A digital control method and system for hail suppression and rain enhancement cannons

Abstract

This application provides a digital control method and system for hail suppression and rain enhancement artillery. By pre-training a meteorological prediction neural network and inputting real-time meteorological observation data into it, the real-time meteorological evolution type is obtained. Based on this real-time meteorological evolution type, the launch strategy of the hail suppression and rain enhancement equipment is set. The resulting meteorological prediction neural network exhibits high accuracy during training, thus demonstrating excellent predictive performance and aiding in intelligent decision-making for hail suppression and rain enhancement.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of meteorological forecasting technology, and in particular to a digital control method and system for a hail suppression and rain enhancement cannon. Background Technology

[0002] Hail suppression and rain enhancement are artificial weather intervention techniques designed to mitigate or alter weather phenomena to protect crops, reduce damage, or increase precipitation. For example, the shockwave generated by gunpowder explosions can counteract updrafts in hailstorms, slowing the formation or development of hail nuclei and thus preventing hail damage. In the workflow of hail suppression and rain enhancement, real-time observation and analysis of atmospheric conditions, precipitation potential, and thunderstorm development are prerequisites for determining the suitability for such operations. Developing appropriate hail suppression and rain enhancement operational plans based on real-time weather data, forecasting models, and expert judgment forms the basis for formulating hail suppression and rain enhancement artillery firing strategies. Accurate meteorological forecasting is a key technical concern. Summary of the Invention

[0003] In view of this, the embodiments of this application provide at least one digital control method and system for hail suppression and rain enhancement cannons.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] On one hand, embodiments of this application provide a digital control method for hail suppression and rain enhancement cannons, including:

[0006] A weather prediction neural network is trained based on meteorological data examples from a collection of meteorological data examples;

[0007] Acquire real-time meteorological observation data sent by a preset terminal, wherein the real-time meteorological observation data includes at least atmospheric environment data and radar satellite data;

[0008] The real-time meteorological observation data is input into the meteorological prediction neural network to predict the real-time meteorological evolution type based on the meteorological prediction neural network;

[0009] The launch strategy for hail suppression and rain enhancement equipment is set based on the real-time weather evolution type.

[0010] The process of training a meteorological prediction neural network based on meteorological data examples in the collected meteorological data example set includes: acquiring multiple sets of target meteorological data examples in the first meteorological data example set, the first meteorological data example set including meteorological data examples corresponding to multiple meteorological evolution types, each set of meteorological data examples recording a first meteorological data cluster including a meteorological intervention trigger state and example supervision information corresponding to the first meteorological data cluster, the multiple sets of target meteorological data examples belonging to multiple meteorological evolution types respectively;

[0011] The first tensor representation of the first meteorological data cluster of each set of target meteorological data examples is mined based on the initialization neural network.

[0012] Based on the initialization neural network, each set of target meteorological data examples includes, at multiple mining granularities, a second meteorological data cluster in the meteorological intervention trigger state, a second tensor representation corresponding to each second meteorological data cluster, and data supervision information corresponding to each second meteorological data cluster;

[0013] Tensor loss is obtained by using the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples, the second meteorological data cluster in the target meteorological data examples, the second tensor representation of each second meteorological data cluster, and the data supervision information of each second meteorological data cluster;

[0014] The learnable variables of the initial neural network are updated using the tensor loss, and the weather prediction neural network is obtained when the network converges.

[0015] In some embodiments, tensor loss is obtained through a first meteorological data cluster, a first tensor representation of the first meteorological data cluster, and example supervision information of the first meteorological data cluster in each set of target meteorological data examples; a second meteorological data cluster in the target meteorological data examples; a second tensor representation corresponding to each second meteorological data cluster; and data supervision information corresponding to each second meteorological data cluster. This includes:

[0016] The target meteorological data cluster is determined from multiple second meteorological data clusters corresponding to the target meteorological data example using the first meteorological data cluster in the target meteorological data example.

[0017] Loss1 is obtained by using the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data clusters of each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples;

[0018] Loss2 is obtained by using the first meteorological data cluster and the example supervision information corresponding to the first meteorological data cluster for each set of target meteorological data examples, as well as the second meteorological data cluster and the data supervision information corresponding to each second meteorological data cluster for each set of target meteorological data examples.

[0019] The loss Loss1 and the loss Loss2 are weighted and summed to obtain the tensor loss.

[0020] In some embodiments, obtaining the loss Loss1 through the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data cluster for each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples includes:

[0021] The target tensor representation of the target meteorological data example is obtained by averaging the first tensor representation corresponding to the target meteorological data example and the second tensor representation of the target meteorological data cluster of the target meteorological data example.

[0022] Tensor similarity is calculated for the target tensor representations of target meteorological data examples corresponding to the same meteorological evolution type to obtain the first example similarity;

[0023] Tensor similarity is calculated for the target tensor representations of target meteorological data examples corresponding to different meteorological evolution types to obtain the second example similarity;

[0024] Loss1 is obtained by dividing the similarity between the first example and the similarity between the second example.

[0025] In some embodiments, determining the target meteorological data cluster from a plurality of second meteorological data clusters corresponding to the target meteorological data examples through a first meteorological data cluster in each set of target meteorological data examples includes:

[0026] The first meteorological data cluster in the target meteorological data example is matched with each of the second meteorological data clusters corresponding to the target meteorological data example to determine the inter-cluster matching, and the matching score of each of the second meteorological data clusters corresponding to the target meteorological data example is obtained.

[0027] Obtain the second meteorological data cluster with a matching score greater than the set matching score, and determine it as the target meteorological data cluster of the target meteorological data example;

[0028] The loss Loss2 is obtained by using the first meteorological data cluster and the example supervision information corresponding to the first meteorological data cluster for each set of target meteorological data examples, as well as the data supervision information corresponding to each second meteorological data cluster and each second meteorological data cluster.

[0029] By using the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples, the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data examples are obtained;

[0030] The example adjustment parameters of the target meteorological data example are obtained by using the number of second meteorological data clusters corresponding to each set of target meteorological data examples and the number of target meteorological data clusters;

[0031] The prediction reliability of the true supervision information of the target meteorological data example is obtained by using the example supervision information corresponding to the first meteorological data cluster and the data supervision information corresponding to each second meteorological data cluster in the target meteorological data example.

[0032] Loss2 is obtained by using the loss adjustment parameters corresponding to each second meteorological data cluster in each target meteorological data example, the example adjustment parameters of the target meteorological data example, and the prediction confidence of the real supervision information.

[0033] In some embodiments, loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data examples are obtained through a first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples, including:

[0034] The overlap rate of the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples is calculated to obtain the overlap rate of each second meteorological data cluster in each set of target meteorological data examples. The overlap rate of each set of second meteorological data clusters is the loss adjustment parameter corresponding to the second meteorological data cluster.

[0035] In some embodiments, obtaining the example adjustment parameters of the target meteorological data examples by using the number of second meteorological data clusters corresponding to each group of target meteorological data examples and the number of target meteorological data clusters includes:

[0036] Calculate the difference between the number of the second meteorological data clusters corresponding to the target meteorological data example and the number of the target meteorological data clusters;

[0037] The example adjustment parameter of the target meteorological data example is obtained by dividing the result of the subtraction of the target meteorological data example by the result of the division between the number of the target meteorological data clusters corresponding to the target meteorological data example.

[0038] In some embodiments, obtaining the true supervision information prediction confidence of the target meteorological data example through the example supervision information corresponding to the first meteorological data cluster and the data supervision information corresponding to each second meteorological data cluster in the target meteorological data example includes:

[0039] Obtain the number of data supervision information entries in the data supervision information corresponding to each second meteorological data cluster in the target meteorological data example that are the same as the example supervision information corresponding to the first meteorological data cluster;

[0040] The number of data supervision information is divided by the number of second meteorological data clusters corresponding to the target meteorological data example. The result of the division is determined as the prediction credibility of the true supervision information of the target meteorological data example.

[0041] In some embodiments, obtaining the loss Loss2 by using the loss adjustment parameters corresponding to each second meteorological data cluster in each set of target meteorological data examples, the example adjustment parameters of the target meteorological data examples, and the prediction confidence of real supervision information includes:

[0042] The loss adjustment parameters of the target meteorological data example are obtained by averaging the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data example.

[0043] The mean square error is calculated for the prediction reliability of the real supervision information of the target meteorological data example. The calculated mean square error is multiplied by the loss adjustment parameter and the example adjustment parameter of the target meteorological data example to obtain the loss corresponding to the target meteorological data example.

[0044] The loss of each set of target meteorological data examples is weighted and summed to obtain the loss Loss2.

[0045] In some embodiments, this application further includes the initialization process for initializing the neural network:

[0046] Obtain a second meteorological data example set, which includes meteorological data examples corresponding to multiple meteorological evolution types. Each set of meteorological data examples records a first meteorological data cluster including the meteorological intervention trigger state and example supervision information corresponding to the first meteorological data cluster.

[0047] Each meteorological data example in the second meteorological data example set is input into the artificial intelligence model, and the artificial intelligence model is trained based on each meteorological data example to obtain the initialized neural network.

[0048] Secondly, embodiments of this application provide a digital control system, including a memory and a processor. The memory stores a computer program that can run on the processor, and the processor executes the program to implement the steps in the method described above.

[0049] The beneficial effects of this application are:

[0050] This application provides a digital control method and system for hail suppression and rain enhancement artillery. By pre-training a meteorological prediction neural network, real-time meteorological observation data is input to obtain real-time meteorological evolution types, thereby setting the launch strategy of the hail suppression and rain enhancement equipment based on these types. During the training of the meteorological prediction neural network, multiple sets of target meteorological data examples are obtained from a first set of meteorological data examples. The first tensor representation of the first meteorological data cluster of each set of target meteorological data examples is mined based on the initialization neural network. The second meteorological data clusters containing meteorological intervention trigger states, the second tensor representations corresponding to each second meteorological data cluster, and data supervision information are also mined at multiple mining granularities for each set of target meteorological data examples. Tensor loss is obtained through the first meteorological data clusters, the first tensor representations and example supervision information of the first meteorological data clusters, the second meteorological data clusters, and the second tensor representations and data supervision information corresponding to each second meteorological data cluster in each set of target meteorological data examples. The learnable variables of the initialization neural network are updated using the tensor loss. The meteorological prediction neural network is obtained when the network converges. Based on this, the obtained meteorological prediction neural network has excellent predictive performance, helping to make intelligent decisions regarding hail suppression and rain enhancement.

[0051] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of this application. Attached Figure Description

[0052] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0053] Figure 1 This is a schematic diagram illustrating the implementation process of a digital control method for a hail suppression and rain enhancement cannon provided in an embodiment of this application.

[0054] Figure 2 This is a schematic diagram of the composition structure of a digital control device provided in an embodiment of this application.

[0055] Figure 3 This is a schematic diagram of the hardware entity of a digital control system provided in an embodiment of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] In the following description, references to "some embodiments" refer to a subset of all possible embodiments. It is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein is for descriptive purposes only and is not intended to limit the scope of this application.

[0059] This application provides a digital control method for a hail suppression and rain enhancement cannon, which can be executed by a processor of a digital control system. The digital control system can refer to devices with data processing capabilities, such as servers, laptops, tablets, desktop computers, smart TVs, and mobile devices (e.g., mobile phones, portable video players, personal digital assistants, dedicated messaging devices, portable gaming devices).

[0060] Figure 1 This application provides a schematic diagram illustrating the implementation process of a digital control method for a hail suppression and rain enhancement cannon. Figure 1 As shown, the method includes the following steps:

[0061] S10, a meteorological prediction neural network is trained based on meteorological data examples in the collected meteorological data example set.

[0062] Specifically, the meteorological data examples in the meteorological data example set are sample data used to train the meteorological prediction neural network. These meteorological data examples can be significant meteorological observation data collected from historical meteorological observations. The meteorological observation data includes at least atmospheric environmental data and radar / satellite data. For example, atmospheric environmental data can be atmospheric conditions and environmental parameters measured regularly by meteorological observation stations, such as temperature, humidity, air pressure, wind speed, and wind direction. Radar / satellite data are observation data based on satellite and radar detection over a wide area, such as cloud images, thunderstorms, and precipitation data, including the location, intensity, development trend, and trajectory of the corresponding observed objects.

[0063] In this embodiment, the training of the weather forecasting neural network can be performed using a pre-set computer device, such as a server. The weather forecasting neural network can be a deep neural network, such as a ResNet residual network or a convolutional neural network. The training process is described in detail below:

[0064] S11, Obtain multiple sets of target meteorological data examples from the first meteorological data example set.

[0065] The first meteorological data example set includes meteorological data examples corresponding to multiple meteorological evolution types (i.e., training sample data). Each set of meteorological data examples records the first meteorological data cluster (a set of meteorological data recorded, which can be recorded according to time periods or according to a meteorological phenomenon) including the meteorological intervention trigger state and the example supervision information corresponding to the first meteorological data cluster (the supervision information can be guaranteed to be a label). Multiple sets of target meteorological data examples belong to meteorological data examples corresponding to multiple meteorological evolution types. The meteorological intervention trigger state is a meteorological state that requires meteorological intervention (such as suppressing the formation of ice crystals and hail in hail clouds by firing shells, shooting laser beams, or releasing chemical agents to interfere with the hail formation process and thus reduce and prevent the impact of hail on the ground; or using artificial means to stimulate water droplets in clouds to combine into larger precipitation particles, thereby increasing precipitation, such as by spraying aerosols or condensers to promote the combination of water droplets in clouds into precipitation particles, or by using anti-aircraft gun launchers to release catalysts or condensers into clouds to directly affect the precipitation process within the clouds). The meteorological state is specifically an evolutionary prediction state obtained through analysis, which is referred to as the meteorological evolution type in this application. The meteorological evolution type is obtained by using a meteorological prediction neural network to predict real-time meteorological observation data. If the prediction indicates that the ice crystal condensation exceeds the limit and the evolutionary state that is about to produce hail is predicted, then artificial hail prevention intervention is carried out. If the prediction indicates a critical state of precipitation, then artificial rain enhancement is carried out when necessary. It can be understood that the more accurate the predicted meteorological evolution type, the more accurate the control result.

[0066] It should be noted that the first set of meteorological data examples is data obtained within the scope permitted by laws and regulations and without infringing on the public interest, such as data obtained from official or credible meteorological data sources.

[0067] S12, based on the initialization of the neural network, mine the first tensor representation of the first meteorological data cluster of each set of target meteorological data examples from multiple sets of target meteorological data examples.

[0068] The initialization neural network is an artificial intelligence model that has been initialized, such as various conventional neural network models (e.g., CNN, RNN, LSTM, Transformer, etc.). It is trained based on meteorological data examples from the second meteorological data example set. Each meteorological data example from the second meteorological data example set is input into the artificial intelligence model, and the initialization neural network is trained based on each meteorological data example to obtain the initialization neural network. When mining the first tensor representation of the first meteorological data cluster for each set of target meteorological data examples based on the initialization neural network, the mined first tensor representation of the first meteorological data cluster can be a tensor of any order, specifically related to the data dimension. For example, it can be a first-order tensor, which is a feature vector, or a second-order tensor, which is a feature matrix; the specific order is not limited. It can be understood that the tensor representation corresponds to the feature information of the meteorological data cluster. Before performing tensor representation mining, i.e., feature information mining, the discrete data (such as region, season, weather conditions, etc.) in the meteorological data cluster can be one-hot encoded first, and then the tensor representation mining can be performed.

[0069] S13, based on the initialization neural network, each set of target meteorological data examples includes a second meteorological data cluster with meteorological intervention triggering state, a second tensor representation corresponding to each second meteorological data cluster, and data supervision information corresponding to each second meteorological data cluster at multiple mining granularities.

[0070] Initializing the neural network allows for scaling of each target meteorological data example, resulting in different mining granularities (i.e., feature extraction from different data scales). The scaled target meteorological data examples are then detected to obtain a second meteorological data cluster containing meteorological intervention trigger states. This second meteorological data cluster is then classified to obtain corresponding data supervision information. The data supervision information can include the meteorological evolution type within the second meteorological data cluster, thus obtaining the second meteorological data cluster for each target meteorological data example at each mining granularity and the corresponding data supervision information for that cluster. It should be noted that the second meteorological data cluster can include only a portion of the first meteorological data cluster, or it can include the entire first meteorological data cluster; in other words, the first meteorological data cluster can be a subset of the second meteorological data cluster.

[0071] In one implementation, the initialization neural network may include multiple feature layers and a downsampling layer, where the filter sizes of each feature layer are different. The initialization neural network can output a first tensor representation and a second tensor representation of the target meteorological data example as follows: Multiple feature layers include a first feature layer and a second feature layer. Feature mining is performed on the target meteorological data example using the first feature layer to obtain a first tensor representation corresponding to a first meteorological data cluster. The target meteorological data example after scale transformation is processed by the downsampling layer to obtain a second meteorological data cluster in the transformed meteorological data example. A second tensor representation is obtained by feature mining the second meteorological data cluster in each of the transformed meteorological data examples based on the second feature layer. The initialization neural network also includes a result output layer, which outputs the confidence level (e.g., probability, confidence coefficient) of the meteorological evolution type corresponding to each second meteorological data cluster mined from the tensor representation of each second meteorological data cluster, and then outputs the result, thereby obtaining the data supervision information corresponding to the second meteorological data cluster of each target meteorological data example at each mining granularity.

[0072] The output layer is essentially a classifier, such as Softmax or SVM. Its input information can be a second tensor representation. After performing matrix multiplication, vector addition, and standardization on the second tensor representation, the classification result of the meteorological evolution type in the second meteorological data cluster of the transformed target meteorological data example is obtained. The meteorological evolution type in the second meteorological data cluster is the first confidence level of the preset type. In other words, the output layer obtains the data supervision information corresponding to the second meteorological data cluster in the scale-transformed target meteorological data example.

[0073] S14. Tensor loss is obtained by using the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples, the second meteorological data cluster in the target meteorological data examples, the second tensor representation of each second meteorological data cluster, and the data supervision information of each second meteorological data cluster.

[0074] Optionally, step S14 specifically involves obtaining loss Loss1 through the first tensor representation corresponding to the first meteorological data cluster and the second tensor representation corresponding to the second meteorological data cluster in the target meteorological data examples, as well as the meteorological evolution type corresponding to each set of target meteorological data examples. Loss2 is obtained through the first meteorological data cluster and the example supervision information corresponding to the first meteorological data cluster for each set of target meteorological data examples, and the data supervision information corresponding to each second meteorological data cluster for each set of target meteorological data examples. Tensor loss is obtained through loss Loss1 and loss Loss2.

[0075] At this point, Loss1 can be obtained by refining the initial neural network through similarity learning added to it, while Loss2 can be obtained based on the loss function used for classification in the initial neural network. This similarity learning algorithm adjusts the initial neural network's representation of data related to meteorological evolution types, making the adjusted initial neural network more similar to the tensor representations of data of the same type, while showing greater distance from different types during similarity analysis.

[0076] For example, the first tensor representation and the second tensor representation corresponding to the target meteorological data examples are averaged to obtain the target tensor representation of each target meteorological data example. The similarity of the target tensor representations of target meteorological data examples corresponding to the same meteorological evolution type is calculated (e.g., the distance between tensors) to obtain the first similarity. The first similarity can represent the tensor similarity of the same type (the distance between tensors, such as the first-order vector distance). The similarity of the target tensor representations of target meteorological data examples corresponding to different meteorological evolution types is calculated to obtain the second similarity. The second similarity can represent the tensor similarity of different types. The first similarity is divided by the second similarity, and the result of the division is the loss Loss1.

[0077] As another implementation, the above operations can also involve obtaining target meteorological data clusters in the second meteorological data cluster corresponding to each set of target meteorological data examples whose overlap rate with the first meteorological data cluster is greater than a set overlap rate and which have consistent supervision information; obtaining the target tensor representation of the meteorological data example through the second tensor representation and the first tensor representation corresponding to each target meteorological data cluster corresponding to the same meteorological data example; calculating the tensor similarity of the target tensor representations of each target meteorological data example corresponding to the same meteorological evolution type based on the meteorological evolution type of each target meteorological data example to obtain the first similarity; calculating the tensor similarity of the tensor representations of each meteorological data example not corresponding to different meteorological evolution types to obtain the second similarity; and obtaining the tensor loss by dividing the first similarity and the second similarity.

[0078] The method for obtaining tensor loss through the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples, the second meteorological data cluster in the target meteorological data examples, the second tensor representation of each second meteorological data cluster, and the data supervision information of each second meteorological data cluster can also be any other feasible method, which will not be introduced here.

[0079] S15, the learnable variables of the initialized neural network are updated by tensor loss, and the weather prediction neural network is obtained when the network converges.

[0080] Optionally, when updating the learnable variables (parameters such as weights, biases, and learning rate) of the neural network by tensor loss, the learnable variables of the neural network are updated based on gradient descent. After each update, the process jumps to S11 for iterative iteration until the network reaches the convergence condition, at which point the target weather prediction neural network is obtained.

[0081] As can be seen, the training in this embodiment is divided into two stages: network pre-training and detail refinement. Pre-training yields an initial neural network, followed by detail refinement. The initial neural network is then trained based on target meteorological data examples to obtain the final network. This application mines the first tensor representation of the first meteorological data cluster for each set of target meteorological data examples based on the initial neural network, and mines the second meteorological data clusters containing meteorological intervention trigger states, the second tensor representations corresponding to each second meteorological data cluster, and data supervision information at multiple mining granularities for each set of target meteorological data examples. Thus, by using the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and example supervision information for each set of target meteorological data examples, the second meteorological data cluster, and the second tensor representations and data supervision information corresponding to each second meteorological data cluster, tensor loss is obtained. The learnable variables of the initial neural network are updated using the tensor loss, and a meteorological prediction neural network is obtained when the network converges. Based on this, the initial neural network can be trained with fewer meteorological data examples, reducing training costs and improving efficiency. At the same time, during training, feature information is mined after scaling each target meteorological data example, ensuring that the second tensor representation of each target meteorological data example is accurate, resulting in higher accuracy of the final network.

[0082] In other embodiments, a digital control method for a hail suppression and rain enhancement cannon is also provided, which may specifically include:

[0083] S21, Obtain multiple sets of target meteorological data examples from the first meteorological data example set.

[0084] The first meteorological data example set includes meteorological data examples corresponding to multiple meteorological evolution types. Each set of meteorological data examples records the first meteorological data cluster including the meteorological intervention trigger state and the example supervision information corresponding to the first meteorological data cluster. Multiple sets of target meteorological data examples belong to meteorological data examples corresponding to multiple meteorological evolution types.

[0085] S22, based on the initialization neural network, mine the first tensor representation of the first meteorological data cluster of each set of target meteorological data examples from multiple sets of target meteorological data examples.

[0086] S23, based on the initialization neural network, each set of target meteorological data examples includes a second meteorological data cluster with meteorological intervention triggering state, a second tensor representation corresponding to each second meteorological data cluster, and data supervision information corresponding to each second meteorological data cluster at multiple mining granularities.

[0087] S24, using the first meteorological data cluster in the target meteorological data example, determine the target meteorological data cluster from the multiple second meteorological data clusters corresponding to the target meteorological data example.

[0088] The target meteorological data cluster can be determined from the first meteorological data cluster in the target meteorological data example among the corresponding multiple second meteorological data clusters by using a similarity algorithm.

[0089] Optionally, S24 can determine the inter-cluster matching between the first meteorological data cluster in the target meteorological data example and each corresponding second meteorological data cluster in the target meteorological data example to obtain the matching score of each second meteorological data cluster corresponding to the target meteorological data example; and obtain the second meteorological data cluster with a matching score greater than a set matching score, and determine it as the target meteorological data cluster of the target meteorological data example. The set matching score is determined according to the actual situation. The matching score can refer to the overlap rate (also called the intersection-union ratio, please refer to the existing formula for the calculation method). Therefore, in S24, the first meteorological data cluster in the target meteorological data example is identified from the second meteorological data clusters corresponding to the target meteorological data example, and the meteorological data clusters with an overlap rate greater than a preset overlap rate with the first meteorological data cluster are determined.

[0090] S25, the loss Loss1 is obtained through the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data cluster of each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples.

[0091] Specifically, the target tensor representation of a target meteorological data example can be determined by the first tensor representation corresponding to each set of target meteorological data examples and the second tensor representation of the target meteorological data cluster of the target meteorological data examples. The first tensor similarity (e.g., cosine distance) between the target tensor representations of target meteorological data examples of the same type is obtained through the meteorological evolution type corresponding to each target meteorological data example, and the second tensor similarity between the target tensor representations of target meteorological data examples of different meteorological evolution types is calculated. The result of dividing the first tensor similarity by the second tensor similarity is used as the loss Loss1.

[0092] The target tensor representation of the target meteorological data example is determined by averaging the first tensor representation of the target meteorological data example and the second tensor representation of the corresponding target meteorological data cluster.

[0093] The calculation of the first tensor similarity for target meteorological data examples of the same type can be achieved by averaging the tensor similarity calculations performed on the target tensor representations of the corresponding target examples of the same type, or by summing the tensor similarity calculations performed on the target tensor representations of the corresponding target examples of the same type. Similarly, the calculation of the second tensor similarity for target meteorological data examples of different meteorological evolution types can be achieved by averaging the tensor similarity calculations performed on the target tensor representations of the corresponding target meteorological data examples of different meteorological evolution types, or by summing the tensor similarity calculations performed on the target tensor representations of the corresponding target meteorological data examples of different meteorological evolution types.

[0094] The specific process of S25 includes, for example:

[0095] Sa is the target tensor representation of the target meteorological data example obtained by averaging the first tensor representation corresponding to the target meteorological data example and the second tensor representation of the target meteorological data cluster of the target meteorological data example.

[0096] Sb is used to calculate the first example similarity by performing tensor similarity calculation on the target tensor representation of the target meteorological data example corresponding to the same meteorological evolution type.

[0097] Sc is used to calculate the second example similarity by performing tensor similarity calculation on the target tensor representation of the target meteorological data examples corresponding to different meteorological evolution types.

[0098] Sd is the loss Loss1 obtained by dividing the similarity between the first example and the similarity between the second example.

[0099] Sd, for example, uses the division between the similarity of the first example and the similarity of the second example as the loss Loss1.

[0100] Since the tensor similarity between target tensor representations of the same type of target meteorological data examples is large (corresponding to small pre-distance), and the tensor similarity between target tensor representations of different types of target meteorological data examples is also large, the larger the division result obtained by dividing the first example similarity between target meteorological data examples of the same type and the second example similarity between target meteorological data examples of different types, the more accurate the data supervision information obtained by initializing the neural network and the second tensor representation of the target meteorological data cluster will be.

[0101] S26, the loss Loss2 is obtained by using the first meteorological data cluster corresponding to each set of target meteorological data examples and the example supervision information corresponding to the first meteorological data cluster, as well as the second meteorological data cluster corresponding to each set of target meteorological data examples and the data supervision information corresponding to the second meteorological data cluster.

[0102] Specifically, the loss Loss2, obtained by substituting the first meteorological data cluster corresponding to the target meteorological data example and the example supervision information corresponding to the first meteorological data cluster, as well as the data supervision information corresponding to each second meteorological data cluster corresponding to each group of target meteorological data examples, into a pre-determined loss function (such as the mean square error loss function or the relative entropy loss function), is calculated based on the loss function.

[0103] Optionally, S26 may specifically include:

[0104] S(1) obtains the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data example by using the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples.

[0105] Specifically, the overlap rate of the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples is calculated to obtain the overlap rate of each second meteorological data cluster in each set of target meteorological data examples. The overlap rate of each set of second meteorological data clusters is the loss adjustment parameter corresponding to the second meteorological data cluster.

[0106] Alternatively, similarity calculations can be performed on the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples to obtain the similarity of each second meteorological data cluster corresponding to each set of target meteorological data examples. The similarity of each set of second meteorological data clusters is the loss adjustment parameter corresponding to the second meteorological data cluster.

[0107] S(2) is obtained by using the number of second meteorological data clusters corresponding to each set of target meteorological data examples and the number of target meteorological data clusters to obtain the example adjustment parameters of the target meteorological data examples.

[0108] The target meteorological data cluster corresponding to the target meteorological data example can be a second meteorological data cluster in the second meteorological data cluster corresponding to the target meteorological data example whose overlap rate with the first meteorological data cluster is greater than a preset overlap rate. Therefore, the example adjustment parameter can be the result of dividing the number of target meteorological data clusters in each set of target meteorological data examples by the number of second meteorological data clusters in that example. Alternatively, the example adjustment parameter can be the result of dividing the difference between the number of target meteorological data clusters in each set of target meteorological data examples and the number of second meteorological data clusters in that example, and the number of target meteorological data clusters.

[0109] Optionally, S(2) can be the result of subtracting the number of the second meteorological data clusters corresponding to the target meteorological data example from the number of the target meteorological data clusters; the example adjustment parameter of the target meteorological data example is obtained by dividing the result of subtracting the target meteorological data example from the number of the target meteorological data clusters corresponding to the target meteorological data example.

[0110] The example adjustment parameter of the target meteorological data example is obtained by dividing the result of subtracting the target meteorological data example by the number of target meteorological data clusters corresponding to the target meteorological data example. Alternatively, the example adjustment parameter can be obtained by multiplying the result of subtracting the target meteorological data example by the number of target meteorological data clusters corresponding to the target meteorological data example by a set value, or by using the result of subtracting the target meteorological data example by the number of target meteorological data clusters corresponding to the target meteorological data example as the example adjustment parameter of the target meteorological data example.

[0111] S(3) obtains the prediction credibility of the real supervision information of the target meteorological data example by using the example supervision information corresponding to the first meteorological data cluster in the target meteorological data example and the data supervision information corresponding to each second meteorological data cluster.

[0112] Specifically, the number of supervisory information entries in the example supervisory information corresponding to the second meteorological data cluster in the target example that are identical to the example supervisory information corresponding to the first example is obtained, and this number is divided by the number of second meteorological data clusters in the target example to obtain the true supervisory information prediction confidence. In other words, S(3) includes: obtaining the number of data supervisory information entries in the data supervisory information corresponding to each second meteorological data cluster in the target meteorological data example that are identical to the example supervisory information corresponding to the first meteorological data cluster; dividing this number by the number of second meteorological data clusters corresponding to the target meteorological data example, and using the division result as the true supervisory information prediction confidence of the target meteorological data example.

[0113] S(4) is obtained by using the loss adjustment parameters corresponding to each second meteorological data cluster in each target meteorological data example, the example adjustment parameters of the target meteorological data example, and the prediction credibility of the real supervision information.

[0114] Specifically, the loss adjustment parameters corresponding to each second meteorological data cluster in each set of meteorological data examples are averaged to obtain the loss adjustment parameters of the target meteorological data examples. The loss adjustment parameters, example adjustment parameters, and prediction confidence of the real supervision information for each set of target meteorological data examples are multiplied and then added together to obtain the loss Loss2.

[0115] Optionally, S(4) specifically involves averaging the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data example to obtain the loss adjustment parameters of the target meteorological data example; calculating the root mean square error of the prediction credibility of the real supervision information of the target meteorological data example; multiplying the calculated root mean square error with the loss adjustment parameters and example adjustment parameters of the target meteorological data example to obtain the loss corresponding to the target meteorological data example; and weighted summing the losses of each group of target meteorological data examples to obtain the loss Loss2.

[0116] S27, weighted sum of loss Loss1 and loss Loss2 to obtain tensor loss.

[0117] S28, the learnable variables of the initial neural network are updated by tensor loss, and the weather prediction neural network is obtained when the network converges.

[0118] In this embodiment, a target meteorological data cluster is determined from multiple second meteorological data clusters corresponding to a first meteorological data cluster in the target meteorological data example. Loss1 is obtained through the first tensor representation corresponding to each target meteorological data example, the second tensor representation of the target meteorological data cluster in each target meteorological data example, and the meteorological evolution type corresponding to each target meteorological data example. Loss2 is obtained through the first meteorological data cluster and its corresponding example supervision information, as well as each second meteorological data cluster and its corresponding data supervision information. The tensor loss obtained by weighted summing of Loss1 and Loss2 is used to update and initialize the neural network. Thus, by updating and initializing the neural network with two losses together, the resulting meteorological prediction neural network has higher accuracy and more accurate meteorological evolution type identification, aiding in effective decision-making for hail prevention and rain enhancement control.

[0119] In another embodiment, a different method for training a weather forecasting neural network is provided, which may specifically include:

[0120] S31, input meteorological data examples of multiple meteorological evolution types from the second meteorological data example set into the residual network respectively, train based on the residual network structure, and obtain the initial neural network.

[0121] Optionally, the residual structure includes a first feature layer, a second feature layer, an affine layer, a tensor projection layer, and a tensor transformation layer. The first feature layer is used to mine the first tensor representation of the meteorological data examples. The second feature layer is used to mine the second tensor representation of the meteorological data examples at multiple mining granularities. The affine layer is used to select the corresponding data range of meteorological evolution types in the meteorological data examples at multiple mining granularities to obtain the second meteorological data cluster. The tensor projection layer is used to establish the relationship between the first tensor representation, the second tensor representation, and the second meteorological data cluster corresponding to each group of meteorological data examples. The tensor transformation layer performs vector transformation of the features. During network training, the residual network outputs a location distribution loss and a classification loss, and updates the learnable variables of the network through the two losses. The initialized neural network is obtained upon convergence.

[0122] After the neural network is initialized through training, a storage set and a similarity learning mechanism are set up for network detail refinement. The storage set contains meteorological data examples from a first meteorological data example set. Since the first set contains relatively few examples, a predetermined number of meteorological data examples are determined from each type of meteorological data example in a second set, and a predetermined number of meteorological data examples are determined from each new type in a new set of meteorological data examples. The resulting first set of meteorological data examples includes examples from multiple types. The similarity learning mechanism includes a tensor projection layer and a tensor transformation layer. The tensor projection layer performs feature mining and encoding on the first and second tensor representations corresponding to each target meteorological data example, while the tensor transformation layer performs feature transformation.

[0123] The process of training a network to represent meteorological evolution types based on a similarity learning mechanism can include:

[0124] S32, based on the first feature layer of the initialized neural network, mine the first tensor representation of the first meteorological data cluster of each set of target meteorological data examples in multiple sets of target meteorological data examples.

[0125] S33, the downsampling layer based on the initialization neural network mines an example of each group of target meteorological data, including the second meteorological data cluster containing the meteorological intervention trigger state at multiple mining granularities, and the second feature layer based on the initialization neural network mines the second tensor representation corresponding to each second meteorological data cluster, and the result output layer based on the initialization neural network outputs the data supervision information corresponding to each second meteorological data cluster.

[0126] S34, Based on the similarity learning mechanism, the target meteorological data cluster is determined from the first meteorological data cluster in the target meteorological data example and from the multiple second meteorological data clusters corresponding to the target meteorological data example. The loss Loss1 is obtained through the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data cluster in each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples.

[0127] The similarity learning mechanism can determine the inter-cluster matching between the first meteorological data cluster in the target meteorological data example and each corresponding second meteorological data cluster based on the tensor transformation layer, obtaining the matching score of each second meteorological data cluster corresponding to the target meteorological data example; the second meteorological data cluster with a matching score greater than a set matching score is identified as the target meteorological data cluster of the target meteorological data example. The target tensor representation of the target meteorological data example is obtained by averaging the first tensor representation corresponding to the target meteorological data example and the second tensor representation of the target meteorological data cluster; the first example similarity is obtained by calculating the tensor similarity of the target tensor representations of target meteorological data examples corresponding to the same meteorological evolution type; the second example similarity is obtained by calculating the tensor similarity of the target tensor representations of target meteorological data examples corresponding to different meteorological evolution types; and the loss Loss1 is obtained by dividing the first example similarity and the second example similarity.

[0128] When training the differences between different types of tensor representations based on a similarity learning mechanism, the classification loss for classifying target meteorological data examples can also be calculated, specifically including:

[0129] S35, the loss Loss2 is obtained by using the first meteorological data cluster corresponding to each set of target meteorological data examples and the example supervision information corresponding to the first meteorological data cluster, as well as the second meteorological data cluster corresponding to each set of target meteorological data examples and the data supervision information corresponding to the second meteorological data cluster.

[0130] Specifically, based on existing loss functions (such as Cross Entropy), loss adjustment parameters and example adjustment parameters can be added to help determine the classification loss. For example, the overlap rate of the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples is calculated to obtain the overlap rate of each second meteorological data cluster corresponding to each set of target meteorological data examples. The overlap rate of each set of second meteorological data clusters corresponds to the loss adjustment parameter of that second meteorological data cluster. The number of second meteorological data clusters corresponding to the target meteorological data examples is obtained, and the result of subtracting the number of target meteorological data clusters is used to divide the result of subtracting the number of second meteorological data clusters corresponding to the target meteorological data examples by the number of target meteorological data clusters corresponding to the target meteorological data examples, thus obtaining the example adjustment parameter of the target meteorological data examples. The number of data supervision information in the data supervision information corresponding to each second meteorological data cluster in the target meteorological data examples that is the same as the example supervision information corresponding to the first meteorological data cluster is obtained; this number is divided by the number of second meteorological data clusters corresponding to the target meteorological data examples, and the result of this division is determined as the prediction confidence of the true supervision information of the target meteorological data examples.

[0131] Based on this, a loss adjustment parameter is used to balance the loss of each second meteorological data cluster determined in the target meteorological data example. Since the number of second meteorological data clusters with matching scores greater than the set matching score is small when the second meteorological data cluster with matching scores greater than the set matching score is obtained as the target meteorological data example, and most of the matching scores are less than 1, an example adjustment parameter is added when determining the loss Loss2. During the training process, the number of meteorological data examples of the target meteorological data cluster begins to increase, and the loss adjustment parameter increases accordingly. Therefore, the perturbation of the weight of the loss adjustment parameter on the loss of the target meteorological data example is reduced based on the example adjustment parameter, so that the accuracy of the classification loss is improved.

[0132] Loss2 is obtained by using the loss adjustment parameters corresponding to each second meteorological data cluster in each target meteorological data example, the example adjustment parameters of the target meteorological data example, and the prediction credibility of the real supervision information.

[0133] After obtaining Loss1 and Loss2, the method also includes:

[0134] S36, weighted sum of loss Loss1 and loss Loss2 to obtain tensor loss.

[0135] S37 updates the learnable variables of the initialized neural network using tensor loss.

[0136] After updating and initializing the learnable variables of the neural network, the convergence status is determined, and the meteorological prediction neural network is obtained when the network converges.

[0137] In summary, the training of the weather forecasting neural network has been completed.

[0138] S20: Obtain real-time meteorological observation data sent by a preset terminal. The real-time meteorological observation data includes at least atmospheric environmental data and radar satellite data.

[0139] Atmospheric environmental data can be atmospheric conditions and environmental parameters measured regularly by meteorological observation stations, such as temperature, humidity, air pressure, wind speed, and wind direction. Radar and satellite data are observational data based on satellite and radar detection over a wide area, such as cloud images, thunderstorms, and precipitation data, including the location, intensity, development trend, and trajectory of the corresponding observed objects.

[0140] S30 inputs real-time meteorological observation data into a meteorological prediction neural network to predict the real-time weather evolution type based on the meteorological prediction neural network.

[0141] S40 sets the launch strategy for hail suppression and rain enhancement equipment based on real-time weather evolution.

[0142] For example, if the predicted weather evolution indicates that ice crystal condensation has exceeded the limit and hail is imminent, then hail suppression and rain enhancement artillery can be used for artificial hail intervention. This can be achieved by firing shells, shooting laser beams, or releasing chemical agents to suppress ice crystal formation and hail formation in hail clouds, thereby interfering with the hail formation process and reducing or preventing the impact of hail on the ground. If the predicted weather evolution indicates that the current state is approaching the critical rainfall threshold, then artificial rain enhancement can be carried out as needed. Catalysts or condensers can be released into the clouds using artillery launchers to directly affect the precipitation process within the clouds. Of course, in practical applications, more abundant meteorological data examples (containing more weather evolution types) can be collected for training, enabling the weather prediction neural network to predict more weather evolution types and allowing for the formulation of hail suppression and rain enhancement equipment launch strategies as needed.

[0143] Based on the foregoing embodiments, this application provides a digital control device. The units and modules included in the device can be implemented by a processor in a computer device; of course, they can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0144] Figure 2 This is a schematic diagram of the composition structure of a digital control device provided in an embodiment of this application, as shown below. Figure 2 As shown, the digital control device 200 includes:

[0145] Network training module 210 is used to train a weather prediction neural network based on meteorological data examples in the collected meteorological data example set;

[0146] Data acquisition module 220 is used to acquire real-time meteorological observation data sent by a preset terminal, wherein the real-time meteorological observation data includes at least atmospheric environment data and radar satellite data;

[0147] The type prediction module 230 is used to input the real-time meteorological observation data into the meteorological prediction neural network to predict the real-time meteorological evolution type based on the meteorological prediction neural network.

[0148] The strategy formulation module 240 is used to set the launch strategy of the hail suppression and rain enhancement equipment based on the real-time weather evolution type.

[0149] The process of training a meteorological prediction neural network based on meteorological data examples in the collected meteorological data example set includes: acquiring multiple sets of target meteorological data examples in the first meteorological data example set, the first meteorological data example set including meteorological data examples corresponding to multiple meteorological evolution types, each set of meteorological data examples recording a first meteorological data cluster including a meteorological intervention trigger state and example supervision information corresponding to the first meteorological data cluster, the multiple sets of target meteorological data examples belonging to multiple meteorological evolution types respectively;

[0150] The first tensor representation of the first meteorological data cluster of each set of target meteorological data examples is mined based on the initialization neural network.

[0151] Based on the initialization neural network, each set of target meteorological data examples includes, at multiple mining granularities, a second meteorological data cluster in the meteorological intervention trigger state, a second tensor representation corresponding to each second meteorological data cluster, and data supervision information corresponding to each second meteorological data cluster;

[0152] Tensor loss is obtained by using the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples, the second meteorological data cluster in the target meteorological data examples, the second tensor representation of each second meteorological data cluster, and the data supervision information of each second meteorological data cluster;

[0153] The learnable variables of the initial neural network are updated using the tensor loss, and the weather prediction neural network is obtained when the network converges.

[0154] The descriptions of the apparatus embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. In some embodiments, the functions or modules included in the apparatus provided in this application can be used to perform the methods described in the method embodiments above. For technical details not disclosed in the apparatus embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0155] It should be noted that, in the embodiments of this application, if the above-mentioned digital control method for hail suppression and rain enhancement cannons is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

[0156] This application provides a digital control system, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above-described method.

[0157] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements some or all of the steps in the above-described method. The computer-readable storage medium can be transient or non-transient.

[0158] This application provides a computer program including computer-readable code, wherein when the computer-readable code is executed in a computer device, a processor in the computer device performs some or all of the steps in the above-described method.

[0159] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium; in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0160] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between them, while their similarities or commonalities can be referred to interchangeably. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0161] Figure 3 A schematic diagram of the hardware entity of a digital control system provided in this application embodiment is shown below. Figure 3 As shown, the hardware entity of the digital control system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

[0162] The memory 1002 stores computer programs that can run on the processor. The memory 1002 is configured to store instructions and applications that can be executed by the processor 1001. It can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data and video communication data) in the processor 1001 and various modules in the digital control system 1000. It can be implemented by flash memory or random access memory (RAM).

[0163] When the processor 1001 executes the program, it implements the steps of the digital control method for the hail suppression and rain enhancement cannon described above. The processor 1001 typically controls the overall operation of the digital control system 1000.

[0164] This application provides a computer storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the digital control method for hail suppression and rain enhancement cannons as described in any of the above embodiments.

[0165] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding. The processor described above can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that the electronic device implementing the above processor function can also be other types, and this application does not specifically limit the specific types.

[0166] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0167] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence number of the above-described steps / processes does not imply the order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "including one..." does not preclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0169] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0170] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0171] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0172] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, magnetic disks, or optical disks.

[0173] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A digital control method for a hail suppression and rain enhancement cannon, characterized in that, include: A weather prediction neural network is trained based on meteorological data examples from a collection of meteorological data examples; Acquire real-time meteorological observation data sent by a preset terminal, wherein the real-time meteorological observation data includes at least atmospheric environment data and radar satellite data; The real-time meteorological observation data is input into the meteorological prediction neural network to predict the real-time meteorological evolution type based on the meteorological prediction neural network; The launch strategy for hail suppression and rain enhancement equipment is set based on the real-time weather evolution type. The process of training a meteorological prediction neural network based on meteorological data examples in a collection of meteorological data examples includes: acquiring multiple sets of target meteorological data examples from a first set of meteorological data examples, wherein the first set of meteorological data examples includes meteorological data examples corresponding to multiple meteorological evolution types, and each set of meteorological data examples records a first meteorological data cluster including a meteorological intervention trigger state and example supervision information corresponding to the first meteorological data cluster, wherein the multiple sets of target meteorological data examples belong to meteorological data examples corresponding to multiple meteorological evolution types. The first tensor representation of the first meteorological data cluster of each set of target meteorological data examples is mined based on the initialization neural network. Based on the initialization neural network, each set of target meteorological data examples includes, at multiple mining granularities, a second meteorological data cluster in the meteorological intervention trigger state, a second tensor representation corresponding to each second meteorological data cluster, and data supervision information corresponding to each second meteorological data cluster; Tensor loss is obtained by using the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples, the second meteorological data cluster in the target meteorological data examples, the second tensor representation of each second meteorological data cluster, and the data supervision information of each second meteorological data cluster; The learnable variables of the initial neural network are updated using the tensor loss, and the weather prediction neural network is obtained when the network converges.

2. The method as described in claim 1, characterized in that, Tensor loss is obtained through the first meteorological data cluster, the first tensor representation of the first meteorological data cluster, and the example supervision information of the first meteorological data cluster in each set of target meteorological data examples; the second meteorological data cluster in the target meteorological data examples; the second tensor representation corresponding to each second meteorological data cluster; and the data supervision information corresponding to each second meteorological data cluster. This includes: The target meteorological data cluster is determined from multiple second meteorological data clusters corresponding to the target meteorological data example using the first meteorological data cluster in the target meteorological data example. Loss1 is obtained by using the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data clusters of each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples; Loss2 is obtained by using the first meteorological data cluster and the example supervision information corresponding to the first meteorological data cluster for each set of target meteorological data examples, as well as the second meteorological data cluster and the data supervision information corresponding to each second meteorological data cluster for each set of target meteorological data examples. The loss Loss1 and the loss Loss2 are weighted and summed to obtain the tensor loss.

3. The method as described in claim 2, characterized in that, The loss Loss1 is obtained by using the first tensor representation corresponding to each set of target meteorological data examples, the second tensor representation of the target meteorological data cluster for each set of target meteorological data examples, and the meteorological evolution type corresponding to each set of target meteorological data examples, including: The target tensor representation of the target meteorological data example is obtained by averaging the first tensor representation corresponding to the target meteorological data example and the second tensor representation of the target meteorological data cluster of the target meteorological data example. Tensor similarity is calculated for the target tensor representations of target meteorological data examples corresponding to the same meteorological evolution type to obtain the first example similarity; Tensor similarity is calculated for the target tensor representations of target meteorological data examples corresponding to different meteorological evolution types to obtain the second example similarity; Loss1 is obtained by dividing the similarity between the first example and the similarity between the second example.

4. The method as described in claim 2, characterized in that, The step of determining the target meteorological data cluster from multiple second meteorological data clusters corresponding to the target meteorological data example through the first meteorological data cluster in each set of target meteorological data examples includes: The first meteorological data cluster in the target meteorological data example is matched with each of the second meteorological data clusters corresponding to the target meteorological data example to determine the inter-cluster matching, and the matching score of each of the second meteorological data clusters corresponding to the target meteorological data example is obtained. Obtain the second meteorological data cluster with a matching score greater than the set matching score, and determine it as the target meteorological data cluster of the target meteorological data example; The loss Loss2 is obtained by using the first meteorological data cluster corresponding to each set of target meteorological data examples and the example supervision information corresponding to the first meteorological data cluster, as well as the data supervision information corresponding to each second meteorological data cluster, including: By using the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples, the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data examples are obtained; The example adjustment parameters of the target meteorological data example are obtained by using the number of second meteorological data clusters corresponding to each set of target meteorological data examples and the number of target meteorological data clusters; The prediction reliability of the true supervision information of the target meteorological data example is obtained by using the example supervision information corresponding to the first meteorological data cluster and the data supervision information corresponding to each second meteorological data cluster in the target meteorological data example. Loss2 is obtained by using the loss adjustment parameters corresponding to each second meteorological data cluster in each target meteorological data example, the example adjustment parameters of the target meteorological data example, and the prediction confidence of the real supervision information.

5. The method as described in claim 4, characterized in that, By using the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples, the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data examples are obtained, including: The overlap rate of the first meteorological data cluster and each second meteorological data cluster corresponding to each set of target meteorological data examples is calculated to obtain the overlap rate of each second meteorological data cluster in each set of target meteorological data examples. The overlap rate of each set of second meteorological data clusters is the loss adjustment parameter corresponding to the second meteorological data cluster.

6. The method as described in claim 4, characterized in that, The step of obtaining the example adjustment parameters of the target meteorological data examples by using the number of second meteorological data clusters corresponding to each set of target meteorological data examples and the number of target meteorological data clusters includes: Calculate the difference between the number of the second meteorological data clusters corresponding to the target meteorological data example and the number of the target meteorological data clusters; The example adjustment parameter of the target meteorological data example is obtained by dividing the result of the subtraction of the target meteorological data example by the result of the division between the number of the target meteorological data clusters corresponding to the target meteorological data example.

7. The method as described in claim 4, characterized in that, The step of obtaining the prediction confidence of the true supervisory information of the target meteorological data example through the example supervisory information corresponding to the first meteorological data cluster and the data supervisory information corresponding to each second meteorological data cluster in the target meteorological data example includes: Obtain the number of data supervision information entries in the data supervision information corresponding to each second meteorological data cluster in the target meteorological data example that are the same as the example supervision information corresponding to the first meteorological data cluster; The number of data supervision information is divided by the number of second meteorological data clusters corresponding to the target meteorological data example. The result of the division is determined as the prediction credibility of the true supervision information of the target meteorological data example.

8. The method as described in claim 4, characterized in that, The loss Loss2 is obtained by using the loss adjustment parameters corresponding to each second meteorological data cluster in each set of target meteorological data examples, the example adjustment parameters of the target meteorological data examples, and the prediction confidence of real supervision information, including: The loss adjustment parameters of the target meteorological data example are obtained by averaging the loss adjustment parameters corresponding to each second meteorological data cluster in the target meteorological data example. The mean square error is calculated for the prediction reliability of the real supervision information of the target meteorological data example. The calculated mean square error is multiplied by the loss adjustment parameter and the example adjustment parameter of the target meteorological data example to obtain the loss corresponding to the target meteorological data example. The loss of each set of target meteorological data examples is weighted and summed to obtain the loss Loss2.

9. The method according to any one of claims 1 to 8, characterized in that, It also includes the initialization process for the neural network: Obtain a second meteorological data example set, which includes meteorological data examples corresponding to multiple meteorological evolution types. Each set of meteorological data examples records a first meteorological data cluster including the meteorological intervention trigger state and example supervision information corresponding to the first meteorological data cluster. Each meteorological data example in the second meteorological data example set is input into the artificial intelligence model, and the artificial intelligence model is trained based on each meteorological data example to obtain the initialized neural network.

10. A digital control system, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 9.

Images