Maximum wave height intelligent forecasting method, system and equipment based on deep learning and storage medium

Abstract

The invention relates to the technical field of marine environment prediction, and discloses a maximum wave height intelligent forecasting method, system and device based on deep learning and a storage medium. The system is internally provided with an acquisition unit, a fusion unit, a preprocessing unit, a training unit, an error correction unit, an evaluation unit, an early warning unit, a guarantee unit and a visualization unit, and all the units are sequentially connected through standardized data interfaces and work cooperatively so as to realize full-chain communication from multi-source heterogeneous data to precise intelligent forecasting and then to industry deep application. Finally, the system solves the problems of insufficient data utilization, insufficient model precision and timeliness, and disjunction between a forecasting result and an application scene in maximum wave height forecasting of a traditional method, and finally provides scientific, reliable and efficient decision support for ocean disaster prevention and reduction, engineering safety operation and navigation safety guarantee.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine environment prediction, in particular to a maximum wave height intelligent prediction method, system, device and storage medium based on deep learning. BACKGROUND

[0002] In the offshore wind power environment, accurate prediction of sea wave conditions is crucial for safe operation, operation and scheduling of wind turbines. However, although the traditional WAM (Wave Model) and SWAN (Simulating Waves Nearshore) numerical prediction model is based on physical equations, its calculation process involves complex spatio-temporal evolution problems, and it is highly dependent on supercomputing resources, resulting in long calculation time and difficulty in supporting high-resolution and real-time business prediction needs. Although the traditional empirical statistical method is simple to calculate, its formula is too simplified to accurately depict the complex nonlinear physical mechanisms of swell propagation, and the prediction deviation is particularly prominent in extreme weather conditions such as typhoons. Although the pure data-driven deep learning model has the ability to learn from data, it also has obvious limitations: first, the model is a "black box" feature, and the decision logic lacks physical interpretability and is easily disturbed by false correlations in the data; second, the prediction results are often systematically lagging behind actual observations during the training process; third, the adaptability to rare extreme events in the training sample is poor, and the prediction error rises sharply in extreme cases.

[0003] Therefore, the existing technology generally has problems such as insufficient data utilization, difficulty in balancing model accuracy and timeliness, and disconnection between prediction output and actual offshore wind power operation scene. It is urgent to build an intelligent wave height prediction scheme that can integrate multi-source data, combine physical mechanisms and data-driven advantages, and directly support terminal decision-making for offshore wind power. SUMMARY

[0004] In order to solve the problems of insufficient data utilization, insufficient model accuracy and timeliness, and disconnection between prediction results and application scenarios in the maximum wave height prediction of traditional methods, the present application provides a maximum wave height intelligent prediction method based on deep learning, which has the advantages of comprehensive multi-source data fusion, accurate signal preprocessing, intelligent and efficient prediction model, and complementary advantages of AI and physical model coupling.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a maximum wave height intelligent prediction method based on deep learning, comprising the following steps: Collecting multi-source marine observation data, and fusing the collected multi-source marine observation data through data format conversion and spatio-temporal registration; Decompose the original wave height signal, extract the period and amplitude characteristics through wavelet transform, reconstruct the preprocessed data using the EEMD (Ensemble Empirical Mode Decomposition) wave height signal reconstruction formula, and enhance the extreme wave height samples; Construct a width learning-LSTM hybrid network, embed an attention mechanism, train the model using preprocessed data, calculate the wave height prediction value in the future period through the wave height prediction formula, and correct the prediction lag; Start the Fortran-Python coupling framework, read the initial prediction results of the WAM numerical model, extract the error characteristics through the spatio-temporal attention neural network, and calculate the AI correction amount using the width learning-LSTM hybrid network prediction formula to correct the WAM prediction error; Input the corrected wave height data, combine the required engineering structure parameters, and evaluate the risk of the project affected by sea waves; According to the corrected wave height, issue warning information, and provide wave avoidance route suggestions for past ships based on the prediction results; Show the wave height change trend in the future period with a time series chart, show the wave height distribution in the sea area with a spatial heat map, generate an intelligent prediction report of the maximum wave height, and push it to the marine management department, shipping company and engineering unit.

[0006] Further, the collection of multi-source marine observation data includes: Control the buoys, satellites, radars and other measurement sensors distributed throughout the ocean to monitor multi-source marine observation data such as significant wave height, period, wind speed, wind direction, sea water temperature salinity, sea current flow rate and direction, and other atmospheric pressure, sea wave spectrum characteristics in real time; Mark the monitored significant wave height data, wind speed data, wind direction data, and sea current flow rate data.

[0007] Further, the collection of multi-source marine observation data also includes: Through the deployment of buoy sensors and underwater acoustic measurement equipment in the sea area, real-time collection of core wave condition parameters such as significant wave height, significant wave height, peak period and trough period in different sea areas, and synchronous recording of data collection time and corresponding latitude and longitude; Through the reception of remote sensing data returned by marine satellites, obtain macro observation information such as wave height distribution, sea surface wind speed field and sea surface temperature of large-scale sea area; According to the high-frequency ground wave radar and X-band radar deployed along the coastline, real-time monitoring of wave height changes, sea current direction and ship dynamics in the nearshore sea area.

[0008] Further, the EEMD wave height signal reconstruction formula is used for denoising and effective feature reconstruction of the original wave height signal, and the calculation formula is:

[0009] In the formula, This represents the wave height time series after preprocessing and denoising. Let i represent the i-th effective eigenmode function obtained from EEMD decomposition. This represents the residual components after EEMD decomposition. This indicates the number of valid intrinsic mode functions after filtering.

[0010] Furthermore, the predicted wave height for the future period is calculated using the wave height prediction formula, which is as follows:

[0011] In the formula, Indicates the future The predicted maximum wave height at time [time]. For the predicted duration, This indicates that the width learning system applies preprocessed wave height data. Feature mapping results The weight matrix represents the width-learned feature map. This represents the bias term for the width-learned feature map. This indicates the effect of the attention mechanism on preprocessing wave height data. The key features focus on the output. The weighting coefficients representing the attention mechanism. This represents the computational process of a Long Short-Term Memory (LSTM) network.

[0012] Furthermore, the error correction for the predicted wave height of the WAM numerical model is performed based on the wide-learning-LSTM hybrid network prediction formula, and the calculation formula is as follows:

[0013]

[0014] In the formula, This indicates the final accurate wave height after AI error correction. This indicates the initial predicted wave height of the WAM numerical wave model. This represents the wave height error correction amount generated by AI, used to correct forecast biases in the WAM model. This represents the output of the spatiotemporal attention mechanism, which predicts wave height by fusing WAM models. Compared with the measured wave height spatiotemporal correlation information, This represents the WAM wave height error mapping function. This represents the measured wave height data for the same period.

[0015] This invention also provides a deep learning-based intelligent maximum wave height forecasting system, the system comprising: an acquisition unit, a fusion unit, a preprocessing unit, a training unit, an error correction unit, an evaluation unit, an early warning unit, and a visualization unit. The acquisition unit is configured to acquire multi-source ocean observation data; The fusion unit is configured to fuse multi-source ocean observation data acquired through data format conversion and spatiotemporal registration. The preprocessing unit is configured to decompose the original wave height signal, extract period and amplitude features through wavelet transform, reconstruct the preprocessed data using the EEMD wave height signal reconstruction formula, and enhance extreme wave height samples. The training unit is configured to build a width learning-LSTM hybrid network, embed an attention mechanism, train the model using preprocessed data, calculate the predicted wave height for future time periods using the wave height prediction formula, and correct for prediction lag. The error correction unit is configured to start the Fortran-Python coupled framework, read the initial forecast results of the WAM numerical model, extract error features through a spatiotemporal attention neural network, calculate the AI ​​error correction amount using the wide learning-LSTM hybrid network prediction formula, and correct the WAM forecast error. The assessment unit is configured to take in corrected wave height data and, in combination with the required engineering structural parameters, assess the risk of the project being affected by sea waves. The early warning unit is configured to issue early warning information based on the corrected wave height and, in conjunction with the forecast results, provide wave avoidance route suggestions for passing vessels. The visualization unit is configured to display the wave height change trend in future periods using a time series graph, display the wave height distribution in the sea area using a spatial heat map, generate a maximum wave height intelligent forecast report, and push it to marine management departments, shipping companies and engineering units simultaneously.

[0016] Furthermore, the system also includes a protection unit. The support unit is configured to provide vessels with route optimization suggestions, wave avoidance area recommendations, and navigation safety level alerts based on real-time wave height forecasts.

[0017] The present invention also provides an electronic device, the electronic device comprising at least one processor and at least one memory, wherein the processor and the memory are data connected, wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of the present invention.

[0018] The present invention also provides a computer storage medium storing one or more instructions that, when executed by one or more computers, cause the one or more computers to perform the method described in the present invention.

[0019] Compared with existing technologies, this invention provides a method, system, device, and storage medium for intelligent maximum wave height prediction based on deep learning, which has the following beneficial effects: 1. This invention utilizes an integrated air-space-sea three-dimensional monitoring network composed of monitoring stations (buoys, satellites, and radars) to achieve continuous observation of a vast sea area in multiple dimensions with high spatiotemporal resolution. Ultimately, it solves the problems of large monitoring blind spots and limited data in traditional methods, providing a solid data foundation for accurate system forecasting.

[0020] 2. In this invention, the preprocessing unit performs denoising and effective feature reconstruction of the original wave height signal according to the EEMD wave height signal reconstruction formula. By screening and integrating the intrinsic mode functions and residual components after the original wave height signal is decomposed, noise interference is eliminated and key wave condition features are retained, thus obtaining a preprocessed wave height time series that meets the model input requirements.

[0021] 3. In this invention, the training unit predicts the maximum wave height for future time periods based on the wave height prediction formula. It integrates the fast feature mapping capability of width learning with the temporal dependency capture capability of LSTM, and combines the attention mechanism to focus on key influencing factors, outputting the maximum wave height prediction value for different future time periods (such as 12 hours, 24 hours, and 48 hours).

[0022] 4. In this invention, the error correction unit corrects the wave height forecast error of the WAM numerical model based on the wide learning-LSTM hybrid network prediction formula. It captures the forecast deviation pattern of the WAM model under extreme weather conditions through a spatiotemporal attention neural network, and generates the error correction amount by combining the measured wave height data. Finally, it obtains the corrected wave height with strong physical consistency and high accuracy.

[0023] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This diagram illustrates the specific process of the deep learning-based maximum wave height intelligent prediction method in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of the deep learning-based maximum wave height intelligent forecasting system is shown in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0027] This invention provides a deep learning-based intelligent maximum wave height prediction method. Figure 1 This diagram illustrates the specific process of the deep learning-based maximum wave height intelligent prediction method in an embodiment of the present invention. Figure 1 The method includes the following steps: Collect multi-source ocean observation data, and fuse the multi-source ocean observation data through data format conversion and spatiotemporal registration; The original wave height signal is decomposed, and the period and amplitude features are extracted by wavelet transform. The preprocessed data is reconstructed using the EEMD wave height signal reconstruction formula, and the extreme wave height samples are enhanced. A wide learning-LSTM hybrid network is constructed, with an attention mechanism embedded. The model is trained using preprocessed data, and the predicted wave height for future time periods is calculated using the wave height prediction formula, while correcting for prediction lag. The Fortran-Python coupled framework is started, the initial forecast results of the WAM numerical model are read, error features are extracted through a spatiotemporal attention neural network, and the AI ​​error correction amount is calculated using the wide learning-LSTM hybrid network prediction formula to correct the WAM forecast error. Input the corrected wave height data and combine it with the required engineering structural parameters to assess the risk of the project being affected by sea waves; Based on the revised wave height warning information and combined with the forecast results, we provide wave avoidance route suggestions for passing vessels; The system uses time series graphs to show the wave height variation trend in future periods and spatial heat maps to show the wave height distribution in the sea area. It generates intelligent forecast reports of maximum wave height and pushes them to marine management departments, shipping companies and engineering units simultaneously.

[0028] Specifically, the collection of multi-source ocean observation data includes: Control buoys, satellites, radars and other measurement sensors distributed throughout the ocean to monitor in real time multi-source ocean observation data, including significant wave height, period, wind speed, wind direction, sea temperature and salinity, ocean current speed and direction, and other atmospheric pressure and wave spectrum characteristics; The effective wave height data, significant wave height data, wind speed data, wind direction data, and ocean current velocity data monitored are marked with special emphasis.

[0029] Specifically, the collection of multi-source ocean observation data also includes: By deploying buoy sensors and underwater acoustic measurement equipment in the sea, core wave condition parameters such as effective wave height, significant wave height, wave crest period, and wave trough period in different sea areas are collected in real time, and the data collection time and corresponding latitude and longitude are recorded simultaneously. By receiving remote sensing data transmitted back by ocean satellites, macroscopic observation information on wave height distribution, sea surface wind speed field and sea surface temperature over a large area of ​​sea can be obtained. Based on high-frequency ground wave radar and X-band radar deployed along the coastline, the system monitors the dynamic information of wave height changes, ocean current direction, and ship movement in nearshore waters in real time.

[0030] Specifically, the following steps are included: Step 1: System Establishment: Establish an intelligent forecasting system, which includes a data acquisition unit, a fusion unit, a preprocessing unit, a training unit, an error correction unit, an evaluation unit, an early warning unit, a support unit, and a visualization unit. Each unit is connected and works collaboratively through standardized data interfaces. Step 2: Data Acquisition: The acquisition unit is equipped with a monitoring station to collect multi-source ocean observation data in real time. The sampling frequency is set to once every 10 minutes to continuously acquire the most original monitoring data. Step 3: Perform data integration: After the fusion unit removes outliers, it preprocesses the monitoring data through data format conversion and spatiotemporal registration, unifying it into an analysis and calculation format; Step 4, Signal Optimization: The preprocessing unit begins to decompose the original wave height signal, extracts the period and amplitude features through wavelet transform, reconstructs the preprocessed data using the EEMD wave height signal reconstruction formula, and enhances extreme wave height samples (such as wave height > 7m) to ensure that the data meets the model input requirements. Step 5, Wave Height Prediction: The training unit is responsible for building a width learning-LSTM hybrid network, embedding an attention mechanism, training the model using preprocessed data, calculating the predicted wave height value for the next 48 hours using the wave height prediction formula, and correcting for prediction lag. Step 6: Perform error correction: The error correction unit starts the Fortran-Python coupled framework (Fortran and Python co-computation framework), reads the initial forecast results of the WAM numerical model, extracts error features through the spatiotemporal attention neural network, calculates the AI ​​error correction amount using the wide learning-LSTM hybrid network prediction formula, and corrects the WAM forecast error; Step 7: Risk Assessment: The assessment unit inputs the corrected wave height data and, in conjunction with the required engineering structural parameters, assesses the risk of the project being affected by sea waves. Step 8: Dual-unit linkage application: The early warning unit issues early warning information based on the corrected wave height, while the support unit combines the forecast results to provide wave avoidance route suggestions for passing vessels; Step 9: Present the final results: The visualization unit displays the wave height change trend over the next 48 hours using a time series graph, shows the wave height distribution in the sea area using a spatial heat map, generates a maximum wave height intelligent forecast report, and pushes it simultaneously to marine management departments, shipping companies, and engineering units.

[0031] This invention also provides a deep learning-based intelligent maximum wave height prediction system. Figure 2 A schematic diagram of the maximum wave height intelligent prediction system based on deep learning in an embodiment of the present invention is shown. Figure 2 The system includes: a data acquisition unit, a fusion unit, a preprocessing unit, a training unit, an error correction unit, an evaluation unit, an early warning unit, and a visualization unit. The acquisition unit is configured to acquire multi-source ocean observation data; The fusion unit is configured to fuse multi-source ocean observation data acquired through data format conversion and spatiotemporal registration. The preprocessing unit is configured to decompose the original wave height signal, extract period and amplitude features through wavelet transform, reconstruct the preprocessed data using the EEMD wave height signal reconstruction formula, and enhance extreme wave height samples. The training unit is configured to build a width learning-LSTM hybrid network, embed an attention mechanism, train the model using preprocessed data, calculate the predicted wave height for future time periods using the wave height prediction formula, and correct for prediction lag. The error correction unit is configured to start the Fortran-Python coupled framework, read the initial forecast results of the WAM numerical model, extract error features through a spatiotemporal attention neural network, calculate the AI ​​error correction amount using the wide learning-LSTM hybrid network prediction formula, and correct the WAM forecast error. The assessment unit is configured to take in corrected wave height data and, in combination with the required engineering structural parameters, assess the risk of the project being affected by sea waves. The early warning unit is configured to issue early warning information based on the corrected wave height and, in conjunction with the forecast results, provide wave avoidance route suggestions for passing vessels. The visualization unit is configured to display the wave height change trend in future periods using a time series graph, display the wave height distribution in the sea area using a spatial heat map, generate a maximum wave height intelligent forecast report, and push it to marine management departments, shipping companies and engineering units simultaneously.

[0032] The data acquisition unit includes wave height and wave period monitoring stations, satellite remote sensing data monitoring stations, and shore-based radar data monitoring stations. By controlling buoys, satellites, radars, and other measurement sensors distributed throughout the ocean, it monitors multi-source ocean observation data in real time, including significant wave height, wave period, wind speed, wind direction, seawater temperature and salinity, ocean current velocity and direction, and other atmospheric pressure and wave spectrum characteristics. It also highlights the monitored significant wave height data, significant wave height data, wind speed data, wind direction data, and ocean current velocity data (supplemented with key marked data, all of which are core parameters affecting the forecast of the maximum wave height). The monitored data is then transmitted to the fusion unit for data preprocessing.

[0033] Wave height and wave period monitoring stations collect core wave condition parameters such as effective wave height, significant wave height, wave crest period, and wave trough period in real time in different sea areas through buoy sensors and underwater acoustic measurement equipment deployed in the sea area. They also record the data collection time and corresponding latitude and longitude to ensure the spatiotemporal accuracy of wave height and period data, and provide basic wave condition basis for maximum wave height prediction. Satellite remote sensing data monitoring stations receive remote sensing data transmitted back by ocean satellites (such as the HY-2 series ocean satellites) to obtain macroscopic observation information on wave height distribution, sea surface wind speed field and sea surface temperature over a wide range of sea areas, making up for the monitoring range limitations of buoys and radar equipment, and achieving monitoring coverage of the open sea and open waters. Based on high-frequency ground wave radar and X-band radar deployed along the coastline, shore-based radar data monitoring stations monitor the dynamic information of wave height changes, ocean current direction and ship movement in nearshore waters in real time. They have the advantage of high spatiotemporal resolution and can capture sudden wave conditions nearshore (such as nearshore diffracted waves and reflected waves), providing accurate data support for nearshore maximum wave height forecasting and navigation safety assurance.

[0034] The integrated air-space-sea monitoring network, consisting of monitoring stations (buoys, satellites, and radar), enables continuous observation of vast sea areas in multiple dimensions with high spatiotemporal resolution. This ultimately solves the problems of large monitoring blind spots and limited data in traditional methods, providing a solid data foundation for accurate system forecasting.

[0035] The preprocessing unit performs denoising and effective feature reconstruction of the original wave height signal according to the EEMD wave height signal reconstruction formula. The calculation formula is as follows:

[0036] In the formula, This represents the preprocessed, denoised wave height time series, which is the core data output by the module and used for subsequent training and prediction of the deep learning model module. The i-th effective intrinsic mode function (IMF) obtained by EEMD (Ensemble Empirical Mode Decomposition) reflects the wave characteristics at different scales in the original wave height signal (such as short-period wind waves and long-period swell characteristics). It needs to be screened through correlation test (to remove noise-dominated IMFs). This represents the residual components after EEMD decomposition, reflecting the overall trend change of the wave height signal. This indicates the number of valid intrinsic mode functions (IMFs) after filtering, ensuring that the reconstructed signal retains key features while removing redundant noise.

[0037] The preprocessing unit performs denoising and effective feature reconstruction of the original wave height signal according to the EEMD wave height signal reconstruction formula. By screening and integrating the intrinsic mode functions (IMF) and residual components after the original wave height signal is decomposed, noise interference is eliminated and key wave condition features are retained, resulting in a preprocessed wave height time series that meets the model input requirements.

[0038] The training unit predicts the maximum wave height for future time periods based on the wave height prediction formula, which is as follows:

[0039] In the formula, Indicates the future Maximum wave height prediction at time ( For the forecast duration, such as =12h (which represents the wave height for the next 12 hours) is the core output of the module. This indicates that the width learning system applies preprocessed wave height data. The feature mapping results transform high-dimensional preprocessed data into low-dimensional and information-rich feature vectors. The weight matrix represents the width-learned feature map. This represents the bias term of the width-learned feature map; both are optimized through model training to improve feature extraction accuracy. This indicates the effect of the attention mechanism on preprocessing wave height data. The key features are focused on output, enhancing attention to important information such as extreme wave heights and sudden wave conditions. The weights representing the attention mechanism are adjusted during training to assign importance to different features. This represents the computation process of a Long Short-Term Memory network, used to capture the temporal dependencies of wave height data (such as the influence of preceding wave heights on subsequent wave heights).

[0040] The training unit predicts the maximum wave height for future periods based on the wave height prediction formula. It combines the fast feature mapping capability of width learning with the temporal dependency capture capability of LSTM, and focuses on key influencing factors with the attention mechanism to output the maximum wave height prediction value for different future durations (such as 12 hours, 24 hours, and 48 hours).

[0041] The error correction unit corrects the wave height prediction error of the WAM numerical model based on the wide-learning-LSTM hybrid network prediction formula. The calculation formula is as follows:

[0042]

[0043] In the formula, This indicates the final accurate wave height after AI error correction, balancing the physical plausibility of the numerical model with the precise error correction capabilities of AI, and is used for subsequent engineering assessments and disaster early warning. This represents the initial predicted wave height of the WAM numerical wave model and is the basic data for error correction. This represents the wave height error correction amount generated by AI, used to correct forecast biases in the WAM model. This represents the output of the spatiotemporal attention mechanism, which predicts wave height by fusing WAM models. Compared with the measured wave height The spatiotemporal correlation information, and the spatiotemporal regions with significant positioning deviations (such as sea areas affected by typhoons). This represents the WAM wave height error mapping function, which converts the predicted WAM wave height into a corresponding deviation estimate, enabling quantitative calculation of the deviation. This represents the measured wave height data from the same period (such as buoy and radar observation data), used to train the attention mechanism and error mapping function, and to provide a reference for deviation correction.

[0044] The error correction unit corrects the wave height forecast error of the WAM numerical model based on the wide learning-LSTM hybrid network prediction formula. It captures the forecast deviation pattern of the WAM model under extreme weather (such as typhoon) through a spatiotemporal attention neural network, and generates the error correction amount by combining the measured wave height data. Finally, it obtains the corrected wave height with strong physical consistency and high accuracy.

[0045] Through an innovative technical approach that combines signal preprocessing, hybrid deep learning models, and AI-numerical models, the core shortcomings of traditional numerical models—such as low computational efficiency, poor accuracy, and significant lag in forecasting complex nonlinear processes (especially extreme weather and swells) using traditional experience and single models—we finally solved these problems.

[0046] Based on the corrected wave height data, the assessment unit evaluates the wave resistance of engineering structures such as port terminals and offshore platforms, and outputs low / medium / high risk levels, providing a unified risk criterion and decision-making basis for the assessment unit, early warning unit, and support unit.

[0047] The early warning unit combines wave height thresholds with weather trends to generate graded wave disaster early warning information in blue, yellow, orange, and red, which is then pushed to relevant departments to enable them to make early disaster prevention deployments, evacuate personnel and equipment, and activate emergency responses.

[0048] Based on real-time wave height forecasts, the support unit provides vessels with route optimization suggestions, wave avoidance area recommendations, and navigation safety level alerts.

[0049] The visualization unit outputs forecast results in the form of charts (wave height time series plots, spatial distribution maps) and reports, and supports access from multiple terminals.

[0050] By implementing a modular, interconnected application from forecasting to assessment, early warning, and support, the problem of the disconnect between forecast results and the needs of end users (such as engineers, decision-makers, and crew members) was finally solved. The original wave height values ​​were transformed into intuitive risk levels, early warning information, and action guidelines, thereby significantly enhancing the operational application value of the system.

[0051] This invention also provides an electronic device, which includes at least one processor and at least one memory, wherein the processor and the memory are data-connected. The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the deep learning-based maximum wave height intelligent forecasting method described in this embodiment of the invention.

[0052] This invention also provides a computer storage medium storing one or more instructions, which, when executed by one or more computers, cause the one or more computers to implement the deep learning-based maximum wave height intelligent forecasting method described in this invention.

[0053] Through the collaborative operation and functional closed loop of the acquisition unit, fusion unit, preprocessing unit, training unit, error correction unit, evaluation unit, early warning unit, support unit and visualization unit, the system achieves full-chain integration from multi-source heterogeneous data to accurate intelligent forecasting and then to in-depth industry applications. Finally, the system solves the problems of insufficient data utilization, inadequate model accuracy and timeliness, and disconnect between forecast results and application scenarios faced by traditional methods in maximum wave height forecasting. It provides scientific, reliable and efficient decision support for marine disaster prevention and mitigation, engineering safety operation and navigation safety assurance.

[0054] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based intelligent maximum wave height prediction method, characterized in that, Includes the following steps: Collect multi-source ocean observation data, and fuse the multi-source ocean observation data through data format conversion and spatiotemporal registration; The original wave height signal is decomposed, and the period and amplitude features are extracted by wavelet transform. The preprocessed data is reconstructed using the EEMD wave height signal reconstruction formula, and the extreme wave height samples are enhanced. A wide learning-LSTM hybrid network is constructed, with an attention mechanism embedded. The model is trained using preprocessed data, and the predicted wave height for future time periods is calculated using the wave height prediction formula, while correcting for prediction lag. The Fortran-Python coupled framework is started, the initial forecast results of the WAM numerical model are read, error features are extracted through a spatiotemporal attention neural network, and the AI ​​error correction amount is calculated using the wide learning-LSTM hybrid network prediction formula to correct the WAM forecast error. Input the corrected wave height data and combine it with the required engineering structural parameters to assess the risk of the project being affected by sea waves; Based on the revised wave height warning information and combined with the forecast results, we provide wave avoidance route suggestions for passing vessels; The system uses time series graphs to show the wave height variation trend in future periods and spatial heat maps to show the wave height distribution in the sea area. It generates intelligent forecast reports of maximum wave height and pushes them to marine management departments, shipping companies and engineering units simultaneously.

2. The deep learning-based maximum wave height intelligent prediction method according to claim 1, characterized in that: The collection of multi-source ocean observation data includes: Control buoys, satellites, radars and other measurement sensors distributed throughout the ocean to monitor in real time multi-source ocean observation data, including significant wave height, period, wind speed, wind direction, sea temperature and salinity, ocean current speed and direction, and other atmospheric pressure and wave spectrum characteristics; The effective wave height data, significant wave height data, wind speed data, wind direction data, and ocean current velocity data monitored are marked with special emphasis.

3. The deep learning-based maximum wave height intelligent prediction method according to claim 2, characterized in that: The collection of multi-source ocean observation data also includes: By deploying buoy sensors and underwater acoustic measurement equipment in the sea, core wave condition parameters such as effective wave height, significant wave height, wave crest period, and wave trough period in different sea areas are collected in real time, and the data collection time and corresponding latitude and longitude are recorded simultaneously. By receiving remote sensing data transmitted back by ocean satellites, macroscopic observation information on wave height distribution, sea surface wind speed field and sea surface temperature over a large area of ​​sea can be obtained. Based on high-frequency ground wave radar and X-band radar deployed along the coastline, the system monitors the dynamic information of wave height changes, ocean current direction, and ship movement in nearshore waters in real time.

4. The deep learning-based maximum wave height intelligent prediction method according to claim 1, characterized in that: The original wave height signal is denoised and its effective features are reconstructed using the EEMD wave height signal reconstruction formula. The calculation formula is as follows: In the formula, This represents the wave height time series after preprocessing and denoising. Let i represent the i-th effective eigenmode function obtained from EEMD decomposition. This represents the residual components after EEMD decomposition. This indicates the number of valid intrinsic mode functions after filtering.

5. The deep learning-based maximum wave height intelligent prediction method according to claim 1, characterized in that: The predicted wave height for the future period is calculated using the wave height prediction formula, which is as follows: In the formula, Indicates the future The predicted maximum wave height at time [time]. For the predicted duration, This indicates that the width learning system applies preprocessed wave height data. Feature mapping results The weight matrix represents the width-learned feature map. This represents the bias term for the width-learned feature map. This indicates the effect of the attention mechanism on preprocessing wave height data. The key features focus on the output. The weighting coefficients representing the attention mechanism. This represents the computational process of a Long Short-Term Memory (LSTM) network.

6. The deep learning-based maximum wave height intelligent prediction method according to claim 1, characterized in that: Error correction for wave height prediction in WAM numerical models is performed based on the wide-learning-LSTM hybrid network prediction formula, and the calculation formula is as follows: In the formula, This indicates the final accurate wave height after AI error correction. This indicates the initial predicted wave height of the WAM numerical wave model. This represents the wave height error correction amount generated by AI, used to correct forecast biases in the WAM model. This represents the output of the spatiotemporal attention mechanism, which predicts wave height by fusing WAM models. Compared with the measured wave height spatiotemporal correlation information, This represents the WAM wave height error mapping function. This represents the measured wave height data for the same period.

7. A deep learning-based intelligent maximum wave height prediction system, characterized in that, The system includes: an acquisition unit, a fusion unit, a preprocessing unit, a training unit, an error correction unit, an evaluation unit, an early warning unit, and a visualization unit. The acquisition unit is configured to acquire multi-source ocean observation data; The fusion unit is configured to fuse multi-source ocean observation data acquired through data format conversion and spatiotemporal registration. The preprocessing unit is configured to decompose the original wave height signal, extract period and amplitude features through wavelet transform, reconstruct the preprocessed data using the EEMD wave height signal reconstruction formula, and enhance extreme wave height samples. The training unit is configured to build a width learning-LSTM hybrid network, embed an attention mechanism, train the model using preprocessed data, calculate the predicted wave height for future time periods using the wave height prediction formula, and correct for prediction lag. The error correction unit is configured to start the Fortran-Python coupled framework, read the initial forecast results of the WAM numerical model, extract error features through a spatiotemporal attention neural network, calculate the AI ​​error correction amount using the wide learning-LSTM hybrid network prediction formula, and correct the WAM forecast error. The assessment unit is configured to take in corrected wave height data and, in combination with the required engineering structural parameters, assess the risk of the project being affected by sea waves. The early warning unit is configured to issue early warning information based on the corrected wave height and, in conjunction with the forecast results, provide wave avoidance route suggestions for passing vessels. The visualization unit is configured to display the wave height change trend in future periods using a time series graph, display the wave height distribution in the sea area using a spatial heat map, generate a maximum wave height intelligent forecast report, and push it to marine management departments, shipping companies and engineering units simultaneously.

8. The deep learning-based maximum wave height intelligent forecasting system according to claim 7, characterized in that, The system also includes a protection unit. The support unit is configured to provide vessels with route optimization suggestions, wave avoidance area recommendations, and navigation safety level alerts based on real-time wave height forecasts.

9. An electronic device, characterized in that, The electronic device includes at least one processor and at least one memory, wherein the processor and the memory are data connected. The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

10. A computer storage medium, characterized in that, The computer storage medium stores one or more instructions, which, when executed by one or more computers, cause the one or more computers to perform the method of any one of claims 1-6.

Images