A method and system for stable real-time data transmission over extremely far-shore distances

Abstract

This invention relates to the field of ultra-far-shore data transmission technology, and discloses a method and system for stable real-time data transmission over ultra-far-shore distances. The steps are as follows: 1. Acquire target transmission link data and extract dynamic fluctuation features containing signal strength distribution information and time delay jitter relationships. 2. Call a trained channel estimation network to perform signal localization processing on the transmission link data, generating a set of signal transmission paths containing channel type identifiers and phase coding mapping relationships. 3. Based on this set, call a waveform analysis network to perform multi-modal feature fusion on the signal strength distribution information, obtaining a fusion result containing time-frequency domain features and modulation mode features. 4. Input the fusion result into a transmission strategy model for dynamic path selection, outputting real-time data transmission elements containing control commands generated based on the signal transmission sequence and phase self-calibration parameters. This method can achieve stable real-time data transmission over ultra-far-shore distances.

Description

Technical Field

[0001] This invention relates to the field of offshore wind power technology, specifically to a method and system for stable real-time data transmission at extremely far offshore distances. Background Technology

[0002] With the rapid development of marine resource development, ocean shipping, and deep-sea exploration, the demand for real-time data transmission over extremely long distances from shore is becoming increasingly urgent. In such scenarios, data transmission faces extremely complex environmental challenges; transmission links are often subject to interference from various factors, making it difficult to guarantee the stability and real-time performance of data transmission.

[0003] In traditional offshore data transmission, commonly used satellite communication methods suffer from limited bandwidth and high transmission latency. Especially at ultra-long distances, signal attenuation is severe, and the data is susceptible to natural phenomena such as cloud cover and ionospheric fluctuations, leading to increased data packet loss rates. This makes it unsuitable for applications with high real-time requirements, such as real-time monitoring data transmission in deep-sea scientific research and dynamic navigation information exchange for ocean-going vessels.

[0004] While terrestrial microwave relay technology can achieve high-quality transmission over a certain distance, in ultra-far-shore scenarios, due to the curvature of the Earth, a large number of relay stations need to be set up. This not only results in high construction costs, but also makes the relay equipment susceptible to salt spray corrosion and wave impact in the marine environment, making maintenance extremely difficult. At the same time, the transmission distance is also significantly limited, making it difficult to cover vast offshore areas.

[0005] Existing wireless ad hoc network technologies face challenges in handling ultra-long-distance transmission, including limited node energy and drastic dynamic changes in topology. During transmission, signals pass through different types of channels, such as sea surface reflection channels and atmospheric scattering channels. These channels exhibit significantly different transmission characteristics, with varying signal attenuation rates and latency jitter, making traditional fixed-path transmission methods unable to adapt to dynamic channel changes.

[0006] Furthermore, during signal transmission, the presence of multipath effects causes phase shifts and interference. Traditional signal processing methods struggle to accurately extract and effectively fuse multimodal features, resulting in transmission strategy adjustments lagging behind channel changes and impacting data transmission stability. Simultaneously, current technologies often rely on static parameters for signal transmission path selection, failing to dynamically adjust based on real-time signal strength distribution and latency jitter. This leads to data transmission interruptions or errors in complex environments.

[0007] With the continuous increase in data transmission volume, the requirements for transmission efficiency and reliability are also becoming increasingly stringent. Existing data transmission methods are gradually revealing problems such as poor adaptability and weak anti-interference capabilities when dealing with complex environments at ultra-far-shore distances, failing to meet the diverse needs of practical applications. Therefore, developing a method that can adapt to complex environments at ultra-far-shore distances and achieve stable real-time data transmission has become an urgent problem to be solved in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for stable real-time data transmission over extremely far-shore distances, in order to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides a method for stable real-time data transmission over extremely far-shore distances, the method comprising:

[0010] Acquire target transmission link data and extract dynamic fluctuation characteristics from the transmission link data, which includes signal strength distribution information and time delay jitter relationship;

[0011] The trained channel estimation network is invoked to perform signal localization processing on the transmission link data, generating a set of signal transmission paths, which includes a mapping relationship between channel type identifiers and phase codes.

[0012] Based on the set of signal transmission paths, a waveform analysis network is invoked to perform multimodal feature fusion on the signal intensity distribution information, generating a feature fusion result that includes time-frequency domain features and modulation mode features;

[0013] The feature fusion result is input into the transmission strategy model for dynamic path selection processing, and the real-time data transmission element is output. The real-time data transmission element includes control commands and phase self-calibration parameters generated based on the signal transmission sequence.

[0014] Preferably, the step of calling the trained channel estimation network to perform signal localization processing on the transmission link data and generating a set of signal transmission paths includes:

[0015] The transmission link data is subjected to signal framing processing to generate multiple signal frame regions and corresponding time-frequency window initial parameters;

[0016] Extract the spectral feature map of each signal frame region, and perform energy attention weighting processing on the spectral feature map to generate a weighted time-domain feature vector;

[0017] The signal-to-noise ratio of the time-domain feature vector is calculated by comparing it with the preset path delay, and candidate paths that meet the reliability threshold are selected.

[0018] Multipath interference suppression processing is performed on the candidate paths to generate a set of signal transmission paths containing the channel type identifier. Each signal transmission path contains a normalized phase code relative to the transmission link data.

[0019] The step of fusing multimodal features of the signal intensity distribution information by calling a waveform analysis network based on the signal transmission path set, generating a feature fusion result containing time-frequency domain features and modulation mode features, includes:

[0020] Based on the normalized phase coding in the signal transmission path set, the transmission link data is subjected to regional segmentation processing to generate multiple signal segments;

[0021] Each signal segment is filtered and denoised to generate optimized waveform analysis input data;

[0022] The bidirectional gated recurrent subnet in the waveform analysis network is invoked to extract sequence features from the waveform analysis input data, generating a time-frequency domain feature vector.

[0023] The modulation classification model is invoked in parallel to perform carrier feature matching on the signal segment and generate a modulation mode probability distribution.

[0024] The time-frequency domain feature vector and the modulation mode probability distribution are concatenated using tensors to generate a multimodal feature fusion result.

[0025] Preferably, the step of inputting the feature fusion result into the transmission strategy model for dynamic path selection processing and outputting real-time data transmission elements includes:

[0026] The contextual dependencies of the current transmission scenario are analyzed to generate transmission priority weights associated with the multimodal feature fusion results;

[0027] A state transition matrix is ​​generated based on the transmission priority weights, and the channel type identifiers in the signal transmission path set are traversed and sorted.

[0028] The optimal transmission sequence is selected from the state transition matrix using a reinforcement learning strategy to generate a basic instruction set containing transmission, relay, and retransmission operations;

[0029] The basic instruction set and the phase self-calibration parameters are integrated into a protocol structure to generate an instruction code segment that conforms to the target transmission protocol.

[0030] Preferably, the method further includes:

[0031] During data transmission, link response data is captured in real time to generate a feedback log containing signal phase offset and status change information.

[0032] Extract the abnormal event features from the feedback log, and perform similarity matching between the abnormal event features and the historical transmission case library to generate adaptive adjustment instructions;

[0033] The path delay parameters of the channel estimation network are dynamically updated based on the adaptive adjustment instructions.

[0034] The updated path delay parameters are injected into the transmission strategy model, and the transmission priority weights in the state transition matrix are recalculated.

[0035] Preferably, the step of extracting abnormal event features from the feedback log and performing similarity matching between the abnormal event features and the historical transmission case library to generate adaptive adjustment instructions includes:

[0036] The abnormal event characteristics are divided into time windows to generate multiple event segments and their corresponding signal screenshot sequences;

[0037] The trained anomaly classification model is invoked to perform root cause analysis on each event segment, generating classification labels including signal recognition errors, phase drift, and transmission timeouts.

[0038] Retrieve solution templates that match the classification labels from the historical transmission case library to generate a set of candidate adjustment strategies;

[0039] Based on the matching degree ranking of the signal screenshot sequence and the candidate adjustment strategy set, the strategy with the highest confidence is selected to generate an adaptive adjustment instruction;

[0040] The call to the trained anomaly classification model performs root cause analysis on each event segment, generating classification labels including signal recognition errors, phase drift, and transmission timeouts, including:

[0041] The event segments are processed by time series segmentation to generate signal screenshot sequence slices containing the start trigger time and the end time;

[0042] The signal screenshot sequence slices are subjected to keyframe sampling processing to generate a set of keyframes for signal state changes and corresponding timestamp indices;

[0043] Extract the signal phase offset between adjacent key frames in the set of key frames for signal state changes, and generate a signal displacement trajectory vector and a time interval sequence;

[0044] The frequency offset of the signal displacement trajectory vector is calculated to generate an abnormal fluctuation mode feature vector.

[0045] The feature vector of the abnormal fluctuation pattern is input into the spatiotemporal convolutional subnet of the abnormal classification model for local feature extraction, generating a spatial abnormality activation map.

[0046] The time interval sequence is subjected to sliding window mean filtering to generate a smoothed response delay time sequence;

[0047] The response delay time series is input into the gated recurrent subnet of the anomaly classification model for periodic pattern matching to generate a time dimension anomaly score; channel max pooling is performed on the spatial anomaly activation map to generate a spatial dimension anomaly score; the spatial dimension anomaly score and the time dimension anomaly score are subjected to feature cross-fusion processing to generate a spatiotemporal joint anomaly probability distribution.

[0048] Peak detection results of signal identification error probability, phase drift probability, and transmission timeout probability are extracted from the spatiotemporal joint anomaly probability distribution; dynamic threshold comparison processing is performed on the preset probability thresholds based on the peak detection results to generate a candidate set of classification labels containing probability ranking;

[0049] The candidate classification labels are subjected to timestamp index alignment verification to generate target classification labels that are consistent with the phase of the abnormal events in the signal screenshot sequence slice; the target classification labels and the signal displacement trajectory vector are subjected to anomaly type reverse verification to generate a final classification label set containing confidence weights.

[0050] Preferably, the method further includes:

[0051] A virtual phase perturbation parameter is injected before data transmission, which is used to simulate a random phase shift scenario of the signal.

[0052] Monitor the fault-tolerant processing results of the transmission strategy model on the disturbed phase, and generate stability evaluation indicators;

[0053] When the stability evaluation index is lower than a preset threshold, the online learning mode of the channel estimation network is triggered;

[0054] The spectral characteristic parameters of the channel estimation network are updated by gradient backpropagation based on the phase difference data before and after the perturbation.

[0055] The monitoring of the transmission strategy model's fault-tolerant processing results for the disturbed phase generates stability evaluation indicators, including:

[0056] The number of positioning failures and the success rate of abnormal capture due to phase errors during the transmission process were statistically analyzed.

[0057] Calculate the ratio of the number of positioning failures to the total number of transmission steps to generate a first evaluation coefficient;

[0058] Extract the proportion of events that match the preset fault tolerance rules from the anomaly capture success rate, and generate a second evaluation coefficient;

[0059] The first evaluation coefficient and the second evaluation coefficient are weighted and summed to generate a comprehensive stability evaluation index.

[0060] Preferably, the step of updating the spectral characteristic parameters of the channel estimation network based on the phase difference data before and after the perturbation through gradient backpropagation includes:

[0061] The phase difference data before and after the perturbation are timestamped to generate a matching sequence containing the phase set before the perturbation and the phase set after the perturbation.

[0062] Extract the phase offset from the matching pair sequence to generate the horizontal offset vector and vertical offset vector for each signal segment;

[0063] The Euclidean distance between the horizontal offset vector and the vertical offset vector is calculated to generate a set of phase difference vectors for each signal segment;

[0064] A position regression loss function is constructed based on the set of phase difference vectors. The position regression loss function includes the mean square error between the predicted phase and the actual phase after perturbation by the channel estimation network.

[0065] The location regression loss function is subjected to a differentiable transformation to generate the loss value tensor required for gradient backpropagation;

[0066] Iterate through the spectral feature parameters of the channel estimation network, calculate the partial derivative of the loss tensor with respect to each spectral feature, and generate the spectral feature gradient matrix.

[0067] The spectral feature gradient matrix is ​​subjected to momentum smoothing to generate a decayed parameter update direction vector;

[0068] Based on the updated direction vector and the preset learning rate parameter, the spectral feature parameters of the channel estimation network are iteratively and incrementally adjusted.

[0069] During the incremental adjustment process, the rate of change of the location regression loss function is monitored in real time to generate a loss convergence status indicator.

[0070] When the loss convergence state indicator reaches a stable threshold, the gradient backpropagation update is terminated and the spectral feature parameters are frozen.

[0071] The updated spectral feature parameters are injected into the forward propagation path of the channel estimation network, and the signal framing and signal localization processes are re-executed to verify the improvement in phase localization accuracy.

[0072] Preferably, the method further includes:

[0073] Construct a cross-protocol transport adapter and use the cross-protocol transport adapter to parse the instruction syntax differences of different transport protocols;

[0074] The real-time data transmission elements are converted into underlying driver instructions supported by the target protocol;

[0075] The timing dependencies of the control instructions and the exception handling context are preserved during the conversion process;

[0076] Inject the performance optimization parameters that match the target protocol to generate transmission instruction bytecode that meets the cross-link execution conditions.

[0077] Preferably, the construction of the cross-protocol transport adapter and the parsing of instruction syntax differences between different transport protocols through the cross-protocol transport adapter include:

[0078] Establish a transport protocol syntax rule base and store the keyword mapping table and parameter passing path for each protocol;

[0079] The real-time data transmission elements are parsed using an abstract syntax tree to generate intermediate representation layer data;

[0080] Based on the intermediate presentation layer data, the keyword mapping table is traversed and queried to generate a protocol-compatible instruction replacement scheme;

[0081] Dependency injection is performed on conflicting parameter passing paths to generate unambiguous instruction conversion results;

[0082] The process of performing dependency injection on conflicting parameter passing paths to generate unambiguous instruction conversion results includes:

[0083] Identify conflicting nodes in the parameter passing path that have type ambiguity or overlapping scope, and generate a set of conflicting path identifiers;

[0084] Extract the context metadata of each node in the set of conflict path identifiers. The context metadata includes parameter type declarations, lifecycle markers, and call stack fingerprints.

[0085] Based on the context metadata, an interface proxy class name corresponding to the conflict node is generated. The interface proxy class name inherits the native interface of the target protocol and injects type casting constraints.

[0086] The dependency chain of the interface proxy class name in the target protocol is parsed to generate a dynamic binding strategy that includes parameter scope isolation boundaries;

[0087] Based on the dynamic binding strategy, conflicting nodes in the parameter passing path are mapped to protocol-compatible interface method instances, generating a dynamic binding queue of parameter instances.

[0088] Traverse the interface method instances in the dynamic binding queue, verify the compatibility status of the interface method instances with the target protocol instruction syntax, and generate parameter binding validity status codes;

[0089] Based on the parameter binding validity status code, filter out interface method instances without type conflicts and generate a protocol-independent intermediate instruction sequence;

[0090] The intermediate instruction sequence is filled with instruction templates according to the syntax rules of the target protocol to generate unambiguous instruction conversion results.

[0091] Preferably, the present invention also includes a real-time stable data transmission system for ultra-far-shore distances, comprising a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the aforementioned real-time stable data transmission method for ultra-far-shore distances.

[0092] Compared with the prior art, the beneficial effects of the present invention are:

[0093] This real-time stable data transmission method for ultra-long-distance shore-to-shore data transmission demonstrates several advantages in handling complex scenarios. First, by acquiring target transmission link data and extracting dynamic fluctuation characteristics, it can accurately capture signal changes during transmission, including signal strength distribution and latency jitter. This provides a more comprehensive understanding of the transmission link, moving beyond simple signal parameters and thus offering rich foundational information for subsequent processing.

[0094] The trained channel estimation network is invoked to generate a set of signal transmission paths, which includes the mapping relationship between channel type identifiers and phase codes. This process can clearly distinguish different types of channels and clarify the correspondence between phase codes and channel types. It avoids the transmission path selection errors caused by the ambiguity of channel type judgment in traditional methods, making the determination of signal transmission paths more targeted and able to select appropriate transmission paths according to different channel characteristics.

[0095] Based on the signal transmission path set, a waveform analysis network is invoked to perform multimodal feature fusion on the signal intensity distribution information, generating a feature fusion result that includes time-frequency domain features and modulation mode features. This allows for a deeper understanding of the signal. The fusion of time-frequency domain features and modulation mode features breaks through the limitations of single feature analysis, comprehensively reflecting the characteristics of the signal in different dimensions. This enables a more accurate grasp of the signal transmission state and provides a more comprehensive basis for subsequent transmission strategy formulation.

[0096] The feature fusion results are input into the transmission strategy model for dynamic path selection, outputting real-time data transmission elements, including control commands generated based on the signal transmission sequence and phase self-calibration parameters. This step enables dynamic adjustment of the transmission strategy. The control commands can instantly regulate the transmission process based on real-time signal conditions, while the phase self-calibration parameters can automatically correct phase deviations in the signal during transmission, reducing signal interference and data errors caused by phase shifts. This allows data transmission to adapt to dynamic changes in the channel and maintain stability in complex environments far from shore.

[0097] Compared to traditional transmission methods, this method forms a complete dynamic processing flow, from acquiring and analyzing link data to selecting paths and formulating transmission strategies. Each stage is tightly integrated, enabling collaborative responses to various interference factors at ultra-far-shore distances. Whether facing the influence of different channels such as sea surface reflection and atmospheric scattering, or fluctuations in signal strength and latency jitter, it can effectively cope with these challenges through its own processing mechanism. This improves the stability and reliability of real-time data transmission at ultra-far-shore distances, meeting the high data transmission requirements of fields such as marine development and ocean shipping. Attached Figure Description

[0098] Figure 1 This is a schematic diagram illustrating the working principle of the real-time stable data transmission method for ultra-far-shore distances described in this invention.

[0099] Figure 2 A flowchart for signal localization and feature fusion;

[0100] Figure 3 A flowchart for dynamic path selection and instruction generation;

[0101] Figure 4 A flowchart for adaptive adjustment of the link. Detailed Implementation

[0102] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0103] Please see Figures 1-4 The present invention provides a method for stable real-time data transmission over extremely far offshore distances, the method comprising:

[0104] The target transmission link data is acquired, and dynamic fluctuation characteristics are extracted from it. This data includes signal strength distribution information and time delay jitter. The transmission link data can be collected through a sensor network deployed between offshore equipment and shore-based terminals. Signal strength distribution information reflects signal attenuation at different transmission nodes, and time delay jitter reflects the changing pattern of time delay during data transmission. Dynamic fluctuation characteristics are extracted based on these two factors, covering the instantaneous peak and valley frequency of signal strength, as well as the amplitude range of time delay jitter.

[0105] The channel estimation network, having completed its training, is invoked to perform signal localization processing on the transmission link data, generating a set of signal transmission paths. This set of paths includes a mapping relationship between channel type identifiers and phase codes. The channel estimation network employs a deep neural network architecture and possesses signal localization capabilities after training on a large amount of historical transmission data. Through analysis of the transmission link data, possible signal transmission paths are determined, and each path is assigned a corresponding channel type identifier (such as a wireless channel, satellite channel, etc.) and a phase code. The phase code is used to distinguish the signal characteristics of different paths.

[0106] Based on the set of signal transmission paths, a waveform analysis network is invoked to perform multimodal feature fusion on the signal intensity distribution information, generating a feature fusion result that includes time-frequency domain features and modulation mode features. The waveform analysis network consists of multiple subnetworks, capable of extracting the time-frequency domain features and modulation mode features of the signal respectively. Time-frequency domain features reflect the signal's distribution characteristics in the time and frequency domains, while modulation mode features reflect the signal's modulation method (such as amplitude modulation, frequency modulation, phase modulation, etc.). Multimodal feature fusion integrates these two types of features to form a more comprehensive signal description.

[0107] The feature fusion result is input into the transmission strategy model for dynamic path selection, outputting real-time data transmission elements. These elements include control commands generated based on the signal transmission sequence and phase self-calibration parameters. Based on the feature fusion result and the actual conditions of the current transmission environment, the transmission strategy model selects the optimal signal transmission path and generates corresponding control commands to guide data transmission and reception operations. The phase self-calibration parameters are used to compensate for phase shifts during signal transmission, ensuring the accuracy of data transmission.

[0108] In Example 1, when the trained channel estimation network is invoked to perform signal localization processing on transmission link data to generate a set of signal transmission paths, the transmission link data is first subjected to signal framing processing. The transmission link data is a continuous data stream, which is divided into multiple independent signal frame regions by setting a fixed time interval or data length as the framing standard. Each signal frame region corresponds to a set of initial parameters for a time-frequency window. These parameters include the start and end times of the time window, the coverage area of ​​the frequency window, etc., and their values ​​are determined based on the carrier frequency and bandwidth characteristics of the transmitted signal.

[0109] The spectral feature map of each signal frame region is extracted. A Fast Fourier Transform (FFT) is performed on the signal frame region to convert the time-domain signal into a frequency-domain representation, resulting in a spectral feature map with frequency on the horizontal axis and energy on the vertical axis. Energy attention weighting is then applied to the spectral feature map, calculating weighting coefficients based on the energy values ​​of different frequency components. Frequency components with higher energy values ​​are assigned greater weights. These weighting coefficients are then multiplied by the corresponding frequency components in the spectral feature map, and finally converted back to the time domain using an Inverse Fourier Transform (IFT) to generate a weighted time-domain feature vector.

[0110] The signal-to-noise ratio (SNR) is calculated by comparing the time-domain feature vector with a preset path delay. The path delay is a possible delay range estimated in advance based on the transmission distance and signal propagation speed. The SNR is calculated as the ratio of signal energy to noise energy in the time-domain feature vector. A reliability threshold is set, and candidate paths with SNR values ​​higher than this threshold are selected. These candidate paths are considered to have basic transmission reliability.

[0111] Multipath interference suppression is performed on candidate paths, and an adaptive equalization algorithm is used to eliminate mutual interference between signals from different paths, resulting in multiple independent signal transmission paths. A corresponding channel type identifier is added to each path, including information such as the channel's transmission medium and frequency band. Simultaneously, a normalized phase code relative to the transmission link data is calculated for each path. Normalization is achieved by mapping the phase value to the range of 0 to 2π, ultimately forming a set of signal transmission paths containing the mapping relationship between channel type identifiers and phase codes.

[0112] When using a waveform analysis network to perform multimodal feature fusion of signal intensity distribution information based on a set of signal transmission paths, the time position and frequency range of each path signal in the transmission link data are determined according to the normalized phase encoding in the set of signal transmission paths. Based on this, the transmission link data is processed by regional truncation to obtain multiple signal segments corresponding to different paths.

[0113] Each signal segment undergoes filtering and denoising. Filtering employs a bandpass filter, with its passband range set according to the signal's carrier frequency and bandwidth, allowing only signals within this range to pass. Denoising utilizes a wavelet thresholding algorithm, performing multi-scale decomposition on the signal segment, setting the high-frequency coefficients below the threshold to zero, and then reconstructing the optimized waveform analytical input data.

[0114] The bidirectional gated recurrent subnet in the waveform analysis network is invoked to extract sequence features from the waveform analysis input data. The bidirectional gated recurrent subnet contains two gated recurrent units: a forward unit and a backward unit. The forward unit processes data from the beginning of the signal segment backward, while the backward unit processes data from the end of the signal segment forward. The outputs of both units are concatenated to form a time-frequency domain feature vector, which contains information about the dynamic changes of the signal in the time and frequency dimensions.

[0115] The modulation classification model is invoked in parallel to perform carrier feature matching on signal segments. The modulation classification model stores carrier feature templates for various common modulation modes (such as amplitude shift keying, frequency shift keying, phase shift keying, etc.). By calculating the similarity between the carrier frequency, phase change pattern, and other features of the signal segment and each template, a modulation mode probability distribution is generated. Each value in this distribution corresponds to the probability that the signal segment belongs to a certain modulation mode.

[0116] The time-frequency domain feature vector and the modulation mode probability distribution are concatenated as a tensor. The time-frequency domain feature vector is a three-dimensional tensor, whose dimensions correspond to the time step, feature dimension, and number of channels of the signal segment, respectively; the modulation mode probability distribution is a one-dimensional vector, whose length is equal to the number of modulation mode types. By concatenating the two along the feature dimension, they are integrated into a new three-dimensional tensor, namely the multimodal feature fusion result, which simultaneously contains the time-frequency domain features and modulation mode features of the signal.

[0117] Throughout the process, the parameters of both the channel estimation network and the waveform analysis network were trained using a large amount of historical transmission data. The training data for the channel estimation network covered transmission link data under different offshore distances and weather conditions. By continuously adjusting the network weights, the deviation between the generated set of signal transmission paths and the actual paths was gradually reduced. The waveform analysis network was trained using signal segment data labeled with time-frequency domain features and modulation modes. The network parameters were optimized through a backpropagation algorithm to improve the accuracy of the multimodal feature fusion results.

[0118] Furthermore, in signal framing, if the end of the transmission link data is shorter than a complete signal frame, it is padded with zeros to ensure that all signal frame regions have consistent lengths. In multipath interference suppression, for candidate paths with severe overlap, their signal strength and stability are compared to retain the better path and eliminate paths with significant interference. In the calculation of modulation mode probability distribution, when the maximum probability value is lower than a set confidence level, the corresponding modulation mode is marked as unknown for further processing.

[0119] In Example 2, when the feature fusion results are input into the transmission strategy model for dynamic path selection and output as real-time data transmission elements, the contextual dependencies of the current transmission scenario are first analyzed. The contextual information of the transmission scenario includes the geographical location of the offshore device, real-time meteorological data (such as wind speed and precipitation intensity), the distribution of electromagnetic interference sources, and link stability data from past transmission records. By structuring this information, an association mapping with the multimodal feature fusion results is established. For example, when strong electromagnetic interference is detected, the transmission priority weight corresponding to the channel type with strong anti-interference capability is increased; when the offshore distance exceeds a preset threshold, the proportion of satellite channels in the priority weight is increased. The transmission priority weights are represented in vector form, with each element corresponding to the priority value of a signal transmission path.

[0120] A state transition matrix is ​​generated based on transmission priority weights. The rows and columns of the state transition matrix correspond to different signal transmission paths, and the elements in the matrix represent the probability of switching from one path to another. This probability is calculated based on the transmission priority weights; the greater the priority difference, the lower the corresponding transition probability. Simultaneously, the channel type identifiers in the signal transmission path set are traversed and sorted. The sorting criteria include the channel's transmission rate, historical bit error rate, and current load. The sorting result is used to determine the initial traversal order during path selection.

[0121] A reinforcement learning strategy is employed to select the optimal transmission sequence from the state transition matrix. The agent in the reinforcement learning strategy aims to maximize the cumulative transmission efficiency, which comprehensively considers factors such as transmission delay, data integrity, and energy consumption. Through interaction with the transmission environment, the agent continuously updates its evaluation of the transmission performance of each path. When the transmission efficiency of a path falls below expectations, a path switching mechanism is triggered. Through multiple iterative learning iterations, a basic instruction set containing sending, relaying, and retransmission operations is generated. The sending instruction specifies the initial transmission time and transmission power; the relay instruction specifies the sequence of nodes participating in signal forwarding and the forwarding time of each node; the retransmission instruction includes the retransmission trigger condition (e.g., the receiver has not received an acknowledgment signal) and the maximum number of retransmissions.

[0122] The basic instruction set and phase self-calibration parameters are integrated into a protocol structure. The phase self-calibration parameters include phase compensation values, calibration periods, and calibration trigger thresholds, which are determined based on the phase characteristics of the signal transmission path. The integration process follows the frame structure requirements of the target transmission protocol (such as TCP / IP, satellite communication protocols, etc.), encapsulating the basic instruction set and phase self-calibration parameters into a protocol data unit. The basic instruction set serves as the data field, and the phase self-calibration parameters serve as the header options field. After encapsulation, an instruction code segment conforming to the target transmission protocol is generated, which can be directly parsed and executed by the transmission equipment.

[0123] During data transmission, monitoring modules deployed on transmission nodes capture link response data in real time. This data includes signal strength, phase deviation, data verification results, and transmission status codes fed back from the receiving end. Feedback logs are generated based on this data and stored in a time-series database. Each record contains a timestamp, signal phase offset (in radians), and status change information (e.g., changing from "normal transmission" to "transmission interrupted").

[0124] Extract the features of abnormal events from the feedback logs. Abnormal events include sudden drops in signal strength, phase offset exceeding the allowable range, and transmission timeouts. Their features include the time of occurrence, duration, affected area, and associated path identifiers. These features are then matched against a historical transmission case database, which stores the features of past abnormal events and their corresponding handling solutions. The matching process uses a cosine similarity algorithm to calculate the similarity value between the current abnormal event features and the feature vectors in the case database, selecting the top N cases with the highest similarity as references.

[0125] Adaptive adjustment instructions are generated based on the matching results. These instructions include corrections to the channel estimation network parameters and suggestions for adjusting the transmission strategy. For example, when a "phase drift" case is matched, the adjustment instruction will include increasing the phase calibration frequency. The path delay parameters of the channel estimation network are dynamically updated based on these adaptive adjustment instructions. This update process is implemented through an online parameter adjustment interface without interrupting current data transmission.

[0126] The updated path delay parameters are injected into the transmission strategy model, and the transmission priority weights in the state transition matrix are recalculated. During recalculation, paths with significant performance improvements after the parameter update are given priority, enabling the transmission strategy model to quickly adapt to changes in link conditions. This entire dynamic adjustment process forms a closed loop, ensuring that data transmission remains stable even when the link environment fluctuates.

[0127] Furthermore, in the extraction of abnormal event features from feedback logs, a sliding time window technique is employed. The window size is set based on the average duration of the abnormal event. Statistical analysis of the data within the window filters out false anomalies caused by transient interference. During the training process of the reinforcement learning strategy, an exploration factor is introduced, allowing the agent to attempt underutilized paths with a certain probability while ensuring transmission stability, in order to discover better transmission sequences.

[0128] In Example 3, abnormal event features are extracted from the feedback log, and these features are matched with a historical transmission case library for similarity. When generating adaptive adjustment instructions, the abnormal event features are first divided into time windows. The length of the time window is set according to the average duration of the abnormal event, typically covering the complete cycle from the occurrence of the abnormality to the recovery of normalcy. By sliding this time window, continuous abnormal event features are divided into multiple non-overlapping event segments. Each event segment corresponds to a set of continuous signal screenshot sequences. The signal screenshot sequences consist of signal waveform diagrams, spectrum diagrams, and phase distribution diagrams at different times, which can intuitively reflect the development process of the abnormal event.

[0129] The trained anomaly classification model is invoked to perform root cause analysis on each event segment. The anomaly classification model employs a deep neural network architecture. The input layer receives the feature vector of the event segment, the hidden layers extract features through multiple convolutional and pooling operations, and the output layer outputs classification label probabilities including signal recognition errors, phase drift, and transmission timeouts. The classification label is determined based on the output layer probability values; when the probability value of a certain label is higher than that of other labels, it is considered the primary anomaly type for that event segment.

[0130] The system retrieves solution templates that match the classification tags from the historical transmission case library. This library is categorized by anomaly type, with each category containing multiple solution templates. Each template includes parameter adjustment suggestions, operational steps, and applicable scenario descriptions. Based on the matching degree of the classification tags, the templates with the highest relevance are selected to form a candidate adjustment strategy set. The system ranks the candidate adjustment strategies based on the matching degree between the signal screenshot sequence and the set of candidate strategies. The overlap between the anomaly features in the signal screenshot sequence and the preset features of each template is calculated. Higher overlap indicates higher confidence in the corresponding candidate strategy. The strategy with the highest confidence is selected to generate adaptive adjustment instructions.

[0131] The trained anomaly classification model is invoked to perform root cause analysis on each event segment. When generating classification labels, the event segments are first processed by time series segmentation. Based on the inflection point of the characteristic change of the abnormal event (such as the moment when the signal strength suddenly drops), the event segment is divided into multiple signal screenshot sequence slices. Each slice is marked with a start trigger time and an end time, ensuring that each slice contains a complete stage of abnormal characteristic change.

[0132] Keyframe sampling is performed on the signal screenshot sequence slices. Keyframes are selected based on the rate of change of signal characteristics. When the rate of change exceeds a set threshold, the signal screenshot at the current moment is recorded as a keyframe. After sampling, a set of keyframes representing signal state changes and their corresponding timestamp indices are generated. The timestamp indices are accurate to the millisecond level and are used to mark the time position of the keyframes within the entire event segment.

[0133] Extract the signal phase offset between adjacent keyframes in the keyframe set for signal state changes. By comparing the phase distribution maps of adjacent keyframes, calculate the phase difference at each frequency point, and integrate these differences into a signal displacement trajectory vector, with the vector's dimension matching the number of frequency points in the signal. Simultaneously, calculate the timestamp difference between adjacent keyframes to generate a time interval sequence reflecting the time interval changes between keyframes.

[0134] Frequency offset calculation is performed on the signal displacement trajectory vector. The displacement trajectory vector is transformed to the frequency domain through Fourier transform, and the offset amplitude and phase change of each frequency component are analyzed to generate an abnormal fluctuation mode feature vector. This vector contains information such as the main frequency, harmonic components and attenuation coefficient of the abnormal fluctuation.

[0135] The feature vector of the abnormal fluctuation pattern is input into the spatiotemporal convolutional subnetwork in the anomaly classification model for local feature extraction. The spatiotemporal convolutional subnetwork contains temporal convolutional layers and spatial convolutional layers. The temporal convolutional layer captures the variation of features over time, while the spatial convolutional layer extracts the correlation features between different frequency points. The outputs of both are processed by an activation function to generate a spatial anomaly activation map. Regions with higher values ​​in the activation map correspond to frequency ranges where abnormal fluctuations are significant.

[0136] A sliding window mean filter is applied to the time interval sequence. The size of the sliding window is set according to the average fluctuation period of the time intervals. By calculating the average value of the time intervals within the window, high-frequency noise in the sequence is smoothed, generating a smoothed response delay time series. This series can more clearly reflect the time interval variation trend of abnormal events.

[0137] The response delay time series is input into the gated recurrent subnet of the anomaly classification model for periodic pattern matching. The gated recurrent subnet controls the flow of information through a gating mechanism, which can effectively capture long-term dependencies in the time series. By comparing it with the preset anomaly periodic patterns, it outputs an anomaly score in the time dimension. The higher the score, the more significant the periodicity of the anomaly within that time period.

[0138] Channel max pooling is performed on the spatial anomaly activation map. The maximum value is selected on the feature map of each channel to compress the spatial dimension and generate a spatial dimension anomaly score. The score reflects the intensity of the anomaly features corresponding to different channels.

[0139] The spatial dimension anomaly scores and the temporal dimension anomaly scores are subjected to feature cross-fusion processing using the following formula:

[0140]

[0141] in, The comprehensive score represents the spatiotemporal joint anomaly probability distribution. Indicates the spatial dimension anomaly score, Indicates anomaly scores in the time dimension. This is the weighting coefficient, with a value ranging from 0 to 1, set according to the importance of spatial and temporal features in anomaly identification.

[0142] A spatiotemporal joint anomaly probability distribution is generated based on the comprehensive score. Each element in the distribution corresponds to the anomaly probability within a certain frequency range at a certain time.

[0143] Peak detection results for signal identification error probability, phase drift probability, and transmission timeout probability are extracted from the spatiotemporal joint anomaly probability distribution. By traversing the probability distribution, the maximum probability and its location corresponding to each anomaly type are found to determine the peak detection result.

[0144] Dynamic threshold comparison processing is performed on preset probability thresholds based on peak detection results. The preset probability thresholds are set separately for each anomaly type. When the peak probability of a certain anomaly type exceeds its corresponding threshold, the type is included in the classification label candidate set and sorted from high to low probability values ​​to generate a classification label candidate set containing the probability ranking.

[0145] The candidate set of classification labels is subjected to timestamp index alignment verification processing. The timestamps of the candidate labels are checked to see if they are consistent with the actual occurrence time of the abnormal event in the signal screenshot sequence slice. Labels with mismatched times are removed, and target classification labels that are consistent with the phase of the abnormal event are generated.

[0146] The target classification label and the signal displacement trajectory vector undergo reverse anomaly verification. The anomaly type indicated by the target classification label is compared with the features reflected by the signal displacement trajectory vector to determine if they are consistent. For example, the trajectory vector corresponding to phase drift should show a continuous phase shift trend. Based on the verification results, a confidence weight is assigned to each label, generating a final classification label set containing these confidence weights.

[0147] Example 4 involves injecting virtual phase perturbation parameters before data transmission. These parameters are random values ​​generated based on common phase offset ranges in ultra-far-shore transmission scenarios. For example, in an offshore wind power data transmission scenario 1000 kilometers offshore, considering that atmospheric refraction, wave reflection, and other factors may cause a ±0.5π shift in the signal phase, the virtual phase perturbation parameters are randomly generated within this range. By superimposing these parameters on the phase data of the original signal, various phase offset situations that may be encountered in actual transmission are simulated.

[0148] The transmission strategy model's fault-tolerant processing of the phase after disturbance is monitored to generate stability evaluation indicators. Specifically, when a virtual phase disturbance is injected, the transmission strategy model processes the disturbed signal according to a preset algorithm, including phase compensation and path adjustment. During this process, positioning failures caused by phase errors are recorded. For example, if a signal should have been transmitted to shore-based terminal A but was mistakenly located to terminal B due to phase error, these cases are counted as positioning failures. Simultaneously, the number of successful anomaly detections is counted, i.e., the number of times the transmission strategy model successfully identified the phase disturbance and initiated the corresponding processing mechanism. The ratio of these two counts constitutes the basic data for the anomaly detection success rate.

[0149] After statistically analyzing the number of positioning failures and the anomaly capture success rate caused by phase errors during transmission, the ratio of the number of positioning failures to the total number of transmission steps is calculated to generate the first evaluation coefficient. The total number of transmission steps refers to the operational steps included in the complete process from data transmission to reception confirmation. For example, a transmission process including signal encoding, transmission, relay, and reception decoding would have a total of 4 transmission steps. The proportion of events matching the preset fault tolerance rules in the anomaly capture success rate is extracted to generate the second evaluation coefficient. The preset fault tolerance rules may stipulate that when the phase offset is within ±0.2π, the transmission strategy model should complete anomaly capture within 100ms. The proportion of events that meet this rule among all anomaly events is the basis for calculating the second evaluation coefficient.

[0150] The first and second evaluation coefficients are weighted and summed to generate a comprehensive stability evaluation index. The weight allocation is determined based on actual transmission requirements. If the transmission scenario requires high positioning accuracy, the first evaluation coefficient has a larger weight; if more emphasis is placed on timely handling of anomalies, the second evaluation coefficient has a higher weight. When the comprehensive stability evaluation index falls below a preset threshold, the online learning mode of the channel estimation network is triggered, enabling the network to adjust its parameters based on new disturbance data.

[0151] When updating the spectral characteristic parameters of the channel estimation network using gradient backpropagation based on the phase difference data before and after the disturbance, the phase difference data before and after the disturbance are first time-stamp aligned. For example, the phase data at 10:00:00.000 before the disturbance is matched with the phase data at the same time point after the disturbance to generate a sequence of matching pairs containing the phase set before the disturbance and the phase set after the disturbance, ensuring that each pair of data corresponds one-to-one in time.

[0152] The phase offset is extracted from the matched sequence to generate a horizontal offset vector and a vertical offset vector for each signal segment. The horizontal offset vector reflects the phase offset on the frequency axis, and the vertical offset vector reflects the phase offset on the time axis. For example, if a signal segment has a phase offset of 0.1π at a frequency of 1GHz and a phase offset of 0.05π in the time interval t1 to t2, the corresponding horizontal and vertical offset vectors will record these specific values.

[0153] The Euclidean distance between the horizontal and vertical offset vectors is calculated. The result of the Euclidean distance calculation quantifies the phase shift of each signal segment, generating a set of phase difference vectors for each signal segment. Each vector in this set corresponds to a signal segment, and the magnitude of the vector represents the total phase shift amplitude of that segment.

[0154] A location regression loss function is constructed based on the set of phase difference vectors. This function takes the phase data predicted by the channel estimation network and the actual measured phase data after perturbation as inputs, and calculates the mean square error between the two. The smaller the error value, the more accurate the network prediction. The location regression loss function is then subjected to a differentiable transformation to convert it into a form suitable for gradient calculation, generating the loss value tensor required for gradient backpropagation. The dimension of this tensor is consistent with the dimension of the network output layer.

[0155] The spectral feature parameters of the channel estimation network are traversed, including convolution kernel weights and bias terms. The partial derivatives of the loss tensor with respect to each spectral feature are calculated to generate the spectral feature gradient matrix. Each element in the gradient matrix represents the degree of influence of the corresponding parameter on the loss value; a positive value indicates that the loss value increases when the parameter increases, while a negative value indicates that the loss value decreases when the parameter increases.

[0156] The spectral feature gradient matrix is ​​smoothed using momentum, and a momentum factor is introduced. This ensures that the current parameter update direction not only considers the current gradient but also retains part of the previous update direction, reducing gradient fluctuations and generating a decayed parameter update direction vector. For example, if the previous update direction vector is v1, the current gradient-calculated direction vector is v2, and the momentum factor is 0.9, then the new update direction vector might be 0.9v1 + 0.1v2.

[0157] Based on the parameter update direction vector and a preset learning rate parameter, the spectral feature parameters of the channel estimation network are iteratively and incrementally adjusted. The learning rate parameter determines the magnitude of each parameter adjustment; for example, when the learning rate is 0.001, the adjustment amount is the product of the update direction vector and 0.001. During the incremental adjustment process, the rate of change of the location regression loss function is monitored in real time, which is the ratio of the difference in loss values ​​between two adjacent iterations to the difference in the number of iterations. When the absolute value of this rate of change is less than a preset stability threshold, a loss convergence status indicator is generated.

[0158] When the loss convergence status indicator reaches a stable threshold, gradient backpropagation updates are terminated and the spectral feature parameters are frozen. These parameters are considered to be at their current optimal values. The updated spectral feature parameters are then injected into the forward propagation path of the channel estimation network, and signal framing and localization processing are re-executed. By comparing the localization results of the same signal before and after the update, the change in phase localization accuracy is verified. The entire process is fully automated and requires no manual intervention.

[0159] Example 5 describes the construction of a cross-protocol transport adapter. When using this adapter to parse the instruction syntax differences between different transport protocols, a transport protocol syntax rule base is first established. This rule base employs a distributed database storage structure, divided into sub-bases according to protocol type. Each sub-base contains a keyword mapping table and parameter passing paths for the corresponding protocol. The keyword mapping table records the correspondence between keywords with the same or similar functions in different protocols. For example, the keyword "SOF" representing "Start of Data Frame" in satellite communication protocols might be represented as "StartFrame" in wireless communication protocols; the mapping table associates and stores these two keywords. The parameter passing paths are recorded in flowchart form, showing the complete flow of parameters from generation to execution, including the encapsulation format and parsing order of parameters at each layer of the protocol stack.

[0160] Abstract syntax tree (AST) parsing is performed on real-time data transmission elements. These elements contain control commands and phase self-calibration parameters. The parsing process transforms these elements into an AST using a parser. Nodes in the tree represent syntactic units (such as command type, parameter name, and value), and connections between nodes represent syntactic relationships (such as inclusion and dependency). After the AST is generated, the core semantic information is extracted and converted into intermediate representation layer data. This intermediate representation layer data uses a protocol-independent structured format, such as JSON, retaining only core information such as the command's operation type, parameter values, and execution order, while stripping away the syntactic details of specific protocols.

[0161] The keyword mapping table is traversed and queried based on the intermediate presentation layer data. According to the operation type and parameter name in the intermediate presentation layer data, the keyword corresponding to the target protocol is searched in the keyword mapping table, generating a protocol-compatible instruction replacement scheme. For example, the "retransmission count" parameter in the intermediate presentation layer data needs to be represented as "RetransCount" in the target protocol. The replacement scheme records this mapping relationship and ensures that the parameter's value range conforms to the target protocol requirements.

[0162] Dependency injection is performed on conflicting parameter passing paths. When parameter passing paths of different protocols have type ambiguity or overlapping scopes at a certain stage, the conflicting nodes are first identified, and a set of conflicting path identifiers is generated. The conflicting path identifier consists of the protocol type, parameter name, and conflict location, such as "Satellite Protocol - Power Parameter - Transport Layer Encapsulation".

[0163] Extract the context metadata for each node in the conflict path identifier set. Context metadata includes parameter type declarations (e.g., integer, floating-point), lifecycle markers (e.g., temporary variables, global variables), and call stack fingerprints (recording the sequence of functions called for the parameters). Based on this metadata, generate an interface proxy class name corresponding to the conflict node. The interface proxy class name follows the naming convention of "protocol name + parameter name + proxy identifier," for example, "SatellitePowerProxy." This interface proxy class inherits the native interface of the target protocol and injects type casting constraints in the class definition, such as casting floating-point parameters to integers required by the target protocol.

[0164] The dependency chain of the interface proxy class name in the target protocol is resolved. By analyzing the class inheritance relationship and function call order of the target protocol, the position of the interface proxy class in the protocol stack is determined, and a dynamic binding strategy including parameter scope isolation boundaries is generated. Scope isolation boundaries are implemented through namespace partitioning, for example, allocating an independent namespace "ProxySpace" for the proxy class to avoid naming conflicts with the native classes of the target protocol.

[0165] A dynamic binding strategy maps conflicting nodes in the parameter passing path to protocol-compatible interface method instances. The generation of interface method instances must adhere to the function definition specifications of the target protocol, including the number of parameters, return value type, and exception handling mechanism. The generated instances are arranged in execution order, forming a dynamic binding queue of parameter instances.

[0166] Iterate through the interface method instances in the dynamic binding queue and verify their compatibility with the target protocol's instruction syntax. Verification includes checking if the method name conforms to the target protocol's naming conventions, if the parameter types match the protocol requirements, and if the calling method matches the protocol's function calling rules. After verification, generate parameter binding validity status codes: "00" indicates full compatibility, "01" indicates minor adjustments are needed (e.g., parameter order adjustment), and "10" indicates a serious conflict.

[0167] Based on the parameter binding validity status code, interface method instances without type conflicts are selected, and a protocol-independent intermediate instruction sequence is generated. The intermediate instruction sequence is arranged in execution order and contains only method call and parameter passing logic, without involving the specific protocol syntax.

[0168] The intermediate instruction sequence is filled into the instruction template according to the syntax rules of the target protocol. The instruction template of the target protocol predefines the structure of the instruction (such as header, data segment, checksum) and syntax format (such as delimiters, keyword case). The filling process embeds the content of the intermediate instruction sequence into the corresponding position of the template to generate an unambiguous instruction conversion result.

[0169] When converting real-time data transmission elements into underlying driver instructions supported by the target protocol, the timing dependencies of control instructions and the exception handling context are preserved. Timing dependencies are implemented through timestamps, with each instruction accompanied by its execution time, ensuring that instructions are executed in the original order. The exception handling context records retry mechanisms and alternative solutions when instruction execution fails. For example, when the "send instruction" fails, the "switching channel and retransmitting" exception handling process is automatically triggered.

[0170] The system injects performance optimization parameters that match the target protocol. These parameters include transmission rate adjustment values, channel coding methods, and power control thresholds. These parameters are determined based on the characteristics of the target protocol and the transmission environment. For example, in a high-interference environment, the system injects an optimization parameter to "shorten the frequency hopping interval to 50ms" into the wireless communication protocol. The resulting transmission instruction bytecode conforms to the binary format requirements of the target protocol and can be directly parsed and executed by the underlying driver of the transmission device.

[0171] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0172] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for stable real-time data transmission over extremely far-shore distances, characterized in that, include: Acquire target transmission link data and extract dynamic fluctuation features from the transmission link data, wherein the dynamic fluctuation features include signal strength distribution information and time delay jitter relationship; The trained channel estimation network is invoked to perform signal localization processing on the transmission link data, generating a set of signal transmission paths, which includes a mapping relationship between channel type identifiers and phase codes. Based on the set of signal transmission paths, a waveform analysis network is invoked to perform multimodal feature fusion on the signal intensity distribution information, generating a feature fusion result that includes time-frequency domain features and modulation mode features; The feature fusion result is input into the transmission strategy model for dynamic path selection processing, and the real-time data transmission element is output. The real-time data transmission element includes control commands and phase self-calibration parameters generated based on the signal transmission sequence. The channel estimation network, which has completed training, is invoked to perform signal localization processing on the transmission link data, generating a set of signal transmission paths, including: The transmission link data is subjected to signal framing processing to generate multiple signal frame regions and corresponding time-frequency window initial parameters; Extract the spectral feature map of each signal frame region, and perform energy attention weighting processing on the spectral feature map to generate a weighted time-domain feature vector; The signal-to-noise ratio of the time-domain feature vector is calculated by comparing it with the preset path delay, and candidate paths that meet the reliability threshold are selected. Multipath interference suppression processing is performed on the candidate paths to generate a set of signal transmission paths containing the channel type identifier. Each signal transmission path contains a normalized phase code relative to the transmission link data. The step of fusing multimodal features of the signal intensity distribution information by calling a waveform analysis network based on the signal transmission path set, generating a feature fusion result containing time-frequency domain features and modulation mode features, includes: Based on the normalized phase coding in the signal transmission path set, the transmission link data is subjected to regional segmentation processing to generate multiple signal segments; Each signal segment is filtered and denoised to generate optimized waveform analysis input data; The bidirectional gated recurrent subnet in the waveform analysis network is invoked to extract sequence features from the waveform analysis input data, generating a time-frequency domain feature vector. The modulation classification model is invoked in parallel to perform carrier feature matching on the signal segment and generate a modulation mode probability distribution. The time-frequency domain feature vector and the modulation mode probability distribution are concatenated using tensors to generate a multimodal feature fusion result; The step of inputting the feature fusion result into the transmission strategy model for dynamic path selection processing and outputting real-time data transmission elements includes: The contextual dependencies of the current transmission scenario are analyzed to generate transmission priority weights associated with the multimodal feature fusion results; A state transition matrix is ​​generated based on the transmission priority weights, and the channel type identifiers in the signal transmission path set are traversed and sorted. The optimal transmission sequence is selected from the state transition matrix using a reinforcement learning strategy to generate a basic instruction set containing transmission, relay, and retransmission operations; The basic instruction set and the phase self-calibration parameters are integrated into a protocol structure to generate an instruction code segment that conforms to the target transmission protocol; The method further includes: During data transmission, link response data is captured in real time to generate a feedback log containing signal phase offset and status change information. Extract the abnormal event features from the feedback log, and perform similarity matching between the abnormal event features and the historical transmission case library to generate adaptive adjustment instructions; The path delay parameters of the channel estimation network are dynamically updated based on the adaptive adjustment instructions. The updated path delay parameters are injected into the transmission strategy model, and the transmission priority weights in the state transition matrix are recalculated.

2. The method for stable real-time data transmission over ultra-far-shore distances as described in claim 1, characterized in that, The step of extracting abnormal event features from the feedback log and performing similarity matching between the abnormal event features and the historical transmission case library to generate adaptive adjustment instructions includes: The abnormal event characteristics are divided into time windows to generate multiple event segments and their corresponding signal screenshot sequences; The trained anomaly classification model is invoked to perform root cause analysis on each event segment, generating classification labels including signal recognition errors, phase drift, and transmission timeouts. Retrieve solution templates that match the classification labels from the historical transmission case library to generate a set of candidate adjustment strategies; Based on the matching degree ranking of the signal screenshot sequence and the candidate adjustment strategy set, the strategy with the highest confidence is selected to generate an adaptive adjustment instruction; The call to the trained anomaly classification model performs root cause analysis on each event segment, generating classification labels including signal recognition errors, phase drift, and transmission timeouts, including: The event segments are processed by time series segmentation to generate signal screenshot sequence slices containing the start trigger time and the end time; The signal screenshot sequence slices are subjected to keyframe sampling processing to generate a set of keyframes for signal state changes and corresponding timestamp indices; Extract the signal phase offset between adjacent key frames in the set of key frames for signal state changes, and generate a signal displacement trajectory vector and a time interval sequence; The frequency offset of the signal displacement trajectory vector is calculated to generate an abnormal fluctuation mode feature vector. The feature vector of the abnormal fluctuation pattern is input into the spatiotemporal convolutional subnet of the abnormal classification model for local feature extraction, generating a spatial abnormality activation map. The time interval sequence is subjected to sliding window mean filtering to generate a smoothed response delay time sequence; The response delay time series is input into the gated recurrent subnet of the anomaly classification model for periodic pattern matching to generate a time dimension anomaly score; channel max pooling is performed on the spatial anomaly activation map to generate a spatial dimension anomaly score; the spatial dimension anomaly score and the time dimension anomaly score are subjected to feature cross-fusion processing to generate a spatiotemporal joint anomaly probability distribution. Peak detection results of signal identification error probability, phase drift probability, and transmission timeout probability are extracted from the spatiotemporal joint anomaly probability distribution; dynamic threshold comparison processing is performed on the preset probability thresholds based on the peak detection results to generate a candidate set of classification labels containing probability ranking; The candidate classification labels are subjected to timestamp index alignment verification to generate target classification labels that are consistent with the phase of the abnormal events in the signal screenshot sequence slice; the target classification labels and the signal displacement trajectory vector are subjected to anomaly type reverse verification to generate a final classification label set containing confidence weights.

3. The method for stable real-time data transmission over ultra-far-shore distances as described in claim 1, characterized in that, The method further includes: A virtual phase perturbation parameter is injected before data transmission, which is used to simulate a random phase shift scenario of the signal. Monitor the fault-tolerant processing results of the transmission strategy model on the disturbed phase, and generate stability evaluation indicators; When the stability evaluation index is lower than a preset threshold, the online learning mode of the channel estimation network is triggered; The spectral characteristic parameters of the channel estimation network are updated by gradient backpropagation based on the phase difference data before and after the perturbation. The monitoring of the transmission strategy model's fault-tolerant processing results for the disturbed phase generates stability evaluation indicators, including: The number of positioning failures and the success rate of abnormal capture due to phase errors during the transmission process were statistically analyzed. Calculate the ratio of the number of positioning failures to the total number of transmission steps to generate a first evaluation coefficient; Extract the proportion of events that match the preset fault tolerance rules from the anomaly capture success rate, and generate a second evaluation coefficient; The first evaluation coefficient and the second evaluation coefficient are weighted and summed to generate a comprehensive stability evaluation index.

4. The method for stable real-time data transmission over ultra-far-shore distances as described in claim 3, characterized in that, The gradient backpropagation update of the spectral feature parameters of the channel estimation network based on the phase difference data before and after the perturbation includes: The phase difference data before and after the perturbation are timestamped to generate a matching sequence containing the phase set before the perturbation and the phase set after the perturbation. Extract the phase offset from the matching pair sequence to generate the horizontal offset vector and vertical offset vector for each signal segment; The Euclidean distance between the horizontal offset vector and the vertical offset vector is calculated to generate a set of phase difference vectors for each signal segment; A position regression loss function is constructed based on the set of phase difference vectors. The position regression loss function includes the mean square error between the predicted phase and the actual phase after perturbation by the channel estimation network. The location regression loss function is subjected to a differentiable transformation to generate the loss value tensor required for gradient backpropagation; Iterate through the spectral feature parameters of the channel estimation network, calculate the partial derivative of the loss tensor with respect to each spectral feature, and generate the spectral feature gradient matrix. The spectral feature gradient matrix is ​​subjected to momentum smoothing to generate a decayed parameter update direction vector; Based on the updated direction vector and the preset learning rate parameter, the spectral feature parameters of the channel estimation network are iteratively and incrementally adjusted. During the incremental adjustment process, the rate of change of the location regression loss function is monitored in real time to generate a loss convergence status indicator. When the loss convergence state indicator reaches a stable threshold, the gradient backpropagation update is terminated and the spectral feature parameters are frozen. The updated spectral feature parameters are injected into the forward propagation path of the channel estimation network, and the signal framing and signal localization processes are re-executed to verify the improvement in phase localization accuracy.

5. The method for stable real-time data transmission over ultra-far-shore distances as described in claim 1, characterized in that, The method further includes: Construct a cross-protocol transport adapter and use the cross-protocol transport adapter to parse the instruction syntax differences of different transport protocols; The real-time data transmission elements are converted into underlying driver instructions supported by the target protocol; The timing dependencies of the control instructions and the exception handling context are preserved during the conversion process; Inject the performance optimization parameters that match the target protocol to generate transmission instruction bytecode that meets the cross-link execution conditions.

6. The method for stable real-time data transmission over ultra-far-shore distances as described in claim 5, characterized in that, The construction of a cross-protocol transport adapter, and the parsing of instruction syntax differences between different transport protocols through the cross-protocol transport adapter, includes: Establish a transport protocol syntax rule base and store the keyword mapping table and parameter passing path for each protocol; The real-time data transmission elements are parsed using an abstract syntax tree to generate intermediate representation layer data; Based on the intermediate presentation layer data, the keyword mapping table is traversed and queried to generate a protocol-compatible instruction replacement scheme; Dependency injection is performed on conflicting parameter passing paths to generate unambiguous instruction conversion results; The process of performing dependency injection on conflicting parameter passing paths to generate unambiguous instruction conversion results includes: Identify conflicting nodes in the parameter passing path that have type ambiguity or overlapping scope, and generate a set of conflicting path identifiers; Extract the context metadata of each node in the set of conflict path identifiers. The context metadata includes parameter type declarations, lifecycle markers, and call stack fingerprints. Based on the context metadata, an interface proxy class name corresponding to the conflict node is generated. The interface proxy class name inherits the native interface of the target protocol and injects type casting constraints. The dependency chain of the interface proxy class name in the target protocol is parsed to generate a dynamic binding strategy that includes parameter scope isolation boundaries; Based on the dynamic binding strategy, conflicting nodes in the parameter passing path are mapped to protocol-compatible interface method instances, generating a dynamic binding queue of parameter instances. Traverse the interface method instances in the dynamic binding queue, verify the compatibility status of the interface method instances with the target protocol instruction syntax, and generate parameter binding validity status codes; Based on the parameter binding validity status code, filter out interface method instances without type conflicts and generate a protocol-independent intermediate instruction sequence; The intermediate instruction sequence is filled with instruction templates according to the syntax rules of the target protocol to generate unambiguous instruction conversion results.

7. A real-time data stable transmission system for ultra-far-shore distances, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of a real-time stable data transmission method for ultra-far-shore distances as described in any one of claims 1 to 6.

Images