Real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling mode

Abstract

This invention belongs to the field of marine observation technology, specifically a real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling mode. The method includes acquiring inertial measurement data; completing attitude measurement of an unmanned mobile vehicle at sea; acquiring the three-dimensional wavefront displacement sequence required for wave calculation; using the CEEMDAN algorithm to perform spectral separation on the wavefront displacement sequence, estimating the power spectrum, unidirectional spectrum, and turning spectrum of the deconstructed waveform, and extracting wave characteristic parameters; using the steady state of the vehicle as a key guiding criterion, matching the waveform sequences of each corresponding frequency band based on the swaying period and frequency of the vehicle's attitude information, and establishing the weight distribution of each frequency band waveform in the subsequent prediction model; inputting the wave characteristic parameters of each frequency band, the weight distribution of waveform elements, and the three-dimensional wavefront motion parameters into the prediction model to achieve short-time prediction of wavefront morphology. This invention can effectively reduce the complexity of actual waves and improve the stability and predictability of waveforms.

Description

Technical Field

[0001] This invention belongs to the field of marine observation technology, specifically a real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling mode. Background Technology

[0002] Unmanned mobile vessels at sea are crucial operational units for marine scientific research, maritime military operations, marine resource development, and maritime emergency rescue. As these vessels continue to expand into deeper waters, higher demands are being placed on their stable navigation and safe operation in complex sea conditions.

[0003] The complexity and uncertainty of ocean waves are key factors affecting the stability and safety of unmanned mobile vehicles (UMVs) operations. Current research on ocean waves, both domestically and internationally, mainly focuses on the statistical analysis and prediction of characteristics over long periods and large spatial scales, with related prediction results primarily emphasizing long-term wind and wave statistical indicators. However, given the urgent need for UMVs to achieve stable navigation and safe take-off and landing in complex sea conditions, the decision-making process requires a time window on the order of seconds, necessitating precise descriptions of different wave surface morphologies in the current area within the next few seconds. While current wave numerical simulation forecasting methods can achieve real-time prediction, they are all statistical prediction models for complex long-term ocean waves, unable to reflect short-term, precise wave surface morphologies, and thus cannot provide instantaneous and effective decision support for real-time operations of UMVs. Therefore, it is necessary to propose a real-time wave surface reconstruction and prediction method with practical guiding value for the stable operation and safe take-off and landing of UMVs at sea, providing crucial decision-making basis for safe operations of UMVs at sea. Furthermore, the impact of wave field motion on the communication stability of mobile carriers is becoming increasingly significant. The motion state of the carrier in the wave field is closely related to the communication timeliness. In order to significantly improve the stability and success rate of communication of unmanned mobile carriers in the wave field environment, it is necessary to use real-time wave surface morphology prediction methods to accurately know the operating status of the carrier in the wave field, and then dynamically optimize the communication timing strategy to ensure efficient and reliable maritime communication and meet the stringent communication requirements of marine information operations. Summary of the Invention

[0004] This invention addresses the shortcomings of current wave characteristic statistics and predictions, which primarily focus on long-term, large-scale wind and wave statistical indicators. These indicators fail to reflect short-term, precise wave surface morphology and are not correlated with the safe operational status of the vessel, thus hindering timely and effective decision support for real-time operations of unmanned mobile vessels at sea. The invention proposes a real-time wave surface reconstruction and prediction method based on frequency division deconstruction and covariant coupling modes. This method, combined with the actual operational status of the vessel, effectively reconstructs and reasonably predicts waves within a limited time and space, meeting the cross-domain navigation and safe operation requirements of next-generation unmanned mobile vessels at sea.

[0005] The technical solution adopted in this invention includes: a real-time wavefront reconstruction and prediction method based on frequency division deconstruction and covariant coupling mode, comprising the following steps: Step 1: Acquire high-speed data from the inertial measurement sensors: three-axis gyroscope, three-axis accelerometer, and three-axis magnetometer; Step 2: Using an extended Kalman filter as the main framework, the output values ​​of the three-axis gyroscope form the update matrix of the filter's state equation, and the output values ​​of the accelerometer / magnetometer form the observation equation of the filter, thus completing the attitude measurement of the unmanned mobile vehicle at sea. Step 3: Using the attitude rotation matrix of the unmanned mobile carrier at sea, the acceleration of wave motion is transformed from the carrier coordinate system to the local coordinate system, and after high-frequency filtering and strapdown calculation, the three-dimensional wave surface displacement sequence required for wave calculation is obtained in a short time. Step 4: The wavefront displacement sequence is separated into its spectrum using the CEEMDAN algorithm. The power spectrum, unidirectional spectrum, and turning spectrum of the deconstructed waveform are estimated. Wave characteristic parameters are extracted using the periodogram method and Fourier series method. On the other hand, the steady state of the carrier is obtained as a key guiding criterion. The waveform sequences of each corresponding frequency band are matched with the swaying period and frequency of the carrier attitude information to establish the weight distribution of each frequency band waveform in the subsequent prediction model. Step 5: Input the wave characteristic parameters of the frequency band, the weight distribution of waveform elements, and the three-dimensional wave surface motion parameters into the BiGRU prediction model to achieve short-term prediction of wave surface morphology, which has important guiding value for the stable operation of unmanned mobile carriers at sea.

[0006] Preferably, in step four, the matching process includes: First, the wavefront displacement sequence is extracted according to different frequency bands using the CEEMDAN algorithm: H1, H2, H3, ... H n The carrier attitude sway information is separated into different frequency bands, and the sway amplitude of each frequency band is statistically calculated: D1, D2, D3, ... D n Based on the magnitude of the oscillation amplitude, the wavefront displacement sequence is arranged from smallest to largest. p (D), and calculate the wavefront displacement sequence H for each frequency band. p The weighting coefficient of (D) is calculated using the following formula: ; in This is the weighting growth factor. P Index of wavefront displacement sequence in frequency band; The weighting coefficients are normalized to their mean. .

[0007] The beneficial effects of this invention are as follows: (1) The present invention adopts frequency division deconstruction mode for short-time wave measurement and waveform reconstruction, which meets the timeliness of short-time wave measurement and waveform reconstruction of mobile carriers at sea. The process focuses on improving the efficiency and accuracy of wave measurement and reconstruction. Through high-speed acquisition, cyclic filtering, redundant observation and carrier dynamic matching, the accurate capture of short-time wave characteristics and rapid waveform reconstruction are achieved.

[0008] (2) This invention performs spectrum separation on the waveform, establishes a mechanism that matches the real-time motion information of the carrier by deconstructing the covariant coupling effect between the waveform and the carrier motion, and forms a waveform element reconstruction process that has practical guiding value for the smooth operation of the mobile carrier. It introduces a deep learning model to train and generate the corresponding frequency band information separately, and uses multiple parameters to constrain the model prediction. Attached Figure Description

[0009] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a block diagram for inertial measurement data acquisition. Figure 3 A block diagram for attitude measurement of unmanned mobile vehicles; Figure 4 Flowchart for outputting wavefront displacement of unmanned mobile carriers; Figure 5 This is a diagram illustrating the structure of short-time wave measurement and waveform reconstruction. Figure 6 The flowchart shows the wavefront prediction algorithm based on CEEMDAN and BiGRU. Figure 7 The image shows a comparison between the triaxial prediction curve and the actual value at time t1 using the method of this invention; where (a) is the X-axis; (b) is the Y-axis; and (c) is the Z-axis. Figure 8 These are the stepwise evaluation index curves for each axis of the method of this invention; where (a), (b), and (c) are the R values ​​for the X-axis, Y-axis, and Z-axis, respectively. 2 Values; (d), (e), and (f) are the RMSE values ​​for the X-axis, Y-axis, and Z-axis, respectively; (g), (h), and (i) are the MAE values ​​for the X-axis, Y-axis, and Z-axis, respectively. Detailed Implementation

[0010] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] This invention provides a real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling modes, the flowchart of which is shown below. Figure 1 As shown: The specific steps are as follows: Step 1: Inertial Measurement. Inertial measurement sensors, such as triaxial gyroscopes, triaxial accelerometers, and triaxial magnetometers, are calibrated in the laboratory, and the calibration parameters are stored in memory. Then, an embedded processor acquires inertial measurement data, using the calibrated data as input for filtering and fusion. Figure 2 As shown.

[0012] Step 2: Filtering and Fusion. Using an extended Kalman filter as the main framework, the output values ​​of the three-axis gyroscope form the update matrix of the filter's state equation, while the output values ​​of the accelerometer / magnetometer form the observation equation part of the filter. This completes the attitude measurement of the unmanned mobile vehicle at sea, such as... Figure 3 As shown.

[0013] Step 3: Data Generation. Using the attitude rotation matrix of the unmanned mobile platform at sea, the acceleration due to wave motion is transformed from the platform coordinate system to the local coordinate system. After high-frequency filtering and strapdown calculation, the three-dimensional wave surface displacement sequence required for wave calculation is obtained in a short time. For example... Figure 4 As shown.

[0014] Step 4: Reconstruction Processing. The wavefront displacement sequence is subjected to spectral separation using the CEEMDAN algorithm. Frequency division and deconstruction reduce the complexity of the actual waves and enhance the orderliness and predictability of the waveform. On one hand, the power spectrum, unidirectional spectrum, and turning spectrum of the deconstructed waveform are estimated, and wave characteristic parameters are extracted using the periodogram method and Fourier series method. On the other hand, obtaining the steady state of the carrier is used as a key guiding criterion to construct an adaptive matching mechanism between the carrier motion and the deconstructed waveform, and to obtain the weight distribution of each deconstructed waveform in the subsequent prediction model, such as... Figure 5 As shown.

[0015] The matching mechanism adopted is as follows: The unmanned mobile platform needs to acquire real-time wave information to effectively guide its operation and ensure safe operation. Based on this principle, the stable operating state of the mobile platform is mainly reflected by its swaying frequency and amplitude. Attitude information calculated using inertial measurements can directly reflect the swaying state of the mobile platform. By acquiring the swaying period and frequency of the corresponding attitude information, waveform sequences for each corresponding frequency band are matched, and the wavefront displacement sequence weights for each frequency band are calculated. The matching process is as follows: First, the wavefront displacement sequence is extracted according to different frequency bands using the CEEMDAN algorithm: H1, H2, H3, ... H n (H represents the wavefront displacement sequence in different frequency bands, and n represents the sequence label for different frequency bands); simultaneously, the attitude roll information is separated using the same method, and the roll amplitude of each frequency band is statistically calculated: D1, D2, D3, ... D n (D represents the swing amplitude of different frequency bands, and n represents the sequence label of different frequency bands); then, based on the swing amplitude D... mSort H by size from smallest to largest p (D), and calculate the wavefront displacement sequence H for each frequency band. p The weighting coefficient of (D) is calculated using the following formula: ; in is the weighting growth factor, and P is the frequency band sequence index.

[0016] To eliminate differences The impact of the selected values ​​on the prediction is addressed by normalizing the weighting coefficients to the mean. .

[0017] Step 5: Prediction Output. The wave characteristic parameters of each frequency band, the weight distribution of waveform elements, and the three-dimensional wave surface motion parameters are input into the BiGRU prediction model. By continuously optimizing the network parameters, the complex correlation between wave information and the stable motion of the carrier is deeply explored, enabling short-term prediction of wave surface morphology, which is of significant guiding value for the stable operation of unmanned mobile carriers at sea. The prediction model combining the CEEMDAN algorithm and the BiGRU network follows the following technical route: Figure 6 As shown.

[0018] Description of each layer of the prediction model: The input layer is responsible for receiving preprocessed data. In this invention, the input data includes wave data parameters for each frequency band after CEEMDAN decomposition. This data is organized into a sequence and input into the network in chronological order. The BiGRU layer is the core of the network, consisting of two sub-layers: a forward GRU and a backward GRU. The forward GRU processes the input sequence from front to back in chronological order, while the backward GRU processes the input sequence from back to front in chronological order. The outputs of these two sub-layers are concatenated to form the output of the BiGRU layer. In the BiGRU layer, this invention uses multiple GRU units stacked to form a deep network to enhance the model's nonlinear expressive power and feature extraction capabilities. A Dropout layer is added after each GRU unit to prevent overfitting. To ensure the output reaches the target dimension, this invention adds a fully connected layer. This fully connected layer maps the output of the BiGRU layer to the target dimension, which in this invention represents the predicted values ​​of wave feature parameters and wavefront displacement sequences for each frequency band within the next few seconds. Therefore, the number of neurons in the fully connected layer should match the number of predicted values. Furthermore, since the prediction task in this invention is a regression problem, a linear activation function is chosen for the fully connected layer's output. The output layer is the last layer of the network and is responsible for outputting the prediction results. In this invention, the output layer takes the output of the fully connected layer as input and directly outputs the predicted waveform and wave feature parameters.

[0019] Ocean wave surfaces are composed of waveform components of various frequencies, directions, and sizes, exhibiting significant randomness. If the entire wave surface is directly trained and reconstructed using a deep network, it is difficult to achieve ideal results. This invention obtains wave information of each frequency band by frequency division and deconstruction, and matches it with the carrier's motion direction, which can effectively reduce the complexity of actual waves and improve the stability and predictability of waveforms.

[0020] In processing a dataset of unmanned sailing vehicles, the data encompassed the instantaneous attitude (Att), acceleration (Acc), velocity (Vel), and displacement (Disp) time series of the vehicle in three-dimensional space. To demonstrate the feasibility and beneficial effects of the method of this invention, in the data simulation processing, by analyzing the overall swaying frequency of the vehicle, the corresponding frequency band of the three-axis displacement dataset was extracted as the true data. The reconstruction processing and model prediction of this invention were then compared with the predicted output at the predicted time (t1), and the results are as follows. Figure 7 As shown.

[0021] To evaluate the ability of the method of this invention to capture the non-stationary evolution characteristics of complex ocean waves and its robustness under multi-step prediction, step-wise metrics and the overall average evaluation index (coefficient of determination (R²)) were calculated for displacement data in three dimensions: northward drift (X-axis), westward drift (Y-axis), and heave (Z-axis). 2 The mean absolute error (MAE) and root mean square error (RMSE) are used to comprehensively evaluate the accuracy and stability of triaxial displacement prediction. The results are as follows: Figure 8 As shown.

[0022] Table 1 Overall Indicators of the Method of the Invention X 0.8284 0.127 0.099 Y 0.8313 0.2136 0.1506 Z 0.8185 0.1754 0.1496 .

[0023] In multi-step prediction tasks, as time steps ( As time progresses, the model is prone to error accumulation due to iterative historical biases. In the initial prediction stage, the method proposed in this invention demonstrates significant initial prediction confidence in both RMSE and MAE, the two core error evaluation indicators.

[0024] From the overall average perspective, the method proposed in this invention achieved optimal performance across all three core evaluation metrics in all spatial dimensions. The coefficient of determination (COP), which characterizes the approximation degree between the predicted and true sequences, is particularly strong. In terms of the X-axis (Disp), the method of the present invention achieved 0.8284, 0.8313 and 0.8185 on the X-axis (Disp), Y-axis (Disp) and Z-axis (Disp) data, respectively.

Claims

1. A real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling mode, characterized in that, Includes the following steps: Step 1: Acquire high-speed data from the inertial measurement sensors: three-axis gyroscope, three-axis accelerometer, and three-axis magnetometer; Step 2: Using an extended Kalman filter as the main framework, the output values ​​of the three-axis gyroscope form the update matrix of the filter's state equation, and the output values ​​of the accelerometer / magnetometer form the observation equation of the filter, thus completing the attitude measurement of the unmanned mobile vehicle at sea. Step 3: Using the attitude rotation matrix of the unmanned mobile carrier at sea, the acceleration of wave motion is transformed from the carrier coordinate system to the local coordinate system, and after high-frequency filtering and strapdown calculation, the three-dimensional wave surface displacement sequence required for wave calculation is obtained in a short time. Step 4: The wavefront displacement sequence is separated into its spectrum using the CEEMDAN algorithm. The power spectrum, unidirectional spectrum, and turning spectrum of the deconstructed waveform are estimated. Wave characteristic parameters are extracted using the periodogram method and Fourier series method. The steady state of the carrier is obtained as the key guiding criterion. The waveform sequences of each corresponding frequency band are matched with the swaying period and frequency of the carrier attitude information to establish the weight distribution of each frequency band waveform in the subsequent prediction model. Step 5: Input the wave characteristic parameters of the frequency bands, the weight distribution of waveform elements, and the three-dimensional wave surface motion parameters into the BiGRU prediction model to achieve short-term prediction of wave surface morphology.

2. The real-time wavefront reconstruction prediction method based on frequency division deconstruction and covariant coupling mode according to claim 1, characterized in that, In step four, the matching process includes: First, the wavefront displacement sequence is extracted according to different frequency bands using the CEEMDAN algorithm: H1, H2, H3, ... H n The carrier attitude sway information is separated into different frequency bands, and the sway amplitude of each frequency band is statistically calculated: D1, D2, D3, ... D n Based on the magnitude of the oscillation amplitude, the wavefront displacement sequence is arranged in ascending order. p (D), and calculate the wavefront displacement sequence H for each frequency band. p The weighting coefficient of (D) is calculated using the following formula: ； in This is the weighting growth factor. P Index of wavefront displacement sequence in frequency band; The weighting coefficients are normalized to their mean. 。

Images