A multi-source meteorological data fusion navigation risk intelligent early warning method and system

Abstract

The application belongs to the technical field of navigation early warning, and discloses a kind of navigation risk intelligent early warning method and system of multi-source meteorological data fusion, wherein, the application collects weather radar data, electric field intensity data, lightning event data and multi-source heterogeneous data such as meteorological monitoring data, then, data fusion is carried out on the aforementioned multi-source heterogeneous data to obtain the fusion characteristics of multi-source heterogeneous data;Finally, it is input into the lightning prediction model, and the lightning early warning of the aircraft in the target navigation area can be accurately realized when navigating;Therefore, compared with the traditional technology, the multi-source heterogeneous data related to lightning is data fused, and the flight lightning early warning is carried out based on this, which can improve the accuracy of early warning, thereby ensuring the flight safety of the aircraft;Therefore, the application is very suitable for large-scale application and popularization.

Description

Technical Field

[0001] This invention belongs to the field of navigation early warning technology, specifically relating to an intelligent early warning method and system for navigation risks based on multi-source meteorological data fusion. Background Technology

[0002] Civil aviation operations are highly dependent on weather conditions. Adverse weather can lead to flight delays and even jeopardize flight safety. Currently, severe weather conditions for the civil aviation industry mainly include heavy rain, thunderstorms, heavy snow, freezing temperatures, and strong winds. Among these, thunderstorms pose a significant threat to aviation safety, as they can directly strike aircraft, causing damage to the fuselage, electronic equipment malfunctions, communication disruptions, and even loss of flight control. Therefore, real-time monitoring and early warning through flight thunderstorm early warning systems, providing decision support for route planning, flight scheduling, and ground operations, is a key technology for ensuring the safety of modern civil aviation operations.

[0003] Currently, flight lightning warning technologies mainly include: (1) combining space lightning mapping data to calculate lightning risk values ​​and compare them with tolerance thresholds to provide lightning warnings; (2) using three-dimensional weather radar data as a basis, combining convolutional neural networks and SVM to detect and warn of lightning. Among these, the aforementioned existing technologies have the following shortcomings: most of them only rely on one type of monitoring data to generate warning information, so the warning deviation for the range of lightning occurrence is large and the accuracy is low. Therefore, based on the aforementioned shortcomings, how to provide a highly accurate intelligent warning method for flight risks has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent early warning of navigation risks based on the fusion of multi-source meteorological data, in order to solve the problem of low accuracy in the existing technology that uses single monitoring data to generate lightning warning information.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] Firstly, a method for intelligent early warning of navigation risks based on multi-source meteorological data fusion is provided, including:

[0007] Acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data for the target navigation area within a historical preset time period;

[0008] An effective meteorological grid data for the target navigation area was constructed using multiple meteorological monitoring data.

[0009] The lightning event data is transformed into lightning grid data that matches the weather radar data.

[0010] Weather input data is generated using weather radar data, effective meteorological grid data, and lightning grid data. Feature extraction processing is then performed on the meteorological input data to obtain meteorological features.

[0011] The meteorological features and the electric field intensity data are subjected to feature extraction and fusion processing to obtain fused features;

[0012] The fused features are input into the lightning prediction model to obtain the lightning prediction results for the target navigation area, and a navigation risk warning notification for the target navigation area is generated based on the lightning prediction results.

[0013] Based on the aforementioned disclosure, this invention first acquires weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data for the target navigation area within a historical preset time period. Then, using the multiple meteorological monitoring data, it constructs effective meteorological grid data for the target navigation area and performs data transformation processing on the lightning event data to obtain lightning grid data. Next, it uses the weather radar data, effective meteorological grid data, and lightning grid data to generate meteorological input data and extracts features from the meteorological input data to obtain meteorological features. This enables data fusion of meteorological data. Then, it performs feature extraction and fusion processing on the meteorological features and electric field intensity data to achieve the fusion of atmospheric electric field features and meteorological features, obtaining fused features. Finally, it inputs the fused features into a lightning prediction model to obtain lightning prediction results. Based on these results, navigation risk warnings for the target navigation area can be issued. Therefore, this invention fuses multi-source heterogeneous data related to lightning and uses this as a basis for flight lightning warnings, improving the accuracy of warnings and ensuring flight safety. Thus, this invention is highly suitable for large-scale application and promotion.

[0014] In one possible design, each meteorological monitoring data point corresponds to a separate meteorological sensor. The effective meteorological grid data for the target navigation area is constructed using multiple meteorological monitoring data points, including:

[0015] The target navigation area is gridded to obtain the corresponding area grid.

[0016] For any grid cell in the regional grid, meteorological sensors located within that grid cell are selected, and the meteorological monitoring data corresponding to the selected meteorological sensors are used to form the initial meteorological dataset for that grid cell.

[0017] For any meteorological monitoring data in the initial meteorological dataset, calculate the compatibility of each target data with the any meteorological monitoring data, and sum the compatibility of each data to obtain the total compatibility. Here, each target data is any meteorological monitoring data in the initial meteorological dataset excluding the any meteorological monitoring data.

[0018] Based on the overall compatibility, determine whether any of the meteorological monitoring data is valid data;

[0019] If so, then extract any of the meteorological monitoring data, and after polling all the meteorological monitoring data in the initial meteorological dataset, obtain several valid data in the initial meteorological dataset;

[0020] Based on several valid data, valid meteorological data for any grid cell is generated, and after all grid cells have been polled, valid meteorological data for each grid cell is obtained.

[0021] The effective meteorological grid data is constructed by using the effective meteorological data of each grid cell.

[0022] In one possible design, the compatibility of each target data with any of the meteorological monitoring data is calculated, including:

[0023] For any target data, calculate the data difference between any meteorological monitoring data and any target data;

[0024] Obtain the data attenuation factor and compatibility control parameters;

[0025] The compatibility factor is calculated based on the data decay factor and the data difference.

[0026] Perform an exponential operation on the compatibility factor with base e, and take the reciprocal of the result as the initial compatibility.

[0027] Based on the initial compatibility and the compatibility control parameters, the compatibility of any target data with any meteorological monitoring data is calculated;

[0028] Accordingly, generating valid meteorological data for any of the grid cells based on several valid data points includes:

[0029] The weight of each valid data point is calculated based on the overall compatibility of each valid data point.

[0030] Based on the weights of each valid data point, the valid data points are fused together to obtain the valid meteorological data for any given grid cell.

[0031] In one possible design, the lightning event data includes several lightning data points, wherein the lightning event data undergoes data transformation processing to convert it into lightning grid data that matches the weather radar data, including:

[0032] The target navigation area is gridded to obtain the corresponding area grid.

[0033] For any grid cell in the regional grid, lightning data whose occurrence location is within any grid cell is selected from the lightning event data, so as to form a lightning dataset for any grid cell using the selected lightning data;

[0034] Using the lightning dataset, the probability of lightning occurrence corresponding to any grid cell is calculated, and after polling all grid cells in the region grid, the probability of lightning occurrence corresponding to each grid cell is obtained.

[0035] The lightning grid data is constructed using the lightning occurrence probability corresponding to each grid cell.

[0036] In one possible design, any lightning data includes lightning intensity and the latitude and longitude at the time of the lightning strike. Using the lightning dataset, the probability of lightning occurrence corresponding to any given grid cell is calculated, including:

[0037] Obtain the grid index information and lightning impact radius of any of the grid cells;

[0038] For the b-th lightning data in the lightning dataset, based on the latitude and longitude of the b-th lightning data and the grid index information, the distance between the occurrence location of the b-th lightning data and the grid center of any grid cell is calculated;

[0039] Obtain the lightning occurrence sub-probability after any grid cell has undergone the (b-1)th lightning strike;

[0040] Using the interval distance, the lightning influence radius, the lightning intensity corresponding to the b-th lightning data, and the lightning occurrence probability of any grid cell after the (b-1)-th lightning strike, the lightning occurrence probability of any grid cell after the b-th lightning strike is calculated.

[0041] Increment b by 1, and recalculate the distance between the location of the b-th lightning data and the center of the grid based on the latitude and longitude of the b-th lightning data, until b equals B, to obtain the lightning occurrence probability corresponding to any grid cell, where the initial value of b is 1, and B is the total number of data in the lightning dataset.

[0042] In one possible design, the meteorological input data is subjected to feature extraction processing to obtain meteorological features, including:

[0043] The meteorological input data is input into the meteorological feature extraction model for feature extraction processing to obtain the meteorological features;

[0044] The meteorological feature extraction model includes an input encoding layer, a feature extraction layer, and a feature output layer. The input encoding layer includes a first convolutional layer and a downsampling layer, and the feature output layer includes a second convolutional layer and an upsampling layer.

[0045] The first convolutional layer is used to perform 1×1 convolution processing on the meteorological input data to obtain the initial input features;

[0046] The downsampling layer is used to downsample the initial input features to obtain the input features;

[0047] The feature extraction layer is used to perform multiple feature extraction processes on the input features to obtain initial meteorological features;

[0048] An upsampling layer is used to upsample the initial meteorological features to obtain upsampled features, and then add the upsampled features to the input features element by element to obtain the initial fused meteorological features;

[0049] The second convolutional layer is used to perform 1×1 convolution processing on the initial fused meteorological features to obtain meteorological convolutional features, and then add the meteorological convolutional features and the initial input features element by element to obtain the meteorological features.

[0050] In one possible design, the feature extraction layer includes: several depthwise separable convolutional modules connected in sequence, and any depthwise separable convolutional module includes: a first depthwise separable convolutional layer, a second depthwise separable convolutional layer, a first feature concatenation layer, a channel-wise convolutional layer, and a second feature concatenation layer;

[0051] Wherein, the first depthwise separable convolutional layer in any depthwise separable convolutional module is used to receive the first output feature of the previous depthwise separable convolutional module of any depthwise separable convolutional module, and perform 3×3 depthwise separable convolution processing on the first output feature to obtain the first depthwise convolutional feature;

[0052] The second depthwise separable convolutional layer in any depthwise separable convolutional module is used to perform 7×7 depthwise separable convolution processing on the first output features to obtain the second depthwise convolutional features.

[0053] The first feature concatenation layer is used to perform channel concatenation processing on the first depthwise convolutional feature and the second depthwise convolutional feature to obtain the first concatenated feature, and to perform layer normalization processing on the first concatenated feature to obtain the normalized feature.

[0054] Channel-wise convolutional layers are used to perform channel-wise convolution processing on normalized features to obtain channel-wise convolutional features.

[0055] The second feature splicing layer is used to perform element-wise addition of the channel-wise convolutional features and the first output features, so that after the element-wise addition, the second output feature is obtained and output to the next depthwise separable convolutional module. When any depthwise separable convolutional module is the last depthwise separable convolutional module in the feature extraction layer, the second output feature is the initial meteorological feature.

[0056] In one possible design, feature extraction and fusion processing are performed on the meteorological features and the electric field intensity data to obtain fused features, including:

[0057] The meteorological features and the electric field intensity data are input into a feature fusion model for feature extraction and fusion processing to obtain the fused features. The feature fusion model includes: a third convolutional layer, a fourth convolutional layer, a first feature fusion layer, a fifth convolutional layer, a sixth convolutional layer, a second feature fusion layer, a first global average pooling layer, a pointwise convolutional layer, a global max pooling layer, a second global average pooling layer, a third feature fusion layer, a dilated convolutional layer, and a fourth feature fusion layer.

[0058] The input of the third convolutional layer receives the meteorological features, and the output of the third convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer. The input of the fourth convolutional layer receives the electric field strength data, and the output of the fourth convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer.

[0059] The output of the first feature fusion layer is connected to the input of the fifth convolutional layer, the input of the sixth convolutional layer, the input of the global max pooling layer, and the input of the second global average pooling layer, respectively. The outputs of the fifth and sixth convolutional layers are connected to the input of the second feature fusion layer. The output of the second feature fusion layer is connected to the input of the pointwise convolutional layer through the first global average pooling layer, and the output of the pointwise convolutional layer is connected to the input of the fourth feature fusion layer.

[0060] The outputs of the global max pooling layer and the second global average pooling layer are electrically connected to the input of the third feature fusion layer. The output of the third feature fusion layer is connected to the input of the dilated convolutional layer, and the output of the dilated convolutional layer is connected to the input of the fourth feature fusion layer. The output of the fourth feature fusion layer outputs the fused feature.

[0061] Secondly, a navigation risk intelligent early warning system based on multi-source meteorological data fusion is provided, including:

[0062] The data acquisition unit is used to acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data of the target navigation area within a historical preset time period.

[0063] The data processing unit is used to construct an effective meteorological grid data for the target navigation area using multiple meteorological monitoring data.

[0064] The data processing unit is also used to perform data transformation processing on lightning event data to convert it into lightning grid data that matches the weather radar data;

[0065] The multi-source data fusion unit is used to generate meteorological input data using weather radar data, effective meteorological grid data, and lightning grid data, and to perform feature extraction processing on the meteorological input data to obtain meteorological features;

[0066] The multi-source data fusion unit is also used to perform feature extraction and fusion processing on the meteorological features and the electric field intensity data to obtain fused features;

[0067] The navigation risk warning unit is used to input the fused features into the lightning prediction model to obtain the lightning prediction result of the target navigation area, and generate a navigation risk warning notification for the target navigation area based on the lightning prediction result.

[0068] Thirdly, an electronic device is provided, comprising a memory, a processor, and a transceiver connected in sequence and communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the intelligent early warning method for navigation risks based on multi-source meteorological data fusion, as described in the first aspect or any possible design of the first aspect.

[0069] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the intelligent early warning method for navigation risks based on the fusion of multi-source meteorological data as described in the first aspect or any possible design of the first aspect.

[0070] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the intelligent early warning method for navigation risks based on multi-source meteorological data fusion, as described in the first aspect or any possible design of the first aspect.

[0071] Beneficial effects:

[0072] This invention collects multi-source heterogeneous data, including weather radar data, electric field intensity data, lightning event data, and meteorological monitoring data. Then, by fusing this multi-source heterogeneous data, a fusion feature is obtained. Finally, this feature is input into a lightning prediction model, enabling accurate lightning warnings for aircraft navigating in target flight areas. Therefore, compared to traditional technologies, this invention fuses multi-source heterogeneous lightning-related data and uses this fusion as the basis for flight lightning warnings, improving the accuracy of warnings and ensuring flight safety. Thus, this invention is highly suitable for large-scale application and promotion. Attached Figure Description

[0073] Figure 1 A flowchart illustrating the steps of the intelligent early warning method for navigation risks based on multi-source meteorological data fusion provided in this embodiment of the invention;

[0074] Figure 2 This is a network structure diagram of the meteorological feature extraction model provided in an embodiment of the present invention;

[0075] Figure 3 This is a network structure diagram of the feature fusion model provided in an embodiment of the present invention;

[0076] Figure 4 A schematic diagram of the structure of the intelligent early warning system for navigation risks based on multi-source meteorological data fusion provided in this embodiment of the invention;

[0077] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0078] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0079] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0080] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0081] Example:

[0082] See Figure 1 As shown in this embodiment, the intelligent early warning method for navigation risks based on multi-source meteorological data fusion collects multi-source heterogeneous data such as weather radar data, electric field strength data, lightning event data, and meteorological monitoring data. Then, by fusing the aforementioned multi-source heterogeneous data, the fusion characteristics of the multi-source heterogeneous data are obtained. Finally, this data is input into a lightning prediction model, which can accurately provide lightning warnings when the aircraft is navigating in the target navigation area. Therefore, compared with traditional technologies, this method fuses multi-source heterogeneous data related to lightning and uses this as a basis for flight lightning warnings, which can improve the accuracy of the warnings and thus ensure the flight safety of the aircraft. Therefore, this method is very suitable for large-scale application and promotion. For example, this method can be run on the navigation risk warning terminal side. Optionally, the navigation risk warning terminal can be, but is not limited to, a server. It is understood that the aforementioned execution entity does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S6 below.

[0083] S1. Acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data for the target navigation area within a historical preset time period. In this embodiment, the target navigation area may include, but is not limited to, airport areas (such as the area within 5 kilometers around the airport). The weather radar data is presented in grid data format, with each grid representing a sub-region within the target navigation area, containing reflectivity information for that sub-region (indicating the intensity of radar wave reflection by precipitation or other meteorological objects, which can be used to identify various weather phenomena, such as lightning and hail). Meanwhile, the lightning event data includes several lightning events that occurred in the target navigation area within a historical preset time period (such as the past 5 hours, 10 hours, or 12 hours), and each lightning event includes the lightning intensity and the latitude and longitude at the time of the lightning strike. Specifically, a lightning positioning system can be used to receive electromagnetic pulse signals generated by lightning, perform multi-station network positioning and signal processing, and directly output the location (longitude and latitude), intensity (current amplitude, etc.), and time information of each lightning strike.

[0084] Furthermore, electric field strength data can be obtained, but is not limited to, by using an atmospheric electric field meter, which can continuously detect the ground electric field value for a long time. An electric field monitoring network composed of multiple atmospheric electric field meters can measure changes in the atmospheric electric field over a large area. In particular, the electric field strength in the atmosphere changes drastically before lightning occurs. Therefore, the electric field strength data in the target navigation area can be used as an influencing indicator for lightning warning.

[0085] Furthermore, multiple meteorological monitoring data are obtained by meteorological monitoring sensors installed at different locations within the target navigation area. That is, one meteorological monitoring data corresponds to one meteorological monitoring sensor, and one meteorological monitoring data may include, but is not limited to, temperature, humidity, air pressure, wind direction and wind speed.

[0086] Thus, after collecting multi-source data related to lightning warnings in the target navigation area, data fusion can be performed to use the fused data for lightning prediction in the target navigation area, thereby achieving navigation risk warning for the corresponding area. In this embodiment, effective meteorological grid data and lightning grid data are first constructed, and then the two are fused with weather radar data to obtain meteorological features. Next, the meteorological features are fused with electric field intensity data to obtain fused features. Finally, lightning prediction in the target navigation area is achieved by using a pre-trained lightning prediction model.

[0087] Optionally, the process of constructing effective meteorological grid data is as shown in step S2 below.

[0088] S2. Using multiple meteorological monitoring data, construct effective meteorological grid data for the target navigation area; in specific implementation, this embodiment divides the target navigation area into grids; then, construct effective meteorological data for each grid cell; finally, use the effective meteorological data of each grid cell to generate effective meteorological grid data for the entire area; wherein, the aforementioned process may be, but is not limited to, as shown in steps S21 to S27 below.

[0089] S21. The target navigation area is processed into a grid to obtain the region grid corresponding to the target navigation area; in this embodiment, gridding is a common technique for discretization processing, and its principle will not be elaborated here.

[0090] After the target navigation area is gridded, the meteorological sensors in each grid cell can be identified based on their installation locations. This allows for the construction of effective meteorological data for each grid cell based on the meteorological monitoring data from the sensors within each cell. The meteorological monitoring data screening process is shown in step S22 below.

[0091] S22. For any grid cell in the regional grid, select the meteorological sensors located within that grid cell, and use the meteorological monitoring data corresponding to the selected meteorological sensors to form an initial meteorological dataset for that grid cell. In this embodiment, after forming the initial meteorological dataset using the meteorological monitoring data corresponding to the meteorological sensors located within that grid cell, effective data can be extracted so that only effective data can be used to construct effective meteorological grid data. The effective data extraction process can be, but is not limited to, the steps S23 to S25 below.

[0092] S23. For any meteorological monitoring data in the initial meteorological dataset, calculate the compatibility of each target data with the any meteorological monitoring data, and sum the compatibility values ​​to obtain the total compatibility, wherein each target data is any meteorological monitoring data in the initial meteorological dataset excluding the any meteorological monitoring data.

[0093] In practical applications, this embodiment determines whether the meteorological monitoring data transmitted by the current meteorological sensor is valid by calculating the compatibility between the meteorological monitoring data of the other meteorological sensors in the same grid cell and the meteorological monitoring data of the current meteorological sensor (which represents the consistency between the two data). The calculation process of compatibility can be, but is not limited to, the steps S23a to S23e below.

[0094] S23a. For any target data, calculate the data difference between the any meteorological monitoring data and the any target data; in this embodiment, it has been explained above that the any meteorological monitoring data includes temperature, humidity, wind direction and wind speed, etc.; then, for each type of meteorological data in the any meteorological monitoring data, the data difference is calculated. For example, for temperature, the difference between the temperature in the any meteorological monitoring data and the temperature in the any target data is calculated; of course, the difference process for other types of meteorological data is the same as the example above, and will not be repeated here.

[0095] Meanwhile, the aforementioned data difference can represent the distance between any meteorological monitoring data and any target data. That is, the smaller the difference, the closer the two are. Thus, after obtaining the data difference, the compatibility factor can be calculated, as shown in steps S23b and S23c below.

[0096] S23b. Obtain the data attenuation factor and compatibility control parameter; in specific implementation, the data attenuation factor is a positive number (e.g., 0.01), which is used to control the rate at which compatibility changes with distance; while the compatibility control parameter takes a value between (0,1], which is used to control the magnitude of compatibility; based on this, after obtaining the data attenuation factor and compatibility control parameter, the compatibility factor can be calculated by combining the aforementioned data difference, as shown in step S23c below.

[0097] S23c. Calculate the compatibility factor based on the data attenuation factor and the data difference. In this embodiment, the data difference is squared, and then the result of the squared operation is multiplied by the data attenuation factor to obtain the compatibility factor. Then, the initial compatibility can be calculated based on the compatibility factor, as shown in step S23d below.

[0098] S23d. Perform an exponential operation on the compatibility factor with base e, and take the reciprocal of the result as the initial compatibility. In this embodiment, the initial compatibility can be expressed as: In the formula, These respectively represent any meteorological monitoring data and any target data. This represents the data attenuation factor. Thus, after calculating the initial compatibility based on the aforementioned method, the compatibility of any target data with any meteorological monitoring data can be calculated by combining the compatibility control parameters, as shown in step S23e below.

[0099] S23e. Based on the initial compatibility and the compatibility control parameters, calculate the compatibility of any target data with any meteorological monitoring data; in specific implementation, the product between the initial compatibility and the compatibility control parameters is used as the compatibility of any target data with any meteorological monitoring data.

[0100] Thus, through the aforementioned steps S23a to S23e, this embodiment calculates the compatibility between two data points based on the principle of distance attenuation. That is, the closer the distance between two data points, the higher their compatibility. The exponential function... The effect is that compatibility decreases exponentially with increasing distance; that is, the farther the distance, the lower the compatibility.

[0101] Based on the aforementioned step S23 and its sub-steps, after calculating the compatibility of each target data with any meteorological monitoring data, the data can be summed to obtain the total compatibility. Then, the total compatibility can be used to determine whether any meteorological monitoring data is valid data, as shown in steps S24 and S25 below.

[0102] S24. Based on the overall compatibility, determine whether any meteorological monitoring data is valid data; in this embodiment, the overall compatibility is compared with the compatibility threshold. If the overall compatibility is greater than or equal to the compatibility threshold, the meteorological monitoring data can be determined to be valid and data extraction is required. The process is shown in step S25 below.

[0103] S25. If so, extract any meteorological monitoring data, and after polling all meteorological monitoring data in the initial meteorological dataset, obtain several valid data in the initial meteorological dataset; in specific implementation, if the total compatibility is less than the compatibility threshold, then determine that any meteorological monitoring data is invalid data and delete any meteorological monitoring data; in this way, after judging the validity of each remaining meteorological monitoring data in the initial meteorological dataset in the aforementioned manner, the valid data in any grid cell can be extracted; then, the extracted valid data can be used to generate the valid meteorological data of any grid cell, the process of which is shown in step S26 below.

[0104] S26. Based on several valid data, generate valid meteorological data for any grid cell, and obtain the valid meteorological data for each grid cell after polling all grid cells. In this embodiment, the weight of each valid data is first calculated based on the total compatibility of each valid data. Then, based on the weight of each valid data, the valid data is fused to obtain the valid meteorological data for any grid cell.

[0105] Specifically, for any valid data, the total compatibility of that valid data is divided by the sum of the total compatibility of several valid data to obtain the weight of that valid data. Then, based on the weights of each valid data, a weighted sum is performed on each valid data to obtain the valid meteorological data for that grid cell. Of course, the valid meteorological data for that grid cell includes valid temperature, valid humidity, valid wind speed, etc.

[0106] Thus, based on the same method described above, valid meteorological data for each of the remaining grid cells can be generated; finally, valid meteorological grid data can be constructed based on the valid meteorological data of each grid cell, as shown in step S27 below.

[0107] S27. Construct the effective meteorological grid data using the effective meteorological data of each grid cell; in this embodiment, the effective meteorological grid data mainly includes temperature grid data, humidity grid data, wind speed grid data, etc.; among which, the temperature grid data contains the effective temperature of the corresponding sub-region of each grid cell, and the data contained in the other grid data will not be described in detail; thus, the type of the constructed effective meteorological grid data is the same as that of weather radar data, both of which use grids as the basic unit to record the corresponding data.

[0108] Therefore, after obtaining effective meteorological grid data through the aforementioned steps S21 to S27, lightning grid data can be constructed, as shown in step S3 below.

[0109] S3. Perform data transformation processing on the lightning event data to convert it into lightning grid data that matches the weather radar data. In specific applications, the lightning event data and weather radar data differ in data type (lightning data usually exists in the form of discrete point data, while weather radar data exists in the form of grid data), resolution (lightning has high spatial resolution, which can accurately locate lightning events, while radar data usually has low spatial resolution, that is, one grid represents a sub-region), and sparsity (radar data is relatively dense, and each grid contains reflectivity information, while lightning data is usually relatively sparse, and there may only be a few lightning events in a region). Therefore, if the two are directly combined, the model will not be able to effectively learn the features in the data. Therefore, this embodiment needs to convert the lightning event data into lightning grid data with the same data type as the weather radar data, and the process is shown in steps S31 to S34 below.

[0110] S31. The target navigation area is gridded to obtain the corresponding area grid. After the gridding of the target navigation area is completed, the lightning events that occur in each grid cell can be determined. The process is shown in step S32 below.

[0111] S32. For any grid cell in the regional grid, lightning data whose occurrence location is within any grid cell is selected from the lightning event data, so as to form a lightning dataset for any grid cell using the selected lightning data. In this embodiment, as previously explained, the lightning data packet contains the latitude and longitude of the lightning occurrence. Therefore, the location of the lightning occurrence can be determined based on the latitude and longitude. Then, the grid cell to which it belongs is determined based on the location of the lightning occurrence. Thus, based on the aforementioned method, lightning data within any grid cell can be selected from the lightning event data to form a lightning dataset. Then, the probability of lightning occurrence in any grid cell can be calculated using the lightning dataset, as shown in step S33 below.

[0112] S33. Using the lightning dataset, calculate the lightning occurrence probability corresponding to any grid cell, and after polling all grid cells in the region grid, obtain the lightning occurrence probability corresponding to each grid cell; in specific applications, this embodiment provides a method for cumulatively calculating the lightning occurrence probability by using the previous lightning occurrence sub-probability of the grid cell and combining the lightning occurrence location with the grid distance and the lightning influence radius, as shown in the following steps S33a to S33e.

[0113] S33a. Obtain the grid index information and lightning impact radius of any grid cell; in this embodiment, the grid index information includes the index row number and index column number of any grid cell; and the lightning impact radius can be specifically set according to actual use, and this embodiment does not make specific limitations.

[0114] Thus, after obtaining the lightning impact radius and the grid index information of any grid cell, the distance between the lightning occurrence location and the grid center can be calculated, as shown in step S33b below.

[0115] S33b. For the b-th lightning data in the lightning dataset, based on the latitude and longitude of the b-th lightning data and the grid index information, calculate the distance between the location of the b-th lightning data and the grid center of any grid cell; in specific applications, it is possible, but not limited to, first determining the first grid index value corresponding to the longitude of the b-th lightning data, and determining the second grid index value corresponding to the latitude of the b-th lightning data; then, using the index row number and index column number in the grid index information, as well as the first grid index value and the second grid index value, calculate the aforementioned distance.

[0116] Furthermore, the following discloses one method for calculating the first grid index value:

[0117] First, determine the longitude range of the target navigation area and the longitude resolution of each grid cell in the area grid; then, filter out the minimum longitude from the longitude range; next, calculate the difference between the longitude of the b-th lightning data and the minimum longitude; finally, divide the difference by the longitude resolution to obtain the first grid index value; of course, the calculation process of the second grid index value is the same, and will not be repeated here.

[0118] Furthermore, for example, but not limited to, the following formula can be used to calculate the aforementioned interval distance.

[0119] ;

[0120] In the formula, This indicates the interval distance. These represent the row number and column number of any given grid cell, respectively. These represent the first grid index value and the second grid index value, respectively.

[0121] Thus, after calculating the distance between the location of the b-th lightning data and the center of any grid cell based on the aforementioned formula, the lightning occurrence sub-probability after the b-th lightning strike can be calculated for any grid cell, as shown in steps S33c and S33d below.

[0122] S33c. Obtain the lightning occurrence sub-probability of any grid cell after the (b-1)th lightning strike; In this embodiment, when b is 1, the lightning occurrence sub-probability of any grid cell after the (b-1)th lightning strike is 0, that is, when there is no lightning, its lightning occurrence sub-probability is 0; Thus, after obtaining the lightning occurrence sub-probability of any grid cell after the (b-1)th lightning strike, the lightning occurrence sub-probability after the bth lightning strike can be calculated by combining the aforementioned interval distance and lightning influence radius, as shown in step S33d below.

[0123] S33d. Using the interval distance, the lightning influence radius, the lightning intensity corresponding to the b-th lightning data, and the lightning occurrence sub-probability of any grid cell after the (b-1)-th lightning strike, calculate the lightning occurrence sub-probability of any grid cell after the b-th lightning strike.

[0124] Optionally, for example but not limited to, the following formula can be used to calculate the lightning occurrence probability after the b-th lightning strike for any grid cell.

[0125] ;

[0126] In the formula, This represents the sub-probability of lightning occurrence after the b-th lightning strike in any given grid cell. This indicates the interval distance. Indicates the radius of lightning's influence. This represents the lightning intensity corresponding to the b-th lightning strike. This represents the sub-probability of lightning occurrence after the (b-1)th lightning strike in any given grid cell. Both represent the lightning threat level (values ​​are 4 and 1).

[0127] Thus, as can be seen from the aforementioned formula, when the interval distance is less than or equal to the lightning influence radius, the lightning contributes to the lightning occurrence sub-probability of any grid cell, and the influence of all previous lightning strikes needs to be accumulated by using the lightning occurrence sub-probability of the previous strike; when the interval distance is greater than the lightning influence radius, the lightning has no effect on any grid cell, and the probability remains unchanged; and when there is no lightning in any grid cell, the lightning occurrence sub-probability is 0.

[0128] Based on this, after calculating the lightning occurrence sub-probability of any grid cell after the b-th lightning strike, the lightning occurrence sub-probability of any grid cell can be accumulated in the same way as described above, as shown in step S33e below.

[0129] S33e. Increment b by 1, and recalculate the distance between the location of the b-th lightning data and the center of the grid based on the latitude and longitude of the b-th lightning data, until b equals B, to obtain the lightning occurrence probability corresponding to any grid unit, where the initial value of b is 1, and B is the total number of data in the lightning dataset; in this embodiment, when b is polled to B, the occurrence sub-probabilities of all lightning in any grid unit are accumulated. At this time, the lightning occurrence sub-probability of any grid unit after the B-th lightning can be used as the lightning occurrence probability of the entire grid unit.

[0130] Through the aforementioned steps S33a to S33e, the probability of lightning occurrence corresponding to each grid cell can be calculated; then, lightning grid data can be constructed based on this, as shown in step S34 below.

[0131] S34. Construct the lightning grid data using the lightning occurrence probability corresponding to each grid cell.

[0132] After converting the lightning event data into lightning grid data of the same type as the weather radar data based on the aforementioned steps S31 to S34, the fusion and feature extraction of meteorological related data can be performed, as shown in step S4 below.

[0133] S4. Using weather radar data, effective meteorological grid data, and lightning grid data, meteorological input data is generated, and feature extraction processing is performed on the meteorological input data to obtain meteorological features. In specific applications, this embodiment concatenates the aforementioned weather radar data, effective meteorological grid data, and lightning grid data to obtain meteorological input data. At the same time, this embodiment constructs a meteorological feature extraction model example. Thus, by inputting the meteorological input data into the meteorological feature extraction model for feature extraction processing, the meteorological features can be obtained.

[0134] Furthermore, the following is one of the network structures of the publicly disclosed meteorological feature extraction model:

[0135] See Figure 2 As shown, the meteorological feature extraction model described in the example may include, but is not limited to, an input encoding layer, a feature extraction layer, and a feature output layer, wherein the input encoding layer includes a first convolutional layer and a downsampling layer, and the feature output layer includes a second convolutional layer and an upsampling layer.

[0136] In specific implementation, the first convolutional layer is used to perform 1×1 convolution processing on the meteorological input data to obtain initial input features, which are then output to the downsampling layer. The downsampling layer is used to downsample the initial input features to obtain input features. In this embodiment, lightning prediction can essentially be regarded as a time series prediction problem. Therefore, this embodiment uses 1×1 convolution and downsampling to extract the spatiotemporal features of the meteorological input data, namely: extracting the trend of radar reflectivity over time, such as the speed of reflectivity movement and intensity changes, which can reflect the dynamic changes in weather; extracting the spatial distribution features of meteorological elements such as temperature, humidity, and wind speed, which can reflect the state of the atmosphere and weather changes; extracting the trend of meteorological elements over time, such as the increase or decrease in temperature and the change in humidity, which can reflect the dynamic changes in the atmosphere; similarly, extracting the trend of lightning occurrence probability over time, such as the increase or decrease in lightning occurrence probability and the temporal distribution of lightning activity, which can reflect the dynamic changes in lightning weather activity.

[0137] Thus, after completing the input encoding and obtaining the spatiotemporal features of the input meteorological data, a feature extraction layer can be used for in-depth feature extraction. The process is as follows: the feature extraction layer is used to perform multiple feature extraction processes on the input features to obtain the initial meteorological features. In this embodiment, the feature extraction layer mainly uses depthwise separable convolution and channel convolution to extract spatiotemporal change information and features with higher semantic meaning.

[0138] This embodiment discloses one detailed construction of the feature extraction layer:

[0139] See Figure 2As shown, in specific applications, the feature extraction layer may include, but is not limited to, several sequentially connected depthwise separable convolutional modules, wherein any depthwise separable convolutional module includes: a first depthwise separable convolutional layer, a second depthwise separable convolutional layer, a first feature concatenation layer, a channel-wise convolutional layer, and a second feature concatenation layer.

[0140] Specifically, the first depthwise separable convolutional layer in any depthwise separable convolutional module is used to receive the first output feature of the previous depthwise separable convolutional module (when the depthwise separable convolutional module is the first depthwise classifiable convolutional module, it receives the input feature provided by the input encoding layer), and performs 3×3 depthwise separable convolution processing on the first output feature to obtain the first depthwise convolutional feature.

[0141] Simultaneously, the second depthwise separable convolutional layer in any depthwise separable convolutional module is used to perform 7×7 depthwise separable convolution processing on the first output feature to obtain the second depthwise convolutional feature; then, the first feature concatenation layer is used to perform channel concatenation processing on the first depthwise convolutional feature and the second depthwise convolutional feature to obtain the first concatenated feature, and the first concatenated feature is subjected to layer normalization processing to obtain the normalized feature; next, the channel-wise convolutional layer is used to perform channel-wise convolution processing on the normalized feature to obtain the channel-wise convolutional feature; finally, the second feature concatenation layer is used to perform element-wise addition processing on the channel-wise convolutional feature and the first output feature, so that after element-wise addition processing, the second output feature is obtained and output to the next depthwise separable convolutional module; of course, when any depthwise separable convolutional module is the last depthwise separable convolutional module in the feature extraction layer, the second output feature is the initial meteorological feature.

[0142] Thus, based on the aforementioned explanation of the depthwise separable convolution module, its working process is as follows:

[0143] In a depthwise separable convolution module, firstly, depthwise separable convolutions of different sizes are used to capture feature variations. Then, channel concatenation is performed to complete a feature fusion (i.e., all original features are preserved through channel concatenation, and features from different sources are integrated by increasing the number of channels, thus forming a richer representation). After fusion, layer normalization is used to normalize the features. Next, channel convolution is used to extract spatiotemporal variation information and higher semantic features corresponding to the spatial movement and changes in meteorological data, thereby achieving richer feature extraction. Finally, by establishing residual connections between the input and output, the model can more easily learn low-frequency and high-frequency meteorological features, thereby improving the model's ability to predict lightning.

[0144] Based on this, through the continuous convolution processing of the aforementioned multiple depthwise separable convolution modules, multi-level feature extraction can be achieved, thereby obtaining initial meteorological features. Finally, these features can be output to the feature output layer to output the final meteorological features.

[0145] In specific implementation, the upsampling layer is used to upsample the initial meteorological features to obtain upsampled features, and then the upsampled features are added element-wise to the input features to obtain the initial fused meteorological features; finally, the second convolutional layer is used to perform 1×1 convolution on the initial fused meteorological features to obtain meteorological convolutional features, and then the meteorological convolutional features are added element-wise to the initial input features to obtain the meteorological features.

[0146] In practical applications, the feature output layer uses the rich meteorological spatiotemporal features (i.e., initial meteorological features) obtained above to reconstruct the final meteorological features. At the same time, this embodiment also adds the features of the same dimension in the feature output layer and the input coding layer (i.e., the first convolutional layer and the second convolutional layer reuse features, and the downsampling layer and the upsampling layer reuse features), thereby realizing feature reuse of multi-source meteorological data, which can further improve the accuracy of feature extraction.

[0147] Therefore, this embodiment constructs a neural network model for multi-source meteorological data, which integrates feature reuse, spatiotemporal feature extraction, and extraction of spatiotemporal change information and higher semantic features corresponding to the movement and changes of meteorological data over time; thus, feature fusion of multi-source meteorological data can be better achieved.

[0148] After obtaining the meteorological characteristics, they can be fused with the electric field strength data for data fusion and feature extraction. The process can be, but is not limited to, the steps shown in step S5 below.

[0149] S5. Perform feature extraction and fusion processing on the meteorological features and the electric field strength data to obtain fused features. In specific applications, this embodiment pre-constructs a feature fusion model to realize the data fusion and feature extraction of meteorological features and electric field strength data. That is, input the meteorological features and the electric field strength data into the feature fusion model for feature extraction and fusion processing to obtain the fused features.

[0150] Furthermore, one of the network structures of the following publicly disclosed feature fusion model is as follows:

[0151] See Figure 3As shown, the example feature fusion model may include, but is not limited to, the following: third convolutional layer, fourth convolutional layer, first feature fusion layer, fifth convolutional layer, sixth convolutional layer, second feature fusion layer, first global average pooling layer, pointwise convolutional layer, global max pooling layer, second global average pooling layer, third feature fusion layer, dilated convolutional layer, and fourth feature fusion layer.

[0152] The connection structure of each of the aforementioned network layers is as follows:

[0153] The input of the third convolutional layer receives the meteorological features, and the output of the third convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer. The input of the fourth convolutional layer receives the electric field intensity data, and the output of the fourth convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer.

[0154] Specifically, the third convolutional layer is used to perform 1×1 convolution processing on the meteorological features to obtain the processed meteorological features, while the fourth convolutional layer is used to perform 1×1 convolution processing on the electric field strength data to obtain the electric field strength features. In this embodiment, two 1×1 convolutions are used to convert the meteorological features and electric field strength data into features of the same dimension, that is, the electric field strength features and the processed meteorological features have the same dimension, so as to facilitate subsequent feature fusion. That is, the first feature fusion layer is used to perform element-wise addition processing on the electric field strength features and the processed meteorological features to obtain the initial fused features. In this way, features from different layers are merged to retain the feature information at each position.

[0155] Then, the first feature fusion layer inputs the initial fused features to the fifth convolutional layer, the sixth convolutional layer, the global max pooling layer, and the second global average pooling layer, respectively. That is, the output of the first feature fusion layer is connected to the input of the fifth convolutional layer, the sixth convolutional layer, the global max pooling layer, and the second global average pooling layer, respectively.

[0156] Furthermore, the fifth convolutional layer performs a first convolutional process on the initial fused features using a 3×3 convolution with a stride of 1 to obtain a first convolutional fused feature; while the sixth convolutional layer performs a second convolutional process on the initial fused features using a 3×3 convolution with a stride of 3 to obtain a second convolutional fused feature; then, the outputs of the fifth and sixth convolutional layers are connected to the input of the second feature fusion layer, and the first and second convolutional fused features are concatenated through the second feature fusion layer to obtain a first channel fused feature; wherein, the output of the second feature fusion layer is connected to the input of the pointwise convolutional layer through a first global average pooling layer, and the output of the pointwise convolutional layer is connected to the input of the fourth feature fusion layer; thus, the first channel fused feature can be globally average pooled using the first global average pooling layer to obtain a first global average pooled feature, and the first global average pooled feature can be further convolved through the pointwise convolutional layer to obtain a pointwise convolutional feature.

[0157] Thus, feature extraction and fusion of one branch can be completed.

[0158] Then, feature extraction and fusion can be performed in another branch, the process of which is as follows:

[0159] First, a second global average pooling layer is used to perform global average pooling on the initial fused features to obtain second global average pooled features. Then, a global max pooling layer is used to perform global max pooling on the initial fused features to obtain global max pooled features. After global average pooling and global max pooling, the features can be output to the third feature fusion layer for feature fusion. That is, the outputs of the global max pooling layer and the second global average pooling layer are electrically connected to the input of the third feature fusion layer. The output of the third feature fusion layer is connected to the input of the dilated convolutional layer, and the output of the dilated convolutional layer is connected to the input of the fourth feature fusion layer.

[0160] In this embodiment, the third feature fusion layer is used to perform channel concatenation processing on the second global average pooling feature and the global max pooling feature to obtain the second channel fusion feature; while the dilated convolution layer is used to perform dilated convolution processing on the second channel fusion feature with two different dilation factors (such as using dilation factors 2 and 5 to perform dilated convolution on the second channel fusion feature respectively) to obtain the first dilated convolution feature and the second dilated convolution feature respectively, and then perform feature fusion processing on the first dilated convolution feature and the second dilated convolution feature to obtain the dilated convolution fusion feature; in this way, feature extraction can be performed on different receptive fields, thereby realizing the fusion of features on different receptive fields.

[0161] Finally, the fourth feature fusion layer is first used to perform element-wise multiplication of the processed meteorological features and the point-by-point convolutional features to obtain the second initial fusion feature (which highlights certain regions and suppresses others, assigning different weights to features at different locations, so that the network can focus more on regions with high weights), and then to perform element-wise multiplication of the electric field strength features and the dilated convolutional fusion features to obtain the third initial fusion feature; at the same time, it is also used to perform feature fusion (which can be set to element-wise addition) on the second initial fusion feature and the third initial fusion feature to obtain the fusion feature, that is, the output of the fourth feature fusion layer outputs the fusion feature.

[0162] Thus, the detailed structure and working principle of the aforementioned feature fusion model are as follows:

[0163] First, 1×1 convolution is used to convert meteorological features and electric field intensity data into features of the same dimension. Then, the two features of the same dimension are added element by element to complete the initial feature fusion and obtain the initial fused features. Then, the initial fused features are processed through two feature extraction and fusion branches to achieve full feature extraction and information interaction between multi-source data.

[0164] Specifically, the first feature extraction and fusion branch is as follows: the input features are processed through two convolutions with different strides to capture meteorological and electric field features at different scales. Then, global average pooling is used to spatially compress the fused features to reduce the dependence on location information. Finally, pointwise convolution is used to learn channel features to reduce redundant parameters and improve the model's generalization ability.

[0165] The second feature extraction and fusion branch involves performing global average pooling and max pooling on the input features to preserve background information of the feature map and highlight features that influence lightning. Simultaneously, due to the large amount of information, this embodiment employs dilated convolutions with different dilation factors to extract meteorological and electric field features across different receptive fields. Finally, the features from different receptive fields are fused to output to the final feature fusion layer. Thus, through different dilated convolutions, the model enhances the feature response of the region of interest, better learning the changing trends of lightning.

[0166] Finally, the outputs of the two branches are multiplied element-wise with their respective inputs (which are the same), and then the results are added element-wise to output the final fused feature, which is a fused feature containing features from multiple sources such as weather radar data, electric field strength data, lightning event data, and meteorological monitoring data.

[0167] Thus, after the fusion of multi-source data is completed, the obtained fusion features can be used to conduct lightning warnings for the target navigation area, as shown in step S6 below.

[0168] S6. Input the fused features into the lightning prediction model to obtain the lightning prediction results for the target navigation area, and generate a navigation risk warning notification for the target navigation area based on the lightning prediction results.

[0169] In practical implementation, for example, but not limited to, the XGBoost model can be trained by first using the sample fusion features of several sample regions as input and the lightning prediction results (lightning activity frequency and duration) of each sample region within a preset future time period as output. After training, a lightning prediction model is obtained. After obtaining the lightning prediction model, the fusion features obtained in step S5 can be input into the lightning prediction model to obtain the lightning activity frequency and duration of the target navigation area. Then, based on the lightning activity frequency and duration, a lightning warning level is matched in a lightning warning table (this table records the range of lightning activity frequency and duration corresponding to different warning levels). Finally, based on the matched lightning warning level, a navigation risk warning notification for the target navigation area is generated and sent to the dispatch control center and the aircraft, thereby realizing the navigation risk warning for the aircraft in the target navigation area.

[0170] Therefore, through the multi-source meteorological data fusion intelligent early warning method for flight risks described in detail in steps S1 to S6 above, this invention fuses multi-source heterogeneous data related to lightning and uses this as a basis for flight lightning early warning; thus, compared with the traditional method of using single monitoring data for lightning early warning, this invention can improve the accuracy of early warning and ensure the flight safety of aircraft; therefore, it is very suitable for large-scale application and promotion.

[0171] In one possible design, the second aspect of this embodiment provides one method for determining the model parameters of the lightning prediction model in the first aspect of the embodiment:

[0172] Step 1: Initialize the hyperparameters of the XGBoost model to obtain multiple sets of initial model parameters. Use these multiple sets of initial model parameters to form an initial population. Each initial individual in the initial population corresponds to a set of initial model parameters. In this embodiment, example hyperparameters may include, but are not limited to, n_estimators (the number of weak learners (decision trees) to be built in the XGBoost model, which determines the number of boosting iterations and is an important parameter for model complexity and training time), learning rate, max_depth (representing the maximum depth of each tree), subsample (used to control the sampling ratio of the training set), and colsample_bytree (used to control the proportion of features used when training each tree out of all features), etc.

[0173] Step 2: Obtain the individual population at the t-th iteration, where the initial value of t is 1, and when t is 1, the individual population at the t-th iteration is the initial population.

[0174] Step 3: Construct the XGBoost model corresponding to each individual in the population at the t-th iteration, and train each XGBoost model based on the sample fusion features of several sample regions. After training, calculate the fitness of each individual based on the loss function of each XGBoost model. The regression loss commonly used in XGBoost models can be used as its loss function, and the reciprocal of the loss function can be used as the fitness function.

[0175] Step 4: Based on the fitness of each individual, determine the globally optimal individual at the t-th iteration. In this embodiment, first select the largest fitness at the t-th iteration; then, determine whether the largest fitness is greater than the fitness of the globally optimal individual at the (t-1)-th iteration; if so, take the individual corresponding to the largest fitness as the globally optimal individual at the t-th iteration; otherwise, take the globally optimal individual at the (t-1)-th iteration as the globally optimal individual at the t-th iteration.

[0176] Step 5: Determine whether the iteration stopping condition is met; In this embodiment, the iteration stopping condition is that the fitness of the globally optimal individual at the t-th iteration is greater than or equal to the fitness threshold, or t reaches the maximum number of iterations.

[0177] Step 6: If not, sort the individual population at the t-th iteration in descending order of fitness to obtain a sorting sequence, and divide the sorting sequence into a first subpopulation, a second subpopulation, a third subpopulation, and a fourth subpopulation according to a preset ratio, wherein the four subpopulations are of the same size; in this embodiment, the ratio of the aforementioned four subpopulations can be specifically set according to actual use, and this embodiment does not make a specific limitation.

[0178] Step 7: Perform local searches on the first, second, third, and fourth subpopulations respectively to obtain the four updated subpopulations, and use the four updated subpopulations to form the individual population at the (t+1)th iteration.

[0179] In this embodiment, for any first individual in the first subpopulation, a local search is performed using the following formula.

[0180] ;

[0181] In the formula, This refers to the first individual in the first subpopulation. This refers to the first individual after the update. Let any one of the first individuals be represented in the (t-1)th iteration. It represents -1 or 1 (it can generate a random number with a probability; if the probability is less than or equal to 0.5, the value is -1, otherwise the value is 1). This represents the first local search coefficient (with values ​​of (0, 0.2)). Represents a constant between (0,1). This represents the individual with the smallest fitness at the t-th iteration. Let C represent the globally optimal individual at the t-th iteration, and let C represent the second local search coefficient. T represents the maximum iteration coefficient. Represents a random number between [-1, 1]. Represents the first random number between (0,1). It is a second random number between (0.5, 1].

[0182] Optionally, for any second individual in the second subpopulation, a local search can be performed using the following formula.

[0183] ;

[0184] In the formula, This represents the historical best individual corresponding to any second individual. Indicates any second individual. This represents the updated second individual. Represents the mutation operator, , Let represent random numbers that follow a Cauchy distribution and random numbers that follow a Gaussian distribution, respectively. and Both represent random vectors, and their dimensions are the same as those of any second individual. These represent the upper and lower bounds of the local search for the second individual, respectively; specifically, , , This represents the upper and lower bounds of the hyperparameter search (i.e., the upper and lower bounds of the values ​​of each hyperparameter, which are essentially a vector). The aforementioned `max()` function compares the two vectors element-wise, taking the maximum value among the elements at the same position in both vectors. This ultimately generates the upper bound of the local search for the second individual, i.e. It is essentially a vector, of course. The same applies, and will not be elaborated further here; in addition, G represents the search factor, and G=1-t / T.

[0185] Similarly, for any third individual in the third subpopulation, a local search is performed using the following formula.

[0186] ;

[0187] In the formula, This refers to any third individual after the update. This refers to any third individual. Let V represent the probability density function of the T-distribution, and let the iteration number t be the free quantity V in the probability density function of the T-distribution to calculate the probability density function value of the T-distribution; of course, the T-distribution is also called the Student's t-distribution, which is a commonly used probability distribution, and its calculation formula will not be repeated here.

[0188] Finally, for any fourth individual in the fourth subpopulation, the following formula is used for local search.

[0189] ;

[0190] In the formula, This represents any fourth individual after the update. This refers to the fourth individual. Represents a constant. Represents a random vector that follows a normal distribution (its dimension is...). same), This represents the historical best individual corresponding to any fourth individual.

[0191] Through the aforementioned design, this embodiment divides the population into different subpopulations and uses different local search methods to update the positions. This fully ensures population diversity and prevents the population from getting trapped in local optima. Based on this, the accuracy of the algorithm's search can be guaranteed.

[0192] Therefore, by using the aforementioned formula to update the positions of the four subpopulations, we can obtain the individual population at the (t+1)th iteration. Then, we can repeat the aforementioned steps until the iteration stopping condition is met, and then obtain the optimal model parameters of the XGBoost model, that is, the optimal hyperparameters. The process is as follows.

[0193] Step 8: Increment t by 1 and reacquire the individual population at the t-th iteration until the iteration stopping condition is met. Based on the globally optimal individual at the iteration stopping condition, determine the optimal model parameters and construct the lightning prediction model based on the optimal model parameters.

[0194] Therefore, based on the aforementioned method, the optimal model parameters can be obtained, and a lightning prediction model can be constructed based on these optimal model parameters, so as to perform lightning prediction for the target navigation area.

[0195] like Figure 4As shown, the third aspect of this embodiment provides a hardware system for implementing the intelligent early warning method for navigation risks based on multi-source meteorological data fusion as described in the first aspect of the embodiment, comprising:

[0196] The data acquisition unit is used to acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data for the target navigation area within a historical preset time period.

[0197] The data processing unit is used to construct an effective meteorological grid data for the target navigation area using multiple meteorological monitoring data.

[0198] The data processing unit is also used to perform data transformation processing on lightning event data to convert it into lightning grid data that matches the weather radar data.

[0199] The multi-source data fusion unit is used to generate meteorological input data using weather radar data, effective meteorological grid data, and lightning grid data, and to perform feature extraction processing on the meteorological input data to obtain meteorological features.

[0200] The multi-source data fusion unit is also used to perform feature extraction and fusion processing on the meteorological features and the electric field intensity data to obtain fused features.

[0201] The navigation risk warning unit is used to input the fused features into the lightning prediction model to obtain the lightning prediction result of the target navigation area, and generate a navigation risk warning notification for the target navigation area based on the lightning prediction result.

[0202] The working process, working details and technical effects of the system provided in this embodiment can be found in the first and second aspects of the embodiment, and will not be repeated here.

[0203] like Figure 5 As shown, the fourth aspect of this embodiment provides an electronic device, including: a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the intelligent early warning method for navigation risks based on multi-source meteorological data fusion as described in the first and second aspects of the embodiments.

[0204] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0205] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0206] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first and second aspects of the embodiment, and will not be repeated here.

[0207] The fifth aspect of this embodiment provides a storage medium for storing instructions containing the navigation risk intelligent early warning method for multi-source meteorological data fusion as described in the first and second aspects of the embodiments. That is, the storage medium stores instructions, and when the instructions are run on a computer, the navigation risk intelligent early warning method for multi-source meteorological data fusion as described in the first and second aspects of the embodiments is executed.

[0208] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0209] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first and second aspects of the embodiment, and will not be repeated here.

[0210] The sixth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the intelligent early warning method for navigation risks based on multi-source meteorological data fusion as described in the first and second aspects of the embodiments. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0211] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of navigation risks through multi-source meteorological data fusion, characterized in that, include: Acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data for the target navigation area within a historical preset time period; An effective meteorological grid data for the target navigation area was constructed using multiple meteorological monitoring data. The lightning event data is transformed into lightning grid data that matches the weather radar data. Weather input data is generated using weather radar data, effective meteorological grid data, and lightning grid data. Feature extraction processing is then performed on the meteorological input data to obtain meteorological features. The meteorological features and the electric field intensity data are subjected to feature extraction and fusion processing to obtain fused features; The fused features are input into the lightning prediction model to obtain the lightning prediction results for the target navigation area, and a navigation risk warning notice for the target navigation area is generated based on the lightning prediction results. Lightning event data includes several lightning data points. These lightning event data undergo data transformation processing to convert them into lightning grid data that matches weather radar data. This includes: The target navigation area is gridded to obtain the corresponding area grid. For any grid cell in the regional grid, lightning data whose occurrence location is within any grid cell is selected from the lightning event data, so as to form a lightning dataset for any grid cell using the selected lightning data; Using the lightning dataset, the probability of lightning occurrence corresponding to any grid cell is calculated, and after polling all grid cells in the region grid, the probability of lightning occurrence corresponding to each grid cell is obtained. The lightning grid data is constructed using the lightning occurrence probability corresponding to each grid cell; Any lightning data set includes lightning intensity and the latitude and longitude at the time of the lightning strike. Using the lightning dataset, the probability of lightning occurrence corresponding to any given grid cell is calculated, including: Obtain the grid index information and lightning impact radius of any of the grid cells; For the b-th lightning data in the lightning dataset, based on the latitude and longitude of the b-th lightning data and the grid index information, the distance between the occurrence location of the b-th lightning data and the grid center of any grid cell is calculated; Obtain the lightning occurrence sub-probability after any grid cell has undergone the (b-1)th lightning strike; Using the interval distance, the lightning influence radius, the lightning intensity corresponding to the b-th lightning data, and the lightning occurrence probability of any grid cell after the (b-1)-th lightning strike, the lightning occurrence probability of any grid cell after the b-th lightning strike is calculated. The following formula is used to calculate the lightning occurrence probability of any grid cell after the b-th lightning strike; ； In the formula, This represents the sub-probability of lightning occurrence after the b-th lightning strike in any given grid cell. This indicates the interval distance. Indicates the radius of lightning's influence. This represents the lightning intensity corresponding to the b-th lightning strike. This represents the sub-probability of lightning occurrence after the (b-1)th lightning strike in any given grid cell. All represent the lightning threat level; Increment b by 1, and recalculate the distance between the location of the b-th lightning data and the center of the grid based on the latitude and longitude of the b-th lightning data, until b equals B, to obtain the lightning occurrence probability corresponding to any grid cell, where the initial value of b is 1, and B is the total number of data in the lightning dataset; The meteorological input data is subjected to feature extraction processing to obtain meteorological features, including: The meteorological input data is input into the meteorological feature extraction model for feature extraction processing to obtain the meteorological features; The meteorological feature extraction model includes an input encoding layer, a feature extraction layer, and a feature output layer. The input encoding layer includes a first convolutional layer and a downsampling layer, and the feature output layer includes a second convolutional layer and an upsampling layer. The first convolutional layer is used to perform 1×1 convolution processing on the meteorological input data to obtain the initial input features; The downsampling layer is used to downsample the initial input features to obtain the input features; The feature extraction layer is used to perform multiple feature extraction processes on the input features to obtain initial meteorological features; An upsampling layer is used to upsample the initial meteorological features to obtain upsampled features, and then add the upsampled features to the input features element by element to obtain the initial fused meteorological features; The second convolutional layer is used to perform 1×1 convolution processing on the initial fused meteorological features to obtain meteorological convolutional features, and then add the meteorological convolutional features and the initial input features element by element to obtain the meteorological features. The feature extraction layer includes: several depthwise separable convolutional modules connected in sequence, and any depthwise separable convolutional module includes: a first depthwise separable convolutional layer, a second depthwise separable convolutional layer, a first feature concatenation layer, a channel-wise convolutional layer, and a second feature concatenation layer; Wherein, the first depthwise separable convolutional layer in any depthwise separable convolutional module is used to receive the first output feature of the previous depthwise separable convolutional module of any depthwise separable convolutional module, and perform 3×3 depthwise separable convolution processing on the first output feature to obtain the first depthwise convolutional feature; The second depthwise separable convolutional layer in any depthwise separable convolutional module is used to perform 7×7 depthwise separable convolution processing on the first output features to obtain the second depthwise convolutional features. The first feature concatenation layer is used to perform channel concatenation processing on the first depthwise convolutional feature and the second depthwise convolutional feature to obtain the first concatenated feature, and to perform layer normalization processing on the first concatenated feature to obtain the normalized feature. Channel-wise convolutional layers are used to perform channel-wise convolution processing on normalized features to obtain channel-wise convolutional features. The second feature splicing layer is used to perform element-wise addition of the channel-wise convolutional features and the first output features, so that after the element-wise addition, the second output feature is obtained and output to the next depthwise separable convolutional module. When any depthwise separable convolutional module is the last depthwise separable convolutional module in the feature extraction layer, the second output feature is the initial meteorological feature. The meteorological features and the electric field intensity data are subjected to feature extraction and fusion processing to obtain fused features, including: The meteorological features and the electric field intensity data are input into a feature fusion model for feature extraction and fusion processing to obtain the fused features. The feature fusion model includes: a third convolutional layer, a fourth convolutional layer, a first feature fusion layer, a fifth convolutional layer, a sixth convolutional layer, a second feature fusion layer, a first global average pooling layer, a pointwise convolutional layer, a global max pooling layer, a second global average pooling layer, a third feature fusion layer, a dilated convolutional layer, and a fourth feature fusion layer. The input of the third convolutional layer receives the meteorological features, and the output of the third convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer. The input of the fourth convolutional layer receives the electric field strength data, and the output of the fourth convolutional layer is connected to the input of the first feature fusion layer and the input of the fourth feature fusion layer. The output of the first feature fusion layer is connected to the input of the fifth convolutional layer, the input of the sixth convolutional layer, the input of the global max pooling layer, and the input of the second global average pooling layer, respectively. The outputs of the fifth and sixth convolutional layers are connected to the input of the second feature fusion layer. The output of the second feature fusion layer is connected to the input of the pointwise convolutional layer through the first global average pooling layer, and the output of the pointwise convolutional layer is connected to the input of the fourth feature fusion layer. The outputs of the global max pooling layer and the second global average pooling layer are electrically connected to the input of the third feature fusion layer. The output of the third feature fusion layer is connected to the input of the dilated convolutional layer, and the output of the dilated convolutional layer is connected to the input of the fourth feature fusion layer. The output of the fourth feature fusion layer outputs the fused feature.

2. The intelligent early warning method for navigation risks based on multi-source meteorological data fusion according to claim 1, characterized in that, Each meteorological monitoring data point corresponds to a meteorological sensor. The effective meteorological grid data for the target navigation area is constructed using multiple meteorological monitoring data points, including: The target navigation area is gridded to obtain the corresponding area grid. For any grid cell in the regional grid, meteorological sensors located within that grid cell are selected, and the meteorological monitoring data corresponding to the selected meteorological sensors are used to form the initial meteorological dataset for that grid cell. For any meteorological monitoring data in the initial meteorological dataset, calculate the compatibility of each target data with the any meteorological monitoring data, and sum the compatibility values ​​to obtain the total compatibility. Here, each target data is any meteorological monitoring data in the initial meteorological dataset excluding the any meteorological monitoring data. Based on the overall compatibility, determine whether any of the meteorological monitoring data is valid data; If so, then extract any of the meteorological monitoring data, and after polling all the meteorological monitoring data in the initial meteorological dataset, obtain several valid data in the initial meteorological dataset; Based on several valid data, valid meteorological data for any grid cell is generated, and after all grid cells have been polled, valid meteorological data for each grid cell is obtained. The effective meteorological grid data is constructed by using the effective meteorological data of each grid cell.

3. The intelligent early warning method for navigation risks based on multi-source meteorological data fusion according to claim 2, characterized in that, Calculate the compatibility of each target data with any of the meteorological monitoring data, including: For any target data, calculate the data difference between any meteorological monitoring data and any target data; Obtain the data attenuation factor and compatibility control parameters; The compatibility factor is calculated based on the data decay factor and the data difference. Perform an exponential operation on the compatibility factor with base e, and take the reciprocal of the result as the initial compatibility. Based on the initial compatibility and the compatibility control parameters, the compatibility of any target data with any meteorological monitoring data is calculated; Accordingly, generating valid meteorological data for any of the grid cells based on several valid data points includes: The weight of each valid data point is calculated based on the overall compatibility of each valid data point. Based on the weights of each valid data point, the valid data points are fused together to obtain the valid meteorological data for any given grid cell.

4. A navigation risk intelligent early warning system that integrates multi-source meteorological data, characterized in that, The system is used to execute the intelligent early warning method for navigation risks based on the fusion of multi-source meteorological data as described in any one of claims 1 to 3, and the system comprises: The data acquisition unit is used to acquire weather radar data, electric field intensity data, lightning event data, and multiple meteorological monitoring data of the target navigation area within a historical preset time period. The data processing unit is used to construct an effective meteorological grid data for the target navigation area using multiple meteorological monitoring data. The data processing unit is also used to perform data transformation processing on lightning event data to convert it into lightning grid data that matches the weather radar data; The multi-source data fusion unit is used to generate meteorological input data using weather radar data, effective meteorological grid data, and lightning grid data, and to perform feature extraction processing on the meteorological input data to obtain meteorological features; The multi-source data fusion unit is also used to perform feature extraction and fusion processing on the meteorological features and the electric field intensity data to obtain fused features; The navigation risk warning unit is used to input the fused features into the lightning prediction model to obtain the lightning prediction result of the target navigation area, and generate a navigation risk warning notification for the target navigation area based on the lightning prediction result.

5. An electronic device, characterized in that, include: The system comprises a memory, a processor, and a transceiver connected in sequence for communication. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the intelligent early warning method for navigation risks based on multi-source meteorological data fusion as described in any one of claims 1 to 3.

Images