A method for calculating total wave energy based on multi-source fusion data

Abstract

This invention relates to the field of electrical digital data processing technology and discloses a method for calculating the total wave energy based on multi-source fusion data. The method includes: acquiring heterogeneous sampled digital sequences characterizing the hydrodynamic distribution; calculating the local covariance matrix within a local time sliding window to generate symmetric positive definite tensor nodes; mapping the tensor nodes to a Riemannian manifold space and calculating the geodesic evolution gradient; extracting tensor affine biases based on the gradients to correct the sampling timescale, achieving phase locking between sequences; and importing the phase-locked sequences into an energy density kernel function and a non-uniform computing power distribution grid model to obtain the total wave energy. This invention utilizes the Riemannian geometric evolution mechanism to overcome matching failures caused by nonlinear dispersion effects, eliminates sampling feature tearing, and enhances the logical convergence during heterogeneous data fusion.

Description

Technical Field

[0001] This invention belongs to the field of electrical digital data processing technology, and in particular relates to a method for calculating the total wave energy based on multi-source fusion data. Background Technology

[0002] Currently, in the digital twin and energy flow scheduling process of wide-area ocean wave energy power generation arrays, wave observation sequences are collected using sensing terminals such as sea surface buoys, shore-based radar, and satellite altimeters. Because sea surface buoys provide low-frequency, high-precision point data, shore-based radar provides high-frequency, medium-precision surface data, and satellite altimeters provide large-scale data, these multi-source heterogeneous data sources have significant differences in spatial grid scale and temporal sampling frequency. This leads to asynchronous sampling phase tearing problems in the data fusion process. Existing fusion logic is limited by the differences in terminal physical performance and has limitations in processing mechanisms. For example, the transmission... Chinese invention patent CN118816825B discloses an adaptive window fusion method for capturing wave height in large wave processes. It captures peak wave height by adaptively adjusting the time resolution. However, it still relies on a spatiotemporal inverse distance weighted linear or pseudo-linear mapping mechanism. Wave propagation in the wide area has nonlinear dispersion characteristics. The alignment method based on Euclidean space weights cannot correct the tearing of heterogeneous data features from the underlying topology dimension. When dealing with transient changes or sea state distortions, it is easy to cause energy phase cancellation due to the non-physical attribute matching of high and low frequency feature points, which leads to deviations in the assessment of total wave energy.

[0003] Current data alignment logic mainly adopts the method of extracting local data extrema points to achieve time difference compensation. However, wave signals have a strong nonlinear dispersion effect in spatial propagation, which makes it difficult for the peak extrema points in high-frequency radar data streams and low-frequency buoy data streams to correspond to the same wave packet entity in the computational space. This rigid mapping mechanism that relies on a one-dimensional time axis produces phase cancellation of non-physical properties when dealing with wave transient changes, which in turn causes a significant deviation in the calculation of energy density tensor. The global error rate of wave energy total assessment is usually maintained at a high level of 30% to 40%.

[0004] Therefore, the technical problem to be solved by this invention is how to solve the problem of feature tearing of heterogeneous wave observation data under asynchronous sampling conditions and the problem of time mismatch under nonlinear dispersion environment, so as to achieve highly stable digital processing of total wave energy in a wide area of ​​the ocean. Summary of the Invention

[0005] This invention provides a method for calculating the total wave energy based on multi-source fusion data, comprising the following steps: Step S1: Obtain a heterogeneous sampling digital sequence that characterizes the hydrodynamic distribution features of the target computational domain. The heterogeneous sampling digital sequence includes a point source discrete sequence corresponding to local point-like physical properties, a surface array distribution sequence corresponding to two-dimensional planar field distribution, and a large-scale trend term sequence corresponding to the overall dynamics of the sea area. Step S2: Extract the digital sample stream of the heterogeneous sampled digital sequence within the local time sliding window, generate symmetric positive definite tensor nodes by calculating the local covariance matrix within the local time sliding window, and transform the statistical distribution characteristics of the heterogeneous sampled digital sequence into a high-dimensional feature space with Riemannian metric properties. Step S3: Project the symmetric positive definite tensor nodes onto the preset Riemannian manifold space, calculate the geodesic evolution gradient between the high-frequency tensor nodes originating from the area array distribution sequence and the low-frequency tensor nodes originating from the point source discrete sequence, so as to quantitatively characterize the evolution deviation of heterogeneous observation sources during spatial propagation. Step S4: Extract the tensor affine bias representing the differences in data distribution based on the geodesic evolution gradient, and use the tensor affine bias as the phase correction increment to guide the heterogeneous sampled digital sequence to remap the sampling time scale and fit it with the data distribution envelope, thereby completing the phase locking between the sequences in the high-dimensional feature space. Step S5: The phase-locked point source discrete sequence, area array distribution sequence, and large-scale trend term sequence are imported into the energy density kernel function and the non-uniform computing power distribution grid model for coupling calculation. The integral weight of each spatial node is dynamically adjusted through the non-uniform computing power distribution grid model to finally generate the total wave energy data of the target computing domain in the target time period.

[0006] Preferably, the process of generating symmetric positive definite tensor nodes in step S2 includes: determining the step size of the local time sliding window, extracting the amplitude discrete features of each sampling point within the step size range, mapping the amplitude discrete features into topological nodes in the form of symmetric positive definite matrices, wherein the trace operation result of the symmetric positive definite matrix is ​​used to define the overall distribution divergence of the data stream, and converting the absolute amplitude comparison of heterogeneous sampled digital sequences into a geometric comparison of local statistical features of the data.

[0007] Preferably, the process of guiding the heterogeneous sampled digital sequence to perform sampling timescale remapping in step S4 includes: establishing a tangent space mapping relationship in the Riemannian manifold space, determining the tensor affine bias by calculating the minimum geodesic distance between adjacent symmetric positive definite tensor nodes on the manifold surface, and using the tensor affine bias to perform nonlinear phase compensation on the array distribution sequence to eliminate feature tearing of the heterogeneous sampled digital sequence during asynchronous sampling.

[0008] Preferably, the spatial grid density adjustment of the non-uniform computing power allocation grid model in step S5 follows the following rules: ,in, For the adjusted spatial grid density, Here, λ is the preset baseline density, and λ is the preset adaptive adjustment coefficient. The magnitude of the energy gradient vector of the isostatic tensor flow is used; mesh subdivision interpolation is performed in regions where the magnitude of the energy gradient vector is greater than a preset gradient threshold, and mesh aggregation is performed in regions where the magnitude is not greater than the preset gradient threshold.

[0009] Preferably, the energy density kernel function is constructed by performing high-dimensional numerical integration on each sequence after phase locking. Furthermore, the energy density kernel function defines the node attributes of the non-uniform computing power distribution grid model by limiting the algebraic superposition relationship between the potential energy parameter and the kinetic energy parameter per unit area within the computing node, so that the numerical integration operation is concentrated on the high energy density distribution area.

[0010] Preferably, before step S3, the method further includes: extracting intrinsic modes from the heterogeneous sampled digital sequence using the empirical orthogonal function decomposition method, identifying and removing steady-state interference modes that characterize environmental background noise, so as to improve the feature purity of the symmetric positive definite tensor nodes entering the Riemannian manifold space, and recalculating the local covariance matrix based on the reconstructed sequence.

[0011] Preferably, a confidence decay threshold is set, and when the missing duration of the point source discrete sequence exceeds a preset time threshold, the manifold mapping dimensionality reduction logic is automatically triggered, and the current phase locking mode is switched to a time-scaled compensation mode based on local linear interpolation to maintain the logical continuity of the calculation process, and the geodesic evolution gradient is re-established within 5 seconds after the point source discrete sequence is recovered.

[0012] Preferably, the point source discrete sequence in the heterogeneous sampling digital sequence corresponds to the surface buoy sampling data, the area array distribution sequence corresponds to the shore-based radar echo characteristic data, and the large-scale trend term sequence corresponds to the satellite altimetry inversion data. Furthermore, each sequence undergoes standardized preprocessing through a preset data interface to eliminate the initial timescale deviation caused by differences in hardware protocols.

[0013] Preferably, the process of generating total wave energy data includes: globally accumulating the energy density values ​​of each grid node in the non-uniform computing power distribution grid model, and performing time dimension weighting according to the duration of the target period, outputting an energy value with the dimension of J, wherein the time dimension weighting is achieved by a cumulative multiplication operator of discrete time steps.

[0014] Preferably, phase locking is achieved by configuring a virtual synchronization clock reference in the computational space. The virtual synchronization clock reference consists of a highly stable timestamp counter, and the phase correction of the virtual synchronization clock reference is monotonically mapped to the tensor affine bias, which is used to correct the relative spatiotemporal position of each sampling sequence in the high-dimensional feature space.

[0015] Compared with existing technologies, the wave energy total calculation method based on multi-source fusion data of this invention has the following advantages: 1. In the calculation of total wave energy from multi-source fusion data, this invention solves the feature tearing problem caused by asynchronous sampling by constructing an energy topology manifold and mapping heterogeneous wave measurement data sequences to this manifold space from the data architecture level. This invention utilizes the manifold mapping mechanism to transform radar data and buoy data with inconsistent sampling frequencies into a unified tensor to be fused, thereby changing the phase cancellation phenomenon that traditional resampling algorithms are prone to when processing transient change signals. This ensures that multi-source heterogeneous signals maintain high fidelity in the digital fusion stage, laying a stable data foundation for the accurate flow of subsequent energy tensors.

[0016] 2. A tensor covariance alignment mechanism based on the evolution of Riemannian manifold geodesics is adopted to overcome the temporal matching failure caused by nonlinear dispersion effect in the propagation of physical signals in wide-area space. This invention calculates the geodesic distance of the symmetric positive definite tensor in the Riemannian geometric space and extracts the tensor affine bias. It reconstructs the traditional one-dimensional time axis point-to-point rigid matching into a high-dimensional space data distribution envelope fitting. Phase locking is achieved by using local statistical geometric invariants of the data, avoiding feature mismatch that is easy to occur when simply relying on signal extreme points for alignment, and improving the logical convergence and anti-disturbance stability of the calculation matrix under complex sea state interference.

[0017] 3. By relying on the deep coupling of the energy density kernel function and the non-uniform computing power allocation grid model, the invention achieves synergistic optimization of global computational efficiency and local analytical accuracy. The invention dynamically adjusts the spatial grid density according to the energy gradient vector of the equidistant tensor flow. In regions with gentle energy gradients, grid aggregation is performed to reduce redundant calculations, while in regions with drastic energy gradient fluctuations, subdivision interpolation is performed to enhance feature capture. This data-driven adaptive grid control method enables numerical integration calculations to be concentrated on high-value energy density regions, optimizing resource allocation efficiency in the electro-digital processing process while ensuring the accuracy of the total wave energy calculation. Attached Figure Description

[0018] Figure 1 This is a global flowchart of the wave energy total assessment and multi-source data fusion of the present invention; Figure 2 This is the logic diagram of asynchronous wave measurement sequence phase locking and tensor alignment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0021] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0022] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0023] A method for calculating the total wave energy based on multi-source fusion data includes the following steps: Step S1: Obtain a heterogeneous sampling digital sequence that characterizes the hydrodynamic distribution features of the target computational domain. The heterogeneous sampling digital sequence includes a point source discrete sequence corresponding to local point-like physical properties, a surface array distribution sequence corresponding to two-dimensional planar field distribution, and a large-scale trend term sequence corresponding to the overall dynamics of the sea area. Step S2: Extract the digital sample stream of the heterogeneous sampled digital sequence within the local time sliding window, generate symmetric positive definite tensor nodes by calculating the local covariance matrix within the local time sliding window, and transform the statistical distribution characteristics of the heterogeneous sampled digital sequence into a high-dimensional feature space with Riemannian metric properties. Step S3: Project the symmetric positive definite tensor nodes onto the preset Riemannian manifold space, calculate the geodesic evolution gradient between the high-frequency tensor nodes originating from the area array distribution sequence and the low-frequency tensor nodes originating from the point source discrete sequence, so as to quantitatively characterize the evolution deviation of heterogeneous observation sources during spatial propagation. Step S4: Extract the tensor affine bias representing the differences in data distribution based on the geodesic evolution gradient, and use the tensor affine bias as the phase correction increment to guide the heterogeneous sampled digital sequence to remap the sampling time scale and fit it with the data distribution envelope, thereby completing the phase locking between the sequences in the high-dimensional feature space. Step S5: The phase-locked point source discrete sequence, area array distribution sequence, and large-scale trend term sequence are imported into the energy density kernel function and the non-uniform computing power distribution grid model for coupling calculation. The integral weight of each spatial node is dynamically adjusted through the non-uniform computing power distribution grid model to finally generate the total wave energy data of the target computing domain in the target time period.

[0024] Preferably, the process of generating symmetric positive definite tensor nodes in step S2 includes: determining the step size of the local time sliding window, extracting the amplitude discrete features of each sampling point within the step size range, mapping the amplitude discrete features into topological nodes in the form of symmetric positive definite matrices, wherein the trace operation result of the symmetric positive definite matrix is ​​used to define the overall distribution divergence of the data stream, and converting the absolute amplitude comparison of heterogeneous sampled digital sequences into a geometric comparison of local statistical features of the data.

[0025] Preferably, the process of guiding the heterogeneous sampled digital sequence to perform sampling timescale remapping in step S4 includes: establishing a tangent space mapping relationship in the Riemannian manifold space, determining the tensor affine bias by calculating the minimum geodesic distance between adjacent symmetric positive definite tensor nodes on the manifold surface, and using the tensor affine bias to perform nonlinear phase compensation on the array distribution sequence to eliminate feature tearing of the heterogeneous sampled digital sequence during asynchronous sampling.

[0026] Preferably, the spatial grid density adjustment of the non-uniform computing power allocation grid model in step S5 follows the following rules: ,in, For the adjusted spatial grid density, Here, λ is the preset baseline density, and λ is the preset adaptive adjustment coefficient. The magnitude of the energy gradient vector of the isostatic tensor flow is used; mesh subdivision interpolation is performed in regions where the magnitude of the energy gradient vector is greater than a preset gradient threshold, and mesh aggregation is performed in regions where the magnitude is not greater than the preset gradient threshold.

[0027] Preferably, the energy density kernel function is constructed by performing high-dimensional numerical integration on each sequence after phase locking. Furthermore, the energy density kernel function defines the node attributes of the non-uniform computing power distribution grid model by limiting the algebraic superposition relationship between the potential energy parameter and the kinetic energy parameter per unit area within the computing node, so that the numerical integration operation is concentrated on the high energy density distribution area.

[0028] Preferably, before step S3, the method further includes: extracting intrinsic modes from the heterogeneous sampled digital sequence using the empirical orthogonal function decomposition method, identifying and removing steady-state interference modes that characterize environmental background noise, so as to improve the feature purity of the symmetric positive definite tensor nodes entering the Riemannian manifold space, and recalculating the local covariance matrix based on the reconstructed sequence.

[0029] Preferably, a confidence decay threshold is set, and when the missing duration of the point source discrete sequence exceeds a preset time threshold, the manifold mapping dimensionality reduction logic is automatically triggered, and the current phase locking mode is switched to a time-scaled compensation mode based on local linear interpolation to maintain the logical continuity of the calculation process, and the geodesic evolution gradient is re-established within 5 seconds after the point source discrete sequence is recovered.

[0030] Preferably, the point source discrete sequence in the heterogeneous sampling digital sequence corresponds to the surface buoy sampling data, the area array distribution sequence corresponds to the shore-based radar echo characteristic data, and the large-scale trend term sequence corresponds to the satellite altimetry inversion data. Furthermore, each sequence undergoes standardized preprocessing through a preset data interface to eliminate the initial timescale deviation caused by differences in hardware protocols.

[0031] Preferably, the process of generating total wave energy data includes: globally accumulating the energy density values ​​of each grid node in the non-uniform computing power distribution grid model, and performing time dimension weighting according to the duration of the target period, outputting an energy value with the dimension of J, wherein the time dimension weighting is achieved by a cumulative multiplication operator of discrete time steps.

[0032] Preferably, phase locking is achieved by configuring a virtual synchronization clock reference in the computational space. The virtual synchronization clock reference consists of a highly stable timestamp counter, and the phase correction of the virtual synchronization clock reference is monotonically mapped to the tensor affine bias, which is used to correct the relative spatiotemporal position of each sampling sequence in the high-dimensional feature space.

[0033] Example 1: In the energy management of a wide-area marine wave energy power generation array, the system acquires in real time heterogeneous sampled digital sequences that characterize the hydrodynamic distribution features of the target computational domain. The heterogeneous sampled digital sequences include discrete sequences of sea surface buoy point sources corresponding to local point-like physical properties and distribution sequences of shore-based radar arrays corresponding to two-dimensional planar field distributions. Under wave-state fluctuation conditions, the propagation of physical wave space exhibits a nonlinear dispersion effect, causing local elevation extrema in the high-frequency array distribution sequence and the low-frequency point source discrete sequence to be mapped to dissimilar data groups in the topological space. This leads to feature mismatch and energy phase cancellation of the computational matrix based on the time-series mapping of extrema detection. The control unit sets the step size of the local time sliding window, extracts the digital sample stream of the heterogeneous sampled digital sequence within the local time sliding window, extracts the amplitude discrete features of each sampling point within the step size range, calculates the local covariance matrix within the local time sliding window, generates symmetric positive definite tensor nodes, and defines the overall distribution divergence of the data stream based on the trace operation results of the symmetric positive definite matrix. The absolute amplitude comparison of the heterogeneous sampled digital sequences is converted into a geometric comparison of the local statistical features of the data in a high-dimensional feature space.

[0034] Symmetric positive definite tensor nodes are projected onto a predefined Riemannian manifold space to establish a tangent space mapping relationship. The minimum geodesic distance between high-frequency tensor nodes originating from the area array distribution sequence and low-frequency tensor nodes originating from the point source discrete sequence on the manifold surface is calculated. The tensor affine bias characterizing the data distribution difference is extracted along the shortest evolution gradient of the minimum geodesic distance. A virtual synchronization clock reference composed of a highly stable timestamp counter is configured in the computational space. The phase correction of the virtual synchronization clock reference exhibits a monotonically mapped relationship with the tensor affine bias. The nonlinear phase of the area array distribution sequence is compensated based on the tensor affine bias, guiding the heterogeneous sampled digital sequence to remap the sampling time scale and fit the data distribution envelope. The phase between each sequence is locked in the Riemannian manifold space. The phase constraint is maintained according to the evolution gradient corresponding to the minimum geodesic distance, and the phase-aligned equidistant tensor flow is output.

[0035] The phase-locked point source discrete sequence, the area array distribution sequence, and the large-scale trend term sequence representing the overall dynamics of the sea area are imported into the energy density kernel function and coupled within the non-uniform computing power distribution grid model, according to the formula. Adjust the spatial grid density of the non-uniform computing power distribution grid model, where, Characterizing the adjusted spatial grid density, λ represents the preset baseline density, and λ represents the adaptive adjustment coefficient. The magnitude of the energy gradient vector representing the isometric tensor flow is used to output a high-density subgrid for subdivision interpolation in the corresponding region if the magnitude is greater than a preset gradient threshold, and to aggregate and output a low-density subgrid in the corresponding region if the magnitude is not greater than the preset gradient threshold, according to the formula. Calculate the spatial energy density grid matrix, where, The wave energy density scalar represents the energy density of the grid cell in the i-th row and j-th column of the spatial energy density grid matrix, K represents the constant correction coefficient, ρ represents the fluid density constant, and g represents the gravitational acceleration constant. Characterizes the root mean square eigenvalues ​​of the wavefront extracted from the isometric tensor flow. The energy cycle feature value extracted from the equidistant tensor flow is characterized. The wave energy density scalar of each grid cell in the non-uniform computing power distribution grid model is accumulated globally in space. The time dimension is added as a weighting operator based on the duration of the target time period and the discrete time step, and the total wave energy data is output. The energy management device extracts the energy flow scheduling command parameters for the power generation array nodes based on the total wave energy data, and drives the underlying relays and power distribution units to operate based on the energy flow scheduling command parameters.

[0036] Example 2: For physical scenarios where sea state distortion induces extreme wave states accompanied by electromagnetic interference, the technical challenge of the wave energy calculation system shifts to how to filter out clutter and lock the true phase in the underlying spatial topology mapping of asynchronous heterogeneous digital sequences. A verification platform is built based on a measured physical dataset containing shore-based X-band radar echo sequences and similarly scaled sea surface buoy telemetry sequences. To simulate the interference boundary of an industrial electromagnetic environment, the verification platform superimposes Gaussian white noise with a signal-to-noise ratio of 15 dB and low-frequency swell interference harmonics at a frequency of 0.5 Hz onto the input array distribution sequence, constructing a raw input data stream containing sea clutter characteristics. To ensure the stability of feature extraction, the system sets a cyclic FIFO buffer with 256 sampling points, a fixed sampling frequency of 10 Hz, and a sliding window overlap rate of 50%. That is, the tensor node update calculation is triggered once every 1.25s to prevent Nyquist aliasing distortion of high-frequency extreme points during nonlinear mapping. In the preprocessing stage of this data stream, a local time sliding window step size is set. The logic for the value of this step size is to balance the physical fidelity of the statistical distribution characteristics of the tensor nodes with the computational load of energy grid space matrix reconstruction. The internal judgment rule is set so that when the spectral bandwidth broadening rate of the heterogeneous sampled digital sequence is greater than the preset bandwidth broadening threshold, the step size tends to the lower limit of the value range to prevent Nyquist aliasing distortion of high-frequency extreme points. Based on this rule, under the specific working condition that the spectral bandwidth broadening rate is twice the base bandwidth, the step size of the local time sliding window is determined to be 2.5s. This parameter is set to extract the amplitude discrete features and generate the underlying time physical slice determined by the symmetric positive definite tensor nodes.

[0037] Parallel calculations were conducted using a control group employing a one-dimensional local extremum point detection time-series mapping mechanism and an experimental group employing a tensor covariance alignment mechanism. In the initial stage without phase alignment algorithms, the transient phase difference between the peaks of the discrete sequence of point sources and the distributed sequence of the area array in the original noisy data stream was 42.5°, exhibiting nonlinear dispersion misalignment and asynchronous feature tearing. The control group, applying extremum-based time-series alignment, experienced a phase difference convergence to 38.6° and topological distortion of the high-frequency data stream due to Gaussian white noise and surge harmonics disrupting the absolute amplitude anchor point of a single point source. The experimental group used a step size of [missing information]. A symmetric positive definite tensor node was generated from a 2.5s sample stream and projected onto a predefined Riemannian manifold space to establish a tangent space mapping. The tensor affine bias was extracted by evolution along the minimum geodesic distance. This tensor affine bias was used to provide nonlinear phase compensation for the array distribution sequence. Under conditions containing 15dB noise disturbance, the sampling time scale was forcibly remapped. The phase difference between the two types of sequences in the equidistant tensor stream output by the experimental group was reduced to 4.1°. This data phenomenon confirms that the phase locking mechanism based on the geometric comparison of local statistical divergence in the high-dimensional feature space avoids the physical fragility of the rigid alignment of the underlying time series.

[0038] To verify the reliable boundary of the spatial grid density adjustment rule in the non-uniform computing power allocation grid model, the experimental group set four quantization gradient samples of 0.1, 0.5, 1.2, and 1.5 for the adaptive adjustment coefficient λ to conduct nonlinear effect tests. The test input source used the equidistant tensor flow output after phase locking, as described above. The adjustment rule was based on the formula... Execution, among which, Characterizing the adjusted spatial grid density, λ represents the preset baseline density, and λ represents the adaptive adjustment coefficient. The magnitude of the energy gradient vector representing the equidistant tensor flow is used to evaluate the trend of the deviation rate between the output total wave energy value and the standard calibration value. Data shows that when the adaptive adjustment coefficient λ is set to the lower limit of 0.1, the adaptive response of the spatial grid density to the energy gradient of the equidistant tensor flow is sluggish, failing to generate subdivided interpolation subgrids in energy-dense regions, resulting in a wave energy calculation deviation rate of 18.4%. When the adaptive adjustment coefficient λ is increased to the normal median of 0.5 and the absolute upper limit of 1.2, respectively, the system aggregates and outputs high-density and low-density grids in the corresponding regions based on the magnitude of the energy gradient vector, reducing the calculation deviation rate to 3.2% and 2.8%, respectively. When the over-range control state of 1.5 is reached, the grid density evolution trend shows a sudden deterioration. The non-uniform computing power allocation grid model enters the overload zone of high-frequency fragmentation reconstruction. The bottom layer processing delay increases to 450ms and causes data packet loss. The deviation rate of the total wave energy calculation rebounds to 9.5%. The gradient verification data establishes that the adaptive adjustment coefficient λ in the range of 0.1 to 1.2 constructs a working range that ensures the dynamic balance between computing accuracy and computing power. The above data flow results containing multi-dimensional variable constraints show that the geodesic evolution of symmetric positive definite tensor nodes in the Riemannian manifold space is the core mechanism. Combined with the non-uniform computing power allocation grid model controlled by strict threshold boundaries, the digital fusion of multi-source wave observation data is transformed from local amplitude anchoring to topological statistical geometric alignment. This mechanism eliminates the digital splicing tearing phenomenon of asynchronous heterogeneous sampling digital sequences in electromagnetic and nonlinear dispersion environments. A definite causal transmission relationship is established from the steady-state extraction of the bottom layer hydrodynamic distribution characteristics to the dimensional convergence of the top layer total wave energy.

[0039] Example 3: When the system faces the characteristic mismatch of heterogeneous sampled digital sequences caused by wave-like fluctuations, the algorithm module of the wave energy total calculation system determines the sample capacity parameter of the local time sliding window. This parameter value is based on the wave fundamental frequency period of the target computational domain and the processor's concurrent operation limit. The algorithm module extracts the fundamental frequency parameter of the digital sequence output by the signal acquisition device and sets the sample capacity to the total number of discrete sampling points covering three fundamental frequency periods. Under the condition that the wave fundamental frequency is 0.1Hz and the sampling rate is 10Hz, the algorithm module determines that the sample capacity is 300 discrete sampling points. The algorithm module extracts data segments of the heterogeneous sampled digital sequence within this local time sliding window, targeting data that is in a single-dimensional time sequence. For a discrete sequence of point sources with column distribution, the algorithm module calculates the autocorrelation function and extracts the time interval corresponding to the first zero crossing as the delay time constant. Based on the phase space reconstruction principle, it extracts continuously delayed sampling points to generate multidimensional column vectors, so that the dimensionality-reduced single-point source observation data and the two-dimensional planar array distribution sequence have the same column vector feature dimension. It constructs an original observation matrix containing time dimension and observation dimension features. The algorithm module extracts the mean vector of each column sequence of the original observation matrix. It calculates the mean-reduced feature matrix by subtracting the corresponding mean vector from the original observation matrix. The algorithm module calculates the product of the mean-reduced feature matrix and its transpose matrix. The product result is divided by the sample size minus a constant difference to generate a symmetric positive definite tensor node.

[0040] The algorithm module projects the generated symmetric positive definite tensor nodes onto the Riemannian manifold space. It selects low-frequency tensor nodes originating from a discrete sequence of point sources as the reference base points for the manifold tangency. The module then performs matrix logarithmic operations to project high-frequency tensor nodes originating from a distributed array sequence onto a local tangency space established using this reference base point. Within this local tangency space, the module calculates the Frobenius norm of the projection point coordinate matrix and outputs the minimum geodesic distance between the high-frequency and low-frequency tensor nodes. The module also calculates the directional derivative of the projection point coordinate matrix and extracts it as the geodesic evolution gradient. Finally, the module modifies the geodesic evolution gradient with a preset constant learning rate. The product of the sub-subs is set as the tensor affine bias. The virtual synchronous clock reference receives the tensor affine bias. The algorithm module decomposes the tensor affine bias eigenvalues ​​into a symmetric positive definite matrix and extracts the largest real eigenvalue of the matrix as a dimensionless scalar representing the geometric mapping divergence. The algorithm module calculates the mathematical product of this dimensionless scalar and the sampling period of the underlying hardware of the target computation domain to generate an absolute time offset with a definite time dimension. The virtual synchronous clock reference establishes the absolute time offset as the phase correction quantity that is monotonically mapped proportionally to the tensor affine bias. The algorithm module translates the digital timestamps of the array distribution sequence according to the phase correction quantity and outputs the phase-locked equidistant tensor flow.

[0041] The algorithm module calculates the spatial partial derivatives of the isometric tensor flow along the spatial dimension and extracts their absolute values ​​as the magnitude of the energy gradient vector. The system synchronously extracts the maximum recorded energy gradient values ​​of global grid nodes within the target computational domain over the past 24-hour observation period, calculates the algebraic ratio of the current energy gradient vector magnitude to the maximum recorded energy gradient value, and generates a dimensionless relative gradient magnitude. The algorithm module directly replaces the absolute magnitude parameter in subsequent computing power allocation calculations with the relative gradient magnitude, converging the numerical divergence under extreme environmental conditions to a bounded closed interval with an upper limit of 1. The algorithm module then transmits a preset baseline density to the non-uniform computing power allocation grid model. And the adaptive adjustment coefficient λ, the algorithm module according to the formula Calculate the adjusted spatial grid density The algorithm module is based on the output spatial grid density. Within the target computational domain, a spatial energy density grid matrix is ​​reconstructed. Based on the fundamental equations of wave hydrodynamics, the energy density of fluid micro-elements is expressed as the scalar sum of the kinetic energy of particle orbital motion and the potential energy of wave surface deformation. The algorithm module calculates the global variance of the equidistant tensor flow sequence to extract potential energy parameters and simultaneously calculates the first-order time difference root mean square of the sequence to extract kinetic energy parameters. The model limits the algebraic superposition of potential energy parameters and kinetic energy parameters within the computational nodes as the background initial state. The superposition result is used to directly define the node attributes of the non-uniform computing power allocation grid model, driving the macroscopic energy density integral operation based on the root mean square eigenvalue in subsequent steps. The algorithm module outputs equidistant tensor flow entities with absolute phase alignment references based on the aforementioned computational link, eliminating the data alignment deviation state caused by the fixed empirical threshold setting, and providing deterministic reference input data for the pre-computation stage of the non-uniform computing power allocation grid model coupling operation.

[0042] Example 4: In the application scenario where the wave energy total calculation system is first deployed in a target sea area, the wave energy total calculation system acquires a continuous 24-hour historical heterogeneous sampling digital sequence of the target sea area under stable sea conditions to construct an offline calibration dataset. It extracts the digital sample stream from the offline calibration dataset and calculates the benchmark covariance matrix set. It then calculates the eigenvalues ​​of each tensor node within the benchmark covariance matrix set and constructs a diagonal matrix. The affine transformation result of the diagonal matrix is ​​established as the initial metric tensor of the preset Riemannian manifold space. The system calculates the benchmark energy gradient vector magnitude of the isometric tensor flow in the offline calibration dataset, extracts the time distribution variance of the benchmark energy gradient vector magnitude, and, based on the hydrodynamic grid convergence criterion, sets the preset benchmark density... The ratio of the maximum to minimum value of the trace in the benchmark covariance matrix set is extracted and set as the adaptive adjustment coefficient λ of the non-uniform computing power distribution grid model. The above offline calibration step establishes the spatial curvature of the Riemann manifold space and the initial operation parameters of the grid model to match the hydrodynamic background characteristics of the target sea area. When the wave energy total calculation system faces the long-term alternation of wave states, the wave energy total calculation system monitors the time change rate of the minimum geodesic distance between high-frequency tensor nodes originating from the area array distribution sequence and low-frequency tensor nodes originating from the point source discrete sequence on the Riemann manifold surface. The algorithm module extracts the magnitude of the energy gradient vector of the equidistant tensor flow within the past 1-hour time sliding window. The root mean square value of the modulus sequence is calculated, and the product of the root mean square value and the preset dimensionless constant residual is established as the updated preset gradient threshold. Based on the updated preset gradient threshold, the non-uniform computing power allocation grid model is instructed to output a high-density subgrid with fine interpolation or an aggregated low-density grid in the corresponding region. The above dynamic debugging process suppresses the physical disturbance of the long-term accumulated error of multi-source wave observation data on the phase locking accuracy and the density of the non-uniform computing power allocation grid model. The wave energy total calculation system maintains the steady-state characteristics of the underlying spatial topology mapping of the heterogeneous sampling digital sequence during long-term operation.

[0043] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A wave energy total amount calculation method based on multi-source fusion data, characterized in that, Includes the following steps: Step S1: Obtain a heterogeneous sampling digital sequence that characterizes the hydrodynamic distribution features of the target computational domain. The heterogeneous sampling digital sequence includes a point source discrete sequence corresponding to local point-like physical properties, a surface array distribution sequence corresponding to two-dimensional planar field distribution, and a large-scale trend term sequence corresponding to the overall dynamics of the sea area. Step S2: Extract the digital sample stream of the heterogeneous sampled digital sequence within the local time sliding window, generate symmetric positive definite tensor nodes by calculating the local covariance matrix within the local time sliding window, and transform the statistical distribution characteristics of the heterogeneous sampled digital sequence into a high-dimensional feature space with Riemannian metric properties. Step S3: Project the symmetric positive definite tensor nodes onto the preset Riemannian manifold space, calculate the geodesic evolution gradient between the high-frequency tensor nodes originating from the area array distribution sequence and the low-frequency tensor nodes originating from the point source discrete sequence, so as to quantitatively characterize the evolution deviation of heterogeneous observation sources during spatial propagation. Step S4: Extract the tensor affine bias representing the differences in data distribution based on the geodesic evolution gradient, and use the tensor affine bias as the phase correction increment to guide the heterogeneous sampled digital sequences to remap the sampling time scale and fit the data distribution envelope, thereby completing phase locking between sequences in the high-dimensional feature space. Step S5: The phase-locked point source discrete sequence, area array distribution sequence, and large-scale trend term sequence are imported into the energy density kernel function and the non-uniform computing power distribution grid model for coupling calculation. The integral weight of each spatial node is dynamically adjusted through the non-uniform computing power distribution grid model to finally generate the total wave energy data of the target computing domain in the target time period.

2. The wave energy total amount calculation method based on multi-source fusion data according to claim 1, characterized in that, The process of generating symmetric positive definite tensor nodes in step S2 includes: determining the step size of the local time sliding window, extracting the amplitude discrete features of each sampling point within the step size range, mapping the amplitude discrete features into topological nodes in the form of symmetric positive definite matrices, wherein the trace operation result of the symmetric positive definite matrix is ​​used to define the overall distribution divergence of the data stream, and converting the absolute amplitude comparison of heterogeneous sampled digital sequences into the geometric comparison of local statistical features of the data.

3. The wave energy total amount calculation method based on multi-source fusion data according to claim 1, characterized in that, The process of guiding the heterogeneous sampled digital sequence to remap the sampling time scale in step S4 includes: establishing a tangent space mapping relationship in the Riemannian manifold space, determining the tensor affine bias by calculating the minimum geodesic distance between adjacent symmetric positive definite tensor nodes on the manifold surface, and using the tensor affine bias to perform nonlinear phase compensation on the array distribution sequence to eliminate feature tearing of the heterogeneous sampled digital sequence in the asynchronous sampling process.

4. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, The spatial grid density adjustment of the non-uniform computing power distribution grid model in step S5 follows the following rules: wherein, is the adjusted spatial grid density, is a preset reference density, and λ is a preset adaptive adjustment coefficient, is the modulus value of the energy gradient vector of the equidistant tensor flow; grid subdivision interpolation is performed in the region where the modulus value of the energy gradient vector is greater than a preset gradient threshold, and grid aggregation is performed in the region where the modulus value is not greater than the preset gradient threshold.

5. The wave energy total amount calculation method based on multi-source fusion data according to claim 4, characterized in that, The energy density kernel function is constructed by performing high-dimensional numerical integration on each phase-locked sequence. Furthermore, the energy density kernel function defines the node attributes of the non-uniform computing power distribution grid model by limiting the algebraic superposition relationship of potential energy parameters and kinetic energy parameters per unit area within the computing node, so that the numerical integration operation is concentrated on the high energy density distribution region.

6. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, Before step S3, the process also includes: extracting intrinsic modes from heterogeneous sampled digital sequences using the empirical orthogonal function decomposition method, identifying and removing steady-state interference modes that characterize environmental background noise, so as to improve the feature purity of symmetric positive definite tensor nodes entering the Riemannian manifold space, and recalculating the local covariance matrix based on the reconstructed sequence.

7. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, Set a confidence decay threshold. When the missing duration of the point source discrete sequence exceeds a preset time threshold, automatically trigger the manifold mapping dimensionality reduction logic, switch the current phase locking mode to the time-scaled compensation mode based on local linear interpolation to maintain the logical continuity of the calculation process, and re-establish the geodesic evolution gradient within 5 seconds after the point source discrete sequence is recovered.

8. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, The point source discrete sequence in the heterogeneous sampling digital sequence corresponds to the surface buoy sampling data, the area array distribution sequence corresponds to the shore-based radar echo characteristic data, and the large-scale trend term sequence corresponds to the satellite altimetry inversion data. Furthermore, each sequence undergoes standardized preprocessing through a preset data interface to eliminate the initial timescale deviation caused by differences in hardware protocols.

9. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, The process of generating total wave energy data includes: global spatial accumulation of the energy density values ​​of each grid node in the non-uniform computing power distribution grid model, and time dimension weighting according to the duration of the target period, outputting an energy value with the dimension of J, wherein the time dimension weighting is achieved by the cumulative multiplication operator of discrete time steps.

10. The method for calculating the total wave energy based on multi-source fusion data according to claim 1, characterized in that, Phase locking is achieved by configuring a virtual synchronization clock reference in the computational space. The virtual synchronization clock reference consists of a highly stable timestamp counter, and the phase correction of the virtual synchronization clock reference is monotonically mapped to the tensor affine bias, which is used to correct the relative spatiotemporal position of each sampling sequence in the high-dimensional feature space.

Images