Marine environment accurate forecasting method and system based on multi-scene dynamic matching

Abstract

The invention relates to the technical field of marine environment forecasting, in particular to a marine environment accurate forecasting method and system based on multi-scene dynamic matching, and the method comprises the steps: firstly constructing a dynamically updated scene feature library through unsupervised learning, and training a hierarchical management sub-model for each scene to balance the precision and efficiency; during forecasting, through a real-time scene identification and fuzzy matching mechanism, a corresponding model is called, and real-time data is used for dynamic correction and rapid learning; and finally, outputting a forecasting result with confidence, forming closed-loop optimization through continuous evaluation, and realizing precision and adaptive evolution of marine environment forecasting. According to the invention, the accuracy of marine environment accurate forecasting based on multi-scene dynamic matching is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine environment prediction, and in particular to a marine environment accurate prediction method and system based on multi-scene dynamic matching. BACKGROUND

[0002] Traditional marine environment prediction methods mainly rely on fixed physical models and parameterization schemes, which perform well under conventional conditions with rich data and mature experience. However, in the face of extreme weather, data scarce areas or complex terrain, the limitations of such methods are increasingly evident: their prediction ability is limited, and they lack adaptability due to fixed models. In addition, static model parameters and lagging historical data cannot match the dynamically changing actual marine environment, resulting in large prediction errors during sudden weather processes such as typhoons. At the same time, existing methods lack an efficient closed-loop mechanism that can dynamically match the optimal model according to the scene and achieve self-management and continuous optimization of model performance, causing bottlenecks in prediction accuracy and efficiency. SUMMARY

[0003] The present application provides a marine environment accurate prediction method and system based on multi-scene dynamic matching, which solves the problem of limited prediction ability, poor model adaptability and lack of self-optimization capability of traditional marine environment prediction methods in complex scenes such as extreme weather and data scarcity.

[0004] The purpose of the present application can be achieved by the following technical solutions: The first aspect of the present application is to provide a marine environment accurate prediction method based on multi-scene dynamic matching, comprising the following steps: S1: Collecting multi-source historical data of the target area, using unsupervised learning methods to classify the historical data into scenes, and dividing into multiple typical marine environment scenes; extracting key meteorological factors for each scene, constructing a scene feature database, and periodically dynamically updating the scene features based on new data; S2: Selecting and training an independent machine learning sub-model for each typical scene; dividing the sub-models into multiple levels according to their performance indicators on the validation set, and implementing differentiated training frequency and parameter optimization strategies for sub-models of different levels; S3: Input real-time observation data, use the trained scene classification model to identify the scene type it belongs to, if it cannot be accurately matched, enable fuzzy matching mechanism; S4: According to the identified scene, call the corresponding sub-model for prediction calculation, and use real-time observation data to dynamically correct the prediction result; when it is detected that the scene features deviate significantly from the historical records, start the fast learning mechanism to update the model; S5: output the prediction results of key variables and confidence, and evaluate and long-term monitor the prediction results by performance indicators for continuous improvement of the model.

[0005] Further, in the step S1, the unsupervised learning method comprises K-means clustering, DBSCAN or principal component analysis; and the periodic dynamic updating is performed once every quarter.

[0006] Further, in the step S2, the performance indicator is root mean square error; and the hierarchical management specifically comprises: dividing the sub-models into three levels of "excellent", "good" and "to be optimized", and automatically triggering a parameter re-adjustment process for the "to be optimized" sub-models, while reducing the training frequency of the "excellent" sub-models.

[0007] Further, in the step S2, the machine learning sub-models comprise support vector machines, random forests or long short-term memory networks.

[0008] Further, in the step S3, the fuzzy matching mechanism is to call the features of the nearest neighbor scene most similar to the features of the current real-time data for matching.

[0009] Further, in the step S4, the dynamic correction method is bias correction or data fusion.

[0010] The second aspect of the application provides a marine environment precise prediction system based on multi-scene dynamic matching, comprising: A scene classification and feature extraction module is used to collect multi-source historical data of a target area, classify the historical data by using an unsupervised learning method, and divide a plurality of typical marine environment scenes; key meteorological factors are extracted for each scene, a scene feature database is constructed, and the scene features are periodically dynamically updated based on new data; A model training and hierarchical management module is used to select and train independent machine learning sub-models for each typical scene; the sub-models are divided into a plurality of levels according to performance indicators on a validation set, and different levels of sub-models are implemented with differentiated training frequency and parameter optimization strategies; A real-time scene recognition module is used to input real-time observation data, recognize the scene type to which the current belongs by using a trained scene classification model, and enable a fuzzy matching mechanism if accurate matching is not possible; A scene adaptive prediction and optimization module is used to call corresponding sub-models for prediction calculation according to the recognized scene, and dynamically correct the prediction results by using real-time observation data; when it is detected that the scene features significantly deviate from historical records, a fast learning mechanism is started to update the model; The prediction result output and optimization module is used for outputting the prediction result and confidence of the key variable, and evaluating and long-term monitoring the prediction result by using the performance index, and is used for continuous improvement of the model.

[0011] Further, the prediction system supports real-time data access, dynamic allocation of computing tasks, and provides a user interaction interface to support user-defined prediction scenarios or variables.

[0012] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the marine environment accurate prediction method based on multi-scene dynamic matching.

[0013] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to realize the marine environment accurate prediction method based on multi-scene dynamic matching.

[0014] Compared with the prior art, the beneficial effects of the present application are: collecting multi-source historical data of the target area, using an unsupervised learning method to classify the historical data into various typical marine environmental scenes, and extracting key meteorological factors for each scene to construct a scene feature database and periodically dynamically update the scene features based on new data; by constructing a dynamically updated scene feature library, the scene recognition deviation problem caused by lagging historical data is solved, laying a self-adaptive foundation for accurate prediction; an independent machine learning sub-model is selected and trained for each typical scene; according to the performance indicators of the sub-model on the validation set, the sub-model is divided into multiple levels, and different levels of sub-models are implemented with differentiated training frequency and parameter optimization strategies; through hierarchical management of sub-models and differentiated optimization strategies, efficient allocation of limited computing resources is realized, which significantly improves the system running efficiency while ensuring the overall prediction accuracy; input real-time observation data, use the trained scene classification model to identify the scene type it belongs to, if it cannot be accurately matched, enable the fuzzy matching mechanism; with the help of the fuzzy matching mechanism, the continuity of the prediction service is ensured when facing unforeseen or transitional weather, and the robustness and practicality of the method are enhanced; according to the identified scene, the corresponding sub-model is called for prediction calculation, and the real-time observation data is used to dynamically correct the prediction result; when it is detected that the scene features deviate significantly from the historical records, the fast learning mechanism is started to update the model; through real-time correction and fast learning mechanism, the model can dynamically adapt to environmental changes and correct errors in time, thereby significantly improving the real-time and accuracy of the prediction; output the prediction results and confidence of the key variables, and use the performance indicators to evaluate and monitor the prediction results for long-term, for the continuous improvement of the model; by outputting the confidence and establishing an evaluation closed loop, the model provides a data-driven feedback for continuous optimization, ensuring the long-term reliability and evolution ability of the system; the problems of limited prediction ability, poor model adaptability and lack of self-optimization ability of traditional marine environment prediction methods in complex scenes such as extreme weather and data scarcity are solved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, brief descriptions will be given below to the drawings needed to be used in the embodiments or prior art descriptions. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any inventive labor.

[0016] Figure 1 A step flowchart of a marine environment accurate prediction method based on multi-scene dynamic matching is provided for the present application. Figure 2 A module flowchart of a marine environment accurate prediction system based on multi-scene dynamic matching is provided for the present application. DETAILED DESCRIPTION

[0017] In order to better understand the technical scheme of the present application, the technical scheme in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work should belong to the scope of protection of the present application.

[0018] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0019] In view of the problems existing in the background art, the marine environment accurate prediction method and system based on multi-scene dynamic matching are designed and researched, which has important practical significance.

[0020] As shown in Figure 1 The first aspect of the present application is to provide a marine environment accurate prediction method based on multi-scene dynamic matching, comprising the following steps: S1: Collecting multi-source historical data of the target area, using unsupervised learning method to classify the historical data, and dividing into multiple typical marine environment scenes; extracting key meteorological factors for each scene, constructing a scene feature database, and periodically dynamically updating the scene features based on new data.

[0021] Among them, the unsupervised learning method includes K-means clustering (K-means clustering algorithm, K-means clustering algorithm), DBSCAN (density-based spatial clustering of applications with noise, density-based spatial clustering of applications with noise) or principal component analysis; the periodic dynamic update is performed once every quarter.

[0022] Among them, the multi-source historical data includes observation data (wind speed, wind direction, air temperature, precipitation, etc.), model output data and reanalysis data.

[0023] wherein, the typical scenarios include smooth weather, strong wind, typhoon, rainfall, wave, etc.

[0024] wherein, the key meteorological factors include wind speed, sea wave height, pressure gradient, etc.

[0025] It should be noted that the core purpose of this step is to decompose the complex and variable marine environment system into several representative and characteristic typical scenario modes, and to establish a characteristic knowledge base that can keep pace with the times. Specifically, through unsupervised learning of multi-source historical data, key scenario categories such as “typhoon” and “smooth weather” can be automatically induced from massive data. The subsequently extracted and constructed feature database provides a precise data basis and discriminant basis for “tailoring” exclusive prediction models for each scenario. And its periodic dynamic updating mechanism ensures that the knowledge base can continuously learn and adapt to climate change and new extreme weather patterns, fundamentally solving the scenario recognition deviation problem caused by lagging historical data, and laying a solid foundation for the adaptability and long-term accuracy of the entire prediction system.

[0026] S2: Select and train an independent machine learning sub-model for each typical scenario; according to the performance indicators of the sub-model on the validation set, divide it into multiple levels, and implement differentiated training frequency and parameter optimization strategies for sub-models of different levels.

[0027] wherein, the performance indicator is the root mean square error (RMSE); the hierarchical management is specifically: dividing the sub-model into “excellent”, “good”, and “to be optimized” three levels, and automatically triggering the parameter re-adjustment process for the “to be optimized” sub-model, while reducing the training frequency of the “excellent” level sub-model.

[0028] wherein, the machine learning sub-model includes support vector machine, random forest or long short-term memory network.

[0029] The three levels are specifically: excellent (RMSE <0.5); good (0.5≤RMSE<1.0); to be optimized (RMSE≥1.0).

[0030] It should be noted that the core purpose of this step is to achieve the optimal balance between accuracy and efficiency of the prediction system through the strategy of "divide and conquer" and "precise measures". Specifically, training an independent sub-model for each typical scenario can enable each model to focus on learning the physical laws in a specific scenario, thereby significantly improving its prediction accuracy in that scenario. And based on the performance indicators, the sub-models are managed hierarchically, which builds an intelligent resource allocation mechanism: the system can prioritize the limited computing resources for optimizing the "to-be-optimized" models with poor performance, while reducing unnecessary repeated training of the "excellent" models. This closed-loop optimization mechanism not only ensures the steady improvement of overall prediction accuracy, but also effectively avoids the waste of computing resources, making the entire system both high-performance and efficient.

[0031] S3: Input real-time observation data, use the trained scene classification model to identify the current scene type, if it cannot be accurately matched, enable the fuzzy matching mechanism.

[0032] Among them, the fuzzy matching mechanism is to call the features of the nearest scene similar to the current real-time data features for matching.

[0033] It should be noted that the core purpose of this step is to quickly and accurately match the most suitable prediction model for the real-time changing marine environment, ensuring the robustness and practicality of the system in actual operation. Specifically, using the trained scene classification model to identify real-time data can achieve fast and accurate mapping from real-time data stream to the preset optimal sub-model, thereby activating the optimal prediction strategy for the current conditions. And the fuzzy matching mechanism introduced as a key fault-tolerant and generalization means, when encountering abnormal weather not fully covered in the historical database or transitional, the system will not fail, but can call the nearest scene model for approximate prediction, thereby ensuring the continuity and reliability of the prediction service in complex real environment, effectively avoiding the prediction interruption or failure caused by scene misjudgment or mismatch.

[0034] S4: According to the identified scene, call the corresponding sub-model for prediction calculation, and use real-time observation data to dynamically correct the prediction results; when it is detected that the scene features deviate significantly from the historical records, start the fast learning mechanism to update the model.

[0035] Among them, the dynamic correction method is bias correction or data fusion.

[0036] It should be noted that the core purpose of this step is to convert the static prediction capability constructed in the previous steps into a dynamic adaptive system that can learn online and correct itself to cope with the real-time variability and uncertainty of the marine environment. Specifically, according to the identification of the scene, the dedicated sub-model is called to ensure the high pertinence of the prediction algorithm, laying a foundation for generating high-precision initial results. Dynamic correction using real-time observation data is equivalent to installing a "real-time calibrator" for the model, which can immediately correct the cumulative error of the model or the deviation caused by sudden disturbances, significantly improving the accuracy of short-term prediction. Further, when the system detects that the scene characteristics deviate significantly from the historical pattern, the triggered rapid learning mechanism enables it to "draw an analogy from one thing to another" and "evolve", enabling it to quickly integrate new knowledge into the model, thereby effectively adapting to unprecedented or continuously changing extreme conditions, greatly enhancing the robustness and prediction reliability of the entire system in the long run.

[0037] S5: Output the prediction results and confidence of the key variables, and evaluate and monitor the prediction results using performance indicators for continuous improvement of the model.

[0038] It should be noted that the core purpose of this step is to quantify and trust the output results of the prediction system, and establish a closed-loop, data-driven continuous optimization loop, thereby ensuring the reliability of the prediction service and the self-evolution ability of the method. Specifically, the output of the prediction results with confidence not only provides users with intuitive decision-making basis (such as high confidence for decisive action and low confidence for cautious reference), but also transparently manages the uncertainty of the prediction. Long-term evaluation and monitoring based on performance indicators are like installing an "operation dashboard" for the system, which can continuously track the performance of the model in actual application and accurately locate the link of performance degradation or failure. These accumulated evaluation data and error records provide crucial feedback signals for model hierarchical retraining in step S2 and scene feature update in step S1, driving the entire system to evolve from a static prediction tool to an intelligent agent that can continuously learn from practice and self-improve.

[0039] In this scheme, through quarterly scene feature update and sub-model performance grading management, the dynamic adaptation of scene and model is realized, the prediction error under extreme weather scene (such as sudden typhoon) is reduced by more than 25%, and the scene recognition deviation problem caused by historical data lag is solved; In the scene classification stage, the multi-dimensional features of observation data, model output data and reanalysis data are fused, and the potential correlation between data (such as the coupling relationship between wind speed and wave height) is mined through unsupervised learning algorithm, compared with single data source scene classification, the scene recognition accuracy is improved by 30%, which provides more accurate scene basis for subsequent sub-model calling; Performance closed loop is constructed, through automatic triggering parameter re-adjustment and dynamic adjustment of training frequency, the performance improvement of "to be optimized" model is guaranteed, and the invalid training of "optimized" model is reduced, the overall prediction system running efficiency is improved by 40%, and the prediction accuracy stability under multiple scenes is ensured.

[0040] Thus, the embodiment is completed.

[0041] As shown in Figure 2 The second aspect of the present application is to provide a marine environment accurate prediction system based on multi-scene dynamic matching, comprising: The scene classification and feature extraction module 101 is used for collecting multi-source historical data of the target area, classifying the historical data by using an unsupervised learning method, and dividing a plurality of typical marine environment scenes; key meteorological factors are extracted for each scene, a scene feature database is constructed, and the scene features are periodically dynamically updated based on new data; The model training and grading management module 102 is used for selecting and training an independent machine learning sub-model for each typical scene; the performance indicators of the sub-model on the verification set are divided into a plurality of grades, and the sub-models of different grades are implemented with differentiated training frequency and parameter optimization strategies; The real-time scene recognition module 103 is used for inputting real-time observation data, recognizing the scene type to which the current belongs by using the trained scene classification model, and enabling a fuzzy matching mechanism if accurate matching cannot be achieved; The scene adaptive prediction and optimization module 104 is used for calling the corresponding sub-model for prediction calculation according to the recognized scene, and dynamically correcting the prediction result by using real-time observation data; when it is detected that the scene features significantly deviate from the historical record, a fast learning mechanism is started to update the model; The prediction result output and optimization module 105 is used for outputting the prediction result and confidence of the key variable, and evaluating and long-term monitoring the prediction result by using the performance indicators, for continuous improvement of the model.

[0042] Among them, the prediction system supports real-time data access, dynamic allocation of computing tasks, and provides a user interaction interface to support user-defined prediction scenes or variables.

[0043] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the marine environment accurate prediction method based on multi-scene dynamic matching when executing the computer program.

[0044] The fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program implements the marine environment accurate prediction method based on multi-scene dynamic matching when executed by a processor.

[0045] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product implemented on one or more computer readable storage media containing computer usable program code (including but not limited to disk storage, optical storage, etc.).

[0046] The present application is described with reference to flowcharts and / or block diagrams of the method, system, and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flow Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The function specified in one flow or multiple flows and / or blocks

[0047] These computer program instructions can also be stored in a computer readable storage medium capable of guiding a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the flow Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The function specified in one flow or multiple flows and / or blocks

[0048] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer implemented process, so that the instructions executed on the computer or other programmable device provide a process for implementing the flow Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The function specified in one flow or multiple flows and / or blocks

[0049] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit it. Although the present application has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the specific embodiments of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, and any modification or equivalent replacement should be covered within the protection scope of the present application.

Claims

1. A precise marine environment forecasting method based on multi-scenario dynamic matching, characterized in that, Includes the following steps: S1: Collect multi-source historical data of the target area, use unsupervised learning methods to classify the historical data into various typical marine environmental scenarios, extract key meteorological factors for each scenario, construct a scenario feature database, and periodically update the scenario features based on new data. S2: Select and train independent machine learning sub-models for each typical scenario; classify the sub-models into multiple levels based on their performance metrics on the validation set, and implement differentiated training frequency and parameter optimization strategies for sub-models of different levels. S3: Input real-time observation data, use the trained scene classification model to identify the current scene type, and enable fuzzy matching mechanism if an exact match cannot be made. S4: Based on the identified scene, call the corresponding sub-model to perform forecast calculations and use real-time observation data to dynamically correct the forecast results; when the scene features are detected to deviate significantly from the historical records, start the fast learning mechanism to update the model; S5: Output the forecast results and confidence levels of key variables, and use performance indicators to evaluate and monitor the forecast results over the long term for continuous model improvement.

2. The method for accurate marine environment forecasting based on multi-scenario dynamic matching according to claim 1, characterized in that, In step S1, the unsupervised learning method includes K-means clustering, DBSCAN, or principal component analysis; the periodic dynamic update is performed once per quarter.

3. The method for accurate marine environment forecasting based on multi-scenario dynamic matching according to claim 1, characterized in that, In step S2, the performance index is root mean square error; the hierarchical management specifically involves dividing the sub-models into three levels: "excellent", "good", and "needs optimization", and automatically triggering the parameter readjustment process for the "needs optimization" sub-models, while reducing the training frequency of the "excellent" sub-models.

4. The method for accurate marine environment forecasting based on multi-scenario dynamic matching according to claim 1, characterized in that, In step S2, the machine learning sub-model includes support vector machine, random forest or long short-term memory network.

5. The method for accurate marine environment forecasting based on multi-scenario dynamic matching according to claim 1, characterized in that, In step S3, the fuzzy matching mechanism involves calling the features of the nearest scene that are most similar to the current real-time data features for matching.

6. The method for accurate marine environment forecasting based on multi-scenario dynamic matching according to claim 1, characterized in that, In step S4, the dynamic correction method is deviation correction or data fusion.

7. A precise marine environment forecasting system based on multi-scenario dynamic matching, characterized in that, include: Scene classification and feature extraction module: used to collect multi-source historical data of the target area, use unsupervised learning methods to classify the historical data into scenes, and divide them into a variety of typical marine environmental scenes; extract key meteorological factors for each scene, build a scene feature database, and periodically and dynamically update the scene features based on new data; Model training and hierarchical management module: used to select and train independent machine learning sub-models for each typical scenario; divide the sub-models into multiple levels according to their performance metrics on the validation set, and implement differentiated training frequency and parameter optimization strategies for sub-models of different levels; Real-time scene recognition module: Used to input real-time observation data, use the trained scene classification model to identify the current scene type, and enable fuzzy matching mechanism if a precise match cannot be made. Scene Adaptive Forecasting and Optimization Module: This module is used to call the corresponding sub-model to perform forecast calculations based on the identified scene, and to dynamically correct the forecast results using real-time observation data; when a scene feature is detected to deviate significantly from historical records, a fast learning mechanism is activated to update the model; Forecast Results Output and Optimization Module: This module outputs the forecast results and confidence levels of key variables, and uses performance indicators to evaluate and monitor the forecast results over the long term, enabling continuous improvement of the model.

8. The marine environment precision forecasting system based on multi-scenario dynamic matching according to claim 7, characterized in that, The forecasting system supports real-time data access, dynamic allocation of computing tasks, and provides a user interface to support users in customizing forecast scenarios or variables.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the marine environment precision forecasting method based on multi-scenario dynamic matching as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the marine environment precision forecasting method based on multi-scenario dynamic matching as described in any one of claims 1-6.

Images