Tightened mooring buoy end internal wave identification method, system, medium and equipment

Abstract

The invention belongs to the field of ocean monitoring, and provides a method, a system, a medium and equipment for identifying internal waves at a buoy end of a tightened anchor system, and the technical scheme is that a constructed internal wave identification model is trained based on obtained underwater temperature chain historical data and buoy attitude information at corresponding moments; wherein the construction process of the internal wave recognition model comprises the following steps: adding a sampling stride in a convolution layer, and calculating each convolution kernel calculation result corresponding to each piece of temperature chain historical data; correcting the attitude change of the buoy by combining each convolution kernel calculation result corresponding to the temperature chain historical data and the buoy attitude information at the corresponding moment, and performing classification based on the corrected buoy end internal wave features to obtain a classification result; pruning and quantifying the trained internal wave recognition model to obtain a lightweight internal wave recognition model; after the lightweight internal wave recognition model is deployed, the temperature chain data collected in real time is processed, and an internal wave recognition result is obtained. And the calculation power consumption is reduced while the recognition precision is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of ocean monitoring, and in particular relates to a method and system for identifying internal waves at the end of a tautly anchored buoy. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of ocean monitoring, accurate internal wave identification based on buoys is crucial, playing a key role in a variety of applications, including marine environmental monitoring, marine engineering, and marine safety. Deploying an internal wave identification system on buoys not only reduces data transmission latency and reliance on remote servers, but also improves data security and real-time performance, providing more accurate and reliable technical support for marine research and applications.

[0004] Currently, the most commonly used internal wave recognition methods at the edge use threshold methods, but they are limited by rule settings and are difficult to adapt to the complex and changing ocean environment, resulting in insufficient recognition accuracy. Although the introduction of artificial intelligence methods can improve recognition accuracy, the high power consumption problem becomes a major challenge for buoy-side applications due to the large amount of computation required for neural network inference. At the same time, fixed buoys, represented by taut anchor buoys, have anchor cables under high tension. Compared with loose anchor buoys, their attitude changes under flow field disturbances are smaller, which can improve measurement stability to a certain extent. However, even taut anchor buoys may still experience slight tilt, vibration, or attitude drift when affected by external factors such as ocean currents, wind and waves. These changes can cause the measurement depth and angle of the temperature chain (temperature profile chain) sensor to shift, thereby affecting the accuracy of internal wave recognition. Summary of the Invention

[0005] In order to solve at least one of the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for identifying internal waves at the end of a tautly anchored buoy, the purpose of which is to reduce computing power consumption while ensuring identification accuracy.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A first aspect of the present invention provides a method for identifying internal waves at the end of a tautly anchored buoy, comprising the following steps: Obtain underwater temperature chain historical data and buoy attitude information at the corresponding moment; The internal wave recognition model is trained based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time. The construction process of the internal wave recognition model includes: In the convolution layer, the sampling stride is increased to calculate the calculation results of each convolution kernel corresponding to each temperature chain historical data; Combining the calculation results of each convolution kernel corresponding to the historical data of the temperature chain and the buoy attitude information at the corresponding moment, the buoy attitude change is corrected to obtain the corrected internal wave characteristics of the buoy end. Classification is performed based on the corrected internal wave characteristics of the buoy end to obtain the classification result; The trained internal wave recognition model is pruned and quantized to obtain a lightweight internal wave recognition model; A lightweight internal wave recognition model is deployed, and the real-time collected temperature chain data is processed based on the deployed internal wave recognition model to obtain the internal wave recognition results.

[0007] Furthermore, the calculation formula for each convolution kernel corresponding to each temperature chain historical data is: , in, Indicates the The original data corresponding to Strip convolution output, Representative The original data corresponding to convolution kernel, Represents the number of bits of convolution output, Represents the sampling stride of the convolution operation, Represents the length calculation formula, Represents the weight index of the convolution kernel.

[0008] Furthermore, when correcting the buoy's attitude change by combining the calculation results of each convolution kernel corresponding to the temperature chain historical data and the buoy attitude information at the corresponding moment, a fixed time window is defined for each feature data point. The attitude data in the window will be corrected according to the weighted strategy, and the weighted average of the attitude data in the window will be calculated to correct the feature data point.

[0009] Furthermore, classification is performed based on the corrected internal wave characteristics at the buoy end to obtain the following classification results, including: The corrected internal wave characteristics of the buoy end are processed by a multi-layer perceptron to generate an attitude guidance vector; Multiply the corrected internal wave feature of the buoy end and the attitude guidance vector element by element, and then calculate the attention weight through the Sigmoid activation function; The modified buoy end internal wave features and attention weights are weighted element by element to obtain a weighted feature vector; The weighted feature vector is input into the hidden fully connected layer and the final classification probability is output through the Softmax function.

[0010] Furthermore, the trained internal wave recognition model is pruned and quantized to obtain a lightweight internal wave recognition model, including: Prune the connections with small contribution weight to the network output, use the L1 norm-based pruning method to perform sparse analysis on the weights, calculate the L1 norm of each weight, and prune the weights below the set threshold; The input underwater temperature chain data and internal wave recognition model activation value and the internal wave recognition model weight value are quantized, and the activation function input value is dequantized; the dequantized internal wave recognition model is activated to obtain a lightweight internal wave recognition model.

[0011] Furthermore, the quantization formula of the weight value of the internal wave recognition model is: , in, Indicates the quantization bit width, represents the scaling factor, Represents the quantized integer weight value, represents the original weight value, Indicates the minimum value of the weight, Indicates the maximum value of the weight.

[0012] Furthermore, the lightweight internal wave recognition model is deployed, and the real-time collected temperature chain data is processed based on the deployed internal wave recognition model, including: The producer task collects underwater temperature chain data by communicating with the buoy controller, and writes the data into the shared queue after integrity verification and preprocessing. The producer task runs according to a fixed sampling period, and the response order of data processing is determined according to the priority level of the consumer task; When the queue is full, an interrupt signal is triggered by the consumer. A second aspect of the present invention provides a device for identifying internal waves at the end of a taut anchored buoy, comprising: Data acquisition module, which is used to obtain historical data of underwater temperature chain and buoy attitude information at the corresponding moment; The internal wave recognition model training module is used to train the constructed internal wave recognition model based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time. The construction process of the internal wave recognition model includes: In the convolution layer, the sampling stride is increased to calculate the calculation results of each convolution kernel corresponding to each temperature chain historical data; Combining the calculation results of each convolution kernel corresponding to the historical data of the temperature chain and the buoy attitude information at the corresponding moment, the buoy attitude change is corrected to obtain the corrected internal wave characteristics of the buoy end. Classification is performed based on the corrected internal wave characteristics of the buoy end to obtain the classification result; A lightweight module, which is used to prune and quantize the trained internal wave recognition model to obtain a lightweight internal wave recognition model; The model deployment module is used to deploy the lightweight internal wave recognition model, process the real-time collected temperature chain data based on the deployed internal wave recognition model, and obtain the internal wave recognition results.

[0013] A third aspect of the present invention provides a computer-readable storage medium.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying internal waves at the end of a tautly anchored buoy as described above.

[0015] A fourth aspect of the present invention provides a computer device.

[0016] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for identifying internal waves at the end of a tautly anchored buoy as described above are implemented.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. The proposed internal wave identification method for a tautly anchored buoy incorporates an attitude correction module into the identification model to reduce the impact of buoy attitude changes on temperature chain measurement data. Through model quantization, pruning, and hardware optimization, the neural network is compressed and accelerated, enabling efficient operation on low-power single-chip microcomputers. This addresses the issues of insufficient accuracy and excessive power consumption associated with traditional technologies. The use of a neural network-based artificial intelligence approach for internal wave identification significantly improves accuracy and reduces false positives compared to traditional threshold methods, particularly in complex marine environments.

[0018] 2. The feature classification module of the present invention adaptively adjusts the importance distribution of features by introducing an attention mechanism based on posture auxiliary vectors. Unlike the static correction of internal wave features at the buoy end in the feature extraction module, the feature classification module dynamically guides the feature attention direction during the classification stage, further refines the compensation for posture influence, and significantly improves the stability of the system in complex sea conditions.

[0019] 3. This invention reduces power consumption during neural network inference through model quantization, pruning, and hardware optimization, making it suitable for low-power embedded platforms. Combined with a real-time operating system (RTOS) to optimize task management, it improves system efficiency and real-time data processing capabilities, effectively reducing overall power consumption.

[0020] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0022] Figure 1 This is a flow chart of a method for identifying internal waves at the end of a tautly anchored buoy provided by an embodiment of the present invention; Figure 2 is a flow chart of a feature extraction module provided by an embodiment of the present invention; Figure 3 This is a flow chart of a feature classification module provided by an embodiment of the present invention; Figure 4 This is a flowchart of model training, pruning, and quantization provided by an embodiment of the present invention; Figure 5 This is a flow chart of the producer-consumer model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0025] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0026] As mentioned in the background, currently, the most commonly used internal wave recognition methods at the edge rely on threshold methods. However, these methods are limited by set rules and are difficult to adapt to the complex and changing ocean environment, resulting in insufficient recognition accuracy. While the introduction of artificial intelligence methods can improve recognition accuracy, high power consumption, due to the large computational complexity of neural network inference, poses a major challenge for buoy-side applications. Furthermore, fixed buoys, such as tautly anchored buoys, have high anchor tension. Compared to slackly anchored buoys, their attitude changes under flow disturbances are smaller, which can improve measurement stability to a certain extent. However, even tautly anchored buoys can still experience slight tilt, vibration, or attitude drift when affected by external factors such as ocean currents, wind, and waves. These changes can cause the measurement depth and angle of the temperature chain (temperature profile chain) sensor to shift, thereby affecting the accuracy of internal wave recognition.

[0027] To make the system suitable for buoys with limited computing resources, the present invention achieves compression and acceleration of the neural network through model quantization, pruning, and hardware optimization, enabling it to run efficiently on a low-power single-chip microcomputer. Furthermore, the system integrates a real-time operating system (RTOS) to optimize task management, improve computing efficiency, and further reduce power consumption. The entire system, as an independent module, interacts directly with the buoy controller to achieve high-precision, low-power internal wave recognition capabilities and enhance the system's adaptability in complex sea conditions. At the same time, an attitude correction module is introduced into the recognition model to reduce the impact of buoy attitude changes on the temperature chain measurement data, thereby improving data stability and recognition accuracy.

[0028] Example 1 This embodiment provides a method for identifying internal waves at the end of a tautly anchored buoy, comprising the following steps: Step 1: Obtain underwater temperature chain historical data and buoy attitude information at the corresponding time; In this embodiment, the historical data of the lower temperature chain and the buoy posture information at the corresponding moment are obtained at the buoy end; according to the changing law of the vertical distribution of water temperature, the underwater temperature sequence of 14 layers of sensors and 30 sampling time points is finally selected as the input of the feature extraction layer.

[0029] Since the change in ocean temperature is usually a slow and continuous process, especially between adjacent time points, the temperature gradient change trend is highly consistent, so the temperature data has significant correlation in the time dimension. The actual calculation of the Pearson Correlation Coefficient between adjacent time points is: , in, and Indicates the t time points and t +1 temperature vector of each depth layer at a time point, is the covariance function, and For the t time points and t +1 temperature standard deviation at a time point; The results show that at most time points, the temperature distributions between adjacent moments have a strong correlation.

[0030] Step 2: Train the constructed internal wave recognition model based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time; The construction process of the internal wave identification model includes the following steps: Step 201: extracting internal wave features at the buoy end based on the constructed feature extraction model; like Figure 2 As shown in the figure, the feature extraction model uses a convolutional neural network (CNN), which includes an input layer, a convolution kernel, and a convolution output. The specific feature extraction process includes the following steps: Step 2011: input ocean temperature profile data with strong correlation between adjacent moments into the input layer; Step 2012: In the convolutional layer, increase the sampling stride , calculate the calculation results of each convolution kernel corresponding to each data; The input ocean temperature profile data has a strong correlation between adjacent moments, which indicates that the temperature change trends of consecutive depth layers or time points are similar. Therefore, the sampling stride is appropriately increased in the convolution operation. The impact on algorithm performance is small.

[0031] Specifically, no. The original data and its corresponding The strip convolution kernel convolution calculation formula is as follows: (1), in, Indicates the The original data corresponding to Strip convolution output, Representative The original data corresponding to convolution kernel, Represents the number of bits of convolution output, Represents the sampling stride of the convolution operation, Represents the length calculation formula, Represents the weight index of the convolution kernel.

[0032] According to formula (1), the input feature number can be obtained , output feature number and sampling stride Relationship: (2), This suggests increasing stride length It will reduce the number of output features However, due to the strong correlation between the data at adjacent moments, it has little effect on the feature extraction effect.

[0033] Step 2013: Output the extracted data in the convolutional output layer. Internal wave characteristics at each buoy end.

[0034] Step 202: Based on the acquired IMU sensor data on the buoy and the extracted internal wave characteristics at the buoy end, the buoy attitude change is corrected to obtain a corrected internal wave characteristic at the buoy end, and classification is performed based on the corrected internal wave characteristic at the buoy end to obtain a classification result; In order to improve the internal wave recognition capability of the taut moored buoy under disturbances such as ocean currents and wind waves, and to process the influence of the buoy attitude change on the underwater temperature chain sensor data, this embodiment proposes an improved attitude correction method to ensure the accuracy and reliability of the water temperature data. After the convolution feature extraction, the internal wave feature at the buoy end does not take into account the deviation of the temperature data caused by the buoy attitude change (such as tilt, yaw, etc.). An attitude correction module has been added to correct the deviation of temperature data caused by changes in the buoy's attitude (such as tilt, yaw, etc.).

[0035] The specific steps include: Step 2021: Perform weighted correction on the attitude data using a time window to obtain a corrected internal wave characteristic of the buoy end; The input of the attitude correction module includes the IMU sensor data on the buoy (pitch angle , roll angle , heading angle ), and features derived from convolutional feature extraction of the raw underwater temperature chain data. Because the buoy's attitude data is sampled at a high frequency, and the temperature feature data may be reduced during extraction, a time window is used to weight the attitude data. Specifically, a fixed time window is defined for each feature data point. The attitude data within the window is corrected according to a specific weighting strategy, and the weighted average of the attitude data within the window is calculated to correct the feature data point.

[0036] Specifically, the weighted correction method is used to correct the features according to the change of the buoy's posture. The calculation formula of the corrected features is: (3), in, is the corrected internal wave characteristic of the buoy end, is the correction coefficient, which indicates the influence of the buoy attitude change on the characteristics. They are the pitch angle, roll angle and heading angle at the corresponding moment of the output feature respectively. The features are adjusted by the weighted effect of the attitude data to correct the influence of the attitude change on the feature value.

[0037] Step 2022: obtaining a classification result based on the corrected internal wave characteristics of the buoy end and the constructed characteristic classification module; like Figure 3 As shown in Figure 3, the feature classification module consists of a two-layer fully connected neural network (FCNN) and a pose-guided attention module.

[0038] First, the feature classification input layer receives the internal wave features of the buoy end after being processed by the feature extraction module and the attitude correction is completed. The shape is a vector of 1×40 ; The features are then fed into a pose-guided attention module, where the raw pose angles are firstly centered by a small multi-layer perceptron (MLP). Perform dimensionality increase operation to generate a 1×40 attitude guidance vector , this process can be expressed as: , Then, the feature vector F is multiplied element-wise with the posture guidance vector g (Hadamard product), and the attention weight w is calculated through the Sigmoid activation function. The formula is as follows: , Among them, σ(⋅) represents the Sigmoid function and ⊙ represents the element-wise product operation.

[0039] Next, the input feature vector F is weighted element by element with the attention weight w to obtain the weighted feature vector : , The weighted eigenvector It is input into a hidden fully connected layer (FCNN) of size 1×32 to further extract deep features. Subsequently, the features pass through an output fully connected layer of size 1×2 and output the final classification probability through the Softmax function: , in, , , and b1, b2 are the weight matrices and bias terms of the hidden layer and output layer respectively.

[0040] The feature classification module adaptively adjusts the importance distribution of features by introducing an attention mechanism based on attitude auxiliary vectors. Specifically, attitude information is fused with feature extraction results to calculate attention weights, which are then weighted and corrected to the original features. This approach enables the model to highlight more discriminative features under attitude changes, improving classification accuracy and robustness. Unlike the static correction of internal wave features at the buoy end in the feature extraction module, the feature classification module dynamically guides the feature attention direction during the classification phase, further refining the compensation for attitude effects. The combination of the two enables a transition from static compensation to dynamic adaptation, significantly improving the stability of the system in complex sea conditions.

[0041] A fully connected neural network itself does not have independent feature extraction capabilities. Its main function is to achieve feature classification by nonlinearly combining input features. This model achieves classification by adding a Softmax layer to the output layer to calculate the probability of belonging to each category.

[0042] The specific form of Softmax is: (4), in, represents the output of the neural network, Represents two nodes of the output layer. Classification network output for The matrix, are the probabilities of identifying it as not an internal wave and identifying it as an internal wave, respectively.

[0043] The final classification result is obtained by setting the threshold The specific rules are as follows: (5), Based on underwater temperature chain data, a neural network model is used to identify internal waves. At the same time, the deviation of the extracted feature values caused by changes in the buoy's posture (such as tilt, yaw, etc.) is corrected, thereby improving the accuracy of the final internal wave detection.

[0044] Step 3: Prune and quantize the trained internal wave recognition model to obtain a lightweight internal wave recognition model; like Figure 4 As shown in the figure, to meet the needs of the MCU platform, the trained model is pruned and quantized. By removing redundant parts and reducing the parameter bit width, the model's resource usage and computing efficiency are greatly optimized, enabling it to run efficiently on embedded platforms such as MCUs while maintaining a certain level of inference accuracy. Step 301: Prune connections with small network output contribution weights. During the pruning process, we use an L1-norm-based pruning method to perform a sparse analysis on weights that contribute less to the network output. By reducing connections with smaller weights, we make the neural network structure more sparse while effectively maintaining model accuracy.

[0045] Specifically, L1 norm pruning calculates the L1 norm of each weight and prunes weights below a set threshold, thereby removing redundant connections, optimizing the model structure, and improving operational efficiency. The L1 norm formula is as follows: (6), in, Represents the weight matrix The L1 norm of , which represents the sum of the absolute values of all weights, Represents the weight matrix Middle A weight value.

[0046] Step 302: quantify the input value and weight value of the model; During the quantization process, due to the limited resources and memory of the microcontroller platform, directly using 32-bit floating-point numbers for calculations and storage would significantly increase time and energy consumption, and might even make the operation impossible. This contradicts the original intention of efficient and low-power processing on the float side. Therefore, the quantization process of this system for the microcontroller platform is essentially the process of quantizing 32-bit floating-point numbers into 8-bit integers. By operating the data in 8-bit integer form, storage space can be reduced by approximately 75% and inference speed is significantly improved.

[0047] The quantization of the buoy-side internal wave recognition model involves quantifying the input underwater temperature chain data, model activation values, and model weights. This design effectively balances the accuracy and efficiency requirements of the model.

[0048] The quantization formula is as follows: Represents 8-bit width, , respectively represent real-time temperature chain data, activation value and model weight value: (7), in, Indicates the quantization bit width (8 in this embodiment, indicating that the quantization target is an 8-bit integer), represents the scaling factor, Represents the quantized integer weight value, represents the original weight value, Indicates the minimum value of the weight, Indicates the maximum value of the weight.

[0049] Step 303: Dequantize the activation function input value; Specifically, before the model is activated, the input to the activation function is a quantized integer value, which lacks direct mathematical meaning. To ensure the activation function correctly performs nonlinear transformations, the quantized value is briefly dequantized, converted to floating-point, and then the activation function is applied, ultimately resulting in a floating-point output. This process is essential for achieving a trade-off between accuracy and efficiency on microcontroller platforms.

[0050] The dequantization formula of the activation function input value is as follows: (8), in, Represents a dequantized floating-point value.

[0051] Through model quantization, pruning, and hardware optimization, power consumption during neural network inference is reduced, making it suitable for low-power embedded platforms. Combined with a real-time operating system (RTOS) to optimize task management, this improves system efficiency and real-time data processing capabilities, effectively reducing overall power consumption.

[0052] Step 4: Deploy the lightweight internal wave recognition model, and process the real-time collected temperature chain data based on the deployed internal wave recognition model to obtain the internal wave recognition results; like Figure 5 As shown in the figure, this system is based on the buoy end internal wave recognition of the single chip microcomputer, combined with the real-time operating system (RTOS) and the producer-consumer model to build an efficient real-time data processing process; The specific steps include: Step 401: The producer task collects underwater temperature chain data by communicating with the buoy controller, and writes the data into the shared queue after integrity verification and pre-processing. Step 402: The producer task runs according to a fixed sampling period and determines the response order of data processing according to the priority level of the consumer task; To improve real-time performance, the producer task runs according to a fixed sampling period, while the consumer task has a higher priority and can respond and process data immediately after the queue full interrupt is triggered.

[0053] Step 403: When the queue data is full, an interrupt signal is triggered, and the consumer task reads the data from the queue and inputs the internal wave recognition model for inference, and then feeds the recognition results back to the buoy controller for storage and forwarding.

[0054] The producer-consumer model and queue interrupt mechanism are used to optimize data transmission and processing processes, improve the stability and efficiency of data processing, quickly feedback recognition results, and achieve high-precision, low-power internal wave recognition.

[0055] Example 2 This embodiment provides a system for identifying internal waves at the end of a taut anchored buoy, comprising: Data acquisition module, which is used to obtain historical data of underwater temperature chain and buoy attitude information at the corresponding moment; The internal wave recognition model training module is used to train the constructed internal wave recognition model based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time. The construction process of the internal wave recognition model includes: In the convolution layer, the sampling stride is increased to calculate the calculation results of each convolution kernel corresponding to each temperature chain historical data; Combining the calculation results of each convolution kernel corresponding to the historical data of the temperature chain and the buoy attitude information at the corresponding moment, the buoy attitude change is corrected to obtain the corrected internal wave characteristics of the buoy end. Classification is performed based on the corrected internal wave characteristics of the buoy end to obtain the classification result; A lightweight module, which is used to prune and quantize the trained internal wave recognition model to obtain a lightweight internal wave recognition model; The model deployment module is used to deploy the lightweight internal wave recognition model, process the real-time collected temperature chain data based on the deployed internal wave recognition model, and obtain the internal wave recognition results.

[0056] It should be noted that the specific implementation method of the internal wave identification system at the end of a tautly anchored buoy in an embodiment of the present invention is similar to the specific implementation method of the internal wave identification method at the end of a tautly anchored buoy in an embodiment of the present invention. Please refer to the description of the method part for details. In order to reduce redundancy, it will not be repeated here.

[0057] Example 3 This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method for identifying internal waves at the end of a tautly anchored buoy as described above are implemented.

[0058] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for identifying internal waves at the end of a tautly anchored buoy as described above are implemented.

[0059] Example 5 This embodiment provides a program product, which is a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method for identifying internal waves at the end of a tautly anchored buoy as described above are implemented.

[0060] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for identifying internal waves at the end of a taut anchored buoy, characterized in that: include: Obtain underwater temperature chain historical data and buoy attitude information at the corresponding moment; The internal wave recognition model is trained based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time. The construction process of the internal wave recognition model includes: In the convolution layer, the sampling stride is increased to calculate the calculation results of each convolution kernel corresponding to each temperature chain historical data; Combining the calculation results of each convolution kernel corresponding to the historical data of the temperature chain and the buoy attitude information at the corresponding moment, the buoy attitude change is corrected to obtain the corrected internal wave characteristics of the buoy end. Classification is performed based on the corrected internal wave characteristics of the buoy end to obtain the classification result; The trained internal wave recognition model is pruned and quantized to obtain a lightweight internal wave recognition model; A lightweight internal wave recognition model is deployed, and the real-time collected temperature chain data is processed based on the deployed internal wave recognition model to obtain the internal wave recognition results.

2. A method for identifying internal waves at the end of a taut anchored buoy according to claim 1, characterized in that: The calculation formula for each convolution kernel corresponding to each temperature chain historical data is: ， in, Indicates the The original data corresponding to Strip convolution output, Representative The original data corresponding to convolution kernel, Represents the number of bits of the convolution output, Represents the sampling stride of the convolution operation, Represents the length calculation formula, Represents the weight index of the convolution kernel.

3. The method for identifying internal waves at the end of a taut anchored buoy according to claim 1, wherein: When correcting the buoy's attitude change by combining the calculation results of each convolution kernel corresponding to the temperature chain historical data and the buoy attitude information at the corresponding moment, a fixed time window is defined for each feature data point. The attitude data in the window will be corrected according to the weighted strategy, and the weighted average of the attitude data in the window will be calculated to correct the feature data point.

4. The method for identifying internal waves at the end of a taut anchored buoy according to claim 1, wherein: The classification results are obtained based on the corrected internal wave characteristics of the buoy end, including: The corrected internal wave characteristics of the buoy end are processed by a multi-layer perceptron to generate an attitude guidance vector; Multiply the corrected internal wave feature of the buoy end and the attitude guidance vector element by element, and then calculate the attention weight through the Sigmoid activation function; The modified buoy end internal wave features and attention weights are weighted element by element to obtain a weighted feature vector; The weighted feature vector is input into the hidden fully connected layer and the final classification probability is output through the Softmax function.

5. The method for identifying internal waves at the end of a taut anchored buoy according to claim 1, characterized in that: The trained internal wave recognition model is pruned and quantized to obtain a lightweight internal wave recognition model, including: Prune the connections with small contribution weight to the network output, use the L1 norm-based pruning method to perform sparse analysis on the weights, calculate the L1 norm of each weight, and prune the weights below the set threshold; The input underwater temperature chain data and internal wave recognition model activation value and the internal wave recognition model weight value are quantized, and the activation function input value is dequantized; the dequantized internal wave recognition model is activated to obtain a lightweight internal wave recognition model.

6. A method for identifying internal waves at the end of a taut anchored buoy according to claim 2, characterized in that: The quantitative formula of the weight value of the internal wave identification model is: ， in, Indicates the quantization bit width, represents the scaling factor, Represents the quantized integer weight value, represents the original weight value, Indicates the minimum value of the weight, Indicates the maximum value of the weight.

7. The method for identifying internal waves at the end of a taut anchored buoy according to claim 1, characterized in that: The lightweight internal wave recognition model is deployed, and the real-time collected temperature chain data is processed based on the deployed internal wave recognition model, including: The producer task collects underwater temperature chain data by communicating with the buoy controller, and writes the data into the shared queue after integrity verification and preprocessing. The producer task runs according to a fixed sampling period, and the response order of data processing is determined according to the priority level of the consumer task; When the queue data is full, an interrupt signal is triggered. The consumer task reads the data from the queue and inputs it into the internal wave recognition model for inference to obtain the internal wave recognition result.

8. A tensioned anchor buoy end internal wave identification system, characterized in that: include: Data acquisition module, which is used to obtain historical data of underwater temperature chain and buoy attitude information at the corresponding moment; The internal wave recognition model training module is used to train the constructed internal wave recognition model based on the acquired underwater temperature chain historical data and the buoy attitude information at the corresponding time. The construction process of the internal wave recognition model includes: In the convolution layer, the sampling stride is increased to calculate the calculation results of each convolution kernel corresponding to each temperature chain historical data; Combining the calculation results of each convolution kernel corresponding to the historical data of the temperature chain and the buoy attitude information at the corresponding moment, the buoy attitude change is corrected to obtain the corrected internal wave characteristics of the buoy end. Classification is performed based on the corrected internal wave characteristics of the buoy end to obtain the classification result; A lightweight module, which is used to prune and quantize the trained internal wave recognition model to obtain a lightweight internal wave recognition model; The model deployment module is used to deploy the lightweight internal wave recognition model, process the real-time collected temperature chain data based on the deployed internal wave recognition model, and obtain the internal wave recognition results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for identifying internal waves at the end of a tautly anchored buoy are implemented as described in any one of claims 1 to 7.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for identifying internal waves at the end of a tautly anchored buoy are implemented as described in any one of claims 1 to 7.

Images