Mesoscale analysis method for marine meteorology

Abstract

The invention discloses a mesoscale analysis method for marine meteorology, and relates to the technical field of mesoscale analysis for meteorology. According to the mesoscale analysis method for marine meteorology, marine meteorology observation data of a plurality of time periods in a set marine operation platform area are continuously acquired, meteorological disturbance aggregation indexes of the corresponding time periods are analyzed, meteorological image data of each time period are acquired, and a pre-trained image disturbance identification model is combined, so that the mesoscale analysis of marine meteorology is realized. Analyzing a meteorological structure evolution index of a corresponding time period, analyzing a meteorological evolution risk perception index in combination with a meteorological disturbance aggregation index of the corresponding time period, and predicting a meteorological evolution risk perception index of a next time period; according to the method, corresponding response measures are taken for the set marine operation platform area through the meteorological evolution risk perception index of the next time period, so that the accuracy of risk perception is improved, continuous identification of meteorological risks in the operation area is further facilitated, and the operation platform has higher prospective pre-judgment ability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mesoscale weather analysis, in particular to a marine weather mesoscale analysis method. BACKGROUND

[0002] Marine weather mesoscale refers to weather with a spatial scale of about several kilometers to several hundred kilometers and a time scale of about tens of minutes to a day, which includes sea-land wind, local thunderstorm, frontal system, tropical depression and offshore mesoscale convection, etc. Compared with large-scale weather systems, mesoscale has the characteristics of strong burst, dramatic structural change and complex evolution process, and has a significant impact on local wind field, precipitation distribution and atmospheric stability in the marine area. Therefore, in the marine operation scenario, the identification, perception and response of marine weather mesoscale have become an important prerequisite for ensuring the safety of marine operations.

[0003] The prior art such as the patent application with the publication number CN119046879B discloses a marine weather data fusion analysis method and system, which comprises the following steps: step S1: establishing a marine weather data set; step S2: respectively preprocessing the marine weather image data and the marine weather time series data; step S3: constructing a marine weather data fusion analysis model; step S4: training the marine weather data fusion analysis model; step S5: acquiring marine weather image data and marine weather time series data of a target marine area; step S6: inputting the image data and time series data of the target marine area after preprocessing into the marine weather data fusion analysis model; and step S7: visualizing the analysis result. The present application fully utilizes the complementarity of image data and time series data, realizes effective combination and deep analysis of marine weather image data and time series data, and can comprehensively and accurately analyze marine weather.

[0004] Based on the above scheme, it is found that the limitations of the prior art at least include the following problems: the prior art lacks step-by-step analysis and correlation of disturbance factors and structural changes in the marine mesoscale weather change process, so it is difficult to realize the collaborative analysis of disturbance evolution trend and structural change characteristics, which leads to insufficient identification accuracy in the weather risk identification process. For example, in the mesoscale process of rapid change of radar echo and dramatic splitting of cloud structure, it is difficult to realize continuous tracking and dynamic response to risk precursors, which further leads to the difficulty in timely identifying and predicting possible weather risks, and then makes it difficult to trigger the response measures accurately. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a marine weather mesoscale analysis method, which solves the problem that the prior art is difficult to realize the collaborative identification and accurate response of mesoscale weather risks.

[0006] To achieve the above object, the present application is implemented by the following technical scheme: a marine weather mesoscale analysis method, comprising the following steps: continuously acquiring marine weather observation data of a plurality of time periods in a set marine operation platform area, and analyzing the corresponding time period meteorological disturbance aggregation index; acquiring meteorological image data of each time period in the set marine operation platform area, and combining a pre-trained image disturbance identification model to analyze the corresponding time period meteorological structure evolution index; based on the meteorological disturbance aggregation index, the meteorological structure evolution index of each time period in the set marine operation platform area, analyzing the corresponding time period meteorological evolution risk perception index, and predicting the next time period meteorological evolution risk perception index in the set marine operation platform area; based on the next time period meteorological evolution risk perception index, taking corresponding response measures for the set marine operation platform area.

[0007] Further, the marine weather observation data includes a micro-vorticity factor, a wind direction jump rate value, a wind field dispersion index, a sea temperature distribution dispersion factor, a temperature anomaly rate value, and a micro-layer inversion index. The specific steps of analyzing the meteorological disturbance aggregation index of each time period in the set marine operation platform area are as follows: based on the marine weather observation data of each time period in the set marine operation platform area, analyzing the corresponding time period meteorological disturbance evaluation set, including a wind variation disturbance index and a thermal difference disturbance index; acquiring the water vapor disturbance index of each time period in the set marine operation platform area, and combining the corresponding time period meteorological disturbance evaluation set to analyze the corresponding time period meteorological disturbance aggregation index.

[0008] Further, the specific steps of analyzing the meteorological disturbance evaluation set of each time period in the set marine operation platform area are as follows: based on the micro-vorticity factor, the wind direction jump rate value, and the wind field dispersion index of each time period in the set marine operation platform area, analyzing the corresponding time period wind variation disturbance index; based on the sea temperature distribution dispersion factor, the temperature anomaly rate value, and the micro-layer inversion index of each time period in the set marine operation platform area, analyzing the corresponding time period thermal difference disturbance index.

[0009] Further, the specific steps of acquiring the water vapor disturbance index of each time period in the set marine operation platform area are as follows: acquiring humidity response data of each time period in the set marine operation platform area, including a humidity difference aggregation factor, a dew point temperature jump rate value, and a humidity delay response factor, and performing standardization processing; based on the standardized humidity response data of each time period in the set marine operation platform area, analyzing the corresponding time period water vapor disturbance index.

[0010] Further, the meteorological image data includes cloud layer image data and radar echo image data of each time window, and the image disturbance identification model includes a feature extraction sub-network, a time sequence evolution sub-network, and a fusion output sub-network.

[0011] Further, the specific steps of analyzing the weather structure evolution index of each time period in the set offshore operation platform region are as follows: inputting the weather image data of each time period in the set offshore operation platform region into the pre-trained image disturbance recognition model for comprehensive analysis to obtain the structure evolution feature vector set of the corresponding time period, including the cloud layer structure evolution index and the echo shape evolution index; and analyzing the weather structure evolution index of the corresponding time period based on the structure evolution feature vector set of each time period in the set offshore operation platform region.

[0012] Further, the specific steps of analyzing the structure evolution feature vector set of each time period in the set offshore operation platform region are as follows: in the feature extraction subnetwork of the image disturbance recognition model, receiving the cloud image data and the radar echo image data of each time window of each time period in the set offshore operation platform region, and analyzing the cloud and echo feature vector set of the corresponding time window; in the time sequence evolution subnetwork of the image disturbance recognition model, analyzing the evolution feature vector set of the corresponding time period based on the cloud and echo feature vector set of each time window of each time period in the set offshore operation platform region; and in the fusion output subnetwork of the image disturbance recognition model, analyzing the structure evolution feature vector set of the corresponding time period based on the evolution feature vector set of each time period in the set offshore operation platform region.

[0013] Further, the specific formula for calculating the weather evolution risk perception index of a certain time period in the set offshore operation platform region is as follows: ; wherein, , , the weather evolution risk perception index, the weather disturbance aggregation index, and the weather structure evolution index of a certain time period in the set offshore operation platform region are sequentially, , , , the disturbance coefficient, the evolution coefficient, the interaction coefficient, and the proportion coefficient stored in the database are sequentially.

[0014] Further, the specific steps of predicting the weather evolution risk perception index of the next time period in the set offshore operation platform region are as follows: analyzing the weather evolution mean value, the weather evolution standard deviation value, and the weather evolution trend mean value in the set offshore operation platform region based on the weather evolution risk perception index of each time period in the set offshore operation platform region; and analyzing the predicted weather evolution risk perception index in the set offshore operation platform region based on the weather evolution mean value, the weather evolution standard deviation value, and the weather evolution trend mean value in the set offshore operation platform region, and marking it as the weather evolution risk perception index of the next time period.

[0015] Further, the specific steps of taking corresponding response measures on the set offshore platform area based on the next period meteorological evolution risk perception index are as follows: judging whether the next period meteorological evolution risk perception index in the set offshore platform area is higher than the preset meteorological evolution risk perception index threshold value; if higher than the preset meteorological evolution risk perception index threshold value, taking corresponding response measures on the set offshore platform area; if not higher than the preset meteorological evolution risk perception index threshold value, not taking corresponding response measures on the set offshore platform area.

[0016] The present application has the following beneficial effects: (1) The marine weather mesoscale analysis method realizes the synchronous extraction of meteorological disturbances and structural changes by step-by-step processing of marine weather observation data and meteorological image data in different time periods in the offshore platform area, generates a meteorological evolution risk perception index, and predicts the risk index in the next period, so that risk identification and emergency measures form an effective closed loop, thereby improving the accuracy of risk perception, and further helping to realize continuous identification of meteorological risks in the operation area and timely triggering of response actions, avoiding safety hazards caused by sudden weather changes, and thereby making the operation platform have stronger forward-looking prediction ability.

[0017] (2) The marine weather mesoscale analysis method processes each period of marine weather observation data collected by item, is more clear when identifying interference characteristics, and generates a meteorological disturbance aggregation index, so that the classification difficulty caused by data mixing is convenient for making clearer analysis of the meteorological disturbance characteristics of each period, enhances the correlation between meteorological disturbance characteristics, and can accurately distinguish various meteorological characteristics, thereby enabling faster identification and analysis of meteorological disturbances, improving overall analysis efficiency, reducing data processing and response time, and thereby ensuring the accuracy of meteorological risk assessment and emergency response.

[0018] (3) The marine weather mesoscale analysis method introduces an image disturbance identification model and multiple meteorological image data, thereby improving the analysis capability of the model in complex weather environments, and through the collaborative work of the feature extraction sub-network, the time series evolution sub-network and the fusion output sub-network, the model can process image data from different sources in real time and accurately extract meteorological disturbance and structural evolution characteristics, thereby enabling the model to accurately identify weather change trends and make dynamic predictions, thereby enhancing the prediction capability of meteorological risks, and greatly enhancing the response capability of the offshore operation platform.

[0019] Of course, implementing any product of the present application does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1A flow chart of a marine weather mesoscale analysis method of the present application.

[0021] Figure 2 A flow chart of specific steps of setting the weather disturbance aggregation index of each period in the set marine operation platform area in the marine weather mesoscale analysis method of the present application.

[0022] Figure 3 A flow chart of specific steps of setting the weather structure evolution index of each period in the set marine operation platform area in the marine weather mesoscale analysis method of the present application.

[0023] Figure 4 A time sequence diagram of setting the cloud structure evolution index in the set marine operation platform area in the marine weather mesoscale analysis method of the present application.

[0024] Figure 5 A time sequence diagram of setting the echo shape evolution index in the set marine operation platform area in the marine weather mesoscale analysis method of the present application. DETAILED DESCRIPTION

[0025] Please refer to Figure 1 The embodiment of the present application provides a technical scheme: a marine weather mesoscale analysis method, comprising the following steps: continuously acquiring marine weather observation data in a set marine operation platform area (such as an area with a radius of 20 kilometers centered on a marine operation platform) in a plurality of periods (each period is 2 hours), and analyzing the weather disturbance aggregation index of the corresponding period; acquiring meteorological image data in the set marine operation platform area in each period, and combining a pre-trained image disturbance recognition model to analyze the weather structure evolution index of the corresponding period; based on the weather disturbance aggregation index and the weather structure evolution index of each period in the set marine operation platform area, analyzing the weather evolution risk perception index of the corresponding period, and predicting the weather evolution risk perception index of the next period in the set marine operation platform area; and taking corresponding response measures for the set marine operation platform area based on the weather evolution risk perception index of the next period.

[0026] The specific formula for calculating the weather evolution risk perception index of a certain period in the set marine operation platform area is as follows: ; wherein, is the weather evolution risk perception index of a certain period in the set marine operation platform area, is the weather disturbance aggregation index of a certain period in the set marine operation platform area, is the disturbance coefficient stored in the database, is the weather structure evolution index of a certain period in the set marine operation platform area, is the evolution coefficient stored in the database, is a disturbance coefficient stored in the database, is a proportion coefficient stored in the database, and in the embodiment, takes a value of 2.000.

[0027] It needs to be explained that the disturbance coefficient stored in the database is The acquisition steps are as follows: reading the water vapor disturbance index, wind variation disturbance index, and thermal difference disturbance index of the period, extracting the disturbance mean and disturbance standard deviation, and performing ratio processing, i.e., [1- (disturbance standard deviation / disturbance mean)] / 2, and taking the result as the disturbance coefficient . The evolution coefficient stored in the database The acquisition steps are as follows: reading the cloud layer structure evolution index and echo shape evolution index of the period, extracting the evolution mean and evolution standard deviation, and performing ratio processing, i.e., [1- (evolution standard deviation / evolution mean)] / 2, and taking the result as the evolution coefficient . The interaction coefficient stored in the database The acquisition steps are as follows: reading the meteorological disturbance aggregation index and meteorological structure evolution index, and based on the interaction correlation value between the two by Pearson correlation coefficient analysis, and performing ratio processing, i.e., (interaction correlation value+1) / 2, and taking the result as the interaction coefficient .

[0028] Specifically, as Figure 2 shown, the marine meteorological observation data includes a micro-vorticity factor, a wind direction jump rate value, a wind field dispersion index, a sea temperature distribution dispersion factor, a temperature anomaly rate value, and a micro-layer inversion index. The specific steps of analyzing the meteorological disturbance aggregation index of each period in the set marine operation platform region are as follows: based on the marine meteorological observation data of each period in the set marine operation platform region, analyzing the corresponding period meteorological interference evaluation set, including the wind variation disturbance index and the thermal difference disturbance index; obtaining the water vapor disturbance index of each period in the set marine operation platform region, and combining the corresponding period meteorological interference evaluation set, analyzing the corresponding period meteorological disturbance aggregation index, which is specifically: performing weighted processing on the water vapor disturbance index, the wind variation disturbance index, and the thermal difference disturbance index of each period to obtain the corresponding period meteorological disturbance aggregation index. It is pointed out that the weighting processing involved in the embodiment can obtain the weighting coefficient corresponding to each parameter by entropy weight method (except for the specific acquisition process which has been written), and the specific acquisition steps of the weighting coefficient corresponding to the water vapor disturbance index, the wind variation disturbance index and the thermal difference disturbance index are taken as examples: that is, the water vapor disturbance index, the wind variation disturbance index and the thermal difference disturbance index of each period are normalized, and the sum value corresponding to the normalized processing result is obtained, and then the normalized processing result is respectively compared with the sum value corresponding to the normalized processing result to obtain the water vapor disturbance index proportion value, the wind variation disturbance index proportion value and the thermal difference disturbance index proportion value of each period, and the corresponding information entropy value is calculated. Specifically, the water vapor disturbance index proportion value of each period is multiplied by the natural logarithm value of the proportion, the sum of the calculation results of all periods is taken as the negative number, and then divided by the natural logarithm value of the total number of periods, and the information entropy values of the wind variation disturbance index and the thermal difference disturbance index are calculated according to the same logic, then 1 is subtracted from the information entropy value of the water vapor disturbance index to obtain the information utility value of the water vapor disturbance index of each period, and the information utility values of the wind variation disturbance index and the thermal difference disturbance index of each period are calculated according to the same logic. The information utility values of the water vapor disturbance index, the wind variation disturbance index and the thermal difference disturbance index of each period are added to obtain the total utility value of each period; the information utility values of the water vapor disturbance index, the wind variation disturbance index and the thermal difference disturbance index of each period are respectively divided by the total utility value of the corresponding period, and the results are taken as the weighting coefficients corresponding to the water vapor disturbance index, the wind variation disturbance index and the thermal difference disturbance index.

[0029] Wherein, the microvorticity factor is the local rotation degree of the wind field in the period, which can be obtained by arranging a plurality of wind speed measuring points (such as GPS buoy, platform wind meter, offshore radar, etc.) in the set ocean operation platform area, and obtaining two-dimensional wind speed vector data (including wind speed and wind direction angle) of each observation point based on wind speed sensor and wind direction sensor, forming a local area with three observation points in any triangular distribution, analyzing two-dimensional horizontal vorticity in finite difference form, that is, constructing the wind speed variation rate in east-west direction (x axis) and north-south direction (y axis) in the local area, calculating the difference value of east-west wind speed (u) and north-south wind speed (v) between any two observation points, and analyzing the horizontal distance difference of the three observation points in east-west direction (x axis) and north-south direction (y axis) according to their geographical coordinates; the difference value of wind speed component and the corresponding direction distance difference value are used for ratio calculation to approximately estimate the partial derivative, and the two approximate partial derivatives are substituted into the two-dimensional vorticity formula, and the mean value is processed based on the result, which is taken as the microvorticity factor for identifying the rotation disturbance degree in the local wind field.

[0030] The wind direction jump rate value is the rate of the sharp angle change of the wind direction in the period, which can be obtained by the wind direction sensor to obtain the wind direction angle value at several time points (such as 10 minutes interval) in the period, and the wind direction angle values of adjacent time points are processed by ratio (such as the absolute value of the difference of the wind direction angle values of the first time point and the second time point / the time interval), and the weighted average processing is performed, and the result is taken as the wind direction jump rate value, which is used to measure the suddenness and severity of the wind direction change in the period.

[0031] The wind field dispersion index is the spatial distribution degree of the wind speed and the wind direction in the period, the two-dimensional wind speed vector data of each observation point, i.e. the wind speed value and the wind direction angle, are read, and the wind speed mean value, the wind speed standard deviation and the wind direction angle standard deviation (the direction cosine transformation is performed on the wind direction angles of all the observation points, and then the standard deviation is obtained) are extracted, the wind speed mean value and the wind speed standard deviation are processed by ratio, i.e. the wind speed standard deviation / the wind speed mean value, and the result is weighted with the wind direction angle standard deviation, and the result is taken as the wind field dispersion index, which is used to measure the stability of the wind field structure in the period.

[0032] The sea surface temperature distribution dispersion factor is the uniformity of the spatial distribution of the sea surface temperature in the period, which can be obtained by the temperature sensor to obtain the sea surface temperature value of each observation point at each time point, and the sea surface temperature variance value and the sea surface temperature mean value are extracted, and the ratio processing is performed, i.e. the sea surface temperature variance value / the sea surface temperature mean value, and the result is averaged, and then taken as the sea surface temperature distribution dispersion factor, which is used to quantify the non-uniformity degree of the sea surface thermal distribution in the period, and reflects the thermal asymmetric structure existing in the local sea-air interface, which is easy to cause the local evaporation enhancement, the convective uplift or the formation of the mesoscale disturbance.

[0033] The air temperature anomaly rate value is the deviation degree of the air temperature in the marine area in the period, which is obtained by the temperature field angel to obtain the air temperature value and the air temperature reference value (the historical air temperature values at several times are obtained, and the mean value is taken) of each observation point, and the ratio processing is performed, such as |air temperature value-air temperature reference value| / air temperature reference value, and the standard deviation processing is performed, and the result is taken as the air temperature anomaly rate value, which is used to quantify the mutation amplitude of the atmospheric thermal distribution in the period, and reflects the degree of the current air temperature field existing sharp deviation, which can be used as the thermal precursor factor of the mesoscale evolution such as the convective excitation, the wind field disturbance and the local circulation formation.

[0034] The micro-layer inversion index is the intensity of temperature inversion in the lower layer of the near-earth atmosphere in the period, which obtains the temperature values of the two height layers (such as 2 meters and 10 meters) of each observation point through the temperature sensor, extracts the temperature difference values of the two height layers of each observation point, takes the temperature mean value and the temperature standard deviation value, and performs ratio processing, that is, the temperature standard deviation value / the temperature mean value, and takes the result as the micro-layer inversion index, which is used to measure the abnormal positive amplification degree of the local vertical temperature gradient. The greater the index value, the higher the stability of the thermal stratification, and the stronger the inhibition effect on the vertical disturbance.

[0035] The specific steps of analyzing the meteorological disturbance evaluation set of each period in the set offshore platform area are as follows: based on the micro-vorticity factor, the wind direction jump rate value and the wind field dispersion index of each period in the set offshore platform area, the wind variation disturbance index of the corresponding period is analyzed, which is specifically: the micro-vorticity factor, the wind direction jump rate value and the wind field dispersion index of each period are standardized, and the wind variation disturbance index of the corresponding period is obtained based on the weighted processing result of the standardized processing result; based on the sea temperature distribution dispersion factor, the air temperature anomaly rate value and the micro-layer inversion index of each period in the set offshore platform area, the thermal difference disturbance index of the corresponding period is analyzed, which is specifically: the sea temperature distribution dispersion factor, the air temperature anomaly rate value and the micro-layer inversion index of each period are standardized, and the thermal difference disturbance index of the corresponding period is obtained based on the weighted processing result of the standardized processing result.

[0036] The specific steps of obtaining the water vapor disturbance index of each period in the set offshore platform area are as follows: obtaining the humidity response data of each period in the set offshore platform area, including the humidity difference aggregation factor, the dew point temperature jump rate value and the humidity delay response factor, and performing standardization processing; based on the humidity response data of each period in the set offshore platform area after standardization processing, the water vapor disturbance index of the corresponding period is analyzed, which is specifically: the humidity difference aggregation factor, the dew point temperature jump rate value and the humidity delay response factor of each period after standardization processing are weighted, and the water vapor disturbance index of the corresponding period is obtained.

[0037] The humidity difference aggregation factor is the spatial aggregation degree of the humidity difference value between a plurality of observation points in the period, which is used to reflect the unevenness of the water vapor distribution in the local area and the potential water vapor aggregation trend. It can be obtained by a plurality of humidity sensors to obtain the humidity value of each observation point, calculate the humidity difference value between adjacent observation points, mark the adjacent observation points with a humidity difference higher than a set threshold as a high humidity difference pair, count the total humidity difference value and the total humidity difference value of the high humidity difference pair, and perform ratio processing to take the result as the humidity difference aggregation factor. The greater the humidity difference aggregation factor value, the stronger the unevenness of the humidity field in the region, and there is a local moisture convergence phenomenon, which is a precursor of potential sea fog, precipitation or convection triggering.

[0038] The dew point temperature jump rate value is the rate of change of the dew point temperature in the period. The temperature value and the relative humidity value of each observation point at each time point can be obtained by the temperature and humidity integrated sensor. The current dew point temperature value of each observation point is calculated by combining the meteorological dew point temperature formula. The dew point temperature time series is processed by first-order difference processing in a sliding time window. The jump rate (i.e. the absolute value of the difference between the dew point temperatures of adjacent two time points divided by the time interval) is extracted and moving average is performed. The output is the dew point temperature jump rate value, which is used to measure the mutation of the water vapor content in the air. The larger the value is, the more intense the water vapor content fluctuates, which may be the intersection of cold and warm air, the invasion of the front or the change of water vapor flux.

[0039] The humidity delay response factor is the delay degree of the relative humidity response to the change of air temperature in the period, which reflects the thermal inertia or hysteresis of the humidity field. The air temperature and humidity time series (i.e. the air temperature value and humidity value at each time point) of each observation point can be obtained by the temperature and humidity integrated sensor. The temperature change rate sequence and humidity change rate sequence are obtained by performing sliding difference processing on the air temperature and humidity time series respectively. The cross-correlation analysis is performed, i.e. the cross-correlation function analysis is performed, to find the time delay corresponding to the maximum value of the cross-correlation function, i.e. the lag time of the temperature change response to the humidity, and the mean value is taken as the humidity delay response factor. The larger the value is, the more significant the humidity change lags behind the temperature disturbance, which may correspond to the delay of thermal transport and the strong atmospheric stability of the marine weather process.

[0040] In the embodiment, by processing the marine meteorological observation data item by item and performing standardization and weighted calculation, the classification difficulty caused by data mixing can be effectively avoided. In the process of meteorological disturbance feature recognition, the accuracy of distinguishing various meteorological disturbances is significantly improved. Secondly, by analyzing and processing the humidity response data, the identification ability of water vapor disturbance is strengthened, so as to deeply mine the internal correlation of meteorological disturbance features in the marine area and improve the accuracy and comprehensiveness of the analysis. Finally, by reasonably weighting the disturbance index of each period, the performance of meteorological risk perception and prediction is optimized, which helps to accurately judge the potential risk and trigger timely response measures.

[0041] Specifically, as shown in Figure 3 The meteorological image data includes cloud image data (including pixel value and two-dimensional coordinates of each pixel point) and radar echo image data (echo intensity value and echo two-dimensional coordinates of each echo pixel point) of each time window (such as 10 minutes). The image disturbance recognition model includes a feature extraction subnetwork, a time sequence evolution subnetwork and a fusion output subnetwork.

[0042] The specific steps of analyzing the weather structure evolution index of each time period in the set offshore platform area are as follows: inputting the weather image data of each time period in the set offshore platform area into the pre-trained image disturbance recognition model for comprehensive analysis to obtain the structure evolution feature vector set of the corresponding time period, including the cloud layer structure evolution index and the echo pattern evolution index; based on the structure evolution feature vector set of each time period in the set offshore platform area, analyzing the weather structure evolution index of the corresponding time period.

[0043] The specific formula for calculating the weather structure evolution index of a certain time period in the set offshore platform area is as follows: ; wherein, is the weather structure evolution index of a certain time period in the set offshore platform area, is the cloud layer structure evolution index of a certain time period in the set offshore platform area, is the cloud layer coefficient stored in the database, is the echo pattern evolution index of a certain time period in the set offshore platform area, is the echo coefficient stored in the database, is the coordination coefficient stored in the database, is the adjustment coefficient stored in the database, and in the present embodiment example, the value is 2.000.

[0044] It needs to be explained that the cloud layer coefficient stored in the database is The acquisition steps are as follows: reading the cloud cluster edge tearing evolution features, disintegration sign evolution features, and cloud texture distribution evolution features of the time period, extracting the cloud layer mean and cloud layer standard deviation, and performing ratio processing, i.e. [1- (cloud layer standard deviation / cloud layer mean)] / 2, taking the result as the cloud layer coefficient ; The echo coefficient stored in the database is The acquisition steps are as follows: reading the echo cluster abnormal evolution features, radar structure shear distortion evolution features, and echo strip splitting evolution features of the time period, extracting the echo mean and echo standard deviation, and performing ratio processing, i.e. [1- (echo standard deviation / echo mean)] / 2, taking the result as the echo coefficient ; The coordination coefficient stored in the database is The acquisition steps are as follows: reading the cloud layer structure evolution index and the echo pattern evolution index of the time period, and based on the evolution correlation value between the two obtained by Pearson correlation coefficient analysis, performing ratio processing, i.e. (evolution correlation value+1) / 2, taking the result as the coordination coefficient .

[0045] The following is a specific implementation example of calculating the meteorological structure evolution index for a certain time period within the designated marine operation platform area. The available data includes cloud structure evolution indices and echo morphology evolution indices for five randomly selected time periods within the designated marine operation platform area, as detailed in Table 1. Figures 4-5 As shown: Table 1. Example of time-series data for setting structural evolution feature vector sets within the area of ​​an offshore operating platform. Cloud coefficients stored in the database Approximately 0.428; echo coefficients stored in the database Approximately 0.374; Collaboration coefficients stored in the database Approximately 0.493; Adjustment coefficients stored in the database The value is: 2.000; Even if the above coefficient is substituted into the specific formula for calculating the meteorological structure evolution index of a certain period within the designated marine operating platform area, we get: The meteorological structure evolution index for the first time period within the marine operation platform area is set as ((0.205^0.428+0.246^0.374) / 2.000)×(ln1+(√0.205×0.246)^0.493)≈0.764; The meteorological structure evolution index for the second time period within the marine operation platform area is set as ((0.314^0.428+0.294^0.374) / 2.000)×(ln1+(√0.314×0.294)^0.493)≈0.894; The meteorological structure evolution index for the third time period within the marine operation platform area is set as ((0.271^0.428+0.238^0.374) / 2.000)×(ln1+(√0.271×0.238)^0.493)≈0.815; The meteorological structure evolution index for the fourth time period within the marine operation platform area is set as ((0.419^0.428+0.349^0.374) / 2.000)×(ln1+(√0.419×0.349)^0.493)≈1.012; The meteorological structure evolution index for the fifth time period within the marine operation platform area is set as ((0.364^0.428+0.318^0.374) / 2.000)×(ln1+(√0.364×0.318)^0.493)≈0.949.

[0046] In this embodiment, by analyzing the marine weather image data of each period, combining the cloud structure evolution and echo shape evolution characteristics, the accurate calculation of the weather structure evolution index is realized. Secondly, through the analysis of the cloud layer coefficient and the echo coefficient, the standard deviation and the mean of the cloud layer and the echo are extracted, and then the dynamic characteristics of the cloud layer and the echo are obtained through the ratio processing, which helps to quantify the change amplitude and stability of them in different time periods, and then provides a clear measurement for the weather structure evolution. Finally, the introduction of the coordination coefficient can reveal the mutual influence degree of the cloud layer and the echo in the evolution process based on the Pearson correlation coefficient analysis of the evolution relationship between the cloud layer and the echo, so as to help accurately extract the key features of the weather disturbance and the structure change, and then effectively present the complex relationship between the cloud layer and the echo, thereby improving the accuracy of risk prediction.

[0047] Specifically, the specific steps of analyzing the structure evolution feature vector set of each period in the set marine operation platform area are as follows: in the feature extraction sub-network of the image disturbance recognition model, the cloud image data and the radar echo image data of each time window in each period in the set marine operation platform area are received, and the cloud and radar feature vector set of the corresponding time window is analyzed, which is specifically: through the cloud feature extraction layer and the radar echo feature extraction layer, the cloud image data and the radar echo image data of each time window in each period in the set marine operation platform area are extracted to extract the cloud cluster edge tearing feature, the collapse sign feature, the cloud texture distribution feature, the echo block abnormal feature, the radar structure shear disturbance feature and the echo strip splitting feature; in the time sequence evolution sub-network of the image disturbance recognition model, based on the cloud and radar feature vector set of each time window in each period in the set marine operation platform area, the evolution feature vector set of the corresponding period is analyzed, which is specifically: through the cloud evolution layer and the radar echo evolution layer, the cloud and radar feature vector set of each time window in each period in the set marine operation platform area is evolved to extract the cloud cluster edge tearing evolution feature, the collapse sign evolution feature, the cloud texture distribution evolution feature, the echo block abnormal evolution feature, the radar structure shear disturbance evolution feature and the echo strip splitting evolution feature; In the fusion output sub-network of the image disturbance recognition model, based on the evolution feature vector set of each period in the set marine operation platform area, the structure evolution feature vector set of the corresponding period is analyzed, and the cloud layer structure evolution index and the echo shape evolution index are extracted through the weighted processing of the cloud layer structure feature output layer and the echo shape feature output layer on the evolution feature vector set of each period in the set marine operation platform area.

[0048] Wherein, the feature extraction sub-network includes a cloud feature extraction layer and a radar echo feature extraction layer.

[0049] a cloud layer feature extraction layer for extracting cloud cluster edge tearing features, disintegration sign features, and cloud texture distribution features, the specific extraction process being as follows: cloud cluster edge tearing features: based on an image edge detection algorithm (such as Canny), the cloud layer image of each time window in each period in the set marine operation platform region is processed to extract a plurality of cloud cluster edge pixel points of the corresponding time window, and each cloud cluster edge pixel point is divided into a plurality of contour boundary curves (using a connected domain marking algorithm, such as 8-neighbor connected analysis, all cloud cluster edge pixel points are grouped, and continuous pixel blocks are divided into a plurality of edge connected regions, in each connected region, further through a contour tracking algorithm, such as the boundary tracking method proposed by Suzuki and Abe, the cloud cluster edge pixel points are sequentially connected to obtain the complete boundary curve of the region, each curve is a contour boundary curve, including a plurality of cloud cluster edge pixel points); a sliding window curvature estimation operation (such as every 5 pixel points as a group) is performed on each contour boundary curve to extract the edge curvature value in the corresponding sliding window; the number of curvature mutation segments of each contour boundary curve (i.e. the current window curvature value changes more than 2 times the average value of adjacent windows) is extracted, and the total length of the mutation segments (for each pair of cloud cluster edge pixel points on the mutation segment, based on their corresponding two-dimensional coordinates, the Euclidean distance between them is calculated, and the total length of the mutation segment is accumulated) is recorded; finally, the total length of the mutation segments is divided by the total length of the contour boundary curve (consistent with the logic of obtaining the total length of the mutation segments) as the tearing intensity value of the contour, and the maximum value of all boundary tearing intensity values is taken as the cloud cluster edge tearing feature of the current image, which is used to identify the potential instability of the possible structural damage and disturbance splitting of the cloud cluster boundary; Collapse sign feature: based on the brightness distribution analysis method, the cloud image of each time window in each period in the set offshore platform area is subjected to Gaussian blur processing, and the low-frequency brightness background image is extracted. Then, the centroid coordinate point of the main cloud cluster region is located by using the brightness peak clustering algorithm. Taking the centroid coordinate point as the center, a plurality of concentric circular analysis regions are sequentially constructed, and the radius of each concentric circle is increased by a fixed interval (such as 10, 20, 30 pixels). The average brightness value of all pixel points in each concentric circular region is counted, a brightness sequence from inside to outside is constructed, then the brightness sequence is calculated layer by layer, a brightness ratio sequence composed of the average brightness value changes between adjacent concentric circles is obtained. For each pair of adjacent circles, the ratio of the brightness of the next circle to the brightness of the previous circle is calculated. The entire brightness ratio sequence is traversed, the part with a brightness drop of more than 20% is identified and marked as a brightness drop segment, and the existence of two or more continuous brightness drop segments is identified. The average value of the brightness drop ratio of these drop segments is counted, and the result is used as the collapse sign feature to measure the degree of loosening, layering fracture or outward diffusion of the main cloud cluster edge structure. Cloud texture distribution feature: read the main cloud cluster region of each time window in each period in the set offshore platform area, perform high-frequency texture response enhancement processing on the main cloud cluster region, and use a multi-scale Gabor filter set (such as 5 scales x 8 directions) to perform convolution operation on the image. The texture response value of each pixel point under different directions and scales is extracted, and the energy value (i.e. the sum of the squares of the filter response values) of each direction response is calculated to construct a directional texture energy spectrum. Then, based on the constructed texture energy spectrum, the main cloud cluster region is divided into a plurality of equal grid regions (such as 8x8), the maximum value, mean value and standard deviation of the 8-direction texture response energy in each grid region are counted, and the direction concentration degree vector of each grid (i.e. the normalized vector of the proportion of each direction energy in the grid) is constructed. The proportion of the direction with the highest energy in each grid is counted, and the average value in all grids is calculated to obtain the direction concentration index. The number of grids with energy proportion exceeding a threshold value (such as 25%) in all directions is counted, and the ratio to the total number of grids is calculated to obtain the direction density anomaly factor. The energy distribution uniformity value is obtained by performing variance analysis on the energy mean value of all grids. The direction concentration index, direction density anomaly factor and energy distribution uniformity value are weighted to obtain the cloud texture distribution feature, which is used to evaluate the degree of local organization abnormality in the main cloud cluster region.

[0050] The radar echo feature extraction layer is used to extract the echo cluster abnormal feature, the radar structure shear-dilation disturbance feature, and the echo strip splitting feature. The specific extraction process is as follows: Echo blob anomaly feature: threshold segmentation is performed on the radar echo image in a set time window (e.g., the reflectivity threshold is set to 35 dBZ), all high-intensity echo pixel points exceeding the threshold are extracted, based on connected component analysis (e.g., eight-neighbor connectedness Flood-Fill algorithm), each high-intensity blob region (including several high-intensity echo pixel points) is identified, the number of all high-intensity echo pixel points in each blob region is counted, that is, the area of the blob region, a contour tracking algorithm (e.g., Suzuki-Abe contour extraction algorithm) is used to perform boundary extraction on each blob region to obtain the corresponding boundary contour pixel point set (including several echo boundary contour pixel points), the boundary contour pixel point set is connected in sequence to form a boundary polyline, the Euclidean distance between adjacent echo boundary contour pixel points is calculated, and the total perimeter value of the blob region is accumulated, and the corresponding circularity factor calculation is analyzed, that is, (4 x π x blob area) / (blob perimeter squared), based on the extracted boundary contour pixel point set, the minimum circumscribed rectangle of the blob is constructed (i.e., the minimum axis-aligned rectangle that can cover all echo boundary contour pixel points), and the diagonal length is recorded as the minimum rectangle side length, then the perimeter value of the blob region is compared with the minimum circumscribed rectangle side length, that is, boundary concave-convex degree = actual boundary length of the blob / minimum circumscribed rectangle side length, and the circularity factor is weighted, based on the result, the result is processed by mean value, and the result is taken as the echo blob anomaly feature, which is used to measure the geometric distortion degree of high-intensity blob in the radar echo image in each time window, if the value is high, it indicates that there is a large area but irregular boundary, and the structure is poor, which usually corresponds to the echo structure with rapid radar reflectivity rise but disordered shape, which can identify the uneven organization, unstable development direction or broken edge of convective cell in the precipitation cloud cluster; Radar structure shear-dilation disturbance feature: the current radar image is subjected to gray scale standardization processing, and a high reflectivity threshold (e.g., 35 dBZ) is set to extract the main echo region; then, based on the connected component division result of the main echo region, the two-dimensional boundary contour of each blob is constructed, the Sobel operator is used to calculate the boundary direction gradient distribution, the gradient change rate in the horizontal and vertical directions of each blob boundary is counted, then the ratio of the horizontal gradient fluctuation mean value to the main axis length of the vertical extension direction of each blob boundary is calculated to obtain the shear index (horizontal gradient fluctuation mean value / main axis length), and the blob expansion rate (the increase in the number of boundary pixels after dilation / the original number of boundary pixels) is calculated based on the boundary dilation processing (e.g., morphological dilation operation); finally, the shear index and expansion rate of all blobs are weighted and fused (e.g., weighted average), and the result is taken as the radar structure shear-dilation disturbance feature value, which is used to measure the organizational tension deformation degree, non-uniform expansion amplitude of the blob boundary and local shear disturbance intensity in each time window of the radar echo image, and represents the instability of the wind field structure, the vertical upward disturbance intensity and the evolution trend of the confluence boundary; Echo strip splitting feature: The radar echo image in the set time window is subjected to gray scale normalization processing, and a high reflectivity threshold (such as 35 dBZ) is set for initial threshold segmentation, and all strong echo regions with reflectivity exceeding the threshold are extracted. Then, a line target extraction algorithm (such as a probability Hough transform or a line detection neural network) is used to perform line detection on the strong echo region, and strip regions that meet the following characteristics are screened out: length greater than fifty pixels; strip width less than ten pixels; average reflectivity higher than the set threshold; the above strip region is expressed as the input of subsequent principal axis extraction according to its boundary pixel coordinates set. For each identified strip region, a least squares fitting algorithm or a line segment fitting algorithm (such as the RANSAC algorithm) is used to extract its principal axis, and the principal axis is divided into several equal-length sub-sections (such as ten pixels per sub-section) along the strip length direction. The included angle (with the horizontal right as the zero degree reference, counterclockwise as positive) of the principal axis direction of each sub-section is calculated, and the change analysis of the direction sequence is performed. If the direction jump angle between two or more adjacent sub-sections is greater than twenty degrees, or a bidirectional principal axis appears in a sub-section (i.e., the principal axis fitting in the sub-section shows a double-peak direction trend), the strip is recorded as having a direction splitting behavior. The number of strips with the above direction splitting behavior is counted, and the ratio of the total number of strips is calculated to obtain the strip splitting proportion value. The result is used as the echo strip splitting feature to represent the overall disruption, continuity degradation, or instability of the precipitation axis concentration trend of the strip organization structure in the radar echo image in the period.

[0051] The cloud cluster edge tearing feature, the disintegration sign feature, the cloud pattern texture distribution feature, the echo cluster anomaly feature, the radar structure shear expansion disturbance feature, and the echo strip splitting feature are marked as a cloud radar feature vector set.

[0052] The time sequence evolution sub-network includes a cloud layer evolution layer and a radar echo evolution layer.

[0053] The cloud layer evolution layer is used to extract the cloud cluster edge tearing feature, the disintegration sign feature, and the cloud pattern texture distribution feature of each time window of each period in the set offshore platform region, and perform evolution analysis respectively. Taking the evolution analysis of the cloud edge tearing feature of each time window as an example, the cloud edge tearing feature mean and standard deviation of each time window in the period are extracted, and the ratio of the cloud edge tearing feature standard deviation to the cloud edge tearing feature mean is processed, that is, the cloud edge tearing stability feature is obtained. The least square method is used to linearly fit the cloud edge tearing feature of each time window, and the cloud edge tearing evolution rate is obtained. The cloud edge tearing features of adjacent time windows are differentially processed, and a plurality of groups of adjacent time window cloud differences are obtained. The maximum cloud difference value is counted, and the maximum cloud difference value is weighted with the cloud edge tearing stability feature and the cloud edge tearing evolution rate. The cloud edge tearing evolution feature of each period is obtained, and the same logic is used to extract the collapse sign evolution feature and the cloud texture distribution evolution feature of each period.

[0054] The radar echo evolution layer is used to extract the echo block abnormal feature, radar structure shear disturbance feature and echo strip splitting feature of each time window in each period in the set offshore platform area, and evolution analysis is performed respectively. The echo block abnormal evolution feature, radar structure shear disturbance evolution feature and echo strip splitting evolution feature of each period are extracted according to the same logic of extracting the cloud edge tearing evolution feature of each period.

[0055] The cloud edge tearing evolution feature, collapse sign evolution feature, cloud texture distribution evolution feature, echo block abnormal evolution feature, radar structure shear disturbance evolution feature and echo strip splitting evolution feature are marked as an evolution feature vector set.

[0056] The fusion output sub-network includes a cloud layer structure feature output layer and an echo morphology feature output layer.

[0057] The cloud layer structure feature output layer is used to extract the cloud edge tearing evolution feature, collapse sign evolution feature and cloud texture distribution evolution feature of each period in the set offshore platform area, and weighted processing is performed. Based on the weighted processing result, the cloud layer structure evolution index between 0 and 1 is normalized and mapped, which is used to represent the organization structure stability and boundary disturbance development degree of the cloud layer in each period, and reflects the structural integrity change trend and local evolution intensity of the cloud cluster in each period.

[0058] The echo morphology feature output layer is used to extract the echo block abnormal evolution feature, radar structure shear disturbance evolution feature and echo strip splitting evolution feature of each period in the set offshore platform area, and weighted processing is performed. Based on the weighted processing result, the echo morphology evolution index between 0 and 1 is normalized and mapped, which is used to represent the spatial integrity, morphology complexity and dynamic evolution intensity of the radar echo structure in each period, and reflects the trend of weakening of the organization, splitting development or structure recombination of precipitation in the horizontal and vertical scales.

[0059] and the pre-training step of the image disturbance recognition model is as follows: In the pre-training stage of the image disturbance recognition model, first, an image training data set covering multiple types of marine weather disturbance working conditions (organizational tearing, collapse expansion, texture anomaly, shear dilation mutation, echo splitting, etc.) is constructed. Satellite cloud images and radar echo image sequences covering typical sea areas are collected, and artificial labeling or automatic labeling means are combined to record the main cloud area, strong echo structure and evolution feature level (such as tearing level, collapse index, shear expansion intensity, etc.) in each image, which are used to construct a supervised training label set.

[0060] For image samples, standardized gray images, edge response images, Gabor texture images, and frequency disturbance images are constructed as auxiliary feature images; to improve the generalization ability of the model, image enhancement operations are performed, including rotation, brightness disturbance, and local occlusion to generate disturbance scene expansion samples.

[0061] Then, the feature extraction subnetwork and the time series evolution subnetwork are pre-trained in stages. The feature extraction subnetwork takes the original image and the auxiliary feature image as the multi-channel tensor input, and outputs the edge tearing feature image, the collapse gradient image, the texture direction image, the cluster anomaly image, the shear index image, and the strip structure image. In the training, a multi-output MSE loss function is used to optimize the response of each feature layer, and a residual connection is used to improve the stability of deep training.

[0062] After feature extraction, the time series evolution training stage is entered, the input is the feature vector sequence of the continuous time window, and the output is the change curve of each type of evolution feature. The training goal is to minimize the slope deviation and trend cross error between the predicted curve and the real level change trend. Finally, the evolution features are mapped to two normalized indices (cloud structure evolution index and echo morphology evolution index) through weighted synthesis, which are used as the disturbance evaluation basis for the downstream decision module.

[0063] In this embodiment, the image disturbance recognition model is used for multi-level feature extraction and evolution analysis, which can effectively improve the recognition accuracy of meteorological evolution. Secondly, by refining the structure evolution features of the cloud and echo, the model can accurately capture the key changes and provide strong support for early identification of meteorological risks. Thirdly, the weighted processing method is used to fuse different features into cloud structure evolution index and echo morphology evolution index, which can fully represent the spatial structure and dynamic evolution of meteorological disturbance, thereby improving the predictability of meteorological evolution trend. Finally, by using deep learning technology, the learning ability and adaptability of the model are enhanced, and the real-time performance of meteorological risk assessment is significantly improved.

[0064] Specifically, the specific steps of predicting the meteorological evolution risk perception index of the next period in the set offshore platform region are as follows: based on the meteorological evolution risk perception index of each period in the set offshore platform region, analyzing the meteorological evolution mean value, meteorological evolution standard deviation value and meteorological evolution trend mean value in the set offshore platform region, specifically: setting a change time window (such as 5 periods), for the meteorological evolution risk perception index of each period in the change time window, respectively performing mean value processing, standard deviation processing and trend processing (such as the meteorological evolution risk perception index of the first period / the meteorological evolution risk perception index of the second period, obtaining the meteorological evolution trend value of adjacent periods, and performing mean value processing), obtaining the meteorological evolution mean value, meteorological evolution standard deviation value and meteorological evolution trend mean value; based on the meteorological evolution mean value, meteorological evolution standard deviation value and meteorological evolution trend mean value in the set offshore platform region, analyzing the predicted meteorological evolution risk perception index in the set offshore platform region, and marking it as the meteorological evolution risk perception index of the next period, specifically: performing weighted processing on the meteorological evolution standard deviation value and the meteorological evolution trend mean value (and the plus sign is replaced by a minus sign), obtaining a change factor, and performing summation processing with the meteorological evolution mean value, obtaining the predicted meteorological evolution risk perception index.

[0065] The specific steps of taking corresponding response measures for the set offshore platform region based on the meteorological evolution risk perception index of the next period are as follows: judging whether the meteorological evolution risk perception index of the next period in the set offshore platform region is higher than a preset meteorological evolution risk perception index threshold; if it is higher than the preset meteorological evolution risk perception index threshold, taking corresponding response measures for the set offshore platform region, specifically: automatically sending a warning notice to the offshore platform, and suspending or postponing the tasks of high-altitude operation, hoisting operation, precise positioning operation and other meteorological interference sensitive task modules in the affected area, updating the operation time window, and automatically issuing a folding, folding, power-off or warehousing instruction for the crane boom, communication antenna, unmanned aerial vehicle system and other equipment to prevent physical damage caused by meteorological disturbance; if it is not higher than the preset meteorological evolution risk perception index threshold, no corresponding response measures are taken for the set offshore platform region.

[0066] In the embodiment, by analyzing the meteorological evolution risk perception index of each period in detail, the mean, standard deviation and trend analysis are used to capture the dynamic characteristics of meteorological changes, and the meteorological evolution risk perception index of the next period is accurately predicted, thereby providing accurate meteorological trend prediction, helping to identify potential meteorological risks in advance, and by weighting the standard deviation and trend value, the prediction result is optimized, thereby ensuring more accurate risk assessment, and the predicted meteorological evolution risk perception index provides a clear basis for subsequent response measures, ensuring that the operation platform can quickly respond, thereby effectively ensuring operation safety, thereby enhancing the prediction ability and emergency handling efficiency of the operation platform, while reducing the risk of operation delay or equipment damage due to meteorological factors, thereby ensuring the safe operation of offshore operations.

[0067] Although preferred embodiments of the application have been described, those skilled in the art will be able to make additional changes and modifications to these embodiments once they have been given the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the application.

[0068] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A mesoscale analysis method for marine meteorology, characterized in that, Includes the following steps: Continuously acquire marine meteorological observation data for several time periods within the designated marine operation platform area, and analyze the meteorological disturbance aggregation index for the corresponding time periods; Meteorological image data for each time period within the designated marine operation platform area are acquired, and combined with a pre-trained image disturbance recognition model, the meteorological structure evolution index for the corresponding time period is analyzed. Based on the meteorological disturbance aggregation index and meteorological structure evolution index of each time period within the designated marine operation platform area, the meteorological evolution risk perception index of the corresponding time period is analyzed, and the meteorological evolution risk perception index of the next time period within the designated marine operation platform area is predicted. Based on the meteorological evolution risk perception index for the next period, corresponding response measures will be taken for designated marine operation platform areas.

2. The marine meteorological mesoscale analysis method according to claim 1, characterized in that, The marine meteorological observation data includes microedytosis factor, wind direction jump rate, wind field dispersion index, sea surface temperature distribution dispersion factor, air temperature anomaly rate, and micro-layer inversion index. The specific steps for analyzing and setting the meteorological disturbance aggregation index for each time period within the marine operation platform area are as follows: Based on marine meteorological observation data for each time period within the designated marine operation platform area, the meteorological disturbance assessment set for the corresponding time period is analyzed, including wind variation disturbance index and thermal difference disturbance index. Obtain the water vapor disturbance index for each time period within the designated marine operation platform area, and analyze the meteorological disturbance aggregation index for the corresponding time period by combining it with the meteorological disturbance assessment set for that time period.

3. The marine meteorological mesoscale analysis method according to claim 2, characterized in that, The specific steps for analyzing and setting the meteorological disturbance assessment set for each time period within the marine operation platform area are as follows: Based on the microvorticity factor, wind direction jump rate, and wind field dispersion index of each time period within the designated marine operation platform area, the wind disturbance index of the corresponding time period is analyzed. Based on the sea surface temperature distribution dispersion factor, air temperature anomaly rate, and microlayer inversion index for each time period within the designated marine operation platform area, the thermal difference disturbance index for the corresponding time period is analyzed.

4. The marine meteorological mesoscale analysis method according to claim 2, characterized in that, The specific steps for obtaining the moisture disturbance index for each time period within the designated marine operation platform area are as follows: Obtain humidity response data for each time period within the designated marine operation platform area, including humidity difference aggregation factor, dew point temperature jump rate value, and humidity delay response factor, and perform standardization processing. Based on the standardized humidity response data of the designated marine operation platform area for each time period, the water vapor disturbance index of the corresponding time period is analyzed.

5. The marine meteorological mesoscale analysis method according to claim 1, characterized in that, The meteorological image data includes cloud image data and radar echo image data for each time window, and the image disturbance recognition model includes a feature extraction subnetwork, a temporal evolution subnetwork, and a fusion output subnetwork.

6. The marine meteorological mesoscale analysis method according to claim 5, characterized in that, The specific steps for analyzing and setting the meteorological structure evolution index for each time period within the marine operation platform area are as follows: Meteorological image data for each time period within the designated marine operation platform area are input into a pre-trained image disturbance recognition model for comprehensive analysis, resulting in a set of structural evolution feature vectors for the corresponding time period, including cloud structure evolution index and echo morphology evolution index. Based on the set of structural evolution feature vectors for each time period within the designated marine operation platform area, the meteorological structure evolution index for the corresponding time period is analyzed.

7. The marine meteorological mesoscale analysis method according to claim 5, characterized in that, The specific steps for analyzing and defining the structural evolution feature vector set for each time period within the area of ​​the marine operation platform are as follows: In the feature extraction subnetwork of the image disturbance recognition model, cloud image data and radar echo image data for each time window of each time period within the designated marine operation platform area are received, and the cloud and thunder feature vector set of the corresponding time window is analyzed. In the temporal evolution subnetwork of the image disturbance recognition model, based on the cloud and thunder feature vector set of each time window in each time period within the marine operation platform area, the evolution feature vector set of the corresponding time period is analyzed. In the fusion output sub-network of the image perturbation recognition model, the structural evolution feature vector set of the corresponding time period is analyzed based on the evolution feature vector set of each time period within the set marine operation platform area.

8. The marine meteorological mesoscale analysis method according to claim 1, characterized in that, The specific formula for calculating the meteorological evolution risk perception index for a specific time period within the designated area of ​​the marine operation platform is as follows: ； in, , , The three indices are, in order, the meteorological evolution risk perception index, the meteorological disturbance aggregation index, and the meteorological structure evolution index for a specific time period within the designated marine operation platform area. , , , The coefficients are, in order, the disturbance coefficient, the evolution coefficient, the interaction coefficient, and the proportional coefficient stored in the database.

9. The marine meteorological mesoscale analysis method according to claim 1, characterized in that, The specific steps for predicting and setting the meteorological evolution risk perception index for the next time period within the area of ​​the marine operation platform are as follows: Based on the meteorological evolution risk perception index for each time period within the designated marine operation platform area, the mean, standard deviation, and mean trend of meteorological evolution within the designated marine operation platform area are analyzed. Based on the mean, standard deviation, and mean trend of meteorological evolution within the designated marine operation platform area, the predicted meteorological evolution risk perception index within the designated marine operation platform area is analyzed and marked as the meteorological evolution risk perception index for the next period.

10. The marine meteorological mesoscale analysis method according to claim 1, characterized in that, The specific steps for taking corresponding response measures for designated marine operation platform areas based on the meteorological evolution risk perception index for the next time period are as follows: Determine whether the meteorological evolution risk perception index for the next time period within the designated marine operation platform area is higher than the preset meteorological evolution risk perception index threshold. If the risk exceeds the preset threshold for meteorological evolution risk perception index, corresponding response measures will be taken for the designated marine operation platform area. If the risk level is not higher than the preset threshold for meteorological evolution risk perception index, no corresponding response measures will be taken for the designated marine operation platform area.

Images