A sound velocity profile time series prediction method and system based on adaptive multi-scale decomposition

Abstract

The present application relates to the technical field of sound velocity profile time series prediction, and provides a sound velocity profile time series prediction method and system based on adaptive multi-scale decomposition. The sound velocity profile time series prediction method based on adaptive multi-scale decomposition comprises the following steps: determining the optimal VMD decomposition parameter of each depth layer; performing variational mode decomposition on the original sound velocity time series of the depth layer by using the optimal VMD decomposition parameter to obtain K intrinsic mode functions and a residual component; then performing normalization processing and difference feature extraction respectively to obtain K mode components and the first-order difference features corresponding to each mode component; constructing an independent LSTM prediction model for each mode component and the corresponding first-order difference features of each depth layer; performing inverse normalization and summation reconstruction on the prediction results output by the LSTM prediction models to obtain the sound velocity prediction value of the depth layer, and iteratively outputting the complete sound velocity profile layer by layer, thereby improving the prediction accuracy of the sound velocity profile.

Description

Technical Field

[0001] This invention relates to the field of sound velocity profile time series prediction technology, and in particular to a sound velocity profile time series prediction method and system based on adaptive multi-scale decomposition. Background Technology

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

[0003] Ocean sound speed profiles (SSPs) are core environmental parameters for underwater sonar detection, underwater acoustic communication, and high-precision positioning and navigation. Their acquisition primarily involves two methods: in-situ measurement and sound speed inversion. In-situ measurements rely on equipment such as sound speed profilers and temperature-salinity-depth (CTD) profilers, providing high data accuracy, but are limited by observation conditions, are time-consuming, and have limited spatial coverage. Influenced by multiple factors such as solar radiation, tides, seasonal changes, and mesoscale ocean processes, sound speed profiles at fixed geographical locations exhibit multi-scale temporal variations ranging from hourly to seasonal scales.

[0004] In existing technologies, Hierarchical Long Short-Term Memory (H-LSTM) networks retain the individualized features of each deep layer by independently modeling each layer. Subsequent improvements include introducing transfer learning to alleviate overfitting with few samples, and the STNet method based on the Transformer architecture. However, existing methods still have the following shortcomings: First, methods based on recurrent neural networks directly model the original sequence without explicitly decoupling multi-scale temporal fluctuations caused by different physical processes. A single network needs to learn multiple scale patterns simultaneously, making parameter optimization difficult and limiting prediction accuracy. Second, using only the absolute value of sound velocity as input does not fully utilize local rate of change information, especially in abrupt regions such as thermoclines, where the model is not sufficiently aware of the rate of change, easily leading to phase lag or amplitude deviation. Third, although Transformer-like methods can capture long-range dependencies, the computational complexity of self-attention increases with the square of the sequence length, making it prone to overfitting to abrupt points with limited training data. Furthermore, lightweight linear models have limited capacity and are difficult to characterize complex temporal nonlinear dependencies. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a method and system for predicting sound velocity profile time series based on adaptive multi-scale decomposition. Targeting the characteristics of ocean sound velocity profile data, which exhibits strong nonlinearity, non-stationarity, and multi-scale time series features, this invention achieves high-precision sound velocity profile time series prediction through a strategy of layer-by-layer decomposition, independent modeling, and reconstruction.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for predicting the temporal sequence of sound velocity profiles based on adaptive multi-scale decomposition.

[0007] A time-series prediction method for sound velocity profiles based on adaptive multi-scale decomposition includes: The original sound velocity time series of each depth layer is obtained, and the optimal VMD decomposition parameters for each depth layer are determined by adaptive grid search optimization. The original sound velocity time series of this depth layer was subjected to variational mode decomposition using the optimal VMD decomposition parameters, resulting in K eigenmode functions and one residual component. Normalization and differential feature extraction are performed on multiple intrinsic mode functions and one residual component respectively to obtain K mode components and the first-order differential feature corresponding to each mode component; An independent LSTM prediction model is constructed for each modal component and its corresponding first-order difference feature in each depth layer to achieve independent temporal modeling of the layer components; The prediction results output by each LSTM prediction model are inversely normalized and summed to reconstruct the sound velocity prediction value for that depth layer, and the complete sound velocity profile is output iteratively layer by layer.

[0008] Furthermore, the adaptive mesh search optimization determines the optimal VMD decomposition parameters for each depth layer; the method includes: For the number of modes in each depth layer and penalty factor Within the preset range of the number of modes and the preset range of the penalty factor, a two-dimensional parametric mesh is generated according to the respective set step size; For each group of candidate parameters in turn Execution: Mirror the original signal, perform VMD decomposition and trim to the original length, calculate and record the envelope entropy and residual standard deviation for this set of parameters; after traversal, perform Min-Max normalization on all recorded envelope entropy and residual standard deviation values, and calculate the weighted composite score: ;in, and These are the envelope entropy and residual standard deviation after Min-Max normalization, respectively. and The weighting coefficients are used; the set of parameters with the smallest weighted composite score is selected as the optimal VMD parameters for the current depth layer. ,in To determine the optimal number of modes, This is the optimal penalty factor.

[0009] Furthermore, the method involves performing variational mode decomposition on the original sound velocity time series of the depth layer using optimal VMD decomposition parameters to obtain K eigenmode functions and one residual component. The method includes: performing mirror extension processing on the original sound velocity time series of the depth layer; performing VMD decomposition on the extended sequence; and extracting only the middle segment corresponding to the original interval length as the final decomposition result to obtain K eigenmode functions and one residual component.

[0010] Furthermore, the mirror extension includes: assuming the length of the original sound speed time series is... The extension ratio is taken Then the single-sided extension length is set as That is, expanding outwards with the two ends of the sequence as the axis of symmetry respectively. Data points.

[0011] Furthermore, the mean squared error of the normalized space is used as the loss function during the training phase of the LSTM prediction model.

[0012] Further, the method involves inversely normalizing and summing the prediction results output by each LSTM prediction model to reconstruct the sound velocity prediction value for that depth layer; the method includes: The predicted estimates output by each LSTM prediction model are inversely normalized to the original physical dimensions to obtain the predicted values ​​of each modal component and the corresponding residual predicted values ​​of that depth layer. The predicted values ​​of each modal component of the depth layer are summed and reconstructed with the corresponding residual predicted values ​​to obtain the predicted sound velocity of the depth layer.

[0013] A second aspect of the present invention provides a temporal prediction system for sound velocity profiles based on adaptive multi-scale decomposition.

[0014] A sound velocity profile time series prediction system based on adaptive multi-scale decomposition includes: The VMD parameter optimization module is configured to: acquire the original sound velocity time series for each depth layer and determine the optimal VMD decomposition parameters for each depth layer through adaptive grid search optimization. The VMD parameter decomposition module is configured to perform variational mode decomposition on the original sound velocity time series of the depth layer using the optimal VMD decomposition parameters to obtain K eigenmode functions and one residual component. The normalization and feature extraction module is configured to perform normalization and differential feature extraction on multiple intrinsic mode functions and a residual component respectively, to obtain K mode components and the first-order differential feature corresponding to each mode component; The LSTM prediction modeling module is configured to: construct an independent LSTM prediction model for each modal component and its corresponding first-order difference feature for each depth layer, thereby achieving independent temporal modeling of the layered components; The inverse normalization and reconstruction module is configured to: inverse normalize and sum and reconstruct the prediction results output by each LSTM prediction model to obtain the sound velocity prediction value of the depth layer, and iterate layer by layer to output the complete sound velocity profile.

[0015] A third aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the adaptive multi-scale decomposition-based temporal prediction method for sound velocity profiles as described in the first aspect above.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing steps in the adaptive multi-scale decomposition-based temporal prediction method for sound velocity profiles as described in the first aspect above.

[0017] The fifth aspect of the present invention provides a computer program product or computer program.

[0018] This invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the adaptive multi-scale decomposition-based temporal prediction method for sound velocity profiles as described in the first aspect above.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention discloses a method and system for predicting sound velocity profiles based on adaptive multi-scale decomposition. By employing adaptive variational mode decomposition based on a composite index of envelope entropy and residual standard deviation, the original sound velocity sequence is decoupled into multiple intrinsic mode functions, achieving explicit separation of multi-scale physical processes. Furthermore, a hierarchical LSTM is used to independently model each mode along the depth direction, fully leveraging the modeling flexibility brought by depth decoupling. Simultaneously, first-order difference features are introduced as auxiliary input, enabling the model to simultaneously obtain gradient information for amplitude error and rate of change error, enhancing the response sensitivity to abrupt changes in sound velocity. This invention significantly improves the accuracy and robustness of sound velocity profile prediction while ensuring computational efficiency, providing reliable support for underwater forward-looking positioning and navigation services. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart illustrating the sound velocity profile time series prediction method based on adaptive multi-scale decomposition, as shown in an embodiment of the present invention. Figure 2 This is a diagram illustrating the overall architecture of the VMD-DF-HLSTM model in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the VMD parameter grid search optimization process in an embodiment of the present invention; Figure 4 This is a structural diagram of a sound velocity profile time-series prediction system based on adaptive multi-scale decomposition, as shown in an embodiment of the present invention. Figure 5 This is a structural diagram of a computer device shown in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Figure 1 This is a flowchart illustrating a sound velocity profile time-series prediction method based on adaptive multi-scale decomposition, as shown in an embodiment of the present invention; see reference. Figure 1 The method includes: The original sound velocity time series of each depth layer is obtained, and the optimal VMD decomposition parameters for each depth layer are determined by adaptive grid search optimization. The original sound velocity time series of this depth layer was subjected to variational mode decomposition using the optimal VMD decomposition parameters, resulting in K eigenmode functions and one residual component. Normalization and differential feature extraction are performed on multiple intrinsic mode functions and one residual component respectively to obtain K mode components and the first-order differential feature corresponding to each mode component; An independent LSTM prediction model is constructed for each modal component and its corresponding first-order difference feature in each depth layer to achieve independent temporal modeling of the layer components; The prediction results output by each LSTM prediction model are inversely normalized and summed to reconstruct the sound velocity prediction value for that depth layer, and the complete sound velocity profile is output iteratively layer by layer.

[0026] The proposed method for predicting sound velocity profiles based on adaptive multi-scale decomposition (hereinafter referred to as VMD-DF-HLSTM method) can be applied to sound velocity prediction in underwater positioning, navigation and communication systems. It includes: (1) an adaptive variational mode decomposition strategy based on the composite index of envelope entropy and residual standard deviation to achieve automatic optimization of decomposition parameters for each depth layer; (2) decoupling the original sound velocity time series into multiple intrinsic mode functions and a residual component through VMD to achieve explicit separation of multi-scale time series modes; (3) performing Z-score normalization and first-order difference feature extraction on each modal component and residual to construct a dual-channel LSTM input of "normalized value + difference feature"; (4) constructing an independent LSTM prediction model for each modal component and residual of each depth layer to achieve independent time series modeling of layered components; (5) performing inverse normalization and summation reconstruction on the prediction results of each component to obtain the sound velocity prediction value of the depth layer, and iteratively outputting the complete sound velocity profile layer by layer. To address the characteristics of ocean sound velocity profile data, which exhibit strong nonlinearity, non-stationarity, and multi-scale temporal features, a strategy of layer-by-layer decomposition, independent modeling, and reconstruction is employed to achieve high-precision time-series prediction of sound velocity profiles. The specific embodiments of this invention are described in detail below with reference to the accompanying drawings.

[0027] Figure 2 This is a diagram illustrating the overall architecture of the VMD-DF-HLSTM model of this invention. Figure 2 As shown, for each depth layer, this invention first adaptively determines the optimal VMD decomposition parameters for that layer through grid search, and then uses these parameters to perform variational mode decomposition on the original sound velocity time series to obtain... The system generates one intrinsic mode function and one residual component; then, Z-score normalization and differential feature extraction are performed on each component; next, an independent LSTM prediction model is constructed for each mode component and residual; finally, the prediction results output by each LSTM are inversely normalized and summed to reconstruct the sound velocity prediction value for that depth layer. The above process is executed independently layer by layer, ultimately obtaining a complete sound velocity profile prediction result. The detailed steps and explanations are as follows: (1) Data organization and preprocessing. The historical sound velocity data is organized into a time-depth matrix. :

[0028] Among them, row index Corresponding to depth layers from shallow to deep, Total number of depth layers, column index Corresponding to discrete sampling points in historical time series, The total number of samples in the historical time series. Indicates the first The depth layer at the ... The speed of sound at time point n is expressed in meters per second (m / s). The dataset is divided into a training set and a test set. For the nth time point in the training set... Each depth layer is used to extract its time series data. , for , The length of the training set time series.

[0029] (2) Adaptive grid search optimization of VMD parameters. Since the sound velocity sequences at different depths differ significantly, fixed VMD parameters cannot guarantee optimal decomposition results. Therefore, this invention designs an adaptive optimization method for VMD parameters based on grid search, the specific process of which is as follows: Figure 3 As shown. This method uses the weighted sum of the envelope entropy (EE) and residual standard deviation (Rstd) of the modal components as the evaluation index for decomposition quality. For the number of modes... and penalty factor Within the preset range and Inner step length and Generate a two-dimensional parameter mesh. For each set of candidate parameters... Perform the following operations: Mirror the original signal, perform VMD decomposition and trim it to its original length, calculate and record the envelope entropy and residual standard deviation for this set of parameters. After traversal, perform Min-Max normalization on all recorded envelope entropy and residual standard deviation values, and calculate the weighted composite score using the following formula:

[0030] in, and These are the envelope entropy and residual standard deviation after Min-Max normalization, respectively, and are dimensionless. and The weighting coefficient has a range of values. In this embodiment, take This indicates that both contribute equally to the overall score; The weighted composite score is dimensionless; a smaller value indicates better decomposition quality. The set of parameters with the lowest composite score is selected as the optimal VMD parameters for the current depth layer. ,in To determine the optimal number of modes, This is the optimal penalty factor.

[0031] (3) Mirror extension and VMD decomposition.

[0032] When decomposing a finite-length signal, truncation at both ends introduces boundary effects, causing distortion of the decomposed modal components at the endpoints, which in turn contaminates the input features of subsequent LSTM. To mitigate these boundary effects, this invention performs mirror extension processing on the original sequence before each VMD decomposition. Traditional mirror extension methods typically use extreme points as axes of symmetry, but considering that the latest endpoint values ​​of the sound velocity sequence need to be preserved as initial conditions for LSTM prediction, this invention adopts a mirror extension strategy using the endpoints themselves as axes of symmetry. Specifically, let the length of the original sequence be... The extension ratio is taken Then the single-sided extension length is set as That is, expanding outwards with the two ends of the sequence as the axis of symmetry respectively. Data points. After performing VMD decomposition on the extended sequence, only the middle segment corresponding to the original interval length is taken as the final decomposition result, thus obtaining... eigenmode functions and a residual term :

[0033] in, For the first The depth layer One modal component, For the first The residual components of each depth layer. Mirror continuation can effectively suppress the endpoint effect caused by endpoint discontinuities during VMD decomposition, thereby improving decomposition accuracy. For the first Each depth layer at time The original sound speed time series; The number of optimal modes determined in step (2); For the first The first depth layer decomposition yields the... Each intrinsic mode function has the same dimension as the original sequence, which is meters per second (m / s), and represents the oscillation component at a specific frequency scale in the sound velocity sequence of this depth layer; For the first The residual components of each depth layer, measured in meters per second (m / s), represent the remaining trend or noise components in the original sequence that were not captured by the modal components. All subsequent depth layers of VMD decomposition are processed in this manner.

[0034] (4) Normalization and differential feature extraction. To accelerate model convergence and enhance generalization ability, Z-score normalization is performed on each modal component and residual:

[0035] in, For the first The depth layer Each modal component at time... The value of is expressed in meters per second (m / s); and The modal component is represented as a whole time series. The mean and standard deviation of the above, with units of meters per second (m / s) and meters per second (m / s), respectively. The normalized value is dimensionless. For the first Each depth layer residual component at time... The value of is measured in meters per second (m / s); and These are the mean and standard deviation of the residual component over the entire time series, respectively, with units of meters per second (m / s) and meters per second (m / s). These are the normalized residuals, dimensionless. After normalization, each sequence has zero mean and unit variance.

[0036] To further enhance the model's ability to perceive local abrupt changes and dynamic trends in sound speed, first-order difference features are constructed for each normalized component:

[0037]

[0038] in, For the first The depth layer The normalized modal components at time... The first-order difference value is dimensionless and reflects the magnitude and direction of the change of the component between adjacent time points; For the first The normalized residual of each depth layer at time... The first-order difference value is dimensionless. The difference feature reflects the changing trend of the sound speed sequence in adjacent time steps, and together with the normalized value, it constitutes the input feature of LSTM, enabling the model to capture both the absolute amplitude information and the relative change information of the sequence.

[0039] (5) LSTM Prediction Modeling. The sound velocity time series is influenced by the continuity of marine environmental parameters and exhibits strong temporal autocorrelation; simultaneously, sound velocity changes significantly at water layers such as the thermocline, and its first-order difference can effectively characterize the location information of such abrupt changes. Therefore, this invention combines the normalized modal components with their first-order differences to construct a dual-channel input feature for LSTM, enabling the model to simultaneously learn the absolute magnitude and local variation trend of sound velocity, thereby improving its ability to track abrupt changes. Taking the first... The depth layer Taking one modal component as an example, the input vector is constructed as follows:

[0040] in, For the first Layer Each modal component at time... The LSTM input is a two-dimensional vector; For the previous moment The normalized modal values ​​are dimensionless. For the previous moment The first-order difference value is dimensionless. This input vector utilizes... Information at any given moment predicts the current situation. The normalized speed of sound at any given moment.

[0041] LSTM units implement temporal memory and updates through input gates, forget gates, and output gates. The network output layer provides a prediction estimate in the normalized space. :

[0042] in, For the first Layer Each modal component at time... The normalized predicted value is dimensionless; For the LSTM unit at time The hidden state vector; This is the weight matrix of the output layer. These are the bias terms for the output layer; both are trainable parameters learned by the LSTM network during training via backpropagation. The mean squared error (MSE) in the normalized space is used as the loss function during training.

[0043] in, is the mean squared error loss value, which is dimensionless and represents the average squared deviation between the model's predicted value and the actual value. The length of the training set time series; For the model at time Normalized predicted values; For a moment The normalized true value. During training, the goal is to minimize this loss function, and the network parameters are iteratively updated using the Adam optimizer. Residual components. The normalized residual prediction values ​​were obtained by independently training using the exact same method. Change to .

[0044] (6) Prediction denormalization and intra-layer reconstruction. The normalized prediction values ​​of each LSTM output are denormalized to the original physical dimensions:

[0045] in, For the inverse normalization of the first Layer Each modal component at time... The predicted value, in meters per second (m / s), is restored to its original dimension; The normalized predicted value output by LSTM in step (5) is dimensionless; and These are the standard deviation and mean of the modal component in step (4), respectively, with units of meters per second (m / s). The predicted residual value after inverse normalization is expressed in meters per second (m / s). These are the normalized residual predictions, which are dimensionless. and These are the standard deviation and mean of the residuals in step (4), respectively, with units of meters per second (m / s).

[0046] The current depth layer at time The final sound velocity prediction is reconstructed by summing the predictions of all modal components and the residual predictions:

[0047] in, For the first Each depth layer at time The final predicted speed of sound is expressed in meters per second (m / s). This represents the total number of modal components. For the first The inverse normalized prediction values ​​of each modal component; This is the inverse normalized predicted value of the residual.

[0048] (7) Layer-by-layer iteration and complete profile output. Complete the current depth layer (the... After predicting the depth of the layer, determine whether all depth layers have been traversed: if ,make If so, continue extracting the next depth layer sequence and repeat steps (2) to (6); Once all depth layers have been processed, the prediction results of each layer are combined in depth order to output a complete sound velocity profile prediction result.

[0049] This invention decomposes non-stationary sound velocity sequences into multiple relatively stationary modal components using VMD, reducing the modeling difficulty of LSTM for complex temporal patterns. By combining differential features with the original normalized values ​​as input, the LSTM network can simultaneously obtain gradient information for amplitude and slope errors during backpropagation, thereby enhancing its sensitivity to abrupt changes in sound velocity. Through independent layer-by-layer modeling and parameter optimization, the interference of differences in sound velocity variation patterns between shallow and deep layers on the optimization of the same model parameters is avoided. The combined effect of these technical solutions significantly improves the accuracy and stability of sound velocity profile prediction.

[0050] In addition, other alternative solutions that can achieve the purpose of this invention include: (1) a sound speed prediction method based on the combination of empirical mode decomposition and recurrent neural network, which first decomposes the original sound speed time series into several intrinsic mode functions through empirical mode decomposition, then establishes a recurrent neural network prediction model for each component, and finally reconstructs the prediction result; (2) a sound speed prediction method based on wavelet transform and temporal convolutional network, which uses wavelet transform to decompose the sound speed sequence into multiple scales, and uses temporal convolutional network to perform temporal modeling of the coefficients of each scale to capture the change features of different time scales; (3) an end-to-end sound speed sequence prediction method based on Transformer, which directly constructs a Transformer model for the original sound speed profile time series, and uses the self-attention mechanism to capture the global time dependency relationship without explicit multi-scale decomposition; (4) a sound speed mutation prediction method based on multilayer perceptron and differential feature input, which uses the original sound speed sequence and the first-order differential sequence as dual-channel input, and uses stacked multilayer perceptrons to perform nonlinear mapping to enhance the perception ability of sound speed mutation.

[0051] The above combination Figure 1 The method for predicting sound velocity profiles based on adaptive multi-scale decomposition based on the embodiments of the present invention has been described in detail. Next, the system for predicting sound velocity profiles based on adaptive multi-scale decomposition based on the embodiments of the present invention will be described in conjunction with the accompanying drawings.

[0052] Figure 4 This is a schematic diagram of the structure of a sound velocity profile time series prediction system based on adaptive multi-scale decomposition, as shown in an embodiment of the present invention. (Refer to...) Figure 4 The system described in this invention includes: The VMD parameter optimization module is configured to: acquire the original sound velocity time series for each depth layer and determine the optimal VMD decomposition parameters for each depth layer through adaptive grid search optimization. The VMD parameter decomposition module is configured to perform variational mode decomposition on the original sound velocity time series of the depth layer using the optimal VMD decomposition parameters to obtain K eigenmode functions and one residual component. The normalization and feature extraction module is configured to perform normalization and differential feature extraction on multiple intrinsic mode functions and a residual component respectively, to obtain K mode components and the first-order differential feature corresponding to each mode component; The LSTM prediction modeling module is configured to: construct an independent LSTM prediction model for each modal component and its corresponding first-order difference feature for each depth layer, thereby achieving independent temporal modeling of the layered components; The inverse normalization and reconstruction module is configured to: inverse normalize and sum and reconstruct the prediction results output by each LSTM prediction model to obtain the sound velocity prediction value of the depth layer, and iterate layer by layer to output the complete sound velocity profile.

[0053] In some embodiments, the VMD parameter optimization module is specifically configured to: for the number of modes in each depth layer and penalty factor Within a preset range of the number of modes and a preset range of the penalty factor, a two-dimensional parameter mesh is generated according to its set step size; for each group of candidate parameters... Execution: Mirror the original signal, perform VMD decomposition and trim to the original length, calculate and record the envelope entropy and residual standard deviation for this set of parameters; after traversal, perform Min-Max normalization on all recorded envelope entropy and residual standard deviation values, and calculate the weighted composite score: ;in, and These are the envelope entropy and residual standard deviation after Min-Max normalization, respectively. and The weighting coefficients are used; the set of parameters with the smallest weighted composite score is selected as the optimal VMD parameters for the current depth layer. ,in To determine the optimal number of modes, This is the optimal penalty factor.

[0054] In some embodiments, the VMD parameter decomposition module is specifically configured to: perform mirror extension processing on the original sound velocity time series of the depth layer, perform VMD decomposition on the extended sequence, and only extract the middle segment corresponding to the original interval length as the final decomposition result to obtain K eigenmode functions and one residual component.

[0055] In some embodiments, the mirror extension includes: assuming the length of the original sound velocity time sequence is... The extension ratio is taken Then the single-sided extension length is set as That is, expanding outwards with the two ends of the sequence as the axis of symmetry respectively. Data points.

[0056] In some embodiments, the mean squared error of the normalized space is used as the loss function during the training phase of the LSTM prediction model.

[0057] In some embodiments, the denormalization and reconstruction module is specifically configured to: denormalize the predicted estimates output by each LSTM prediction model to the original physical dimensions to obtain the predicted value of each modal component and the corresponding residual predicted value of the depth layer; and sum and reconstruct the predicted value of each modal component and the corresponding residual predicted value of the depth layer to obtain the predicted value of the sound velocity of the depth layer.

[0058] According to embodiments of the present invention, the sound velocity profile time series prediction system based on adaptive multi-scale decomposition can correspond to the execution of the method described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the sound velocity profile time series prediction system based on adaptive multi-scale decomposition are respectively for implementing Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0059] See Figure 5 The diagram shows the structure of a computer device, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium are connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory. The computer-readable storage medium stores computer programs, including program instructions, and the processor executes the program instructions stored in the computer-readable storage medium. The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device, adapted to implement one or more instructions, specifically adapted to load and execute one or more instructions to implement the corresponding steps in the embodiment of the sound velocity profile time-series prediction method based on adaptive multi-scale decomposition.

[0060] This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device.

[0061] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0062] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above embodiment of the sound velocity profile time series prediction method based on adaptive multi-scale decomposition.

[0063] This embodiment provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps in the above-described embodiment of the sound velocity profile time-series prediction method based on adaptive multi-scale decomposition.

[0064] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0065] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the temporal series of sound velocity profiles based on adaptive multi-scale decomposition, characterized in that, include: The original sound velocity time series of each depth layer is obtained, and the optimal VMD decomposition parameters for each depth layer are determined by adaptive grid search optimization. The original sound velocity time series of this depth layer was subjected to variational mode decomposition using the optimal VMD decomposition parameters, resulting in K eigenmode functions and one residual component. Normalization and differential feature extraction are performed on multiple intrinsic mode functions and one residual component respectively to obtain K mode components and the first-order differential feature corresponding to each mode component; An independent LSTM prediction model is constructed for each modal component and its corresponding first-order difference feature in each depth layer to achieve independent temporal modeling of the layer components; The prediction results output by each LSTM prediction model are inversely normalized and summed to reconstruct the sound velocity prediction value for that depth layer, and the complete sound velocity profile is output iteratively layer by layer. The adaptive grid search optimization determines the optimal VMD decomposition parameters for each depth layer; the method includes: For the number of modes in each depth layer and penalty factor Within the preset range of the number of modes and the preset range of the penalty factor, a two-dimensional parametric mesh is generated according to the respective set step size; For each group of candidate parameters in turn Execution: Mirror the original signal, perform VMD decomposition and trim to the original length, calculate and record the envelope entropy and residual standard deviation for this set of parameters; after traversal, perform Min-Max normalization on all recorded envelope entropy and residual standard deviation values, and calculate the weighted composite score: ;in, and These are the envelope entropy and residual standard deviation after Min-Max normalization, respectively. and The weighting coefficients are used; the set of parameters with the smallest weighted composite score is selected as the optimal VMD parameters for the current depth layer. ,in To determine the optimal number of modes, This is the optimal penalty factor.

2. The sound velocity profile time series prediction method based on adaptive multi-scale decomposition according to claim 1, characterized in that, The method involves performing variational mode decomposition on the original sound velocity time series of the depth layer using optimal VMD decomposition parameters to obtain K eigenmode functions and one residual component. The method includes: performing mirror extension processing on the original sound velocity time series of the depth layer; performing VMD decomposition on the extended sequence; and extracting only the middle segment corresponding to the original interval length as the final decomposition result to obtain K eigenmode functions and one residual component.

3. The sound velocity profile time series prediction method based on adaptive multi-scale decomposition according to claim 2, characterized in that, The mirror extension includes: assuming the length of the original sound velocity time sequence is... The extension ratio is taken Then the single-sided extension length is set as That is, expanding outwards with the two ends of the sequence as the axis of symmetry respectively. Data points.

4. The sound velocity profile time series prediction method based on adaptive multi-scale decomposition according to claim 1, characterized in that, In the training phase of the LSTM prediction model, the mean squared error of the normalized space is used as the loss function.

5. The sound velocity profile time series prediction method based on adaptive multi-scale decomposition according to claim 1, characterized in that, The method involves inversely normalizing and summing the prediction results output by each LSTM prediction model to reconstruct the predicted sound velocity value for that depth layer; the method includes: The predicted estimates output by each LSTM prediction model are inversely normalized to the original physical dimensions to obtain the predicted values ​​of each modal component and the corresponding residual predicted values ​​of that depth layer. The predicted values ​​of each modal component of the depth layer are summed and reconstructed with the corresponding residual predicted values ​​to obtain the predicted sound velocity of the depth layer.

6. A sound velocity profile time series prediction system based on adaptive multi-scale decomposition, characterized in that, include: The VMD parameter optimization module is configured to: acquire the original sound velocity time series for each depth layer and determine the optimal VMD decomposition parameters for each depth layer through adaptive grid search optimization. The VMD parameter decomposition module is configured to perform variational mode decomposition on the original sound velocity time series of the depth layer using the optimal VMD decomposition parameters to obtain K eigenmode functions and one residual component. The normalization and feature extraction module is configured to perform normalization and differential feature extraction on multiple intrinsic mode functions and a residual component respectively, to obtain K mode components and the first-order differential feature corresponding to each mode component; The LSTM prediction modeling module is configured to: construct an independent LSTM prediction model for each modal component and its corresponding first-order difference feature for each depth layer, thereby achieving independent temporal modeling of the layered components; The inverse normalization and reconstruction module is configured to: inverse normalize and sum and reconstruct the prediction results output by each LSTM prediction model to obtain the sound velocity prediction value of the depth layer, and iterate and output the complete sound velocity profile layer by layer. The adaptive grid search optimization determines the optimal VMD decomposition parameters for each depth layer; the method includes: For the number of modes in each depth layer and penalty factor Within the preset range of the number of modes and the preset range of the penalty factor, a two-dimensional parametric mesh is generated according to the respective set step size; For each group of candidate parameters in turn Execution: Mirror the original signal, perform VMD decomposition and trim to the original length, calculate and record the envelope entropy and residual standard deviation for this set of parameters; after traversal, perform Min-Max normalization on all recorded envelope entropy and residual standard deviation values, and calculate the weighted composite score: ;in, and These are the envelope entropy and residual standard deviation after Min-Max normalization, respectively. and The weighting coefficients are used; the set of parameters with the smallest weighted composite score is selected as the optimal VMD parameters for the current depth layer. ,in To determine the optimal number of modes, This is the optimal penalty factor.

7. A computer device, characterized in that, A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the sound velocity profile time-series prediction method based on adaptive multi-scale decomposition as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the steps of the sound velocity profile time-series prediction method based on adaptive multi-scale decomposition as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the sound velocity profile time series prediction method based on adaptive multi-scale decomposition as described in any one of claims 1-5.

Images