Marine heat wave prediction method

Abstract

The invention relates to an ocean heat wave prediction method, and the method comprises the steps: obtaining the physical variable historical time series data of each region of a target sea area, generating a physical variable prediction sequence containing the physical variable prediction value of a future time point through a composite model obtained through training, and carrying out the mining to obtain the physical variable relation data. Then, the physical variable prediction sequence, the physical variable relation data and the preset task instruction are input into a heat wave analysis module, the input data are accurately analyzed through the heat wave analysis module, and conclusion deviation caused by non-uniform standards during manual interpretation is avoided. Through automatic analysis, direct manual participation is reduced, and the analysis efficiency and the standardization degree are improved. And finally, obtaining ocean heat wave information output by the heat wave analysis module and obtaining a prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine heat waves, in particular to a marine heat wave prediction method. BACKGROUND

[0002] Under the trend of global warming, marine heat waves, as an extreme marine meteorological phenomenon, occur more frequently and intensify, causing multiple impacts on marine ecosystems and human society. Marine heat waves refer to the situation that the sea surface temperature in a specific sea area is significantly higher than the historical level in a short period of time, with a duration of several days to several months and a spatial coverage range from nearshore areas to ocean basins. This abnormal warming phenomenon can cause a series of serious consequences, such as causing large-scale bleaching of coral reefs, disrupting the balance of marine biological communities, and further causing ecological and economic losses such as reduced fishery production; it can also induce extreme weather events such as typhoons and heavy rains by affecting air-sea interaction, seriously threatening the safety of production and life in coastal areas.

[0003] Currently, there are certain limitations in the prediction technology of marine heat waves. The existing prediction methods have insufficient accuracy and reliability of prediction results. Some solutions highly rely on human experience for interpretation, which not only consumes time and effort, but also easily leads to biased conclusions due to the lack of unified interpretation standards, making it difficult to support the standardized application of marine heat wave prediction. The existing marine heat wave prediction technology cannot meet the demand for accurate prediction of marine heat waves to take early measures and reduce their negative impacts on marine ecosystems and human society. Therefore, it is of great practical significance to develop a marine heat wave prediction method that can improve prediction accuracy and reliability and reduce human dependence. SUMMARY

[0004] Based on this, the purpose of the present application is to provide a marine heat wave prediction method to achieve accurate prediction of marine heat waves so as to take early measures and reduce their negative impacts on marine ecosystems and human society.

[0005] The marine heat wave prediction method described in the embodiments of the present application includes the following steps: Obtain physical variable historical time series data of each region in the target sea area; the physical variable historical time series data includes physical variable observation values at a plurality of historical time points; Input the physical variable historical time series data of each region into a preset composite model to obtain physical variable prediction sequences and physical variable relationship data of each region; wherein the composite model is used to generate physical variable prediction sequences and physical variable relationship data according to the physical variable historical time series data; the physical variable prediction sequences include physical variable prediction values at a plurality of future time points; inputting the physical variable prediction sequence of each region, the physical variable relationship data, and a preset task instruction into a preset heat wave analysis module; wherein the preset task instruction is used to prompt the heat wave analysis module to analyze the ocean heat wave information according to the physical variable prediction sequence of each region and the physical variable relationship data; obtaining the ocean heat wave information output by the heat wave analysis module; and obtaining the ocean heat wave prediction result of the target sea area according to the ocean heat wave information.

[0006] The embodiment of the present application obtains the physical variable historical time series data of each region of the target sea area, and inputs the data into a preset composite model. The model deeply mines the internal relationship between physical variables, generates a physical variable prediction sequence containing the predicted value of the physical variable at a future time point, and outputs physical variable relationship data. The rich information of the historical data is fully utilized. Compared with the traditional method which relies on limited artificial experience, the overall and accuracy of data processing are greatly improved, so that the prediction result is more reliable and accurate. Subsequently, the physical variable prediction sequence, the physical variable relationship data, and the preset task instruction are input into the heat wave analysis module. The preset task instruction provides a clear analysis direction for the heat wave analysis module, so that the heat wave analysis module accurately analyzes the input data based on the analysis direction, and avoids the conclusion deviation caused by the non-uniform standard when manually interpreting. This automatic analysis method reduces the direct participation of artificial, improves the analysis efficiency and standardization. Finally, the ocean heat wave information output by the heat wave analysis module is obtained, and the prediction result is obtained. The embodiment of the present application realizes the accurate prediction of the ocean heat wave, can provide a strong basis for the relevant departments to take measures in advance, thereby effectively reducing the negative impact of the ocean heat wave on the marine ecological system and human society, and has a significant advantage in ensuring the balance of the marine ecology and the safety of production and life in the coastal area.

[0007] In order to better understand and implement, the present application will be described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 a flowchart of the ocean heat wave prediction method of the embodiment of the present application; Figure 2 a flowchart of the step of training the composite model of the embodiment of the present application; Figure 3 a flowchart of the step of converting the historical observation sequence into a high-dimensional feature tensor of the embodiment of the present application. DETAILED DESCRIPTION

[0009] In order to make the objects, technical solutions and advantages of the present application clearer, the following will further describe the embodiments of the present application with reference to the accompanying drawings. Wherein, the following description relates to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated.

[0010] It should be clear that the implementations described in the following described embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0011] The terms used in the present application are merely for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a," "said," and "the" used in the present application are also intended to include the plural forms unless the context clearly indicates otherwise. In addition, in the description of the present application, "a plurality of" means two or more unless otherwise stated. It should also be understood that the term "and / or" used herein means and includes any or all possible combinations of one or more associated listed items, for example, A and / or B can represent the three cases of A existing alone, A and B existing together, and B existing alone; the character " / " generally represents an "or" relationship between the associated objects before and after.

[0012] It should be understood that although the terms first, second, third, etc. can be used in the present application to describe various information, these information should not be limited to these terms, and these terms are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can be understood as indicating or implying relative importance. For a person of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances. Depending on the context, the word "if" used in the present application can be interpreted as "when" or "when" or "in response to determining".

[0013] Under the trend of global warming, marine heatwaves, as an extreme marine meteorological phenomenon, occur more frequently and intensively, causing multiple impacts on marine ecosystems and human society. Marine heatwaves refer to the situation that the sea surface temperature in a specific sea area is significantly higher than the historical level in a short period of time, with a duration of several days to several months, and a spatial coverage range from nearshore areas to ocean basins. This abnormal warming phenomenon will trigger a series of serious consequences, such as causing large-scale bleaching of coral reefs, disrupting the balance of marine biological communities, and further causing ecological and economic losses such as fishery yield reduction; it will also affect the interaction between sea and air, induce extreme weather events such as typhoon and heavy rain, and seriously threaten the safety of production and life in coastal areas.

[0014] Currently, there are certain limitations in the prediction technology of marine heat waves. The accuracy and reliability of the prediction results of existing prediction methods are insufficient. Some solutions highly rely on manual experience for interpretation, which not only consumes time and effort, but also easily leads to biased conclusions due to the lack of unified interpretation standards, making it difficult to support the standardized application of marine heat wave prediction. The existing marine heat wave prediction technology cannot meet the demand for accurate prediction of marine heat waves to take early measures and reduce their negative impact on marine ecosystems and human society. Therefore, it is of great practical significance to develop a marine heat wave prediction method that can improve prediction accuracy and reliability and reduce reliance on manual work.

[0015] The embodiment of the present application provides a marine heat wave prediction method, which realizes accurate prediction of marine heat waves, so as to take early measures and reduce their negative impact on marine ecosystems and human society.

[0016] Please refer to Figure 1 The marine heat wave prediction method described in the embodiment of the present application includes the following steps: S101: Obtain physical variable historical time series data of each region in a target sea area; the physical variable historical time series data includes physical variable observation values at a plurality of historical time points; S102: Input the physical variable historical time series data of each region into a preset composite model to obtain physical variable prediction sequences of each region and physical variable relationship data; wherein the composite model is used to generate physical variable prediction sequences and physical variable relationship data according to the physical variable historical time series data; the physical variable prediction sequence includes physical variable prediction values at a plurality of future time points; S103: Input the physical variable prediction sequences of each region, the physical variable relationship data and a preset task instruction into a preset heat wave analysis module; wherein the preset task instruction is used to prompt the heat wave analysis module to analyze marine heat wave information according to the physical variable prediction sequences of each region and the physical variable relationship data; S104: Obtain marine heat wave information output by the heat wave analysis module; obtain marine heat wave prediction results of the target sea area according to the marine heat wave information.

[0017] The embodiment of the present application generates a physical variable prediction sequence containing the predicted values of physical variables at future time points and outputs physical variable relationship data by obtaining the physical variable historical time series data of each region in the target sea area and inputting it into the preset composite model, and deeply mining the internal relationship between physical variables through the model. The rich information of historical data is fully utilized, compared with the traditional method which relies on limited artificial experience, the comprehensiveness and accuracy of data processing are greatly improved, and the prediction result is more reliable and accurate. Subsequently, the physical variable prediction sequence, the physical variable relationship data and the preset task instruction are input into the heat wave analysis module. The preset task instruction provides a clear analysis direction for the heat wave analysis module, so that the heat wave analysis module can accurately analyze the input data based on this, and avoid the conclusion deviation caused by the non-uniform standard when manually interpreting. This automatic analysis method reduces the direct participation of artificial, improves the analysis efficiency and standardization degree. Finally, the marine heat wave information output by the heat wave analysis module is obtained and the prediction result is obtained. The embodiment of the present application realizes accurate prediction of marine heat wave, can provide a strong basis for relevant departments to take measures in advance, thereby effectively reducing the negative impact of marine heat wave on marine ecological system and human society, and has significant advantages in ensuring the safety of marine ecological balance and production and life in coastal areas.

[0018] The marine heat wave prediction method described in the embodiment of the present application takes a computer as the execution subject, and each step is described in detail below.

[0019] For step S101, the physical variable historical time series data of each region in the target sea area is obtained; the physical variable historical time series data includes physical variable observation values at a plurality of historical time points.

[0020] The target sea area refers to a specific marine area to be studied or monitored for marine heat wave, and the range and boundary of the area can be determined according to actual needs and research purposes, such as a sea area near a certain coastal city, or a sea area with specific ecological characteristics, etc. A plurality of sub-regions are further divided in the target sea area, and the division of these sub-regions can be based on geographical location, marine environmental characteristics, etc. For example, it is divided into nearshore area, middle shore area and offshore area according to the distance from the coastline; or it is divided into shallow sea area and deep sea area according to the depth of seawater, etc.

[0021] The physical variable historical time series data is a collection of physical variable observation values of each region in the target sea area at different historical time points.

[0022] The physical variable refers to various parameters that can reflect the state of marine environment, such as seawater temperature, which reflects the distribution of marine heat and has an important influence on the survival of marine organisms and ocean circulation, etc.

[0023] In this embodiment, appropriate physical variables can be selected according to the situation, including but not limited to sea surface temperature, 2-meter air temperature, total precipitation, significant wave height, sea level pressure, wind speed and direction, relative humidity, atmospheric precipitable water, evaporation, sea surface salinity, sea current (flow rate and direction), ocean heat content, tidal level / height, and mixed layer depth. Sea surface temperature refers to the temperature of seawater within a certain depth range on the ocean surface, which is a direct manifestation of ocean heat and has a crucial impact on the survival, distribution of marine life, and ocean circulation. For example, different sea surface temperature regions will attract different types of marine life, and changes in sea surface temperature will trigger adjustments in ocean circulation. 2-meter air temperature is the air temperature at a height of 2 meters above the sea surface, which reflects the thermal conditions of the atmospheric environment near the ocean surface. 2-meter air temperature and sea surface temperature interact with each other and jointly affect the ocean-atmosphere interaction process, thereby influencing the formation and changes of weather and climate. For example, an increase in sea surface temperature may cause an increase in 2-meter air temperature, leading to local weather changes. Total precipitation refers to the total amount of precipitation falling on the ocean surface within a certain time and spatial range. Total precipitation affects the salinity balance and water cycle of the ocean, and excessive precipitation may dilute seawater, reduce salinity, and affect the survival environment of marine life. Significant wave height refers to the average value of one-third of the maximum wave height in a certain period of time, which reflects the energy size and intensity of sea waves. Significant wave height has an important impact on marine navigation, offshore operations, and coastal erosion, and larger significant wave height may make ship navigation more difficult and increase the risk of offshore operations. Sea level pressure refers to the atmospheric pressure at sea level, which is an important basic element in meteorology. The distribution and changes of atmospheric pressure are closely related to the formation, movement, and development of weather systems. For example, low-pressure areas are usually accompanied by cloud aggregation, precipitation, and other weather phenomena, while high-pressure areas are often sunny. Wind speed represents the distance that air moves in a unit of time, usually measured in meters per second (m / s) or kilometers per hour (km / h). The size of wind speed affects the intensity of meteorological phenomena, such as strong winds that may cause storms, hurricanes, and other severe weather. Wind direction refers to the direction in which the wind blows, usually represented by 16 compass points or angles. Wind direction has an important influence on the movement of weather systems and the distribution of meteorological elements, such as monsoons that bring specific seasonal climate characteristics. Relative humidity refers to the percentage of water vapor pressure in the air compared to the saturated water vapor pressure at the same temperature, which reflects the degree of air saturation. Relative humidity has an impact on human comfort, plant transpiration, and some industrial production processes. When relative humidity is high, people feel hot and stuffy; when relative humidity is low, the air is dry and can cause skin and respiratory problems. Atmospheric precipitable water refers to the total amount of water vapor contained in a vertical column of air from the ground to the top of the atmosphere per unit area. If all the water vapor condenses into water and falls to the ground, the amount of precipitation obtained is the atmospheric precipitable water. It is an important indicator for assessing the potential of water resources in a region and has important significance for precipitation forecasting, water resource management, and other aspects.Evaporation refers to the total amount of water evaporated from the surface (such as water surface, soil, plant surface, etc.) into the atmosphere within a certain period of time. Evaporation is influenced by factors such as temperature, humidity, wind speed, solar radiation, etc. In the field of agriculture, understanding evaporation helps to arrange irrigation reasonably; in water resources management, evaporation is an important parameter for calculating water resources balance. Sea surface salinity refers to the salinity concentration of the surface layer of seawater, usually expressed in parts per thousand. Sea surface salinity is influenced by factors such as precipitation, evaporation, river runoff, glacier melting, etc. Changes in salinity affect the density of seawater, which in turn affects ocean circulation. For example, high-salinity seawater has high density and will sink, driving the vertical movement of the ocean. Ocean currents are large-scale water movements in the ocean with relatively stable flow speed and direction. The flow speed represents the distance moved by the current in unit time, usually in centimeters per second (cm / s) or knots (1 knot = 1.852 km / h). The size of the flow speed of the ocean current affects the transportation of marine materials and the transmission of energy. The flow direction refers to the direction of the ocean current flow, which is expressed in terms of direction. The flow direction and flow speed of the ocean current have important influences on global climate, marine ecosystems, and navigation, etc. For example, warm currents make the coastal areas warm and humid, while cold currents make the coastal areas cold and dry. Ocean heat content refers to the total amount of heat stored in the ocean, which is related to the temperature, volume, and specific heat capacity of seawater. Changes in ocean heat content have an important regulating effect on global climate. The ocean can absorb and store a large amount of heat, and when the ocean heat content changes, it will affect atmospheric circulation and climate patterns, for example, the El Niño phenomenon is closely related to the abnormal changes in the ocean heat content of the Pacific Ocean. Tidal level refers to the vertical position of seawater on the ocean surface, usually with a certain reference surface (such as mean sea level) as the reference. Tidal height refers to the height of the tidal level relative to a certain reference surface. The changes in tidal level and tidal height are caused by the gravitational effects of the moon and the sun and other factors (such as weather conditions, topography, etc.). Tidal phenomena have important influences on port navigation, coastal engineering, fishing activities, etc. in coastal areas. For example, ships can smoothly enter and exit the port during high tide, and some shoals may be exposed during low tide. Mixed layer depth refers to the depth of a layer of seawater in the ocean where temperature, salinity, etc. are relatively uniform due to turbulent mixing. The mixed layer depth is influenced by factors such as wind stress, solar radiation, ocean heat flux, etc. The size of the mixed layer depth affects the exchange of heat, momentum, and matter between the ocean and the atmosphere, and plays an important role in the upper ocean ecosystem and climate. For example, a deeper mixed layer can promote the exchange of water between the upper and lower layers of the ocean, affecting the distribution of nutrients and the growth of plankton.

[0024] Among them, the historical time series data emphasizes that these observation values are arranged in chronological order, covering multiple historical time points, and through long-term data accumulation, it can reflect the change law of the marine environment over time.

[0025] For step S102, the physical variable historical time series data of each region is input into a preset composite model to obtain physical variable prediction sequences and physical variable relationship data of each region; the composite model is used to generate physical variable prediction sequences and physical variable relationship data according to the physical variable historical time series data; the physical variable prediction sequences include physical variable prediction values at a plurality of future time points.

[0026] The preset composite model is a mathematical model that is pre-trained and set, which establishes a mapping relationship between the physical variable historical time series data and the future state through learning and analysis of a large amount of historical data. When the physical variable historical time series data of each region is input into the composite model, the model will process and predict the data using its internal algorithm to generate a sequence of physical variable prediction values at a plurality of future time points. For example, the values of physical variables such as sea surface temperature, 2-meter air temperature, total precipitation, and significant wave height in the next week are predicted, which can help us understand the trend of changes in the marine environment in advance.

[0027] The composite model generates prediction sequences while also mining relationship data between different physical variables. These relationship data reveal the internal relationship between each physical variable, such as how changes in sea water temperature may affect changes in 2-meter air temperature, total precipitation, and significant wave height, or what the correlation between sea water temperature and total precipitation is, etc. By understanding these relationships, we can more deeply understand the dynamic change mechanism of the marine environment.

[0028] In one embodiment, the composite model includes an input adapter, a pre-trained large model backbone network, a prediction module, and a relationship mining module; the pre-trained large model backbone network includes a trainable low-rank matrix; the input adapter is connected to the pre-trained large model backbone network, and the pre-trained large model backbone network is connected to the prediction module and the relationship mining module; The input adapter is used to decompose and embed the historical observation sequence data to obtain high-dimensional features, and transmit the high-dimensional features to the pre-trained large model backbone network; The pre-trained large model backbone network is used to receive the high-dimensional features and process them to obtain high-dimensional hidden states, and transmit the high-dimensional hidden states to the prediction module and the relationship mining module; The prediction module is used to convert the high-dimensional hidden states into physical variable prediction sequences through a linear mapping layer; The relationship mining module is used to convert the high-dimensional hidden states into physical variable relationship data through an attention mechanism.

[0029] The embodiment realizes the comprehensive improvement of marine heat wave prediction in accuracy, efficiency and interpretability through composite model architecture innovation and training process optimization. The high-dimensional feature processing capability of the input adapter combined with the low-rank adaptation technology of the pre-trained large model backbone network significantly reduces the computational load and energy consumption of the marine monitoring equipment. The cooperative working mechanism of the prediction module and the relationship mining module enables the model not only to output high-precision physical variable prediction sequences, but also to reveal the hidden associations between variables.

[0030] In one embodiment, the relationship mining module is configured to extract a feature vector corresponding to a prediction starting time point from the high-dimensional hidden state as a query vector; extract feature vectors corresponding to all historical time points in the high-dimensional hidden state as key vectors; perform transpose processing on the last two dimensions of the key vectors to obtain transposed key vectors; calculate the dot product similarity between the query vector and the transposed key vectors to obtain an association score; perform normalization processing on the association score in the historical time dimension to obtain an influence score corresponding to each historical time point; and obtain physical variable relationship data according to the influence score.

[0031] In the attention mechanism, the query vector represents the feature of a specific time point that needs to be focused on, and the key vector represents the feature library of the historical time sequence. The similarity is quantified through dot product operation to accurately extract and predict the information strongly related to the target from the historical data.

[0032] Dot product operation directly reflects the direction matching degree between vectors, and normalization processing (such as Softmax) converts the original similarity into an influence score in the range of 0-1, which clearly defines the contribution weight of each historical time to the prediction starting point and avoids analysis deviation caused by numerical difference.

[0033] The embodiment enables the model to accurately locate the key time window before the heat wave occurs and the abnormal combination of environmental variables through the query-key vector mechanism combined with dot product similarity calculation. For example, in a certain heat wave event, the module successfully identifies the strong association between the continuous high sea surface temperature 30 days before the heat wave starting point and the abnormal fluctuation of significant wave height, providing a scientific basis for issuing a heat wave warning 20 days in advance. The normalized influence score generates physical variable relationship data, which not only reveals the hidden coupling law between variables, but also quantifies the contribution weight of each factor to the heat wave. This interpretable output enables the meteorological department to develop targeted protection strategies, such as adjusting the irrigation plan in coastal areas according to the sea temperature-rainfall association strength.

[0034] Please refer to Figure 2 In one embodiment, before the step of inputting the physical variable historical time series data of each region into the preset composite model to obtain the physical variable prediction sequence of each region and the physical variable relationship data in step S102, the method further comprises: In step S1021, historical observation data of the target sea area is obtained; the historical observation data includes time series data of physical variables of each region of the target sea area; the time series data of physical variables includes observation values of physical variables sorted in time sequence.

[0035] The historical observation data is a record of various environmental parameters in the target sea area over a period of time. The time series data of physical variables is an important part of the historical observation data, which includes observation values of sea surface temperature, 2-meter air temperature, total precipitation, and effective wave height sorted in time sequence. By collecting these long time series of data, the variation of physical variables in the target sea area over time can be understood, providing a basis for subsequent model training and prediction. For example, by analyzing the time series data of sea surface temperature for many years, the variation trend in different seasons and years can be found.

[0036] In step S1022, a plurality of historical observation sequences and target sequences corresponding to each historical observation sequence are obtained according to the historical observation data; the historical observation sequence includes a plurality of continuous physical variable observation values at first historical time points in each region; the target sequence includes a plurality of continuous physical variable observation values at second historical time points in each region, and the second historical time points are subsequent continuous time points of the first historical time points.

[0037] The historical observation sequence is a sequence of physical variable observation values in a continuous time period selected from the historical observation data. The physical variable observation values at the first historical time points reflect the environmental state of the target sea area in a certain time period. For example, a historical observation sequence is composed of observation values of sea surface temperature, 2-meter air temperature, etc. for each day in a continuous week.

[0038] The target sequence is a sequence of physical variable observation values at subsequent continuous time points corresponding to the historical observation sequence. It represents the environmental change of the target sea area in a period of time after the historical observation sequence. For example, if the historical observation sequence is one week of data, the target sequence may be the next week of data. By this splitting method, the relationship between past environmental state and future environmental change can be established, providing data support for the model to learn the temporal dependence.

[0039] In step S1023, a training data set is generated according to each historical observation sequence and the target sequence corresponding to each historical observation sequence; the training data set includes a plurality of training samples, each training sample corresponds to a group of historical observation sequences, and the training label of each training sample is the target sequence corresponding to the group of historical observation sequences.

[0040] The training data set is a data set used to train the composite model. Each training sample is composed of a set of historical observation sequences, which contain physical variable information of the target sea area in the past period of time. The training label is the target sequence corresponding to the set of historical observation sequences, i.e., the physical variable observation value in the future period of time. In this way, the model can learn the ability to predict future environmental changes from historical environmental states. For example, a training sample may contain historical observation sequences of sea surface temperature, 2-meter air temperature, etc. in the past week, and its training label is the corresponding physical variable observation value in the next week.

[0041] In step S1024, the training data set is input into the pre-trained composite model for training to obtain a trained composite model.

[0042] The pre-trained composite model is a model framework that already has an initial structure and parameters, and has a certain ability to process and analyze data, but still needs to be optimized and adjusted through specific training data.

[0043] In this embodiment, historical observation data of the target sea area is first obtained, which contains time series data of key physical variables such as sea surface temperature, 2-meter air temperature, total precipitation, and significant wave height. These rich and detailed data provide a comprehensive information base for model training, enabling the model to learn the complex variation rules of physical variables in the target sea area over time. Then, the historical observation data is split to obtain historical observation sequences and corresponding target sequences, and a training data set is generated. This processing method cleverly establishes the relationship between past and future environmental states, enabling the model to learn the skill of predicting future changes from historical data. By inputting the training data set into the pre-trained composite model for training, the model continuously optimizes its parameters, improving the accuracy of future physical variable prediction.

[0044] In one embodiment, the step of obtaining historical observation data of the target sea area in step S1021 includes: In step S201, the off-shore distance and the seafloor topography gradient of each region of the target sea area are obtained.

[0045] The off-shore distance refers to the horizontal distance of a region in the target sea area from the coastline. Different off-shore distances may result in differences in marine environmental characteristics. For example, the nearshore area may be more affected by land, while the offshore area may have more prominent marine characteristics.

[0046] The seafloor topography gradient represents the degree of change of the seafloor topography in the horizontal direction. A large seafloor topography gradient means that the seafloor topography is highly undulating, such as the presence of trenches and seamounts. A small gradient indicates that the seafloor topography is relatively flat.

[0047] The off-shore distance and the gradient of the seabed topography of each region in the target sea area are obtained through various ways. For example, the off-shore distance can be measured by satellite remote sensing technology, and the gradient of the seabed topography can be obtained by measuring the water depth at different positions of the seabed using a multi-beam sounding system or other marine exploration equipment. These data are the basis for determining the region type and the geographical resolution.

[0048] In step S202, the region type and the corresponding geographical resolution of each region are determined according to the off-shore distance and the gradient of the seabed topography of each region. The region type includes a first type and a second type. The off-shore distance of a region of the first type is less than a preset distance or the fluctuation degree of the gradient of the seabed topography is greater than a preset fluctuation degree. The first type region corresponds to a first geographical resolution, and the second type region corresponds to a second geographical resolution. The first geographical resolution is greater than the second geographical resolution.

[0049] The geographical resolution is an index in geographical data that describes the size of the geographical region represented by the data. The higher the geographical resolution, the smaller the data region represented, and the more detailed geographical information can be reflected. The lower the geographical resolution, the larger the data region represented, and the more macro geographical information can be reflected.

[0050] The off-shore distance and the gradient of the seabed topography are used as the division criteria. When the off-shore distance of a region is less than a preset distance, it indicates that the region is close to the coastline and may be greatly affected by land factors (such as river runoff, human activities, etc.). When the fluctuation degree of the gradient of the seabed topography is greater than a preset fluctuation degree, it indicates that the seabed topography of the region is complex and may have special ocean dynamic processes. A region that meets one of these two conditions is classified as a region of the first type, and the remaining regions are classified as regions of the second type.

[0051] The first type region corresponds to a first geographical resolution, and the second type region corresponds to a second geographical resolution, and the first geographical resolution is greater than the second geographical resolution. This is because the environment of the first type region changes relatively more complexly and violently, and a higher geographical resolution is needed to accurately describe its physical variable characteristics. The environment of the second type region is relatively stable and simple, and a lower geographical resolution can meet the research needs. For example, a smaller grid size can be used to represent the geographical region of the first type region to capture more detailed changes, and a larger grid size can be used for the second type region.

[0052] In step S203, the observation sequence data of the target sea area at the original resolution is obtained. The observation sequence data includes the physical variable values of each region of the target sea area at several historical time points.

[0053] The original resolution observation sequence data is the observation value of the physical variable (sea surface temperature, 2-meter air temperature, total precipitation, effective wave height) of the target sea area at different historical time points in each region within a period of time, which is obtained by various marine observation means (such as buoy observation, satellite remote sensing observation, etc.). These data have the original geographical resolution, which is the basis for subsequent data sampling and processing. For example, the buoy can continuously measure the sea surface temperature and other physical variables at a specific location, and the satellite remote sensing can quickly obtain the relevant information of the ocean surface in a large area.

[0054] In step S204, the first type region physical variable time series data is sampled according to the first geographical resolution, and the second type region physical variable time series data is sampled according to the second geographical resolution, according to the original resolution observation sequence data; wherein the first geographical resolution is greater than the second geographical resolution.

[0055] According to the region type and the corresponding geographical resolution determined in the foregoing, the original resolution observation sequence data is sampled. For the first type region, since the environment is complex, more detailed data description is required, so the first geographical resolution is sampled to select the physical variable time series data meeting the resolution requirement from the original data; for the second type region, the second geographical resolution is sampled. Through this sampling method, the data of different regions can be reasonably processed under the premise of ensuring data quality, so that the data is more in line with the environmental characteristics and research needs of each region. For example, in the first type region, data points are selected at a smaller spatial interval; in the second type region, data points are selected at a larger spatial interval.

[0056] In step S205, the first type region physical variable time series data and the second type region physical variable time series data obtained by sampling are used to obtain the historical observation data of the target sea area.

[0057] The first type region physical variable time series data and the second type region physical variable time series data obtained by sampling are integrated to form the complete historical observation data of the target sea area. These data contain the physical variable information of each region of the target sea area at different historical time points, and the geographical resolution of the data matches the environmental characteristics of each region, providing accurate and appropriate data basis for subsequent model training and marine heat wave prediction.

[0058] The embodiment first obtains the off-shore distance and the seafloor topography gradient of each region in the target sea area. The two key parameters can accurately reflect the marine environmental characteristics of different regions. According to the off-shore distance and the seafloor topography gradient, the region type and the corresponding geographical resolution are determined. This differentiated processing method fully considers the environmental differences in different regions. The first type of region adopts a higher geographical resolution due to its proximity to the coastline or complex seafloor topography, which can more accurately capture the subtle changes in its physical variables. The second type of region has a relatively stable environment and adopts a lower geographical resolution, which reduces the data processing amount while ensuring data effectiveness. This reasonable geographical resolution setting enables the data to accurately reflect the actual situation of each region and improve data processing efficiency. After obtaining the observation sequence data at the original resolution, sampling is performed according to the corresponding geographical resolution for different region types, further optimizing the data quality. The sampled data is more consistent with the environmental characteristics of each region, avoiding errors caused by mismatched data resolution. The final integrated historical observation data of the target sea area contains rich and accurate physical variable time series information, providing a solid foundation for subsequent composite model training.

[0059] In one embodiment, the step S204 of sampling the physical variable time series data of the first type of region according to the first geographical resolution and sampling the physical variable time series data of the second type of region according to the second geographical resolution further comprises: Step S2041, obtaining the weather information of the target sea area at several historical time points.

[0060] Weather information covers various factors that affect the environment of the sea area, such as wind speed, wind direction, air pressure, and types of precipitation (such as strong convective precipitation). These weather factors will cause different degrees of disturbance to the sea area, and then affect the physical variables such as sea surface temperature, 2-meter air temperature, total precipitation, and significant wave height.

[0061] This step can obtain the weather information of the target sea area at several historical time points through various means. Weather satellites can monitor the weather conditions above the sea in a large area and in real time, providing information such as wind speed, wind direction, and cloud distribution. In addition to measuring physical variables, some buoys are equipped with weather sensors to obtain local weather data. Furthermore, coastal weather observation stations can also provide relevant weather information. Through data integration and processing, comprehensive historical weather information of the target sea area can be obtained.

[0062] Step S2042, determining the time resolution corresponding to each historical time point according to the weather information of each historical time point; wherein the first type of weather information corresponds to the first time resolution, and the second type of weather information corresponds to the second time resolution; the disturbance degree of the first type of weather information to the sea area is greater than that of the second type of weather information; the first time resolution is greater than the second time resolution.

[0063] The first type of weather information refers to those weather factors combination or extreme weather conditions that disturb the sea area greatly, such as strong wind (large wind speed and long duration), low pressure system (may cause storm surge, etc.), strong convective precipitation (large amount of precipitation in a short time and may be accompanied by lightning and other strong weather phenomena), etc. These weather conditions will cause rapid and drastic changes in the physical variables of the sea area. The second type of weather information is the relatively stable weather condition that disturbs the sea area less, such as light wind, gentle pressure change, light to moderate rain, etc.

[0064] Due to the drastic changes in the physical variables of the sea area caused by the first type of weather information, in order to accurately capture these rapid changes, a higher time resolution is required, i.e. the first time resolution is larger, which means that data sampling is performed multiple times in a short period of time. For example, under strong wind weather, the sea surface temperature may change significantly in a few minutes, so data needs to be collected every few minutes. Under the second type of weather information, the physical variables of the sea area change relatively slowly, and a lower time resolution (second time resolution is smaller) can meet the needs, such as collecting data every hour.

[0065] In step S2043, the physical variable time series data of the first type of area is sampled according to the first geographical resolution and the time resolution, and the physical variable time series data of the second type of area is sampled according to the second geographical resolution and the time resolution.

[0066] For the first type of area, the physical variable time series data is sampled in combination with the first geographical resolution and the first time resolution. The first geographical resolution ensures that the characteristics of the area can be finely described in space, and the first time resolution ensures that the rapid physical variable changes caused by the first type of weather information can be captured in time. For example, in a first type of area close to the coast and with complex seabed topography, when encountering strong wind weather, the physical variable data such as sea surface temperature and significant wave height is collected every few minutes according to the small area divided by the first geographical resolution. For the second type of area, the second geographical resolution and the second time resolution are used for sampling. The second geographical resolution adapts to the relatively stable environmental characteristics of the area, and the second time resolution matches the slow physical variable changes of the area under the second type of weather information. For example, in a second type of area far from the coast and with flat seabed topography, under the weather conditions of light wind and stable pressure, the relevant physical variable data is collected every hour according to the large area divided by the second geographical resolution.

[0067] The embodiment can more accurately reflect the actual changes of physical variables in different regions under different weather conditions by combining weather information to determine the time resolution and sampling. In the case of large weather disturbance, high time resolution sampling can capture the rapid fluctuations of physical variables; when the weather is relatively stable, low time resolution sampling avoids data redundancy while ensuring the representativeness of the data. This makes the historical observation data obtained by sampling more truly reflect the actual situation of the target sea area.

[0068] In one embodiment, before the step of generating a training data set according to each of the historical observation sequences and the target sequence corresponding to each of the historical observation sequences in step S1023, the step includes: Step S301, converting each of the historical observation sequences into a historical observation sequence in the form of a high-dimensional feature tensor; Before the step of inputting the physical variable historical time series data of each region into a preset composite model to obtain the physical variable prediction sequence of each region and the physical variable relationship data in step S102, the step includes: Step S302, converting the physical variable historical time series data into physical variable historical time series data in the form of a high-dimensional feature tensor.

[0069] The high-dimensional feature tensor refers to a data structure that converts original sequence data into a multi-dimensional array form through feature engineering or deep learning encoder, and its dimensions include time, space, physical variable type and other multi-dimensional feature information, which can more comprehensively depict the spatio-temporal evolution law of the marine environment state. For example, the time series data of physical variables such as sea surface temperature and 2-meter air temperature are mapped into a three-dimensional tensor, where one dimension represents time step, one dimension represents spatial grid position, and one dimension represents different physical variable types.

[0070] The embodiment significantly enhances the feature extraction capability and pattern recognition accuracy of the marine heat wave prediction model by converting the historical observation sequence and the physical variable historical time series data into a high-dimensional feature tensor. The data representation in the form of a high-dimensional tensor can comprehensively capture the complex correlations of physical variables such as sea surface temperature, 2-meter air temperature, total precipitation, and significant wave height in the spatio-temporal dimension, enabling the composite model to learn more profound marine environment evolution laws during the training process. For example, by representing the information of time evolution, spatial distribution, and variable type through a three-dimensional tensor structure, the model can accurately identify the multivariate abnormal coupling signals before the occurrence of a marine heat wave. In the prediction phase, the high-dimensional tensor input enables the composite model to output more interpretable physical variable relationship data, providing more reliable decision basis for the heat wave analysis module.

[0071] Please refer to Figure 3In an embodiment, the step of converting each of the historical observation sequences into a historical observation sequence in the form of a high-dimensional feature tensor in step S301 comprises: In step S3011, a sliding average processing is performed on the historical observation sequence to obtain a trend component; a frequency domain conversion is performed on the historical observation sequence and a self-correlation analysis is performed to obtain a seasonal component; and the historical observation sequence is subtracted by the trend component and the seasonal component to obtain a residual component.

[0072] The sliding average processing is to calculate the moving average value of the sequence through a sliding window, which is used to smooth the data and extract the long-term trend component, and weaken the short-term fluctuation interference. For example, the sliding average with a 30-day window can reflect the monthly variation trend of the sea surface temperature.

[0073] The frequency domain conversion (such as Fourier transform) is to convert the time series data from time domain to frequency domain to identify the periodic characteristics; and the self-correlation analysis is to detect the time interval of the periodic pattern by calculating the correlation between the sequence and the lagged version of itself. The combination of the two can accurately extract the seasonal fluctuation rule of the marine environment variable.

[0074] In this step, a sliding window average is performed on the historical observation sequence, such as calculating the long-term trend of the sea surface temperature with a 60-day window, filtering out the daily scale fluctuations to highlight the change trend above the month. The main cycle frequency is identified by Fourier transform, and the seasonal cycle is determined by combining the autocorrelation function. For example, the total precipitation sequence may exhibit seasonal characteristics with a period of one year, while the significant wave height may have a tidal-related fluctuation with a period of half a month. After subtracting the trend and seasonal components from the original sequence, the residual component retains the high-frequency random fluctuations and abnormal event signals, which provides key input for subsequent anomaly detection.

[0075] In step S3012, the trend component is input into a first feature embedding unit, and numerical embedding, position embedding and time stamp embedding operations are sequentially performed to obtain a first high-dimensional embedded sequence; the seasonal component is input into a second feature embedding unit, and numerical embedding, position embedding and time stamp embedding operations are sequentially performed to obtain a second high-dimensional embedded sequence; and the residual component is input into a third feature embedding unit, and numerical embedding, position embedding and time stamp embedding operations are sequentially performed to obtain a third high-dimensional embedded sequence.

[0076] The feature embedding unit includes modular processing units of numerical embedding (mapping continuous physical variable values into high-dimensional vectors), position embedding (encoding spatial grid coordinate information) and time stamp embedding (annotating time features of time points), which realizes the fusion representation of multi-modal features.

[0077] In step S3013, the first high-dimensional embedded sequence, the second high-dimensional embedded sequence and the third high-dimensional embedded sequence are fused to obtain the historical observation sequence in the form of a high-dimensional feature tensor.

[0078] The three high-dimensional embedding sequences are fused by three-dimensional splicing or attention mechanism to form a high-dimensional feature tensor with channel dimension containing trend, season, and residual characteristics, spatial dimension corresponding to regional grid, and time dimension marking timestamp.

[0079] The embodiment significantly improves the spatio-temporal feature extraction capability and prediction reliability of the marine heat wave prediction model through the synergistic optimization of trend-season-residual decomposition and multi-dimensional feature embedding. The sliding average and frequency domain analysis accurately strip the multi-scale variation characteristics of physical variables such as sea surface temperature and 2-meter air temperature, enabling the composite model to learn the evolution laws of long-term trend, periodic fluctuation, and random anomaly respectively. The numerical-position-timestamp three-dimensional embedding mechanism realizes the deep fusion of spatial distribution, temporal evolution, and numerical features, for example, in the complex topography of the coastal area, high-resolution position embedding can accurately capture local sea temperature anomalies, and timestamp embedding can enhance the heat wave prediction capability during the seasonal transition period. The high-dimensional tensor generated by feature fusion breaks through the linear expression limit of traditional sequence models, enabling the composite model to simultaneously process spatio-temporal dependence and multivariate coupling through a convolution-recurrent hybrid architecture. In the heat wave analysis module, this high-dimensional feature representation can accurately locate the heat wave occurrence area, quantify the intensity level, and predict the duration. Similarly, step S302 converts the physical variable historical time series data into high-dimensional feature tensor form, for details, please refer to the embodiment.

[0080] In one embodiment, the step of inputting the training data set into the pre-trained composite model for training to obtain the trained composite model in step S1024 comprises: Step S10241, inputting the training data set into the pre-trained composite model to obtain the physical variable prediction sequence corresponding to each training sample generated by the prediction module.

[0081] The training sample containing the historical observation sequence and the corresponding target sequence is input into the composite model. The input adapter first performs trend-season-residual decomposition and three-dimensional embedding processing on the historical observation sequence to generate a high-dimensional feature tensor input into the backbone network. The backbone network generates a hidden state after processing the high-dimensional feature, and the prediction module converts it into a physical variable prediction sequence through a linear mapping layer. For example, the sea surface temperature and significant wave height for the next 7 days are predicted daily to form an output matching the dimension of the target sequence.

[0082] Step S10242, according to the loss value of the physical variable prediction sequence corresponding to each training sample and the corresponding target sequence, updating the parameters of the input adapter and the trainable low-rank matrix and updating the linear mapping layer parameters of the prediction module.

[0083] The MSE loss value of the predicted sequence and the target sequence, such as sea surface temperature prediction error, 2-meter temperature deviation, etc., is calculated to form a multivariate joint loss value. Based on the loss value back propagation, the feature embedding parameters of the input adapter, the trainable low-rank matrix of the backbone network, and the linear layer weights of the prediction module are updated in priority. For example, the learning rate is dynamically adjusted by the Adam optimizer to ensure efficient and stable parameter updating.

[0084] In step S10243, the training data input, loss value calculation and parameter update steps are repeatedly performed until the loss value is less than the preset loss threshold, and it is determined that the composite model training is completed.

[0085] The training data input, loss calculation, and parameter update steps are repeatedly performed. In each iteration, the model learns from batch data to gradually adjust the parameters and reduce the prediction error. When the loss values of consecutive iterations are all lower than the preset threshold (such as 0.05), it is determined that the model training is completed. At this time, the model has fully learned the spatiotemporal evolution law in the historical data and has high-precision prediction capability.

[0086] Through the fine design of the training process, the composite model realizes a breakthrough in both prediction accuracy and computational efficiency. The multi-dimensional feature processing of the input adapter combined with the low-rank parameter optimization of the backbone network effectively solves the overfitting problem caused by parameter redundancy in traditional models. In the iterative training mechanism, the loss value-driven parameter update strategy ensures the accurate capture of the coupling relationship between multiple physical variables.

[0087] For step S103, the physical variable prediction sequence of each region, the physical variable relationship data, and the preset task instruction are input into the preset heat wave analysis module. The preset task instruction is used to prompt the heat wave analysis module to analyze the marine heat wave information according to the physical variable prediction sequence of each region and the physical variable relationship data.

[0088] The preset heat wave analysis module is a program module or neural network model specifically designed for analyzing marine heat wave information. It has specific analysis algorithms and logic and can make comprehensive judgments and processing based on input data.

[0089] The preset task instruction is used to instruct the heat wave analysis module to analyze the marine heat wave information based on the input physical variable prediction sequence of each region and the physical variable relationship data. Specifically, the preset task instruction can instruct the heat wave analysis module to determine whether a marine heat wave will occur in a target sea area in a future time period, and if so, the intensity, duration, and impact range of the marine heat wave.

[0090] In one embodiment, the preset heat wave analysis module in step S103 is obtained by the following steps: Step S1031, obtaining the physical variable prediction sequence corresponding to each training sample output by the trained composite model and the physical variable relationship data.

[0091] The trained composite model generates physical variable prediction sequences and physical variable relationship data (such as the correlation weight matrix of sea surface temperature and significant wave height) by inputting historical observation sequences. These output data serve as input sources for the heat wave analysis module, ensuring that the analysis is based on high-precision, multi-dimensional environmental prediction results.

[0092] Step S1032, according to the physical variable prediction sequence corresponding to each training sample and the physical variable relationship data, the first marine heat wave information corresponding to each training sample is analyzed.

[0093] Based on the prediction sequence and relationship data output by the composite model, the first marine heat wave information is generated by a regularized algorithm (such as threshold judgment, trend analysis). For example, when the sea surface temperature prediction value of a certain area exceeds 28℃ for 3 consecutive days and is positively correlated with the significant wave height, it is determined that there is a heat wave risk in that area. This step serves as a benchmark for the heat wave analysis module and is used for subsequent training effect evaluation.

[0094] Step S1033, inputting the physical variable prediction sequence corresponding to each training sample and the physical variable relationship data and the preset task instruction into the pre-trained heat wave analysis module.

[0095] The prediction sequence, relationship data and preset task instruction output by the composite model are input into the pre-trained heat wave analysis module. The module performs deep analysis on the input data through its internal multi-layer perceptron or decision tree structure, and generates second marine heat wave information such as heat wave occurrence probability, impact range, etc.

[0096] Step S1034, obtaining the second marine heat wave information corresponding to each training sample output by the heat wave analysis module; updating the training parameters of the heat wave analysis module according to the second marine heat wave information corresponding to each training sample and the loss value of the corresponding first marine heat wave information; re-inputting the physical variable prediction sequence corresponding to each training sample and the physical variable relationship data and the preset task instruction into the pre-trained heat wave analysis module; until the loss value is less than a preset threshold, obtaining the trained heat wave analysis module.

[0097] By comparing the differences between the first marine heat wave information and the second marine heat wave information, the loss value (such as mean square error) is calculated. The parameters (such as weight matrix, decision threshold) of the heat wave analysis module are updated through the back propagation algorithm, and the input-compute-update process is repeated until the loss value is lower than the preset threshold, completing the module training. This process ensures that the heat wave analysis module is highly matched with the output of the composite model, improving the prediction consistency.

[0098] The embodiment realizes the whole-process optimization of marine heat wave prediction from data generation to analysis decision through the cooperative training mechanism of the composite model and the heat wave analysis module. The high-precision physical variable prediction sequence and the relationship data output by the composite model provide a multi-dimensional input basis for heat wave analysis, enabling the analysis module to capture key heat wave precursor signals such as abnormal rise of sea surface temperature and coordinated change of effective wave height. The training process of the heat wave analysis module ensures that its output is highly consistent with the prediction results of the composite model through parameter optimization driven by loss value, avoiding the subjective bias of manual analysis in traditional methods.

[0099] In one embodiment, the preset task instruction is also used to prompt the heat wave analysis module to output the marine heat wave information according to a preset structured analysis report template.

[0100] Among them, the preset task instruction adds a structured report generation function on the basis of traditional heat wave analysis. The instruction analyzes user needs through natural language processing technology, automatically fills the marine heat wave information (such as occurrence time, intensity, and impact range) output by the heat wave analysis module into the preset template, and generates a standardized analysis report containing graphics, tables, and trend charts. For example, the instruction can specify that the report should include chapters such as "Heat Wave Event Overview", "Physical Variable Correlation Analysis", and "Early Warning Suggestions", and require output in PDF or HTML format.

[0101] The embodiment realizes the automatic generation of structured reports, converts heat wave analysis results into directly usable decision-making basis, and reduces manual collation time. By forcing the report structure to be uniform through the preset template, key information omission or expression differences caused by manual writing are avoided.

[0102] For step S104, the marine heat wave information output by the heat wave analysis module is obtained; and the marine heat wave prediction result of the target sea area is obtained according to the marine heat wave information.

[0103] Among them, the marine heat wave information is the specific content about marine heat wave output by the heat wave analysis module after analysis, which can specifically include whether the marine heat wave will occur, the specific time of occurrence, the duration, the range of influence, and the intensity level of the heat wave, etc.

[0104] The marine heat wave prediction result is the overall judgment and summary of the future marine heat wave situation of the target sea area based on the obtained marine heat wave information. This prediction result can provide decision-making basis for relevant departments and personnel. For example, the fishery department can adjust the fishing plan according to the prediction result to avoid work during the marine heat wave and reduce economic losses; the disaster prevention and mitigation department of the coastal area can make preparations in advance, such as reinforcing coastal protection facilities and issuing warning information, to protect the safety of people's lives and property.

[0105] The above embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but cannot be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and the present application also intends to include these modifications and improvements.

Claims

1. A method for predicting marine heat waves, characterized in that, Includes the following steps: Acquire historical time-series data of physical variables for various areas of the target sea area; the historical time-series data of physical variables includes observed values ​​of physical variables at several historical time points; Historical time-series data of physical variables in each region are input into a preset composite model to obtain predicted sequences of physical variables and relational data of physical variables in each region; wherein, the composite model is used to generate predicted sequences of physical variables and relational data of physical variables based on the historical time-series data of physical variables; the predicted sequences of physical variables include predicted values ​​of physical variables at several future time points; The predicted sequences of physical variables for each region, the relationship data of the physical variables, and the preset task instructions are input into the preset heat wave analysis module; wherein, the preset task instructions are used to prompt the heat wave analysis module to analyze and obtain marine heat wave information based on the predicted sequences of physical variables for each region and the relationship data of the physical variables. Obtain the marine heat wave information output by the heat wave analysis module; based on the marine heat wave information, obtain the marine heat wave prediction result for the target sea area.

2. The marine heat wave prediction method according to claim 1, characterized in that, Before the step of inputting the historical time-series data of physical variables of each region into a preset composite model to obtain the predicted sequence of physical variables and the relationship data of physical variables of each region, the following steps are included: Acquire historical observation data of the target sea area; the historical observation data includes time-series data of physical variables in various regions of the target sea area; the time-series data of physical variables includes observed values ​​of physical variables sorted by time series. The historical observation data is divided to obtain several historical observation sequences and a target sequence corresponding to each historical observation sequence; wherein, the historical observation sequence includes several consecutive first historical time points of physical variable observation values ​​of each region; the target sequence includes several consecutive second historical time points of physical variable observation values ​​of each region, and the second historical time points are consecutive time points following the first historical time points; A training dataset is generated based on each of the historical observation sequences and the target sequence corresponding to each of the historical observation sequences; wherein, the training dataset includes several training samples, each training sample corresponds to a set of historical observation sequences, and the training label of each training sample is the target sequence corresponding to the set of historical observation sequences; The training dataset is input into the pre-trained composite model for training, resulting in a trained composite model.

3. The marine heat wave prediction method according to claim 2, characterized in that, The steps for obtaining historical observation data of the target sea area include: Obtain the offshore distance and seabed topographic gradient of various areas in the target sea area; Based on the offshore distance and seabed topographic gradient of each region, the region type and corresponding geographic resolution of each region are determined; wherein, the region type includes a first type and a second type; the offshore distance of the first type region is less than a preset distance or the fluctuation of the seabed topographic gradient is greater than a preset fluctuation; the first type region corresponds to a first geographic resolution, and the second type region corresponds to a second geographic resolution; the first geographic resolution is greater than the second geographic resolution; Obtain observation sequence data of the target sea area at its original resolution; the observation sequence data includes the physical variable values ​​of various regions of the target sea area at several historical time points; For the observation sequence data at the original resolution, time-series data of physical variables of a first type of region are sampled according to the first geographic resolution, and time-series data of physical variables of a second type of region are sampled according to the second geographic resolution; wherein, the first geographic resolution is greater than the second geographic resolution; Based on the time-series data of physical variables from the first type of region and the time-series data of physical variables from the second type of region obtained through sampling, historical observation data of the target sea area are obtained.

4. The marine heat wave prediction method according to claim 3, characterized in that, The step of sampling time-series physical variable data of a first type of region according to a first geographic resolution and sampling time-series physical variable data of a second type of region according to a second geographic resolution further includes: Obtain meteorological information of the target sea area at several historical time points; Based on meteorological information at various historical time points, the corresponding time resolution for each historical time point is determined; wherein, the first type of meteorological information corresponds to the first time resolution, and the second type of meteorological information corresponds to the second time resolution; the first type of meteorological information has a greater disturbance effect on the sea area than the second type of meteorological information; the first time resolution is greater than the second time resolution; Based on the first geographic resolution and the time resolution, sample the time series data of physical variables of the first type of region; based on the second geographic resolution and the time resolution, sample the time series data of physical variables of the second type of region.

5. The marine heat wave prediction method according to claim 2, characterized in that, Before the step of generating a training dataset based on each of the historical observation sequences and the target sequence corresponding to each of the historical observation sequences, the following steps are included: Each of the aforementioned historical observation sequences is converted into a historical observation sequence in the form of a high-dimensional feature tensor. Before the step of inputting the historical time-series data of physical variables of each region into a preset composite model to obtain the predicted sequence of physical variables and the relationship data of physical variables of each region, the following steps are included: The historical time series data of the physical variables are converted into historical time series data of the physical variables in the form of high-dimensional feature tensors.

6. The marine heat wave prediction method according to claim 5, characterized in that, The step of converting each of the historical observation sequences into a high-dimensional feature tensor form includes: The historical observation sequence is processed by moving average to obtain the trend component; the historical observation sequence is transformed by frequency domain and autocorrelation analysis is performed to obtain the seasonal component; the residual component is obtained by subtracting the trend component and the seasonal component from the historical observation sequence. The trend component is input into the first feature embedding unit, and numerical embedding, location embedding, and time stamp embedding operations are performed sequentially to obtain the first high-dimensional embedding sequence; the seasonal component is input into the second feature embedding unit, and numerical embedding, location embedding, and time stamp embedding operations are performed sequentially to obtain the second high-dimensional embedding sequence; the residual component is input into the third feature embedding unit, and numerical embedding, location embedding, and time stamp embedding operations are performed sequentially to obtain the third high-dimensional embedding sequence. The first high-dimensional embedding sequence, the second high-dimensional embedding sequence, and the third high-dimensional embedding sequence are fused to obtain the historical observation sequence in the form of a high-dimensional feature tensor.

7. The marine heat wave prediction method according to claim 2, characterized in that, The composite model includes an input adapter, a pre-trained large-scale model backbone network, a prediction module, and a relationship mining module; the pre-trained large-scale model backbone network includes a trainable low-rank matrix; the input adapter is connected to the pre-trained large-scale model backbone network, and the pre-trained large-scale model backbone network is connected to the prediction module and the relationship mining module; The input adapter is used to decompose and embed the historical observation sequence data to obtain high-dimensional features, and then transmit the high-dimensional features to the pre-trained large model backbone network. The pre-trained large model backbone network is used to receive the high-dimensional features and process them to obtain the high-dimensional hidden state, and transmit the high-dimensional hidden state to the prediction module and the relationship mining module; The prediction module is used to convert the high-dimensional hidden state into a physical variable prediction sequence through a linear mapping layer; The relationship mining module is used to convert the high-dimensional hidden state into physical variable relationship data through an attention mechanism.

8. The marine heat wave prediction method according to claim 7, characterized in that, The relationship mining module is used to extract the feature vector corresponding to the prediction start time point from the high-dimensional hidden state as the query vector; and to use the feature vectors corresponding to all historical time points in the high-dimensional hidden state as the key vectors. The last two dimensions of the key vector are transposed to obtain the transposed key vector; the dot product similarity between the query vector and the transposed key vector is calculated to obtain the relevance score. The correlation score is normalized along the historical time dimension to obtain the influence score corresponding to each historical time point; the physical variable relationship data is obtained based on the influence score.

9. The marine heat wave prediction method according to claim 7, characterized in that, The step of inputting the training dataset into the pre-trained composite model for training to obtain the trained composite model includes: Input the training dataset into the pre-trained composite model to obtain the physical variable prediction sequence corresponding to each training sample generated by the prediction module; Based on the loss values ​​of the predicted sequence and the corresponding target sequence corresponding to the physical variables of each training sample, update the parameters of the input adapter and the trainable low-rank matrix, as well as update the parameters of the linear mapping layer of the prediction module. Repeat the steps of inputting training data, calculating loss value, and updating parameters until the loss value is less than the preset loss threshold, at which point the training of the composite model is considered complete.

10. The marine heat wave prediction method according to claim 2, characterized in that, Before the step of inputting the predicted sequences of physical variables for each region, the relational data of the physical variables, and the preset task instructions into the preset heat wave analysis module, the method further includes: Obtain the predicted sequence of physical variables and the relationship data of physical variables corresponding to each training sample output by the composite model after training is completed; Based on the predicted sequence of physical variables and the relationship data of physical variables corresponding to each training sample, the information of the first ocean heat wave corresponding to each training sample is obtained by analysis. The physical variable prediction sequence and physical variable relationship data corresponding to each training sample, as well as the preset task instructions, are input into the pre-trained heat wave analysis module. Obtain the second marine heat wave information corresponding to each training sample output by the heat wave analysis module; update the training parameters of the heat wave analysis module based on the second marine heat wave information corresponding to each training sample and the loss value of the corresponding first marine heat wave information; re-input the physical variable prediction sequence and physical variable relationship data corresponding to each training sample and the preset task instruction into the pre-trained heat wave analysis module; until the loss value is less than the preset threshold, the trained heat wave analysis module is obtained.

Images