Point-to-region wave height short-term prediction method and system for offshore construction early warning

Abstract

This invention relates to the field of marine engineering safety early warning and intelligent prediction, and in particular to a point-to-area short-term wave height prediction method and system for offshore construction early warning. The invention takes historical pattern wave height, historical reference monitoring wave height, and future pattern wave height of a measurement point as inputs. It constructs a short-term prediction model based on a Transformer encoder and introduces a joint loss function including absolute error, large value underestimation penalty, trend asymmetry, residual, and peak time terms to automatically output the future correction results for each measurement point. Then, based on the mapping relationship between the measurement point and the corresponding coverage area, the correction results of the measurement point are extended to the entire target area to form a corrected wave height field. Combined with construction thresholds, it automatically determines the levels of normal, attention, alert, and work stoppage. Without requiring continuous manual interpretation, it forms a complete closed loop from data acquisition, regional analysis, prediction, regional correction to graded early warning and result storage and push.

Description

Technical Field

[0001] This invention relates to the field of marine engineering safety early warning and intelligent prediction technology, and in particular to a point-to-area short-term wave height prediction method and system for early warning of offshore construction. Background Technology

[0002] Offshore construction activities, including offshore foundation construction, platform hoisting, submarine cable laying, and offshore wind power installation and maintenance, are highly sensitive to sea conditions. For a specific construction area, changes in wave height over the next few hours directly affect the stability of operational equipment, the safety of vessel operations, and whether construction needs to be suspended. Therefore, establishing a short-term wave height prediction and early warning mechanism for construction areas is of great significance for ensuring the safety of offshore construction.

[0003] Current offshore construction safety management typically relies on numerical model outputs or human experience for interpretation. On the one hand, raw model wave height results may have deviations in key characteristics such as high wave height, rapid rise process, and peak time, which can easily lead to false alarms or missed alarms if directly used for construction early warning. On the other hand, manual interpretation often requires on-duty personnel to continuously monitor data changes, making it difficult to achieve automated, continuous, and precise risk early warning.

[0004] Furthermore, existing machine learning correction methods focus more on reducing the overall average error, lacking specific constraints on high wave height underestimation, wave height rise phase changes, and peak moment offset—issues of greater concern for construction safety. Most solutions remain at the offline algorithm level, failing to form an integrated system encompassing automatic data acquisition, automatic model prediction, automatic threshold determination, and automatic result storage and display. Meanwhile, limitations in equipment cost, deployment conditions, and maintenance difficulty typically make it difficult to deploy a large number of wave height meters across the entire target area, allowing for the acquisition of accurate wave height data only at a limited number of measurement points.

[0005] Chinese patent applications CN119578311A, CN113283588A, CN104021308A, CN 117909666 A, and CN113704693 A disclose technologies in similar fields. However, these technical solutions have the following problems: First, they focus more on the comprehensive wave height numerical prediction of a single construction area and have not established a mapping relationship between "limited measurement points and coverage areas," making it difficult to generate a corrected wave height field from a point to a region. Second, they mainly rely on historical similar day matching and real-time physical prediction weighting, and have not constructed a short-term temporary sequence learning model based on the historical pattern wave height, supervised wave height, and future pattern wave height of the measurement points. Third, they have not set specific constraints for the underestimation of high wave height, the rapid rise process, and the deviation of the peak time. Fourth, they have not formed an integrated closed loop for regional risk assessment, graded early warning, result storage, and push. Therefore, there is still room for improvement in the regionalization and automation of early warning applications for marine construction safety. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems in the prior art, such as reliance on manual interpretation for short-term wave height early warning in offshore construction areas, insufficient characterization of high-value wave height predicted by the original model, limited number of measured points making it difficult to cover the entire area, and low degree of automation in early warning. This invention provides a point-to-area short-term wave height prediction method and system for offshore construction early warning.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A point-to-area short-term wave height prediction method for early warning of offshore construction includes the following steps: Step 1: Obtain historical mode wave height data of the target area, and determine several measurement points that can cover the target area and their corresponding coverage areas based on spatial correlation analysis of the historical mode wave height field; obtain the historical mode wave height data and reference supervision wave height data of the measurement points, and establish a training dataset; Step 2: Perform time alignment, length consistency verification, and sliding time window sample construction on the historical pattern wave height data and reference supervision wave height data of the measurement points to obtain model input samples and supervision labels; Step 3: Establish a short-term correction prediction model for wave height. The input of the short-term correction prediction model for wave height is the sample constructed in Step 2, and the output is the wave height prediction results for each measurement point at multiple future times. Step 4: Construct a joint loss function and train the wave height short-term correction prediction model using the training dataset; Step 5: During the online operation phase, the model wave height data of the target area for the future forecast period is automatically acquired, and the corrected wave height results of each measurement point at multiple future times are output according to the short-term correction prediction model of wave height. The correction results of each measurement point are applied to the corresponding coverage area to generate the corrected wave height field of the target area at multiple future times. Step 6: Compare the corrected wave height field with the preset construction safety threshold and automatically generate an early warning level; Step 7: Store the prediction results and early warning results, and output them to the display interface and / or message push terminal.

[0008] Furthermore, in step 1, the spatial correlation relationship between regional grid points is constructed based on the historical pattern wave height data of the target area, and the measurement points and their corresponding coverage areas are determined according to the correlation threshold, the maximum correlation principle, or the spatial coverage index.

[0009] Furthermore, in step 2, a sliding time window method is used to construct samples for the wave height sequence; the model input consists of three parts: historical mode wave height, historical reference supervision wave height, and future mode wave height, with the supervision label being the future reference supervision wave height sequence.

[0010] Furthermore, in step 3, the wave height short-term correction prediction model is a Transformer encoder-based model, including input projection, multi-head self-attention, feedforward network, layer normalization, position encoding, pooling layer and output layer, which is used to learn the temporal dependencies in the wave height time series and output the correction results for multiple future times.

[0011] Furthermore, in step 4, the joint loss function includes an absolute error term, a large value underestimation penalty term, a trend asymmetry term, a residual term, and a peak time term to enhance the model's ability to learn key features related to construction risks.

[0012] Furthermore, in step 5, the system preferably automatically acquires the pattern wave height data for the future forecast period according to a 1-hour operating cycle, and outputs the measurement point correction results and regional corrected wave height field for the next 1 to 5 hours; in other embodiments, the operating cycle can also be set to 30 minutes, 1 hour or 3 hours according to the construction task requirements.

[0013] Furthermore, in step 6, the system automatically generates an early warning level based on a preset construction safety threshold; the early warning level may include multiple levels such as normal, attention, alert, and work stoppage.

[0014] Furthermore, in step 7, the system uniformly stores the measurement point identifier, coverage area identifier, original mode input, corrected wave height result, regional corrected wave height field, warning level, over-threshold period and generation time, and provides the ability to trace and display historical results.

[0015] To achieve the above objectives, the present invention also employs the following technical solution: A point-to-regional short-term wave height prediction system for early warning of offshore construction includes: The data acquisition module is used to acquire historical model wave height data of the target area, historical model wave height data of the measurement points, reference supervision wave height data of the measurement points, and model wave height data for future forecast periods during the online operation phase. The regional analysis module is used for spatial correlation analysis, measurement point selection, and coverage area division. The data processing module is used for time alignment, length verification, anomaly detection, sliding time window construction, and model input generation. The model training module is used for offline training of the wave height short-term correction prediction model; The wave height prediction module is used to call the trained short-term wave height correction prediction model during the online operation phase and output the corrected wave height results for each measurement point at multiple future times. The regional correction module is used to apply the correction results of each measurement point to the corresponding coverage area to generate the corrected wave height field of the target area at multiple future times. The early warning determination module is used to compare the corrected wave height results with preset construction safety thresholds and output the early warning level. The results storage module is used to store raw input data, prediction results, regional correction results, warning levels, and historical records. The results display and push module is used to display the prediction curve, regional wave height distribution, warning level and time period exceeding the threshold, and push warning information to user terminals.

[0016] This invention takes historical pattern wave height, historical reference monitoring wave height, and future pattern wave height of measurement points as inputs. By constructing a short-term prediction model based on a Transformer encoder and introducing a joint loss function including absolute error, large value underestimation penalty, trend asymmetry, residual, and peak time terms, it achieves automatic output of correction results for each measurement point in the next 1 to 5 hours. Then, based on the mapping relationship between the measurement point and the corresponding coverage area, the correction results of the measurement point are extended to the entire target area to form a corrected wave height field. Combined with construction thresholds, it automatically determines the levels of normal, attention, warning, and work stoppage. Without the need for continuous manual interpretation, it forms a complete closed loop from data acquisition, regional analysis, prediction, regional correction to hierarchical early warning and result storage and push.

[0017] Compared with the prior art, the beneficial effects of this invention are as follows: (1) High degree of systematization. This invention is not a single prediction algorithm, but forms an integrated business system from automatic data acquisition, regional analysis, data processing, model prediction, regional correction, hierarchical early warning to result storage and push, which is suitable for deployment in real marine construction safety early warning scenarios.

[0018] (2) Regional early warning. This invention not only predicts based on measurement points, but also uses a limited number of measurement points to correct the wave height field and assess the risks of the entire construction area, which is closer to the actual needs of offshore construction management.

[0019] (3) Limited measurement points cover the entire area. Measurement points and their corresponding coverage areas are determined by spatial correlation analysis of historical model wave height fields, so as to achieve effective coverage of the target area while reducing the number of wave height meters deployed.

[0020] (4) Strong ability to characterize key risk features. By introducing large value underestimation penalty, trend asymmetry, residual constraint and peak time constraint, the ability to predict high wave height, rapid rise process and peak time is improved.

[0021] (5) Suitable for automated early warning. The prediction results can be directly linked with the construction safety threshold, automatically outputting graded early warning conclusions, reducing the burden of manual interpretation and improving response efficiency.

[0022] (6) Facilitates result management and traceability. The system supports unified storage of prediction inputs, outputs and early warning records, which facilitates post-event review, model evaluation and construction decision tracking.

[0023] (7) High computational efficiency and easy business integration. The computational cost in the inference stage is significantly lower than that of traditional nested numerical simulation, which can meet the requirements of high-frequency updates and rapid response, and is suitable for integration with port areas, offshore engineering and coastal disaster prevention and mitigation business systems. The above-mentioned benefits such as "high degree of systematization, automated early warning and traceable results" expand the application object from "point" to "region". Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention; Figure 2 A schematic diagram showing the layout of the measurement points; Figure 3 A schematic diagram showing the division of the coverage area corresponding to the measurement points; Figure 4 Comparison of short-term correction results and errors for effective wave height at typical measurement points in real-world cases. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] The structures, proportions, and sizes illustrated in the accompanying drawings are merely for illustrative purposes and to aid those skilled in the art in understanding and reading the invention. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, provided they do not affect the effectiveness or purpose of the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the terms "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0027] A point-to-area short-term wave height prediction system for early warning of offshore construction includes: The data acquisition module is used to acquire historical model wave height data of the target area, historical model wave height data of the measurement points, reference supervision wave height data of the measurement points, and model wave height data for future forecast periods during the online operation phase. The regional analysis module is used for spatial correlation analysis, measurement point selection, and coverage area division. The data processing module is used for time alignment, length verification, anomaly detection, sliding time window construction, and model input generation. The model training module is used for offline training of the wave height short-term correction prediction model; The wave height prediction module is used to call the trained short-term wave height correction prediction model during the online operation phase and output the corrected wave height results for each measurement point at multiple future times. The regional correction module is used to apply the correction results of each measurement point to the corresponding coverage area to generate the corrected wave height field of the target area at multiple future times. The early warning determination module is used to compare the corrected wave height results with preset construction safety thresholds and output the early warning level. The results storage module is used to store raw input data, prediction results, regional correction results, warning levels, and historical records. The results display and push module is used to display the prediction curve, regional wave height distribution, warning level and time period exceeding the threshold, and push warning information to user terminals.

[0028] Point-to-regional short-term wave height prediction method for early warning of offshore construction, such as Figure 1 As shown, it includes the following steps: Step 1: Obtain historical pattern wave height data for the target area, determine measurement points and coverage areas, and establish a training dataset. In this embodiment, the historical pattern wave height data of the target area can be the regional wave height field sequence output by the SWAN mode. Let the target area be discretized as... Given N discrete grid points, the mode wave height of the j-th grid point at time t is denoted as: , j =1,2,… N , t =1,2,… T， Wherein, the superscript m represents the model wave height, that is, the wave height value output by the numerical model; T represents the total number of moments in the historical model wave height time series.

[0029] To characterize the consistency of wave height variations at different locations within the region, the Spearman correlation coefficient of the historical mode wave height sequences of any two grid points i and j is calculated and expressed as follows: ; in, R i ( t )and R j ( t ) represent the rank values ​​of the mode wave height sequences of grid points i and j at time t, respectively.

[0030] Based on the correlation analysis results, the coverage area corresponding to each measurement point is divided. Furthermore, the spatial coverage rate of the k-th measurement point is defined as: ; Ω k This represents the coverage area corresponding to the k-th measurement point. A j This represents the area of ​​the j-th grid cell. When a regular grid is used, all grid cells have the same area, and the above formula can be used to measure the coverage of the target area by the measurement point.

[0031] like Figure 2 As shown, the selected measurement points are distributed at different representative locations within the target construction sea area, preferably covering the core construction area, main operation channels, and areas with significant changes in wave propagation characteristics. This deployment method can maximize the characterization of wave height changes in the target area while minimizing the number of observation devices.

[0032] After determining the measurement points, the coverage area is further divided based on the historical wave height correlation between each regional grid point and each measurement point. Specifically, each regional grid cell is assigned to the measurement point with the highest correlation to its historical wave height changes, thus forming a "measurement point - coverage area" correspondence. For example... Figure 3 As shown, the target area is divided into several coverage areas, each corresponding to a measurement point. During subsequent online operation, the correction result of this measurement point will be used for the regional wave height correction of its corresponding coverage area.

[0033] After determining the measurement points, historical model wave height data corresponding to each measurement point are extracted, and reference supervision wave height data for each measurement point is obtained. The reference supervision wave height data can be measured data from a wave height meter. Training data can be data from 2015 to 2022, and validation data can be data from 2023. After reading, the consistency of the lengths of the model wave height sequence and the reference supervision wave height sequence is checked.

[0034] Step 2: Perform time alignment, length consistency verification, and sliding time window sample construction on the historical pattern wave height data and reference supervision wave height data of the measurement points to obtain model input samples and supervision labels.

[0035] In this embodiment, a sliding time window method is used to construct samples. For any time index t, past data is extracted. H PAST The pattern wave height sequence at each moment, past H PAST Reference monitoring wave height sequence at each moment and the future H FUT The pattern wave height sequence at each moment is spliced ​​together to form the future. H FUT The reference supervisory wave height sequence at each time point is used as the supervisory label; the model input sample is represented as: ; The supervisory label is represented as: ; S k This represents the mode wave height sequence at the k-th measurement point. E k Let represent the reference monitoring wave height sequence at the k-th measurement point, and t represent the time index. In one embodiment, H PAST =3, H FUT =5, therefore the length of the input sequence is 11, and the output is the wave height at the next 5 moments.

[0036] Step 3: In this embodiment, the wave height short-term correction prediction model adopts a Transformer encoder-based structure. The model input is a time series sample constructed using a sliding time window. The model first maps the input to the feature space through an input projection layer, and then sequentially passes through multiple TransformerBlock encoding layers. Each TransformerBlock includes a multi-head self-attention layer, a feedforward network, residual connections, a layer normalization layer, and a Dropout layer. To preserve temporal sequence information, the model introduces learnable positional encoding; the encoding result is then passed through a global average pooling layer and a fully connected layer to output the wave height prediction results for multiple future time periods.

[0037] Step 4: Construct a joint loss function and train the wave height short-term correction prediction model using the training dataset.

[0038] In this embodiment, a joint loss function is used during model training, and its expression is as follows: ; Where λ1, λ2, λ3, λ4, and λ5 are the weighting coefficients corresponding to each loss term; L abc Indicates absolute error loss, L under-peak This indicates that the large value underestimates the penalty loss. L trend-asym Indicates trend asymmetric loss, L res Indicates residual loss, L peak-time This indicates the loss at the peak time.

[0039] The large value underestimation penalty loss identifies high wave height samples by using the quantile threshold of the supervised labels in the training set; the trend asymmetric loss assigns greater weight to the error during the wave height rise phase; the residual loss is constrained by the relative error between the actual correction and the predicted correction; and the peak moment loss simultaneously constrains the peak amplitude error and the peak occurrence time error.

[0040] The true correction at the measurement point is expressed as: ; in, This indicates that the k-th measurement point is at time [time]. t Reference monitoring wave height, Indicates the first k Each measurement point at time [time] t The pattern wave height.

[0041] The residual loss is used to constrain the difference between the model's predicted correction and the actual correction. In this embodiment, a high-value threshold is calculated based on the statistical results of the supervision labels in the training set, using the 90th quantile of the supervision labels in the training set as the high-value identification threshold. Model training employs the Adam optimizer, combined with EarlyStopping and ReduceLROnPlateau mechanisms for training control.

[0042] Step 5: During the online operation phase, the system automatically acquires the model wave height data for the target area during future forecast periods and outputs the corrected wave height results for each measurement point at multiple future times based on the short-term wave height correction prediction model. Let the k-th measurement point at a future time... The prediction error correction amount is Then for the area Ω covered by this measurement point k For any grid point j in a region, its corrected wave height is expressed as: ; in, This represents the mode wave height at grid point j in the region. This indicates the corrected wave height.

[0043] Using the above method, the prediction correction results from a limited number of measurement points can be extended to the entire target area, generating corrected wave height fields for multiple future time periods. The system preferably runs once every hour, automatically reading model wave height field data for the future forecast period each time, extracting the corresponding model wave height sequence for each measurement point, and generating regional correction results for the target area for the next 1 to 5 hours.

[0044] Step 6: In step 6, the corrected wave height field is compared with a preset construction safety threshold to automatically generate an early warning level. Let the target construction area be... Ω workThe maximum corrected wave height of the target construction area during the future forecast period is defined as: ; in, L f This indicates the number of forecast steps included in the future forecast period, i.e., the total number of future moments involved in the calculation of the maximum corrected wave height; Let the first threshold, the second threshold, and the third threshold be η1, η2, and η3, respectively, and satisfy: ; The warning level is represented as follows: ; The threshold can be configured based on the type of construction vessel, construction method, equipment operating conditions, and project requirements.

[0045] Step 7: The system writes the results of each run to a database or file system. The stored information includes the measurement point number, coverage area number, run time, original model wave height sequence for multiple future time points, corrected wave height sequence, regional corrected wave height field, warning level, maximum predicted wave height, time period exceeding the threshold, and push status. The results display and push module can output results via a web-based visual interface, a large screen interface, daily reports, or message notifications. When the system determines that a warning or work stoppage level has been reached, it can automatically push a warning message to the construction management personnel's terminal so that timely measures can be taken.

[0046] This invention takes historical pattern wave height, historical reference monitoring wave height, and future pattern wave height of measurement points as inputs. By constructing a short-term prediction model based on a Transformer encoder and introducing a joint loss function including absolute error, large value underestimation penalty, trend asymmetry, residual, and peak time terms, it achieves automatic output of correction results for each measurement point in the next 1 to 5 hours. Then, based on the mapping relationship between the measurement point and the corresponding coverage area, the correction results of the measurement point are extended to the entire target area to form a corrected wave height field. Combined with construction thresholds, it automatically determines the levels of normal, attention, warning, and work stoppage. Without the need for continuous manual interpretation, it forms a complete closed loop from data acquisition, regional analysis, prediction, regional correction to hierarchical early warning and result storage and push.

[0047] To verify the feasibility and effectiveness of the method of this invention, this embodiment selects two independent test periods at a typical measurement point in the target sea area for comparative analysis: from 00:00 on November 27, 2025 to 23:00 on December 3, 2025, and from 00:00 on December 18, 2025 to 23:00 on December 24, 2025. For both periods, real observed buoy data are used as reference monitoring values, and compared with the original SWAN numerical forecast results and the short-term correction results output by the method of this invention.

[0048] like Figure 4 As shown, during both test periods, the original SWAN forecast results deviated significantly from the observed buoy results, especially during the peak wave height stage and the rapid wave height change stage, where the original forecast exhibited a certain degree of amplitude deviation and insufficient process tracking. However, after applying the short-term correction method of this invention, the correction curve more closely approximates the observed buoy results overall, demonstrating better fitting performance near the peak, during the rising phase, and during the falling phase.

[0049] Specifically, during the test period from 00:00 on November 27, 2025 to 23:00 on December 3, 2025, the mean absolute error (MAE) of the original SWAN forecast was 0.211, and the root mean square error (RMSE) was 0.360. After correction using the method of this invention, the MAE decreased to 0.048, and the RMSE decreased to 0.073. During the test period from 00:00 on December 18, 2025 to 23:00 on December 24, 2025, the MAE of the original SWAN forecast was 0.227, and the RMSE was 0.282. After correction using the method of this invention, the MAE decreased to 0.049, and the RMSE decreased to 0.072. Therefore, the method of this invention significantly reduced the prediction error in both independent test periods, indicating that the method has good stability and engineering applicability.

[0050] Depend on Figure 4 It can also be seen that the method of the present invention not only improves the overall error level, but also performs better during high wave height periods when offshore construction risks are sensitive, and can more accurately track wave height peaks and their occurrence process. This shows that the joint loss constraint and short-term correction mechanism based on time series characteristics proposed in this invention have a better characterization ability for high wave heights, upward trends and peak times that are of concern in construction safety early warning.

[0051] Combination Figure 3 The coverage area division results shown can... Figure 4The correction results of the typical measurement points shown are further extended to their corresponding coverage areas, thereby forming a regional corrected wave height field, and based on this, a graded early warning judgment for the target construction area in the next 1 to 5 hours is completed. This shows that the present invention is not only applicable to single-point wave height correction, but can realize short-term corrected prediction of wave height and graded early warning of construction safety for the entire target area under limited measurement point conditions, which has practical application feasibility and innovation.

[0052] In summary, this embodiment demonstrates that the present invention can utilize historical model data and observation data from a limited number of measurement points in the target sea area to establish a mapping relationship between measurement points and the coverage area, thereby enabling short-term wave height correction and regional early warning output. Compared to directly using the original numerical model results, it has higher prediction accuracy and stronger engineering application value.

[0053] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A point-to-regional short-term wave height prediction method for early warning of offshore construction, characterized in that, Includes the following steps: Step 1: Obtain historical mode wave height data of the target area, and determine several measurement points that can cover the target area and their corresponding coverage areas based on spatial correlation analysis of the historical mode wave height field; obtain the historical mode wave height data and reference supervision wave height data of the measurement points, and establish a training dataset; Step 2: Perform time alignment, length consistency verification, and sliding time window sample construction on the historical pattern wave height data and reference supervision wave height data of the measurement points to obtain model input samples and supervision labels; Step 3: Establish a short-term correction prediction model for wave height. The input of the short-term correction prediction model for wave height is the sample constructed in Step 2, and the output is the wave height prediction results for each measurement point at multiple future times. Step 4: Construct a joint loss function and train the wave height short-term correction prediction model using the training dataset; Step 5: During the online operation phase, the model wave height data for the future forecast period of the target area is automatically acquired, and the corrected wave height results for each measurement point at multiple future times are output according to the short-term correction prediction model of the wave height. The correction results of each measurement point are applied to the corresponding coverage area to generate the corrected wave height field for the target area at multiple future times. Step 6: Compare the corrected wave height field with the preset construction safety threshold, and automatically generate an early warning level; Step 7: Store the prediction results and early warning results, and output them to the display interface and / or message push terminal.

2. The point-to-regional short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 1, the spatial correlation between grid points in the target area is constructed based on historical pattern wave height data of the target area, and the measurement points and their corresponding coverage areas are determined according to the correlation threshold, the maximum correlation principle or the spatial coverage index.

3. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 2, characterized in that, In step 1, the target region is discretized into N discrete grid points, and the mode wave height of the j-th grid point at time t is denoted as . , j =1,2,… N , t =1,2,… T， in, The superscript m represents the model wave height, i.e., the wave height value output by the numerical model; T represents the total number of time points in the historical model wave height time series, and the Spearman correlation coefficient between any two grid points i and j is expressed as: ; in, R i ( t )and R j ( t ) represent the rank values ​​of the mode wave height sequences of grid points i and j at time t, respectively; the spatial coverage of the k-th measurement point is expressed as: ; Among them, Ω k This represents the coverage area corresponding to the k-th measurement point. A j This represents the area of ​​the j-th grid cell.

4. The point-to-regional short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 2, a sliding time window method is used to construct samples. For any time index t, past data is extracted. H PAST The pattern wave height sequence at each moment, past H PAST Reference monitoring wave height sequence at each moment and the future H FUT The pattern wave height sequence at each moment is spliced ​​together to form the future. H FUT The reference supervisory wave height sequence at each time point is used as the supervisory label; the model input sample is represented as: ; The supervision label is represented as: ; in, S k This represents the mode wave height sequence at the k-th measurement point. E k This represents the reference monitoring wave height sequence at the k-th measurement point.

5. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 3, the wave height short-term correction prediction model is a time series prediction model based on a Transformer encoder, including an input layer, an input projection layer, multiple TransformerBlock coding layers, a position coding layer, a pooling layer, a fully connected layer, and an output layer; wherein, the TransformerBlock coding layer includes a multi-head self-attention layer, a feedforward network, residual connections, a layer normalization layer, and a Dropout layer.

6. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 4, the joint loss function includes absolute error loss. L abc Underestimating the penalty loss L under-peak Trend asymmetric loss L trend-asym Residual loss L res and peak time loss L peak-time Its total loss function is expressed as: ; Where λ1, λ2, λ3, λ4, and λ5 are the weighting coefficients corresponding to each loss term; the true correction at the measurement point is expressed as: ; in, This indicates that the k-th measurement point is at time [time]. t Reference monitoring wave height, Indicates the first k Each measurement point at time [time] t The pattern wave height.

7. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 5, let the k-th measurement point be at a future time. The prediction error correction amount is Then for the area Ω covered by this measurement point k For any grid point j in a region, its corrected wave height is expressed as: ; in, This represents the mode wave height at grid point j in the region. This indicates the corrected wave height.

8. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 7, characterized in that, In step 6, the target construction area is set as... Ω work The maximum corrected wave height of the target construction area during the future forecast period is defined as: ; in, L f This represents the number of forecast steps included in the future forecast period, i.e., the total number of future moments participating in the calculation of the maximum corrected wave height; let the first threshold, the second threshold, and the third threshold be η1, η2, and η3, respectively, and satisfy: ; The warning level is represented as follows: 。 9. The point-to-area short-term wave height prediction method for early warning of offshore construction as described in claim 1, characterized in that, In step 7, the stored content includes at least the measurement point identifier, coverage area identifier, forecast time, original model wave height sequence, corrected wave height sequence, regional corrected wave height field, warning level, over-threshold period, and result generation time.

10. A point-to-area short-term wave height prediction system for early warning of offshore construction, characterized in that, include: The data acquisition module is used to acquire historical model wave height data of the target area, historical model wave height data of the measurement points, reference supervision wave height data of the measurement points, and model wave height data for future forecast periods during the online operation phase. The regional analysis module is used for spatial correlation analysis, measurement point selection, and coverage area division. The data processing module is used for time alignment, length verification, anomaly detection, sliding time window construction, and model input generation. The model training module is used for offline training of the wave height short-term correction prediction model; The wave height prediction module is used to call the trained short-term wave height correction prediction model during the online operation phase and output the corrected wave height results for each measurement point at multiple future times. The regional correction module is used to apply the correction results of each measurement point to the corresponding coverage area to generate the corrected wave height field of the target area at multiple future times. The early warning determination module is used to compare the corrected wave height results with preset construction safety thresholds and output the early warning level. The results storage module is used to store raw input data, prediction results, regional correction results, warning levels, and historical records. The results display and push module is used to display the prediction curve, regional wave height distribution, warning level and time period exceeding the threshold, and push warning information to user terminals.

Images