A rainfall prediction method and system based on multi-band radar networking

Abstract

This invention relates to the field of prediction model training technology, and discloses a rainfall prediction method and system based on multi-band radar networking. The method includes: performing a preliminary rainfall model update based on extracted radar prediction clients to obtain a local rainfall prediction model; generating a precise update sequence for model layers based on the local rainfall prediction model to obtain a local model layer sequence; iteratively updating the model parameters of the local model layer sequence to obtain a global parameter update dataset; transmitting multiple global parameter update datasets to a command server center to obtain an execution server center; and aggregating the global rainfall prediction model based on the execution server center to obtain the current rainfall prediction model. This invention can improve the effectiveness and training efficiency of rainfall prediction model updates and reduce interference from invalid communication and noisy updates.

Description

Technical Field

[0001] This invention relates to the field of prediction model training technology, and in particular to a rainfall prediction method and system based on multi-band radar networking. Background Technology

[0002] With the increasing frequency of extreme weather events, higher demands are placed on the spatiotemporal accuracy and timeliness of forecasts. Multi-band radar networks can collaboratively acquire multi-level and multi-dimensional atmospheric observation data, providing a rich data foundation for refined forecasts. Against this backdrop, the federated learning framework can be used to collaboratively train prediction models from multiple geographically distributed radar clients without aggregating the original sensitive data from each radar site. This has become an important technical direction for fully utilizing dispersed data resources and improving the generalization ability of models while protecting data privacy and security.

[0003] Traditional implementations typically employ a standard federated learning process, where each client trains locally and then uploads the complete model parameter updates to the server for aggregation. This approach has several drawbacks: first, the client needs to upload all parameter updates, impacting overall training efficiency; second, the aggregation process treats all parameter updates indiscriminately, including minor, anomalous, or unhelpful parameter changes caused by local data noise or distribution differences, which may interfere with the aggregation direction and reduce the model's accuracy and convergence stability. Summary of the Invention

[0004] This invention provides a rainfall prediction method based on multi-band radar networking and a computer-readable storage medium. Its main purpose is to improve the effectiveness and training efficiency of rainfall prediction model updates and reduce interference from invalid communication and noise updates.

[0005] To achieve the above objectives, this invention provides a rainfall prediction method based on multi-band radar networking, comprising: The radar network prediction environment has been identified. This environment includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. A global rainfall prediction model is generated based on the command server center, and then sent to multiple radar control clients to obtain multiple radar prediction clients. The radar prediction clients are extracted sequentially from multiple radar prediction clients. Based on the extracted radar prediction clients, the preliminary rainfall model is updated to obtain the local rainfall prediction model. Based on the local rainfall prediction model, the model layer is accurately updated in the correct order to obtain the local model layer sequence. The model parameters are iteratively updated on the local model layer sequence to obtain the global parameter update dataset; The global parameter update datasets corresponding to each radar prediction client are aggregated to obtain multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model, thus completing the rainfall prediction based on multi-band radar networking.

[0006] Optionally, the preliminary rainfall model update based on the extracted radar prediction client to obtain a local rainfall prediction model includes: The rainfall prediction unit extracted from the radar prediction client is used to obtain the front radar acquisition dataset, which includes multiple front radar acquisition data, and each front radar acquisition data contains an acquisition timestamp. Extract the data collected by the front radar sequentially from the front radar data set, and query the previous rainfall based on the collection timestamp in the extracted front radar data. The data collected by the preceding rainfall is labeled using the preceding rainfall data to obtain labeled radar data. The data collected by the marker radar is summarized to obtain the marker radar dataset; Randomly select from the labeled radar acquisition dataset to obtain the training radar acquisition dataset; The global rainfall prediction model is trained and updated based on the training radar dataset to obtain a local rainfall prediction model.

[0007] Optionally, the step of generating a local model layer sequence by accurately updating the model layer according to the local rainfall prediction model includes: Obtain the set of model update parameter types, which includes multiple types of model update parameters; Perform the following operation for each model update parameter type in the model update parameter type set: Based on the types of model update parameters, local model parameters and global model parameters are extracted from the local rainfall prediction model and the global rainfall prediction model, respectively, and the local parameter deviation value is calculated using the local model parameters and global model parameters. Summarize the local parameter deviation values ​​corresponding to each type of updated parameter in the model to obtain a set of local parameter deviation values; Multiple local model layers were identified in the local rainfall prediction model, where the local model layers are either convolutional layers or fully connected layers; The importance of multiple local model layers is ranked based on the local parameter deviation value set to obtain a local model layer sequence, which contains multiple local model layers.

[0008] Optionally, the step of ranking the importance of multiple local model layers based on the local parameter deviation value set to obtain a local model layer sequence includes: Perform the following operation on each of the multiple local model layers: Multiple current layer parameters of the local model layer are identified, and multiple current parameter deviation values ​​are identified in the local parameter deviation value set based on the multiple current layer parameters; An urgency analysis is performed based on multiple current parameter deviation values ​​to obtain an updated urgency value, which is expressed as:

[0009] in, This indicates that the urgent value needs to be updated. This indicates the number of current layer parameters out of multiple current layer parameters. Represents the hyperbolic sine function. This represents the first of multiple current parameter deviation values. One current parameter deviation value; By summing the update urgency values ​​corresponding to each local model layer, multiple update urgency values ​​are obtained; Multiple local model layers are sorted using multiple update urgency values ​​to obtain a sequence of local model layers. The local model layer with the larger update urgency value is placed earlier in the sequence of local model layers.

[0010] Optionally, the step of iteratively updating the model parameters of the local model layer sequence to obtain the global parameter update dataset includes: Local model layers are extracted sequentially from the local model layer sequence, and the extracted local model layers are denoted as the local model layers to be trained. Based on the local model layer to be trained, identify the set of update parameter types in the same layer from the set of model update parameter types; Search for the set of parameter deviation values ​​corresponding to the set of updated parameter types in the same layer within the local parameter deviation value set; Based on the parameter deviation value set of the same layer, the model update contribution analysis is performed on the same layer update parameter type set to obtain the parameter update contribution value set; The global parameter update dataset is obtained by iterating the parameters using the parameter update contribution value set and the local rainfall prediction model.

[0011] Optionally, the step of performing model update contribution analysis on the same-layer updated parameter type set based on the same-layer parameter deviation value set to obtain the parameter update contribution value set includes: The mean of the parameter deviation values ​​in the same layer is calculated to obtain the average parameter deviation value; Extract the same-level update parameter types sequentially from the same-level update parameter type set, and record the extracted same-level update parameter types as the parameter types to be evaluated; Identify the parameter deviation values ​​corresponding to the types of parameters to be evaluated within the same layer of parameter deviation value set; Based on the types of parameters to be evaluated, the local rainfall prediction model is trained and gradient queries are performed to obtain the local parameter gradients. The parameter update contribution value is calculated based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient. Summarize the parameter update contribution values ​​corresponding to each type of update parameter in the same layer to obtain the parameter update contribution value set.

[0012] Optionally, the step of calculating the parameter update contribution value based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient includes: The parameter update contribution value is calculated using the following formula:

[0013] in, This indicates that the parameter updates the contribution value. This represents a predefined symbolic function. Represents the gradient of local parameters. This indicates taking the absolute value. This represents the deviation value of the parameter to be evaluated. express function, This represents the average parameter deviation value.

[0014] Optionally, the step of using the parameter update contribution value set and the local rainfall prediction model to perform parameter iteration to obtain the global parameter update dataset includes: Based on the parameter update contribution value set and the same-layer update parameter type set, obtain the updatable parameter type set and the single-layer update tag set; The local rainfall prediction model is calibrated using an updatable parameter set to obtain the calibrated rainfall prediction model. A single-layer training dataset is obtained by randomly selecting from the labeled radar acquisition dataset. The calibration rainfall prediction model was iteratively trained using a single-layer training dataset to obtain a locally accurate update model. A model accurate update value set is generated based on the set of updatable parameter types in the global rainfall prediction model and the local accurate update model. The model accurate update value set includes multiple model accurate update values, and the model accurate update values ​​correspond one-to-one with the updatable parameter types in the set of updatable parameter types. The model's precise update value set and the single-layer update tag set are merged to obtain global parameter update data; The local accurate update model is used as the local rainfall prediction model, and the step of extracting local model layers sequentially in the local model layer sequence is returned until all local model layers in the local model layer sequence have been extracted. The global parameter update data is then aggregated to obtain the global parameter update dataset.

[0015] Optionally, obtaining the set of updatable parameter categories and the single-layer update tag set based on the parameter update contribution value set and the same-layer update parameter category set includes: Based on a preset update contribution threshold, the target contribution value set and the non-target contribution value set are identified in the parameter update contribution value set; In the same layer update parameter type set, the updatable parameter type set corresponding to the target contribution value set and the non-updatable parameter type set corresponding to the non-target contribution value set are respectively identified; Based on the set of updatable and non-updatable parameter types, co-positional markers are generated to obtain a single-layer update marker set.

[0016] To achieve the above objectives, the present invention also provides a rainfall prediction system based on multi-band radar networking, comprising: The prediction environment confirmation module is used to confirm the radar network prediction environment, which includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. The local model update module is used to generate a global rainfall prediction model based on the command server center, send the global rainfall prediction model to multiple radar control clients to obtain multiple radar prediction clients, extract radar prediction clients sequentially from multiple radar prediction clients, and perform preliminary rainfall model update based on the extracted radar prediction clients to obtain a local rainfall prediction model. The model layer sequence generation module is used to generate a precise update sequence of model layers based on the local rainfall prediction model, obtain a local model layer sequence, and perform iterative updates of model parameters on the local model layer sequence to obtain a global parameter update dataset. The multi-source model aggregation module is used to summarize the global parameter update datasets corresponding to each radar prediction client, resulting in multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and the multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model.

[0017] To address the above problems, the present invention also provides an electronic device, the electronic device comprising: Memory, storing at least one instruction; The processor executes the instructions stored in the memory to implement the rainfall prediction method based on multi-band radar networking described above.

[0018] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the above-described rainfall prediction method based on multi-band radar networking.

[0019] To address the problems described in the background, this invention first generates a precise update sequence for model layers based on a local rainfall prediction model, resulting in a local model layer sequence. This step calculates the deviation of parameters within each model layer from the global model and uses a hyperbolic sine function to calculate the update urgency value, thereby ranking the importance of each model layer. Compared to existing technologies that indiscriminately process or randomly update all model layers, this method can identify the model layers with the largest deviation from the global model and the most urgent need for adjustment. This provides a basis for the efficient allocation of subsequent computing resources, ensuring that optimization can focus on key parts when resources are limited, thus improving the efficiency of the overall training process. Furthermore, this solution iteratively updates the model parameters of the local model layer sequence to obtain a global parameter update dataset. This step uses a formula that comprehensively considers the contribution value of parameter update direction, magnitude, and consistency to select updatable parameters that are critical to model performance, rather than uploading all parameters. Compared to the existing federated learning approach where the client uploads all or a random portion of the model parameters, this significantly reduces the amount of data that needs to be uploaded, improving the efficiency of the model aggregation stage. Secondly, this filtering mechanism effectively filters out minor or abnormal parameter updates caused by local data noise, resulting in higher quality and more effective updated datasets uploaded to the server, thereby improving the accuracy and stability of subsequent global model aggregation. Therefore, this invention can improve the effectiveness and training efficiency of rainfall prediction model updates while reducing interference from invalid communication and noisy updates. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a rainfall prediction method based on multi-band radar networking provided in an embodiment of the present invention. Figure 2 This is a functional block diagram of a rainfall prediction system based on multi-band radar networking provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device for implementing the rainfall prediction method based on multi-band radar networking, as provided in an embodiment of the present invention.

[0021] Explanation of reference numerals in the attached figures: 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0024] This application provides a rainfall prediction method based on a multi-band radar network. The execution entity of the rainfall prediction method based on a multi-band radar network includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the rainfall prediction method based on a multi-band radar network can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0025] Reference Figure 1 The diagram shown is a flowchart illustrating a rainfall prediction method based on a multi-band radar network according to an embodiment of the present invention. In this embodiment, the rainfall prediction method based on a multi-band radar network includes: S1. The radar network prediction environment is identified. The radar network prediction environment includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit.

[0026] It is clear that the radar network prediction environment refers to a distributed system architecture consisting of multiple radar control clients and a command server center. The radar control clients refer to terminal devices distributed in different regions for collecting and processing radar data. A multi-band radar group refers to a collection of radar devices capable of emitting and receiving electromagnetic waves of different frequencies. This multi-band radar group can detect precipitation particles in the atmosphere, thereby collecting radar data related to rainfall, such as radar reflectivity factors and radial velocity. The rainfall prediction unit refers to a computational module capable of training a model based on the radar data collected by the radar control clients. This rainfall prediction unit also includes storing radar data historically collected by the multi-band radar group, such as subsequent pre-collection radar datasets. The command server center refers to a central server used to coordinate all radar control clients and manage the global model. This command server center is used to generate an initial global model (such as the subsequent global rainfall prediction model), distribute the global model to each radar control client, and aggregate the global parameter update datasets trained by each radar control client.

[0027] S2. Generate a global rainfall prediction model based on the command server center, and send the global rainfall prediction model to multiple radar control clients to obtain multiple radar prediction clients.

[0028] Understandably, the global rainfall prediction model refers to the machine learning model maintained by the command server center for rainfall prediction. This global rainfall prediction model is the current rainfall prediction model obtained after the last model aggregation. If model aggregation has not yet been performed, then the global rainfall prediction model is an initial rainfall prediction model constructed manually. The construction method of this initial rainfall prediction model is as follows: First, multiple segments of radar data from different periods are collected. These multiple segments of radar data have the same composition as the subsequent front-end radar data. Then, the rainfall amounts for these periods are obtained. The multiple segments of radar data are paired with the corresponding rainfall amounts to construct a labeled training dataset. Then, the initial rainfall prediction model is obtained by using this training dataset to train neural network models such as convolutional neural networks (CNN) and recurrent neural networks (RNN) through supervised learning (including forward propagation, loss calculation, backpropagation, and gradient descent optimization). The radar prediction client refers to the radar control client that receives the global rainfall prediction model.

[0029] S3. Extract radar prediction clients sequentially from multiple radar prediction clients, and update the preliminary rainfall model based on the extracted radar prediction clients to obtain a local rainfall prediction model.

[0030] It is clear that the local rainfall prediction model refers to the global rainfall prediction model after the radar prediction client has performed preliminary model training.

[0031] Specifically, the preliminary rainfall model update based on the extracted radar prediction client data to obtain a local rainfall prediction model includes: The rainfall prediction unit extracted from the radar prediction client is used to obtain the front radar acquisition dataset, which includes multiple front radar acquisition data, and each front radar acquisition data contains an acquisition timestamp. Extract the data collected by the front radar sequentially from the front radar data set, and query the previous rainfall based on the collection timestamp in the extracted front radar data. The data collected by the preceding rainfall is labeled using the preceding rainfall data to obtain labeled radar data. The data collected by the marker radar is summarized to obtain the marker radar dataset; Randomly select from the labeled radar acquisition dataset to obtain the training radar acquisition dataset; The global rainfall prediction model is trained and updated based on the training radar dataset to obtain a local rainfall prediction model.

[0032] It should be explained that the "forward radar acquisition dataset" refers to a collection of data from multiple forward radar acquisitions, stored in the rainfall prediction unit. The forward radar acquisition data refers to radar data collected by a multi-band radar group in the radar prediction client within a certain period. This forward radar acquisition data includes radar observation parameters reflecting the intensity and motion of precipitation particles, such as radar reflectivity factor, radial velocity, spectral width, differential reflectivity, and differential phase. The acquisition timestamp refers to the time information when the forward radar acquisition data was acquired. The forward rainfall refers to the rainfall in the period corresponding to the acquisition timestamp, measured by a ground rain gauge. The labeled radar acquisition data refers to the labeled forward radar acquisition data. Labeling the forward radar acquisition data using the forward rainfall means using the forward rainfall as a label for the actual ground rainfall in the period corresponding to the forward radar acquisition data, thereby constructing samples for training subsequent local rainfall prediction models. The training radar acquisition data refers to a subset selected from the labeled radar acquisition dataset, such as by simple random sampling. The method of training and updating the global rainfall prediction model based on the training radar dataset is the same as the training method of the initial rainfall prediction model, and will not be repeated here. This training and updating process is executed by the rainfall prediction unit.

[0033] S4. Generate the local model layer sequence by accurately updating the model layer according to the local rainfall prediction model.

[0034] Understandably, the local model layer sequence refers to the ordered combination of each local model layer in the sorted local rainfall prediction model.

[0035] In detail, the step of generating a local model layer sequence by accurately updating the model layer according to the local rainfall prediction model includes: Obtain the set of model update parameter types, which includes multiple types of model update parameters; Perform the following operation for each model update parameter type in the model update parameter type set: Based on the types of model update parameters, local model parameters and global model parameters are extracted from the local rainfall prediction model and the global rainfall prediction model, respectively, and the local parameter deviation value is calculated using the local model parameters and global model parameters. Summarize the local parameter deviation values ​​corresponding to each type of updated parameter in the model to obtain a set of local parameter deviation values; Multiple local model layers were identified in the local rainfall prediction model, where the local model layers are either convolutional layers or fully connected layers; The importance of multiple local model layers is ranked based on the local parameter deviation value set to obtain a local model layer sequence, which contains multiple local model layers.

[0036] It should be explained that the "model update parameter type set" refers to a collection of multiple model update parameter types, where each type refers to a specific parameter in the global rainfall prediction model and the local rainfall prediction model that needs to be iteratively optimized. The "local model parameter" refers to the parameter in the local rainfall prediction model that corresponds to the model update parameter type, and the "global model parameter" refers to the parameter in the global rainfall prediction model that corresponds to the model update parameter type. The "local parameter deviation value" refers to the absolute difference between the local model parameter and the global model parameter. The "local model layer" refers to a complete neural network structure layer in the local rainfall prediction model, such as a convolutional layer or a fully connected layer, which contains multiple trainable parameters (such as parameters for the current layer).

[0037] Understandably, the purpose of the above-mentioned steps of ranking the importance of multiple local model layers is to determine the order in which each local model layer is updated in the subsequent model parameter iteration update steps. The ranking is based on the update urgency value, which reflects the urgency of the corresponding local model layer as a whole needing to be updated. By forming a sequence of local model layers through ranking, it can be ensured that when computing resources are limited, local model layers that differ more from the global rainfall prediction model and are more in urgent need of adjustment are updated first, thereby improving the efficiency of the entire iterative update process.

[0038] In detail, the step of ranking the importance of multiple local model layers based on the local parameter deviation value set to obtain a local model layer sequence includes: Perform the following operation on each of the multiple local model layers: Multiple current layer parameters of the local model layer are identified, and multiple current parameter deviation values ​​are identified in the local parameter deviation value set based on the multiple current layer parameters; An urgency analysis is performed based on multiple current parameter deviation values ​​to obtain an updated urgency value, which is expressed as:

[0039] in, This indicates that the urgent value needs to be updated. This indicates the number of current layer parameters out of multiple current layer parameters. Represents the hyperbolic sine function. This represents the first of multiple current parameter deviation values. One current parameter deviation value; By summing the update urgency values ​​corresponding to each local model layer, multiple update urgency values ​​are obtained; Multiple local model layers are sorted using multiple update urgency values ​​to obtain a sequence of local model layers. The local model layer with the larger update urgency value is placed earlier in the sequence of local model layers.

[0040] It should be explained that the "current layer parameter" refers to the parameters in the local model layer that need to be updated and optimized. The "current parameter deviation value" refers to the local parameter deviation value corresponding to the current layer parameter in the local parameter deviation value set. The "update urgency value" is a numerical value that quantifies the degree of deviation between the local model layer and the global rainfall prediction model. The larger the update urgency value, the greater the degree of deviation between the local model layer and the global rainfall prediction model, meaning that the corresponding local model layer needs to be updated more preferentially. In the above formula for calculating the update urgency value, since the hyperbolic sine function has an amplifying effect on larger input values ​​(i.e., the input current parameter deviation value), the update urgency value of the local model layer containing more large changes in the current layer parameters (i.e., larger current parameter deviation values) will be higher.

[0041] S5. Iteratively update the model parameters of the local model layer sequence to obtain the global parameter update dataset.

[0042] It is clear that the global parameter update dataset refers to the set of parameters obtained after iterative updating of model parameters, which represents the state of the global rainfall prediction model. This global parameter update dataset will be used to guide the aggregation of the global rainfall prediction model.

[0043] Specifically, the step of iteratively updating the model parameters of the local model layer sequence to obtain the global parameter update dataset includes: Local model layers are extracted sequentially from the local model layer sequence, and the extracted local model layers are denoted as the local model layers to be trained. Based on the local model layer to be trained, identify the set of update parameter types in the same layer from the set of model update parameter types; Search for the set of parameter deviation values ​​corresponding to the set of updated parameter types in the same layer within the local parameter deviation value set; Based on the parameter deviation value set of the same layer, the model update contribution analysis is performed on the same layer update parameter type set to obtain the parameter update contribution value set; The global parameter update dataset is obtained by iterating the parameters using the parameter update contribution value set and the local rainfall prediction model.

[0044] It should be explained that the set of same-layer update parameter types refers to the set of all model update parameter types contained in the local model layer to be trained. The set of same-layer parameter deviation values ​​refers to the set of local parameter deviation values ​​corresponding to each same-layer update parameter type in the set of same-layer update parameter types. The set of parameter update contribution values ​​refers to the set of multiple parameter update contribution values, where the parameter update contribution value refers to the numerical value that quantifies the importance weight of a certain same-layer update parameter type. The larger the parameter update contribution value, the more important the corresponding same-layer update parameter type is to the performance of the local model layer to be trained. That is, in this global parameter iteration, the same-layer update parameter type should be given a higher update priority.

[0045] In detail, the step of performing model update contribution analysis on the same-layer updated parameter type set based on the same-layer parameter deviation value set to obtain the parameter update contribution value set includes: The mean of the parameter deviation values ​​in the same layer is calculated to obtain the average parameter deviation value; Extract the same-level update parameter types sequentially from the same-level update parameter type set, and record the extracted same-level update parameter types as the parameter types to be evaluated; Identify the parameter deviation values ​​corresponding to the types of parameters to be evaluated within the same layer of parameter deviation value set; Based on the types of parameters to be evaluated, the local rainfall prediction model is trained and gradient queries are performed to obtain the local parameter gradients. The parameter update contribution value is calculated based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient. Summarize the parameter update contribution values ​​corresponding to each type of update parameter in the same layer to obtain the parameter update contribution value set.

[0046] It should be explained that the average parameter deviation value refers to the average of all parameter deviation values ​​in the same layer parameter deviation value set. The parameter deviation value to be evaluated refers to the parameter deviation value in the same layer parameter deviation value set corresponding to the parameter type to be evaluated. The local parameter gradient refers to the gradient value of the parameter corresponding to the parameter type to be evaluated in the local rainfall prediction model obtained during this model training process, relative to the local task loss function. The model training process refers to the steps described above of training and updating the global rainfall prediction model based on the training radar acquisition dataset to obtain the local rainfall prediction model. The mean squared error (MSE) can be selected as the local task loss function.

[0047] Specifically, the calculation of the parameter update contribution value based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient includes: The parameter update contribution value is calculated using the following formula:

[0048] in, This indicates that the parameter updates the contribution value. This represents a predefined symbolic function. Represents the gradient of local parameters. This indicates taking the absolute value. This represents the deviation value of the parameter to be evaluated. express function, This represents the average parameter deviation value.

[0049] Clearly, by calculating the parameter update contribution value, the importance of the type of parameter being evaluated to the update of the local rainfall prediction model can be assessed. This parameter update contribution value incorporates the update direction (i.e., Item), update range (i.e.) (item) and consistency with the parameter update behavior of other parameters in the same local model layer (i.e. Compared to traditional federated learning schemes where clients typically upload all or a random portion of parameter updates, this step filters parameters based on their contribution values, selecting only those with high contribution values ​​(i.e., the types of updated parameters) for subsequent training and uploading. This approach reduces the amount of data that needs to be uploaded, thereby improving the efficiency of global model aggregation. It also filters out minor or abnormal updates that may be caused by local data noise, making the subsequently uploaded global parameter update dataset more effective and thus improving the accuracy of subsequent global model aggregation.

[0050] Furthermore, in the calculation formula for the above parameter update contribution value, The term indicates the sign of the local parameter gradient; if the local parameter gradient is positive, then the term... The item is 1, otherwise the The item is -1, The purpose of this item is to preserve the directional information of parameter updates, ensuring that the update direction is correct when performing global model aggregation later. The term represents the absolute value of the deviation of the parameter to be evaluated. This term represents the magnitude of the change in the corresponding parameter's value during this local training; the greater the magnitude of the change, the larger the absolute value of this term. In the item, This represents the absolute difference between the deviation value of the parameter to be evaluated and the average deviation value of the parameter in that local model layer, divided by... The difference in the deviation values ​​of the parameters to be evaluated relative to all other parameters of the local model layer is obtained, and then... The function maps this relative difference to a weight. If the update magnitude of a certain parameter differs significantly from the average update level of the same local model layer (i.e., relative difference), then the weight is assigned. If the value of a parameter is relatively large, the weight is close to 0, thus suppressing the parameter update contribution value of that parameter; conversely, if the value of a parameter is relatively small, the weight is close to 1, thus preserving the parameter update contribution value of that parameter.

[0051] In detail, the process of using the parameter update contribution value set and the local rainfall prediction model to perform parameter iteration to obtain the global parameter update dataset includes: Based on the parameter update contribution value set and the same-layer update parameter type set, obtain the updatable parameter type set and the single-layer update tag set; The local rainfall prediction model is calibrated using an updatable parameter set to obtain the calibrated rainfall prediction model. A single-layer training dataset is obtained by randomly selecting from the labeled radar acquisition dataset. The calibration rainfall prediction model was iteratively trained using a single-layer training dataset to obtain a locally accurate update model. A model accurate update value set is generated based on the set of updatable parameter types in the global rainfall prediction model and the local accurate update model. The model accurate update value set includes multiple model accurate update values, and the model accurate update values ​​correspond one-to-one with the updatable parameter types in the set of updatable parameter types. The model's precise update value set and the single-layer update tag set are merged to obtain global parameter update data; The local accurate update model is used as the local rainfall prediction model, and the step of extracting local model layers sequentially in the local model layer sequence is returned until all local model layers in the local model layer sequence have been extracted. The global parameter update data is then aggregated to obtain the global parameter update dataset.

[0052] Understandably, the set of updatable parameter types refers to a collection of multiple updatable parameter types, where each updatable parameter type refers to a layer-level parameter type that can be numerically updated during subsequent iterative training. The set of single-layer update markers refers to a collection of multiple single-layer update markers, where each single-layer update marker is a Boolean value used to identify whether a layer-level parameter type at a corresponding position needs to be updated. This single-layer update marker set is used in subsequent model aggregation to determine which parameters have been updated (parameters with a Boolean value of 1 have been updated) and which parameters have not been updated (parameters with a Boolean value of 0 have not been updated). The calibrated rainfall prediction model refers to the local rainfall prediction model after being calibrated with updatable parameters. The specific method for calibrating the local rainfall prediction model using the set of updatable parameter types is as follows: traverse all current layer parameters in the corresponding local model layer of the local rainfall prediction model, check whether the parameter type corresponding to each current layer parameter exists in the set of updatable parameter types. If it exists, set the boolean value of the current layer parameter to True (i.e., the value 1), indicating that the current layer parameter needs to calculate the gradient and participate in optimization updates in subsequent training. If it does not exist, set the boolean value of the current layer parameter to False (i.e., the value 0), thereby freezing the current layer parameter and keeping it unchanged in subsequent training. After each current layer parameter has completed the above traversal, the local rainfall prediction model at this time is recorded as the calibrated rainfall prediction model.

[0053] It is clear that the single-layer training dataset refers to a randomly selected dataset used for iterative training of the calibration rainfall prediction model. The selection method for this single-layer training dataset is the same as that for the radar-collected training dataset, and will not be repeated here. The local precision update model refers to the calibration rainfall prediction model trained on the single-layer training dataset. The training method for this local precision update model is the same as that for the local rainfall prediction model, and will not be repeated here.

[0054] It should be explained that the model accurate update value set refers to a collection of multiple model accurate update values. Here, the model accurate update value refers to the magnitude of the change in the numerical amplitude of a certain updatable parameter type in the local accurate update model relative to the global rainfall prediction model. The model accurate update value set is generated as follows: Updatable parameter types are extracted sequentially from the set of updatable parameter types. Then, the global and local parameters corresponding to each updatable parameter type are identified in both the global rainfall prediction model and the local accurate update model. The rate of change of the local parameter relative to the global parameter (i.e., the ratio of the local parameter to the global parameter) is calculated. This rate of change is the model accurate update value. The model accurate update values ​​corresponding to each updatable parameter type are then summarized to obtain the model accurate update value set. The global parameter update data refers to the collection composed of the model accurate update value set and the single-layer update tag set.

[0055] Furthermore, this scheme achieves several beneficial effects compared to the conventional global indiscriminate training approach by training each local model layer sequentially and fixing a portion of the parameters in the local model layer to be trained (these parameters will not accept numerical updates). First, layer-by-layer training optimizes each local model layer in an orderly and focused manner according to the priority determined by the sequence of local model layers, avoiding mutual interference or optimization target conflicts that may occur when parameters of different local model layers are trained simultaneously, making the updates of each local model layer more complete and stable. Second, when training a local model layer, only the set of updatable parameter types selected based on parameter update contribution values ​​is trained. This ensures that the training process only continues to train those updatable parameter types that are determined to be important (i.e., those with high parameter update contribution values), while freezing those parameters with low parameter update contribution values, which may be redundant or unimportant. This method can significantly reduce the computational load of each iteration of training, improve training efficiency, and effectively prevent parameters with lower importance from drifting or overfitting during continued training, protecting the learned features from being destroyed, thereby improving the accuracy of local precise model updates.

[0056] Specifically, the step of obtaining the set of updatable parameter categories and the single-layer update tag set based on the parameter update contribution value set and the same-layer update parameter category set includes: Based on a preset update contribution threshold, the target contribution value set and the non-target contribution value set are identified in the parameter update contribution value set; In the same layer update parameter type set, the updatable parameter type set corresponding to the target contribution value set and the non-updatable parameter type set corresponding to the non-target contribution value set are respectively identified; Based on the set of updatable and non-updatable parameter types, co-positional markers are generated to obtain a single-layer update marker set.

[0057] It should be explained that the update contribution threshold refers to a dynamically set numerical limit. When the parameter update contribution value is greater than this threshold, it indicates that the parameter-degree global rainfall prediction model corresponding to that parameter update contribution value has a higher adjustment weight. The update contribution threshold is set by calculating the average and standard deviation of the parameter update contribution value set; the update contribution threshold is the average plus three times the standard deviation. The target contribution value set refers to a set of multiple target contribution values, where the index of the target contribution value is greater than the parameter update contribution value of the update contribution value. The non-target contribution value set refers to a set of multiple non-target contribution values, where the index of the non-target contribution value is not greater than the parameter update contribution value of the update contribution value. The non-updatable parameter type set refers to the set of update parameter types at the same level corresponding to each non-target contribution value in the non-target contribution value set. The generation of same-position markers based on the updatable and non-updatable parameter types refers to generating a binary marker sequence of the same length as the updatable parameter type set, following a fixed order of all updatable parameter types in the same-layer update parameter type set. In this binary marker sequence, the marker whose position matches that of an updatable parameter type is valued as 1, and the marker whose position matches that of a non-updatable parameter type is valued as 0. For example, suppose a layer has 4 parameters, with a fixed type order of [A1, A2, B1, A3]. After threshold filtering, the updatable parameter type set is determined to be {A1, B1}, and the non-updatable parameter type set is {A2, A3}. Therefore, the generated single-layer update marker set is [1, 0, 1, 0].

[0058] S6. Summarize the global parameter update datasets corresponding to each radar prediction client to obtain multiple global parameter update datasets. Transmit the multiple global parameter update datasets to the command server center to obtain the execution server center.

[0059] It is clear that the execution server center refers to the command server center that receives multiple global parameter update datasets, and the execution server center will update the global rainfall prediction model based on the multiple global parameter update datasets.

[0060] S7. Based on the execution server center and multiple global parameter update datasets, perform model aggregation on the global rainfall prediction model to obtain the current rainfall prediction model, and complete the rainfall prediction based on multi-band radar networking.

[0061] It should be explained that the current rainfall prediction model refers to the new prediction model obtained after aggregating the global rainfall prediction model. Specifically, the aggregation of the global rainfall prediction model is performed as follows: For each model layer in the global rainfall prediction model, the following operations are performed: Based on that model layer, the global parameter update data corresponding to that model layer (denoted as target parameter update data) is found in each of the multiple global parameter update datasets, resulting in multiple target parameter update data. Then, using aggregation methods such as weighted averaging based on data volume, the multiple precise update value sets of the multiple target parameter update data are weighted and summed to obtain an aggregated update value set. This aggregated update value set includes multiple... The model layer aggregates and updates values, with each aggregated update value corresponding to an original parameter in that model layer. The aggregated update value represents the rate of change of the original parameter. If the single-layer update flag corresponding to a precise update value of a model is 0, then that precise update value is not included in the weighted summation step. Finally, each original parameter in the model layer is transformed based on the aggregated update value set to obtain multiple updated parameters. Each updated parameter is the product of the original parameter and its corresponding aggregated update value. These multiple updated parameters are used to replace multiple original parameters at corresponding positions in the model layer, thus completing the replacement of that model layer. When each model layer in the global rainfall prediction model has been replaced, the current global rainfall prediction model is recorded as the current rainfall prediction model. This current rainfall prediction model will serve as the starting point for the next round of federated learning (i.e., the global rainfall prediction model) and can be directly deployed at the command server center. The command server center inputs radar data collected by various multi-band radar groups into this current rainfall prediction model to obtain the rainfall within a certain future time period, thereby completing the rainfall prediction.

[0062] To address the problems described in the background, this invention first generates a precise update sequence for model layers based on a local rainfall prediction model, resulting in a local model layer sequence. This step calculates the deviation of parameters within each model layer from the global model and uses a hyperbolic sine function to calculate the update urgency value, thereby ranking the importance of each model layer. Compared to existing technologies that indiscriminately process or randomly update all model layers, this method can identify the model layers with the largest deviation from the global model and the most urgent need for adjustment. This provides a basis for the efficient allocation of subsequent computing resources, ensuring that optimization can focus on key parts when resources are limited, thus improving the efficiency of the overall training process. Furthermore, this solution iteratively updates the model parameters of the local model layer sequence to obtain a global parameter update dataset. This step uses a formula that comprehensively considers the contribution value of parameter update direction, magnitude, and consistency to select updatable parameters that are critical to model performance, rather than uploading all parameters. Compared to the existing federated learning approach where the client uploads all or a random portion of the model parameters, this significantly reduces the amount of data that needs to be uploaded, improving the efficiency of the model aggregation stage. Secondly, this filtering mechanism effectively filters out minor or abnormal parameter updates caused by local data noise, resulting in higher quality and more effective updated datasets uploaded to the server, thereby improving the accuracy and stability of subsequent global model aggregation. Therefore, this invention can improve the effectiveness and training efficiency of rainfall prediction model updates while reducing interference from invalid communication and noisy updates.

[0063] like Figure 2 The diagram shown is a functional block diagram of a rainfall prediction system based on multi-band radar networking provided in an embodiment of the present invention.

[0064] The rainfall prediction system 100 based on multi-band radar networking described in this invention can be installed in an electronic device. Depending on the functions implemented, the rainfall prediction system 100 based on multi-band radar networking may include a prediction environment verification module 101, a local model update module 102, a model layer sequence generation module 103, and a multi-source model aggregation module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and which are stored in the memory of the electronic device. The prediction environment confirmation module 101 is used to confirm the radar network prediction environment, wherein the radar network prediction environment includes multiple radar control clients and a command server center, wherein each radar control client contains a multi-band radar group and a rainfall prediction unit. The local model update module 102 is used to generate a global rainfall prediction model based on the command server center, send the global rainfall prediction model to multiple radar control clients to obtain multiple radar prediction clients, extract radar prediction clients sequentially from the multiple radar prediction clients, perform preliminary rainfall model update based on the extracted radar prediction clients, and obtain a local rainfall prediction model. The model layer sequence generation module 103 is used to generate a local model layer sequence by generating a precise update order of the model layer according to the local rainfall prediction model, and to perform iterative update of the model parameters on the local model layer sequence to obtain a global parameter update dataset. The multi-source model aggregation module 104 is used to summarize the global parameter update datasets corresponding to each radar prediction client to obtain multiple global parameter update datasets. The multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and the multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model.

[0065] In detail, the modules in the rainfall prediction system 100 based on multi-band radar networking described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the rainfall prediction method based on multi-band radar networking described in the article, and can produce the same technical effect, so it will not be repeated here.

[0066] like Figure 3 The diagram shown is a schematic representation of an electronic device for implementing a rainfall prediction method based on multi-band radar networking, according to an embodiment of the present invention.

[0067] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a rainfall prediction method program based on multi-band radar networking.

[0068] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a rainfall prediction method program based on multi-band radar networking, but also to temporarily store data that has been output or will be output.

[0069] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a rainfall prediction method program based on multi-band radar networking) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0070] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0071] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0072] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0073] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0074] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0075] The rainfall prediction method program based on multi-band radar networking, stored in the memory 11 of the electronic device 1, is a combination of multiple instructions. When run in the processor 10, it can achieve the following: The radar network prediction environment has been identified. This environment includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. A global rainfall prediction model is generated based on the command server center, and then sent to multiple radar control clients to obtain multiple radar prediction clients. The radar prediction clients are extracted sequentially from multiple radar prediction clients. Based on the extracted radar prediction clients, the preliminary rainfall model is updated to obtain the local rainfall prediction model. Based on the local rainfall prediction model, the model layer is accurately updated in the correct order to obtain the local model layer sequence. The model parameters are iteratively updated on the local model layer sequence to obtain the global parameter update dataset; The global parameter update datasets corresponding to each radar prediction client are aggregated to obtain multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model, thus completing the rainfall prediction based on multi-band radar networking.

[0076] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0077] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0078] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following: The radar network prediction environment has been identified. This environment includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. A global rainfall prediction model is generated based on the command server center, and then sent to multiple radar control clients to obtain multiple radar prediction clients. The radar prediction clients are extracted sequentially from multiple radar prediction clients. Based on the extracted radar prediction clients, the preliminary rainfall model is updated to obtain the local rainfall prediction model. Based on the local rainfall prediction model, the model layer is accurately updated in the correct order to obtain the local model layer sequence. The model parameters are iteratively updated on the local model layer sequence to obtain the global parameter update dataset; The global parameter update datasets corresponding to each radar prediction client are aggregated to obtain multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model, thus completing the rainfall prediction based on multi-band radar networking.

[0079] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0080] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, 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 modules.

[0082] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for rainfall prediction based on multi-band radar networking, characterized in that, The method includes: The radar network prediction environment has been identified. This environment includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. A global rainfall prediction model is generated based on the command server center, and then sent to multiple radar control clients to obtain multiple radar prediction clients. The radar prediction client is extracted sequentially from multiple radar prediction clients, and the preliminary rainfall model is updated based on the extracted radar prediction client to obtain the local rainfall prediction model. Based on the local rainfall prediction model, the model layer is accurately updated in the correct order to obtain the local model layer sequence. The model parameters are iteratively updated on the local model layer sequence to obtain the global parameter update dataset; The global parameter update datasets corresponding to each radar prediction client are aggregated to obtain multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model, thus completing the rainfall prediction based on multi-band radar networking. 2.The multi-band radar network-based rainfall prediction method of claim 1, wherein, The preliminary rainfall model update based on the extracted radar prediction client is used to obtain a local rainfall prediction model, including: The rainfall prediction unit extracted from the radar prediction client is used to obtain the front radar acquisition dataset, which includes multiple front radar acquisition data, and each front radar acquisition data contains an acquisition timestamp. Extract the data collected by the front radar sequentially from the front radar data set, and query the previous rainfall based on the collection timestamp in the extracted front radar data. The data collected by the preceding rainfall is labeled to obtain labeled radar data. The data collected by the marker radar is summarized to obtain the marker radar dataset; Randomly select from the labeled radar acquisition dataset to obtain the training radar acquisition dataset; The global rainfall prediction model is trained and updated based on the training radar dataset to obtain a local rainfall prediction model. 3.The method of claim 2, wherein, The process of generating a local model layer sequence based on the accurate update order of the local rainfall prediction model includes: Obtain the set of model update parameter types, which includes multiple types of model update parameters; Perform the following operation for each model update parameter type in the model update parameter type set: Based on the types of model update parameters, local model parameters and global model parameters are extracted from the local rainfall prediction model and the global rainfall prediction model, respectively, and the local parameter deviation value is calculated using the local model parameters and global model parameters. Summarize the local parameter deviation values ​​corresponding to each type of updated parameter in the model to obtain a set of local parameter deviation values; Multiple local model layers were identified in the local rainfall prediction model, where the local model layers are either convolutional layers or fully connected layers; The importance of multiple local model layers is ranked based on the local parameter deviation value set to obtain a local model layer sequence, which contains multiple local model layers. 4.The method of claim 3, wherein, The method of ranking the importance of multiple local model layers based on the local parameter deviation value set to obtain a sequence of local model layers includes: Perform the following operation on each of the multiple local model layers: Multiple current layer parameters of the local model layer are identified, and multiple current parameter deviation values ​​are identified in the local parameter deviation value set based on the multiple current layer parameters; An urgency analysis is performed based on multiple current parameter deviation values ​​to obtain an updated urgency value, which is expressed as: ； wherein, represents an update urgency value, represents a number of current layer parameters in the plurality of current layer parameters, represents a hyperbolic sine function, represents an i-th current parameter bias value in the plurality of current parameter bias values; and represents an i-th current parameter bias value in the plurality of current parameter bias values. By summing the update urgency values ​​corresponding to each local model layer, multiple update urgency values ​​are obtained; Multiple local model layers are sorted using multiple update urgency values ​​to obtain a sequence of local model layers. The local model layer with the larger update urgency value is placed earlier in the sequence of local model layers. 5.The method of claim 4, wherein, The step of iteratively updating the model parameters of the local model layer sequence to obtain the global parameter update dataset includes: Local model layers are extracted sequentially from the local model layer sequence, and the extracted local model layers are denoted as the local model layers to be trained. Based on the local model layer to be trained, identify the set of update parameter types in the same layer from the set of model update parameter types; Search for the set of parameter deviation values ​​corresponding to the set of updated parameter types in the same layer within the local parameter deviation value set; Based on the parameter deviation value set of the same layer, the model update contribution analysis is performed on the same layer update parameter type set to obtain the parameter update contribution value set; The global parameter update dataset is obtained by iterating the parameters using the parameter update contribution value set and the local rainfall prediction model.

6. The rainfall prediction method based on multi-band radar networking as described in claim 5, characterized in that, The method of performing model update contribution analysis on the same-layer parameter type set based on the same-layer parameter deviation value set yields a parameter update contribution value set, including: The mean of the parameter deviation values ​​in the same layer is calculated to obtain the average parameter deviation value; Extract the same-level update parameter types sequentially from the same-level update parameter type set, and record the extracted same-level update parameter types as the parameter types to be evaluated; Identify the parameter deviation values ​​corresponding to the types of parameters to be evaluated within the same layer of parameter deviation value set; Based on the types of parameters to be evaluated, the local rainfall prediction model is trained and gradient queries are performed to obtain the local parameter gradients. The parameter update contribution value is calculated based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient. Summarize the parameter update contribution values ​​corresponding to each type of update parameter in the same layer to obtain the parameter update contribution value set.

7. The rainfall prediction method based on multi-band radar networking as described in claim 6, characterized in that, The calculation of parameter update contribution based on the deviation value of the parameter to be evaluated, the average parameter deviation value, and the local parameter gradient includes: The parameter update contribution value is calculated using the following formula: ； in, This indicates that the parameter updates the contribution value. This represents the predefined symbolic function. Represents the gradient of local parameters. This indicates taking the absolute value. This represents the deviation value of the parameter to be evaluated. express function, This represents the average parameter deviation value.

8. The rainfall prediction method based on multi-band radar networking as described in claim 7, characterized in that, The parameter iteration, which utilizes the parameter update contribution value set and the local rainfall prediction model, yields a global parameter update dataset, including: Based on the parameter update contribution value set and the same-layer update parameter type set, obtain the updatable parameter type set and the single-layer update tag set; The local rainfall prediction model is calibrated using an updatable parameter set to obtain the calibrated rainfall prediction model. A single-layer training dataset is obtained by randomly selecting from the labeled radar acquisition dataset. The calibration rainfall prediction model was iteratively trained using a single-layer training dataset to obtain a locally accurate update model. A model accurate update value set is generated based on the set of updatable parameter types in the global rainfall prediction model and the local accurate update model. The model accurate update value set includes multiple model accurate update values, and the model accurate update values ​​correspond one-to-one with the updatable parameter types in the set of updatable parameter types. The model's precise update value set and the single-layer update tag set are merged to obtain global parameter update data; The local accurate update model is used as the local rainfall prediction model, and the step of extracting local model layers sequentially in the local model layer sequence is returned until all local model layers in the local model layer sequence have been extracted. The global parameter update data is then aggregated to obtain the global parameter update dataset.

9. The rainfall prediction method based on multi-band radar networking as described in claim 8, characterized in that, The process of obtaining the set of updatable parameter categories and the set of single-layer update tags based on the parameter update contribution value set and the same-layer update parameter category set includes: Based on a preset update contribution threshold, the target contribution value set and the non-target contribution value set are identified in the parameter update contribution value set; In the same layer update parameter type set, the updatable parameter type set corresponding to the target contribution value set and the non-updatable parameter type set corresponding to the non-target contribution value set are respectively identified; Based on the set of updatable and non-updatable parameter types, co-positional markers are generated to obtain a single-layer update marker set.

10. A rainfall prediction system based on multi-band radar networking, characterized in that, The system includes: The prediction environment confirmation module is used to confirm the radar network prediction environment, which includes multiple radar control clients and a command server center. Each radar control client contains a multi-band radar group and a rainfall prediction unit. The local model update module is used to generate a global rainfall prediction model based on the command server center, send the global rainfall prediction model to multiple radar control clients to obtain multiple radar prediction clients, extract radar prediction clients sequentially from multiple radar prediction clients, and perform preliminary rainfall model update based on the extracted radar prediction clients to obtain a local rainfall prediction model. The model layer sequence generation module is used to generate a precise update sequence of model layers based on the local rainfall prediction model, obtain a local model layer sequence, and perform iterative updates of model parameters on the local model layer sequence to obtain a global parameter update dataset. The multi-source model aggregation module is used to summarize the global parameter update datasets corresponding to each radar prediction client, resulting in multiple global parameter update datasets. These multiple global parameter update datasets are then transmitted to the command server center to obtain the execution server center. Based on the execution server center and the multiple global parameter update datasets, the global rainfall prediction model is aggregated to obtain the current rainfall prediction model.

Images