Sea ice thickness prediction method and device, storage medium and electronic equipment

Abstract

This disclosure relates to the field of marine photovoltaic technology, specifically to a sea ice thickness prediction method, a sea ice thickness prediction device, a storage medium, and electronic equipment. The sea ice thickness prediction method includes acquiring modal data; the modal data includes pile foundation data, ice anchor data, and meteorological data; each modal data is feature-encoded to obtain feature data, and the weights of each feature data are dynamically adjusted based on the current sea ice density to perform feature fusion to obtain cross-modal spatiotemporal fusion features; the cross-modal spatiotemporal fusion features are projected onto a physical parameter space to obtain physical constraint features, and the predicted sea ice thickness is calculated based on the physical constraint features. The sea ice thickness prediction method provided by this disclosure can improve the prediction accuracy of sea ice thickness under marine photovoltaic backgrounds.

Description

Technical Field

[0001] This disclosure relates to the field of marine photovoltaic technology, specifically to sea ice thickness prediction methods, sea ice thickness prediction devices, storage media, and electronic equipment. Background Technology

[0002] Driven by energy security concerns, photovoltaic power generation has ushered in a new era of development. In recent years, as land resources with good sunlight conditions have gradually decreased, floating photovoltaic (referring to freshwater) power generation systems in rivers, lakes, and reservoirs have developed rapidly.

[0003] Offshore photovoltaic (PV) systems are becoming a research hotspot due to their advantages such as abundant water surface resources, unobstructed views, high light reflection, reduced temperature rise losses, proximity to load centers, and low dust levels. However, offshore PV systems face complex hydrological and climatic environments, especially in winter when sea ice and floating ice appear in the Bohai Sea, increasing the operational risks of pile foundations and making them susceptible to damage from ice loads. Therefore, accurate prediction of sea ice thickness is crucial.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this disclosure is to provide a method for predicting sea ice thickness, which aims to solve the problem of low accuracy in predicting sea ice thickness under marine photovoltaic background.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to one aspect of this disclosure, a method for predicting sea ice thickness is provided, comprising:

[0008] Modal data is collected; the modal data includes pile foundation data, ice anchor data, and meteorological data.

[0009] Each modal data is feature-encoded to obtain each feature data, and the weight of each feature data is dynamically adjusted based on the current sea ice density to perform feature fusion on each feature data to obtain cross-modal spatiotemporal fusion features;

[0010] The cross-modal spatiotemporal fusion features are projected onto the physical parameter space to obtain physical constraint features, and the predicted sea ice thickness is calculated based on the physical constraint features.

[0011] Optionally, the pile foundation data includes pile foundation strain data and tilt sensor data, and the acquired modal data includes:

[0012] The distributed fiber optic strain sequence of the pile foundation is sampled according to a preset sampling rate, and the strain extrema, strain gradient, and bending mode frequency of the distributed fiber optic strain sequence are extracted as the pile foundation strain data; and

[0013] The tilt angle of the pile foundation in three orthogonal directions is measured in real time, and the tilt angle change rate and equivalent rotation angle are calculated based on the tilt angle as the tilt angle sensor data.

[0014] Optionally, the ice anchor data includes sonar point cloud data and ice anchor growth parameter data, and the acquired modal data includes:

[0015] The DBSCAN clustering algorithm was used to segment the ice anchor point cloud, and the volume, anchoring depth, distance from the pile foundation, and contact area of ​​the ice anchor point cloud were extracted as sonar point cloud data; and

[0016] Ice anchor growth parameters were estimated using Kalman filtering.

[0017] Optionally, the meteorological data includes basic data and enhanced data, and the acquired modal data includes:

[0018] Temperature data, wind speed data, and radiation data were collected respectively to serve as the aforementioned basic data; and

[0019] The wind chill index is calculated based on the temperature data and the wind speed data, and is used as the enhanced data.

[0020] Optionally, the feature data includes pile foundation strain time-series features, ice anchor geometric time-series features, and meteorological time-series features. The step of performing feature encoding on each of the modal data to obtain each feature data includes:

[0021] The pile foundation data is extracted using a pile foundation data encoder; the pile foundation data encoder includes a convolutional layer, a pooling layer, and a normalization layer.

[0022] The ice anchor geometric temporal features of the ice anchor data are extracted using an ice anchor data encoder; the ice anchor data encoder is a point cloud-specific Transformer, and the point cloud-specific Transformer includes an attention mechanism.

[0023] The meteorological data time-series features are extracted using a meteorological data encoder; the meteorological data encoder is a standard Transformer.

[0024] Optionally, the step of dynamically adjusting the weights of each feature data based on the current sea ice concentration to perform feature fusion on the feature data to obtain cross-modal spatiotemporal fusion features includes:

[0025] Configure the basic weights corresponding to each of the aforementioned feature data;

[0026] Real-time acquisition of the current density of sea ice;

[0027] When the current density exceeds a preset threshold density, the basic weights are dynamically adjusted based on the adjustment coefficient to obtain the adjustment weights corresponding to each feature data.

[0028] The cross-modal spatiotemporal fusion features are calculated based on the adjusted weights.

[0029] Optionally, the physical constraint features include: energy balance equation regularization and Hamiltonian mechanical constraints.

[0030] According to a second aspect of this disclosure, a sea ice thickness prediction device is provided, comprising:

[0031] The acquisition module is used to acquire modal data; the modal data includes pile foundation data, ice anchor data, and meteorological data.

[0032] The fusion module is used to encode the features of each modal data to obtain each feature data, and dynamically adjust the weight of each feature data based on the current density of sea ice to perform feature fusion on the feature data to obtain cross-modal spatiotemporal fusion features;

[0033] The prediction module is used to project the cross-modal spatiotemporal fusion features onto the physical parameter space to obtain physical constraint features, and calculate the predicted sea ice thickness based on the physical constraint features.

[0034] According to a third aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the sea ice thickness prediction method as described in the above embodiments.

[0035] According to a fourth aspect of the present disclosure, an electronic device is provided, characterized in that it includes: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the sea ice thickness prediction method as described in the above embodiments.

[0036] The exemplary embodiments disclosed herein may have some or all of the following beneficial effects:

[0037] In the technical solutions provided by some embodiments of this disclosure, multiple modal data, including pile foundation data, ice anchor data, and meteorological data, are collected as input. Then, feature extraction and fusion are performed on each modal data. During the fusion process, sea ice density is used as a dynamic attention mechanism, and physical constraint features are embedded to ultimately obtain a predicted sea ice thickness value. Using the above method for sea ice thickness prediction, on the one hand, it designs and extracts cross-modal features, especially capturing the mechanical coupling degree of the ice anchor-pile foundation, which can simultaneously capture temporal evolution and spatial distribution, improving prediction accuracy; on the other hand, it dynamically adjusts weights based on ice conditions, focusing on the characteristics of dense sea ice areas, which can further improve prediction accuracy.

[0038] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0039] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0040] Figure 1 The schematic diagram illustrates a flowchart of a sea ice thickness prediction method according to an exemplary embodiment of the present disclosure;

[0041] Figure 2 The illustration schematically shows a flowchart of a feature cross-modal spatiotemporal fusion method according to an exemplary embodiment of the present disclosure;

[0042] Figure 3 The schematic diagram illustrates a flowchart of a sea ice thickness prediction model in an exemplary embodiment of the present disclosure;

[0043] Figure 4 This schematic diagram illustrates the composition of a sea ice thickness prediction device according to an exemplary embodiment of the present disclosure;

[0044] Figure 5 The schematic diagram illustrates the structure of a computer system of an electronic device according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0045] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0046] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0047] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0048] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0049] The implementation details of the technical solutions of the embodiments of this disclosure are described in detail below.

[0050] Figure 1 The illustration schematically shows a flowchart of a sea ice thickness prediction method according to an exemplary embodiment of this disclosure. Figure 1 As shown, the sea ice thickness prediction method includes steps S101 to S103:

[0051] Step S101: Collect modal data; the modal data includes pile foundation data, ice anchor data, and meteorological data;

[0052] Step S102: Each modal data is feature-encoded to obtain each feature data, and the weight of each feature data is dynamically adjusted based on the current sea ice density to perform feature fusion on each feature data to obtain cross-modal spatiotemporal fusion features;

[0053] Step S103: Project the cross-modal spatiotemporal fusion features onto the physical parameter space to obtain physical constraint features, and calculate the predicted sea ice thickness based on the physical constraint features.

[0054] In the technical solutions provided by some embodiments of this disclosure, multiple modal data, including pile foundation data, ice anchor data, and meteorological data, are collected as input. Then, feature extraction and fusion are performed on each modal data. During the fusion process, sea ice density is used as a dynamic attention mechanism, and physical constraint features are embedded to ultimately obtain a predicted sea ice thickness value. Using the above method for sea ice thickness prediction, on the one hand, it designs and extracts cross-modal features, especially capturing the mechanical coupling degree of the ice anchor-pile foundation, which can simultaneously capture temporal evolution and spatial distribution, improving prediction accuracy; on the other hand, it dynamically adjusts weights based on ice conditions, focusing on the characteristics of dense sea ice areas, which can further improve prediction accuracy.

[0055] The following will describe in more detail each step of the sea ice thickness prediction method in this exemplary embodiment, with reference to the accompanying drawings and embodiments.

[0056] In step S101, modal data is collected; the modal data includes pile foundation data, ice anchor data, and meteorological data.

[0057] In one embodiment of this disclosure, the pile foundation data includes pile foundation strain data and tilt sensor data, and the acquisition of modal data includes:

[0058] The distributed fiber optic strain sequence of the pile foundation is sampled according to a preset sampling rate, and the strain extrema, strain gradient, and bending mode frequency of the distributed fiber optic strain sequence are extracted as the pile foundation strain data; and

[0059] The tilt angle of the pile foundation in three orthogonal directions is measured in real time, and the tilt angle change rate and equivalent rotation angle are calculated based on the tilt angle as the tilt angle sensor data.

[0060] Specifically, regarding the collection of pile foundation strain data:

[0061] High-frequency sampling is used to capture micro-strain fluctuations in the pile foundation. A sampling rate is set, and then a sliding window is used for time-series analysis to acquire a distributed fiber optic strain sequence. For example, a sampling rate of 1Hz and a 10-minute sliding window are used. After obtaining the distributed fiber optic strain sequence, feature extraction is performed to obtain the pile foundation strain data. The maximum strain value within the sliding window is extracted as the strain extremum (σ). max ), σ max It is used to monitor the ultimate stress state of pile foundation structures; the strain gradient (dσ / dz) is obtained by calculating the strain change rate along the depth direction of the pile foundation, and dσ / dz can reflect the deformation distribution characteristics of the structure; the first three main frequencies are extracted by Fast Fourier Transform (FFT) to obtain the bending mode frequency, and the bending mode frequency can be used to analyze the vibration characteristics and bending modes of the pile foundation.

[0062] For collecting tilt sensor data:

[0063] First, real-time measurements of the pile foundation in three orthogonal directions (θ) were taken. x ,θ y ,θ z The tilt angles were used to obtain a three-dimensional tilt angle sequence. Then, feature engineering methods were employed to calculate the tilt sensor data. The rate of change of tilt angle over time in each direction was calculated to obtain the tilt angle change rate (dθ / dt), which can be used to evaluate the dynamic stability of the pile foundation. The equivalent rotation angle (θ) was calculated using a formula. eq ), to comprehensively characterize the overall spatial tilt state of the pile foundation.

[0064]

[0065] Based on the above method, pile foundation strain data and tilt angle sensor data are used as pile foundation data for sea ice thickness prediction. This method can integrate pile foundation strain and tilt angle data to reveal the ice-pile mechanical coupling effect.

[0066] In one embodiment of this disclosure, the ice anchor data includes sonar point cloud data and ice anchor growth parameter data, and the acquired modal data includes:

[0067] The DBSCAN clustering algorithm was used to segment the ice anchor point cloud, and the volume, anchoring depth, distance from the pile foundation, and contact area of ​​the ice anchor point cloud were extracted as sonar point cloud data; and

[0068] Ice anchor growth parameters were estimated using Kalman filtering.

[0069] Specifically, for collecting sonar point cloud data:

[0070] First, data preprocessing is performed, specifically using the DBSCAN clustering algorithm to segment the sonar point cloud, separating the ice anchor point cloud from background noise. This method identifies the geometric structure of the ice anchors through density clustering and is suitable for non-uniformly distributed ice anchor point clouds.

[0071] The key to sonar point cloud segmentation lies in point cloud stitching and coordinate transformation, which unifies point clouds collected from multiple perspectives into the same coordinate system to ensure spatial consistency; noise separation and vegetation removal, which remove outliers (such as noise reflected by floating objects) through statistical filtering and thresholding; and data resampling to reduce point cloud density, improve processing efficiency, and preserve key morphological features of ice anchors.

[0072] Next, sonar point cloud data is collected based on the segmented ice anchor point cloud. The volume (V) of the space occupied by the ice anchor is calculated based on the 3D reconstruction of the point cloud; V reflects the overall size of the ice anchor. The anchoring depth (D) is obtained by measuring the vertical distance between the bottom of the ice anchor and the foundation plane of the pile; D can assess the mechanical strength of the ice anchor embedded in the pile. The shortest spatial distance between the ice anchor's centroid and the pile surface is calculated to obtain the distance to the pile (L); L can be used to analyze the impact risk of the ice anchor. The actual contact area between the ice anchor and the pile is estimated by fitting the point cloud surface (e.g., triangular meshing) to obtain the contact area (S); S can assess the local stress distribution.

[0073] For collecting ice anchor growth parameter data:

[0074] The Kalman filter algorithm is used to estimate the ice anchor growth rate (dh / dt) in real time, and the predicted value is dynamically updated by combining the physical model and observational data. Algorithm flow:

[0075] First, a dynamic model of ice anchor growth is established as the state equation;

[0076]

[0077] In the formula, w(k) represents the process noise.

[0078] Then, the observation equation is obtained by measuring the current ice thickness using sonar point cloud data;

[0079] z(k)=h(k)+v(k) (3)

[0080] In the formula, v(k) represents the measurement noise.

[0081] Finally, a prediction-correction loop is performed to estimate the current ice thickness and covariance matrix based on the state at the previous time step, and to update the estimated value using real-time observations. The Kalman gain matrix is ​​then calculated to optimize the estimation accuracy.

[0082] In one embodiment of this disclosure, the meteorological data includes basic data and enhanced data, and the acquired modal data includes:

[0083] Temperature data, wind speed data, and radiation data were collected respectively to serve as the aforementioned basic data; and

[0084] The wind chill index is calculated based on the temperature data and the wind speed data, and is used as the enhanced data.

[0085] The meteorological data includes basic data. Basic data may include temperature (T), which can be collected by deploying digital temperature sensors or infrared thermometers at different heights of the pile foundation. T directly affects the ice growth rate and phase change process. Basic data may also include wind speed (u), which can be collected using ultrasonic anemometers, cup anemometers, etc. During installation, the sensor orientation is adjusted to align with the prevailing wind direction. u reflects the intensity of airflow and is correlated with the evaporation and heat dissipation from the ice surface. Basic data may also include radiation (Q), which can be measured using a total radiation meter.

[0086] To fully utilize meteorological data, feature enhancement can be performed, which involves calculating enhanced data based on the basic data, such as the wind chill index.

[0087] The Wind Chill Index (WCI) can be used to comprehensively quantify the combined effect of low temperature and wind speed on human perception, indirectly reflecting the heat dissipation rate of the ice surface. Its calculation formula is as follows:

[0088] WCI = 13.12 + 0.6215T - 11.37u 0.16 +0.3965T·u 0.16 (4)

[0089] In the formula, T is the temperature in degrees Celsius (°C) and u is the wind speed in kilometers per hour (km / h).

[0090] It should be noted that meteorological data is inherently spatiotemporally dependent and constrained by physical laws. Spatiotemporal information can be explicitly injected into the model through an encoding mechanism to improve prediction accuracy and interpretability. Therefore, spatiotemporal location encoding can be performed on meteorological data.

[0091] The spatial coordinates (x, y) are constructed based on the latitude and longitude of the pile foundation. Even dimensions use a sine function, and odd dimensions use a cosine function. The timestamp encoding PE is:

[0092] PE (pos,2i) =sin(pos / 10000) 2i / d (5)

[0093] PE (pos,2i+1) =cos(pos / 10000) 2i / d (6)

[0094] In the formula, i is the dimension index, which controls the frequency of different dimensions; d is the total dimension of the position encoding, which is consistent with the model input embedding dimension.

[0095] Simultaneously, add a spatial distance weighting term:

[0096] PE spatial =exp(-γ·||x-x0||) (7)

[0097] In the formula, x is the coordinate vector of the current pile foundation, x0 is the coordinate of the reference point, γ is the attenuation coefficient, which controls the rate attenuation of the weight with distance, and ||x-x0|| is the Euclidean distance from x0.

[0098] In step S102, feature encoding is performed on each modal data to obtain each feature data, and the weight of each feature data is dynamically adjusted based on the current sea ice density to perform feature fusion on each feature data to obtain cross-modal spatiotemporal fusion features.

[0099] In one embodiment of this disclosure, the feature data includes pile foundation strain time-series features, ice anchor geometric time-series features, and meteorological time-series features. The step of performing feature encoding on each of the modal data to obtain each feature data includes:

[0100] The pile foundation data is extracted using a pile foundation data encoder; the pile foundation data encoder includes a convolutional layer, a pooling layer, and a normalization layer.

[0101] The ice anchor geometric temporal features of the ice anchor data are extracted using an ice anchor data encoder; the ice anchor data encoder is a point cloud-specific Transformer, and the point cloud-specific Transformer includes an attention mechanism.

[0102] The meteorological data time-series features are extracted using a meteorological data encoder; the meteorological data encoder is a standard Transformer.

[0103] The pile foundation data encoder takes pile foundation data as input, with a shape of [B,T,5], where B is the batch size, T is the time step, and 5 is the feature dimension, namely strain extrema, strain gradient, bending modal frequency, dip angle change rate, and equivalent rotation angle. The output of the pile foundation data encoder has a shape of [B,64,T], retaining the time dimension to support subsequent time-series modeling.

[0104] The structure of the pile foundation data encoder includes convolutional layers, pooling layers, and normalization layers. The processing procedure is as follows: first, a one-dimensional convolutional layer is used to extract local temporal features; then, a pooling layer is used to downsample and compress the temporal dimension; and finally, a normalization and stabilization training is performed through a normalization layer.

[0105] The input to the ice anchor data encoder is ice anchor data with a shape of [B, N, 4], where B is the batch size, N is the number of point clouds, and 4 is the feature dimension, namely volume, anchoring depth, distance from the pile foundation, and contact area. The output shape of the ice anchor data encoder is [B, 64, N], retaining the point cloud number dimension to support spatial analysis.

[0106] The ice anchor data encoder is a Point Transformer, a point cloud-specific Transformer that supports feature extraction from irregular point clouds. It captures spatial geometric relationships through a local neighborhood attention mechanism in the point cloud and performs dynamic feature aggregation to adapt to the diversity of ice anchor morphology.

[0107] The input to the meteorological data encoder is meteorological data in the shape [B,T,4], where B is the batch size, T is the time step, and 4 is the feature dimension, including temperature, wind speed, radiation, and wind chill index. The output shape is [B,T,64].

[0108] Feature encoding is performed separately by a hierarchical encoder to align pile foundation data, ice anchor data, and meteorological data, enabling it to support cross-modal fusion.

[0109] After obtaining the pile foundation strain time-series characteristics, ice anchor geometric time-series characteristics, and meteorological time-series characteristics through hierarchical encoding, all three feature data are input into the global spatiotemporal Transformer. The specific processing procedure is as follows:

[0110] First, the input feature data is dimensionally aligned. The dimensions of the input feature data are adjusted from [B,C,T] (batch × channel × time step) to [T,B,C] (time step × batch × channel). This is to adapt to the Transformer's requirements for the input sequence format, i.e., the sequence length dimension comes first, which facilitates the handling of dependencies between time steps.

[0111] Then, multi-source feature fusion is performed. Features from different data sources are stitched together along the channel dimension to fuse multimodal spatiotemporal data of pile foundations, ice anchors, and meteorology, preserving the independent information of each feature. Through the above dimensional adjustment and stitching, spatial features and time series can be unified into a spatiotemporal sequence input, conforming to the sequence processing paradigm of Transformer.

[0112] Next, dynamic attention gating is performed. Attention weights are generated through a gating network, which are then used to weight and adjust the fused features, adaptively selecting important spatiotemporal features and suppressing noise or redundant information. The gating network structure is: linear layer (192→64) → GELU activation → linear layer (64→1) → Sigmoid normalization. Compared to traditional multi-head self-attention, using a single-layer gating network reduces computational complexity and is suitable for multi-source heterogeneous feature fusion scenarios.

[0113] Figure 2 The illustration schematically shows a flowchart of a feature cross-modal spatiotemporal fusion method according to an exemplary embodiment of this disclosure. Figure 2 As shown, the method of dynamically adjusting the weights of each feature data based on the current sea ice concentration to perform feature fusion on the feature data to obtain cross-modal spatiotemporal fusion features includes:

[0114] Step S201: Configure the basic weights corresponding to each of the aforementioned feature data;

[0115] Step S202: Obtain the current concentration of sea ice in real time;

[0116] Step S203: When the current density exceeds the preset threshold density, the basic weights are dynamically adjusted based on the adjustment coefficient to obtain the adjustment weights corresponding to each feature data.

[0117] Step S204: Calculate the cross-modal spatiotemporal fusion features based on the adjusted weights.

[0118] Specifically, firstly, each feature data is assigned a corresponding basic weight w. base Then, real-time data on ice thickness and density in areas with dense ice anchors are obtained using monitoring equipment such as radar and drones as C. current This can be expressed as the ratio of the current number of ice anchors to the total number of ice anchors. When the density exceeds the threshold C... threshold When a high-risk, densely populated area is identified by the system, a weight adjustment mechanism is triggered. The formula is as follows:

[0119] w dense =w base +α·(C current -C threshold (8)

[0120] Among them, w dense For the adjusted weights, w base The base weights, α is the adjustment coefficient, and C current For the current density, C threshold For threshold density, C current Given the current density, the weight of the dense ice anchor area increases non-linearly with the density.

[0121] The adjustment coefficient α can be optimized by combining historical ice condition data. For example, the recent density fluctuation rate can be calculated by using the sliding window method, and α can be adjusted adaptively. The larger α is, the more significant the weight of the dense area is increased, reflecting a strong response to high-risk areas.

[0122] Based on the above method, through this dynamic weight adjustment mechanism for ice conditions, the system can adaptively focus on high-risk areas, balance the importance of global and local features, and meet the engineering requirements of dynamic ice condition monitoring.

[0123] Finally, Transformer feature transformation is performed. The weighted fusion features are processed using the standard Transformer architecture to obtain cross-modal spatiotemporal fusion features. Parameter configuration can be set to 6 encoder layers + 3 decoder layers, multi-head attention, 192 hidden layer dimensions, and a default feedforward network dimension of 2048. The encoder captures global spatiotemporal dependencies, while the decoder may be used to generate predicted sequences. Employing the standard Transformer architecture to capture global spatiotemporal dependencies is suitable for sequence-to-sequence prediction tasks, and the fewer decoder layers compared to the encoder reduce the risk of overfitting in downstream prediction task optimization.

[0124] Based on the above methods, a cross-modal feature fusion design is developed, especially to capture the mechanical coupling degree of ice anchor-pile foundation, which can simultaneously capture temporal evolution and spatial distribution, thereby improving the accuracy of prediction.

[0125] In step S103, the cross-modal spatiotemporal fusion features are projected onto the physical parameter space to obtain physical constraint features, and the predicted sea ice thickness is calculated based on the physical constraint features.

[0126] Specifically, after extracting multimodal spatiotemporal features from the global spatiotemporal Transformer and generating a high-dimensional representation, the Physics Projector maps this representation to the physical parameter space, such as ice density and thermal conductivity, establishing a bridge between deep learning and physical equations. The output of the projection layer directly corresponds to physical quantities, enhancing the credibility of the model.

[0127] The model incorporates physical constraints, including regularization of the energy balance equation and Hamiltonian mechanics constraints, to ensure that the model output conforms to the thermodynamic law of conservation of energy and the laws of mass conservation and momentum transport, thus balancing data-driven approaches with physical laws.

[0128] Among them, the energy balance equation is regularized:

[0129]

[0130] Hamiltonian constraints:

[0131]

[0132] Meanwhile, a parameter sharing mechanism is adopted, which uses the Physics Projector to map the 192-dimensional features output by the Transformer to physical parameters such as ice density and thermal conductivity, thereby achieving physical interpretability of the parameters.

[0133] Based on the above method, the physical equations are embedded into the Transformer inductive bias, thereby improving the interpretability of ice thickness prediction.

[0134] Then, based on the aforementioned physical constraints, the predicted sea ice thickness h is calculated. Additionally, a mean prediction μ and a variance estimate σ can be added. 2 The dual-headed output is used to characterize prediction uncertainty.

[0135] Specifically, different risk levels can be defined based on ice thickness, and corresponding response strategies can be formulated. After predicting sea ice thickness, responses can be triggered promptly according to the risk level. The risk level can be output as a probability using Softmax.

[0136] Table 1 Risk Levels and Response Strategies

[0137]

[0138] Figure 3 This illustration schematically shows a flowchart of a sea ice thickness prediction model according to an exemplary embodiment of this disclosure. Figure 3 As shown, after acquiring pile foundation data, ice anchor data, and meteorological data respectively, the corresponding hierarchical encoder Local Encoder is used for feature encoding. Finally, the encoded feature data is input into the global spatiotemporal GlobalTransformer to obtain the ice thickness prediction head.

[0139] Based on the above methods, and leveraging the powerful representation capabilities of the Transformer model, a physically interpretable ice thickness prediction system was constructed, providing an intelligent solution for pile-based offshore photovoltaic systems to cope with extreme ice conditions.

[0140] Figure 4 This schematic diagram illustrates the composition of a sea ice thickness prediction device according to an exemplary embodiment of the present disclosure, such as... Figure 4 As shown, the sea ice thickness prediction device 400 may include a data acquisition module 401, a fusion module 402, and a prediction module 403. Wherein:

[0141] The acquisition module 401 is used to acquire modal data; the modal data includes pile foundation data, ice anchor data, and meteorological data.

[0142] The fusion module 402 is used to perform feature encoding on each of the modal data to obtain each feature data, and dynamically adjust the weight of each feature data based on the current density of sea ice to perform feature fusion on the feature data to obtain cross-modal spatiotemporal fusion features;

[0143] The prediction module 403 is used to project the cross-modal spatiotemporal fusion features onto the physical parameter space to obtain physical constraint features, and calculate the predicted sea ice thickness based on the physical constraint features.

[0144] According to an exemplary embodiment of this disclosure, the pile foundation data includes pile foundation strain data and tilt sensor data. The acquisition module 401 is further configured to sample the distributed optical fiber strain sequence of the pile foundation according to a preset sampling rate, and extract the strain extreme values, strain gradients and bending mode frequencies of the distributed optical fiber strain sequence as the pile foundation strain data; and to measure the tilt angle of the pile foundation in three orthogonal directions in real time, and calculate the tilt angle change rate and equivalent rotation angle based on the tilt angle as the tilt sensor data.

[0145] According to an exemplary embodiment of this disclosure, the ice anchor data includes sonar point cloud data and ice anchor growth parameter data. The acquisition module 401 is further configured to segment the ice anchor point cloud using the DBSCAN clustering algorithm, and extract the volume, anchoring depth, distance from the pile foundation, and contact area of ​​the ice anchor point cloud as the sonar point cloud data; and to estimate the ice anchor growth parameter data through Kalman filtering.

[0146] According to an exemplary embodiment of this disclosure, the meteorological data includes basic data and enhanced data. The acquisition module 401 is further configured to acquire temperature data, wind speed data, and radiation data respectively as the basic data; and to calculate the wind chill index based on the temperature data and the wind speed data as the enhanced data.

[0147] According to an exemplary embodiment of this disclosure, the feature data includes pile foundation strain time-series features, ice anchor geometric time-series features, and meteorological time-series features. The fusion module 402 is further configured to extract the pile foundation strain time-series features of the pile foundation data using a pile foundation data encoder; the pile foundation data encoder includes a convolutional layer, a pooling layer, and a normalization layer; the ice anchor geometric time-series features of the ice anchor data are extracted using an ice anchor data encoder; the ice anchor data encoder is a point cloud-specific Transformer, which includes an attention mechanism; and the meteorological time-series features of the meteorological data are extracted using a meteorological data encoder; the meteorological data encoder is a standard Transformer.

[0148] According to an exemplary embodiment of this disclosure, the fusion module 402 is further configured to configure the basic weights corresponding to each of the feature data; obtain the current concentration of sea ice in real time; when the current concentration exceeds a preset threshold concentration, dynamically adjust the basic weights based on the adjustment coefficient to obtain the adjustment weights corresponding to each feature data; and calculate the cross-modal spatiotemporal fusion features based on the adjustment weights.

[0149] According to an exemplary embodiment of this disclosure, the physical constraint features include: energy balance equation regularization and Hamiltonian mechanical constraints.

[0150] The specific details of each module in the aforementioned sea ice thickness prediction device 400 have been described in detail in the corresponding sea ice thickness prediction method, so they will not be repeated here.

[0151] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0152] In exemplary embodiments of this disclosure, a storage medium capable of implementing the above-described methods is also provided. It may be a portable compact disc read-only memory (CD-ROM) and include program code, and can run on a terminal device, such as a mobile phone. However, the program product of this disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0153] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided. Figure 5 The schematic diagram illustrates the structure of a computer system of an electronic device according to an exemplary embodiment of the present disclosure.

[0154] It should be noted that, Figure 5 The computer system 500 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0155] like Figure 5 As shown, the computer system 500 includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 502 or programs loaded from storage section 508 into Random Access Memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0156] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0157] In particular, according to embodiments of this disclosure, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs various functions defined in the system of this disclosure.

[0158] It should be noted that the computer-readable medium shown in the embodiments of this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0160] The units described in the embodiments of this disclosure can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the unit itself.

[0161] In another aspect, this disclosure also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

[0162] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0163] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this disclosure.

[0164] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

[0165] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for predicting sea ice thickness, characterized in that, include: Modal data is collected; the modal data includes pile foundation data, ice anchor data, and meteorological data; the pile foundation data includes pile foundation strain data and tilt sensor data. The ice anchor data includes sonar point cloud data and ice anchor growth parameter data; the acquired modal data includes: The distributed fiber strain sequence of the pile foundation is sampled according to a preset sampling rate, and the strain extreme value, strain gradient and bending mode frequency of the distributed fiber strain sequence are extracted as the pile foundation strain data; and the tilt angle of the pile foundation in three orthogonal directions is measured in real time, and the tilt angle change rate and equivalent rotation angle are calculated based on the tilt angle as the tilt angle sensor data. The DBSCAN clustering algorithm was used to segment the ice anchor point cloud, and the volume, anchoring depth, distance from the pile foundation, and contact area of ​​the ice anchor point cloud were extracted as sonar point cloud data; and the ice anchor growth parameters were estimated by Kalman filtering. Each modal data is feature-encoded to obtain a feature data, and the weights of each feature data are dynamically adjusted based on the current sea ice concentration to perform feature fusion to obtain cross-modal spatiotemporal fusion features; the feature data includes pile foundation strain time series features, ice anchor geometric time series features, and meteorological time series features; The cross-modal spatiotemporal fusion features are projected onto the physical parameter space to obtain physical constraint features, and the predicted sea ice thickness is calculated based on the physical constraint features.

2. The sea ice thickness prediction method according to claim 1, characterized in that, The meteorological data includes basic data and enhanced data, and the collected modal data includes: Temperature data, wind speed data, and radiation data were collected respectively to serve as the aforementioned basic data; and The wind chill index is calculated based on the temperature data and the wind speed data, and is used as the enhanced data.

3. The sea ice thickness prediction method according to claim 1, characterized in that, The step of performing feature encoding on each of the modal data to obtain each feature data includes: The pile foundation data is extracted using a pile foundation data encoder; the pile foundation data encoder includes a convolutional layer, a pooling layer, and a normalization layer. The ice anchor geometric temporal features of the ice anchor data are extracted using an ice anchor data encoder; the ice anchor data encoder is a point cloud-specific Transformer, and the point cloud-specific Transformer includes an attention mechanism. The meteorological data time-series features are extracted using a meteorological data encoder; the meteorological data encoder is a standard Transformer.

4. The sea ice thickness prediction method according to claim 1, characterized in that, The method of dynamically adjusting the weights of each feature data based on the current sea ice concentration to perform feature fusion on the feature data to obtain cross-modal spatiotemporal fusion features includes: Configure the basic weights corresponding to each of the aforementioned feature data; Real-time acquisition of the current density of sea ice; When the current density exceeds a preset threshold density, the basic weights are dynamically adjusted based on the adjustment coefficient to obtain the adjustment weights corresponding to each feature data. The cross-modal spatiotemporal fusion features are calculated based on the adjusted weights.

5. The sea ice thickness prediction method according to claim 1, characterized in that, The physical constraints include: regularization of the energy balance equation and Hamiltonian mechanical constraints.

6. A sea ice thickness prediction device, characterized in that, include: The acquisition module is used to acquire modal data; the modal data includes pile foundation data, ice anchor data, and meteorological data; the pile foundation data includes pile foundation strain data and tilt sensor data; the ice anchor data includes sonar point cloud data and ice anchor growth parameter data; the acquisition module is used to: sample the distributed optical fiber strain sequence of the pile foundation according to a preset sampling rate, and extract the strain extrema, strain gradient, and bending mode frequency of the distributed optical fiber strain sequence as the pile foundation strain data; The system measures the tilt angle of the pile foundation in three orthogonal directions in real time, and calculates the tilt angle change rate and equivalent rotation angle as the tilt angle sensor data based on the tilt angle; it uses the DBSCAN clustering algorithm to segment the ice anchor point cloud, and extracts the volume, anchoring depth, distance from the pile foundation, and contact area of ​​the ice anchor point cloud as the sonar point cloud data; and it estimates the ice anchor growth parameters through Kalman filtering. The fusion module is used to encode the features of each modal data to obtain each feature data, and dynamically adjust the weight of each feature data based on the current sea ice concentration to perform feature fusion to obtain cross-modal spatiotemporal fusion features; the feature data includes pile foundation strain time series features, ice anchor geometric time series features, and meteorological time series features; The prediction module is used to project the cross-modal spatiotemporal fusion features onto the physical parameter space to obtain physical constraint features, and calculate the predicted sea ice thickness based on the physical constraint features.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the sea ice thickness prediction method as described in any one of claims 1 to 5.

8. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the sea ice thickness prediction method as described in any one of claims 1 to 5.

Images