and sea current collected by the environmental sensor to dynamically adjust the propagation speed in the signal propagation model established in step S5 using the neural network, and correct the error caused by environmental changes. This design is an environmental adaptive calibration mechanism. By collecting data such as wind speed, sea current, and temperature in real time, the positioning algorithm is dynamically adjusted to eliminate the interference of environmental changes on signal propagation.

[0120] The specific steps are as follows:

[0121] Step S61, construct a signal propagation speed model, and the correction formula is

[0122]

[0123] In the formula, is the propagation speed correction value, is the standard propagation speed, E is the environmental factor, and the environmental factor is temperature or humidity or wind speed, is a non-linear function for environmental correction, that is, a neural network training model; the standard propagation speed here is assumed to be the known speed (such as the wireless signal speed is close to the speed of light, and the sound wave speed is about 1500 m / s) propagating in a uniform environment.

[0124] Construct a multi-layer feedforward neural network (MLP), with the environmental factor as the input and output the propagation speed correction value .

[0125] Step S62, evaluate the accuracy of the prediction of the correction formula, and adjust the parameters of the neural network accordingly to make its prediction closer to the true value. Among them, the evaluation formula is

[0126]

[0127] In the formula, is the propagation speed correction value predicted by the correction formula, is the true propagation speed correction value obtained by offline data collection, and ζ is the loss function of the correction formula.

[0128] After training is completed, the model is deployed to the floating particles, and the propagation speed correction result can be inferred in real time, improving the efficiency and accuracy of calibration.

[0129] Step S7: Construct an optimized single floating particle trajectory prediction model based on the signal propagation model corrected by the propagation speed correction formula in step S6 and the target floating buoy node position determined in step S5.

[0130] This step is actually a data fusion process. Combining RSSI (Received Signal Strength Indicator) and multi-modal sensor data, the TDOA calculation result is further optimized through a deep neural network model. The specific steps are as follows:

[0131] Step S71: Use a convolutional neural network (CNN) to perform multi-dimensional feature extraction on the time difference matrix and signal matrix generated when determining the target floating buoy node position, and fuse the extracted features to output an accurate position estimate value. The fusion formula is

[0132]

[0133] In the formula, is the predicted position vector, T is the time difference matrix, R is the signal matrix;

[0134] The network model uses a supervised learning method and is trained based on a dataset with position labels, which can effectively reduce the interference of environmental noise and sensor errors on the positioning result. Step S72: Evaluate the gap between the predicted position and the true position, adjust the parameters of the single floating particle trajectory prediction model, and finally accurately locate the position of the floating particle. Among them, the evaluation formula is

[0135]

[0136] In the formula, is the loss function of the evaluation formula, is the predicted position vector, is the known true position, is the Euclidean distance. Finally, the predicted and relatively accurate position of the floating particle is output.

[0137] In the above description of the positioning method, when it is necessary to separate the characteristics of different floating particles from the mixed signals, some floating particles float relatively far away, and their signals cannot be directly transmitted to the control center. Therefore, it is necessary to rely on the floating particles closer to themselves to forward the signals through the particle signal forwarding module. However, there are various signal forwarding methods, which need to be distinguished and determined. Therefore, this application further provides a distributed computing method. The floating node runs a lightweight neural network (such as the CNN model mentioned above) to preprocess the data and reduce the amount of data transmission.

[0138] Add a dynamic routing selection algorithm to the floating particles, and use a reinforcement learning model (such as DQN) to dynamically optimize the signal forwarding path, with the goal of minimizing communication delay and energy consumption:

[0139]

[0140] In the formula, and are weight parameters.

[0141] Adopt the Federated Learning framework to jointly train the neural network models of multiple floating particles. Each floating particle updates the model weights locally to improve the model generalization ability on the premise of ensuring data privacy:

[0142]

[0143] In the formula, is the weight parameter of the

[0144] Figure 3 As shown, assume floating node A, floating node B, and floating node C, which are relatively far away from the selected positioning particles. Therefore, after implementing the above distributed computing method, a suitable link is successfully selected to form a signal transmission with the positioning particles.

[0145] Through the above optimization, this application has achieved significant improvements in environmental adaptability, complex target processing ability, and computing efficiency, and is particularly suitable for high-precision TDOA positioning scenarios in the marine environment.

[0146] In summary, the floating particles provided in this application support long-term monitoring through LoRa communication and solar power supply, achieving the purpose of low power consumption. The provided high-precision positioning method for multiple floating particles meets the high-precision requirements through multi-float collaboration and dynamic calibration. The distributed network and efficient algorithm meet the real-time positioning needs of floating particles. The environment adaptive calibration mechanism overcomes the interference of complex marine environments and has strong adaptability. At the same time, the floating particles achieve dynamic relay communication through the signal forwarding function, expanding the signal coverage range. Even in long-distance or complex marine environments, the signal can still be effectively transmitted to the float nodes or Internet access devices.

[0147] Those skilled in the art of this technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used here have the same meaning as the general understanding of those of ordinary skill in the field to which this application belongs. It should also be understood that terms defined in general dictionaries should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as here.

[0148] The meaning of "and / or" described in this application refers to the situation where each exists alone or both exist simultaneously.

[0149] The meaning of "connection" described in this application can be a direct connection between components or an indirect connection between components through other components.

[0150] Taking the above ideal embodiments of the present invention as an inspiration, through the above description, relevant staff can make various changes and modifications completely within the scope without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A high-precision positioning method for multiple floating particles, characterized in that: Specifically, it includes the following steps: Step S1: Put a number of floating particles into the ocean. The floating particles include a signal acquisition module, an environmental monitoring module, a positioning module, a communication module, a particle signal forwarding module, and an energy management module. All the foregoing modules are integrated inside a housing. The signal acquisition module includes an optical sensor, an ultrasonic sensor, and an electromagnetic sensor, which are used to detect the signals of the floating particles and transmit the signals to a central processing unit deployed at a coastal station or on a ship. The environmental monitoring module includes an anemometer, a turbidity sensor, an acceleration sensor, and a temperature sensor, which are used to collect environmental data in real time to assist in positioning error correction. The positioning module is used to measure the distance or positional relationship between floating particles. The communication module is based on the LoRa communication protocol for long-distance data transmission. The particle signal forwarding module expands the signal coverage range of the floating particles through an inter-forwarding mechanism. The energy management module includes an integrated solar or wave energy power generation system. Step S2: Activate the floating particles. The signal acquisition module sends the signals of the floating particles and the collected environmental data to the central processing unit through the communication module. Step S3: Construct a multi-target processing model to separate the characteristics of different floating particles from the signals of a number of floating particles, and at the same time assign an identifier to each floating particle, and output the predicted coordinates of a single floating particle. Step S4: Based on the predicted coordinates obtained in Step S3, narrow down the target position range of the floating particles. Step S5: Within the target position range, select a floating particle as the target buoy node, establish a signal propagation model for a single floating particle through the time difference of the signals of the target buoy node reaching different floating particles, and determine the position of the target buoy node. Step S6: Based on the environmental data collected in Step S2, use a neural network to dynamically adjust the propagation speed in the signal propagation model established in Step S5 to correct the error caused by environmental changes. Step S7: Obtain a corrected signal propagation model with the propagation speed corrected in Step S6, and at the same time combine the position of the target buoy node determined in Step S5 to construct an optimized single floating particle trajectory prediction model.

2. The high-precision positioning method for multiple floating particles according to claim 1, wherein: In Step S3, the steps of constructing a multi-target processing model to output the predicted coordinates of a single floating particle are specifically as follows: Step S31: Construct the model architecture of the multi-target processing model, including an input layer, a feature extraction layer, and a fusion classification layer. Step S32, the input layer converts the original signal x ( t ) into a frequency spectrum through fast Fourier transform X ( F ). The feature extraction layer extracts the frequency feature vector and the amplitude feature vector respectively through a dual-branch convolutional network. The fusion classification layer concatenates the dual-branch features and outputs the classification result through a fully connected layer. The classification result output formula is ， In the formula, y is the classification result, is the frequency feature vector, is the amplitude feature vector, W is the weight, b is the bias; Step S33: Treat each floating particle as a target and define it as k , and generate a unique identifier for each single floating particle according to the classification result ID k , that is ， In the formula, ID k is an identifier, F k is the target k frequency, A k is the target k amplitude, is the target k timestamp of the first occurrence, Hash The function ensures uniqueness; Step S34: Use a recurrent neural network to process the normalized frequency and amplitude to obtain the hidden time series data at the current moment. The hidden state update formula is ， In the formula, h t is the target k hidden state at the current moment, h t-1 is the target k hidden state at the previous moment, x t is the target k input at the current moment, i.e., the normalized frequency F t and amplitude A t , W h is the weight at the current moment, b h is the bias at the current moment, σ is the Sigmoid activation function; Step S35: Output the predicted coordinates of a single floating particle. The prediction formula is ， In the formula, is the predicted coordinates including the next moment in the output ( x ′, y ′, z ′), W y is the weight including the next moment, h t is the target k the hidden state at the current moment, b y is the bias including the next moment.

3. The high-precision positioning method for multiple floating particles according to claim 2, wherein: In Step S5, the steps of constructing the signal propagation speed model are specifically as follows: Step S511: Select a floating particle from a number of floating particles and define it as the target buoy node. The formula for calculating the time difference between other floating particles and the target buoy node is ， In the formula, is the time difference, is the time of other floating particles, is the time of the target float node, N is the total number of floating particles; The constructed signal propagation model is ， In the formula, is the distance difference corresponding to the time difference, is the propagation speed, is the time difference.

4. The high-precision positioning method for multiple floating particles according to claim 3, wherein: In Step S5, the steps of determining the position of the target buoy node are specifically as follows: Step S521, construct a hyperbolic equation corresponding to the time difference between every two floating particles, specifically: ， In the formula, ( x i , y i , z i ), ( x j , y j , z j ) are all obtained from the predicted coordinates of a single floating particle output in step S35; If N ≥ 4, N floating particles form multiple hyperboloid equations to establish an overdetermined system of equations: ， Step S522, optimize the result of the constructed hyperbolic equation to determine the position of the target float node. The optimization formula is: ， In the formula, , M is the total number of equations.

5. The high-precision positioning method for multiple floating particles according to claim 4, wherein: In step S6, the specific steps for dynamically adjusting the propagation speed in the signal propagation model are: Step S61, construct a signal propagation speed model. The correction formula is ， In the formula, is the propagation speed correction value, is the standard propagation speed, E is the environmental factor, and the environmental factor is temperature or humidity or wind speed, is the non - linear function for environmental correction; Step S62, evaluate the accuracy of the prediction of the correction formula, adjust the parameters of the correction formula through the evaluation results, take the environmental factor as the input, and finally output the propagation speed correction value. Among them, the evaluation formula is ， In the formula, is the corrected value of the propagation speed predicted by the correction formula, is the true corrected value of the propagation speed obtained by offline data collection, and ζ is the loss function of the correction formula.

6. The high-precision positioning method for multiple floating particles according to claim 5, characterized in that: In step S7, the specific steps for constructing an optimized single floating particle trajectory prediction model are: Step S71, use a convolutional neural network to perform multi-dimensional feature extraction on the time difference matrix and the signal matrix generated when determining the position of the target float node, and fuse the extracted features to output an accurate position estimate value. The fusion formula is ， In the formula, is the predicted position vector, T is the time difference matrix, R is the signal matrix; Step S72, evaluate the gap between the predicted position and the actual position, adjust the parameters of the single floating particle trajectory prediction model, and finally accurately locate the position of the floating particle. Among them, the evaluation formula is ， In the formula, is the loss function for evaluating the formula, is the predicted position vector, is the known true position, is the Euclidean distance.

Images