A meteorological monitoring method and system based on a GRU neural network and differential privacy federated learning

Abstract

The application provides a meteorological monitoring method based on a GRU neural network and differential privacy federated learning, and comprises the following steps: collecting local meteorological data to generate meteorological time series data sequences; selecting a suitable model structure according to the heterogeneity of the client itself, performing local meteorological monitoring model training based on a GRU neural network, and updating the parameters of the local model through an LPDP-FedProx federated learning algorithm; after the central server obtains the model parameters of the client, performing model aggregation and updating according to a global model weighted average aggregation algorithm; the client receives the updated global model parameters to perform the next round of local training until the best monitoring model is obtained. The application combines the personalized differential privacy algorithm with the FedProx federated learning algorithm, proposes an LPDP-FedProx federated learning algorithm, and constructs a federated learning framework capable of effectively processing heterogeneous data, effectively protecting the privacy of local data, and improving the security of federated learning.

Description

Technical Field

[0001] This invention relates to the field of meteorological monitoring technology, specifically but not limited to a meteorological monitoring method and system based on GRU neural network and differential privacy federated learning. Background Technology

[0002] Meteorological monitoring is an important field of atmospheric science. Meteorological monitoring refers to the activities of meteorological monitoring agencies in comprehensively monitoring and issuing early warnings about meteorological environmental conditions through meteorological monitoring systems. These systems monitor and report indicators reflecting meteorological quality to determine local meteorological environmental data such as rainfall, wind speed, and wind direction. The core objective of meteorological monitoring is to provide data on the current status and changing trends of meteorological elements, predict meteorological changes, effectively address current major meteorological problems, and serve meteorological management. With the development of microelectronics, computer technology, satellite technology, and materials science, many technologies have been applied to the automation of meteorological observation, such as in modern agricultural development, forest fire prevention, highway environmental monitoring, and school environmental monitoring, greatly improving the detection accuracy and maintainability of meteorological monitoring. However, with the continuous development of meteorological monitoring systems, the traditional centralized data processing model is not flexible and efficient enough in dealing with the challenges of multi-regional, multi-device, and multi-modal data. Furthermore, traditional meteorological monitoring systems often face problems such as data silos, privacy protection, and limited computing resources. On the one hand, for more complex weather events, some existing meteorological monitoring systems can only monitor data in a single mode; on the other hand, meteorological monitoring involves a large amount of meteorological data and information sharing, making the security and privacy of this data a significant challenge. If the security issues of meteorological monitoring systems are not resolved, it could lead to problems such as data leakage and tampering at weather stations. Meteorological monitoring systems need more intelligent, secure, and adaptable solutions to better cope with the challenges of weather changes and complex weather events.

[0003] Current common meteorological monitoring systems, such as precipitation monitoring, wind detection, and humidity monitoring, only utilize partial characteristics of the current regional weather environment to monitor single events. This results in the waste of a large number of effective features. Monitoring these meteorological events separately wastes significant space and time resources, and meteorological monitoring equipment in different regions needs to transmit these monitored data to a central server, inevitably leading to latency and privacy issues, and increasing communication overhead, thus reducing the efficiency of meteorological data processing. Compared with traditional meteorological monitoring methods, federated learning can better correlate the monitoring of various meteorological events, improving the model's generalization ability and accuracy, and has been widely used in many fields. As an emerging big data-driven technology, federated learning can fully utilize meteorological data collected by different devices, train meteorological monitoring models more efficiently, and more accurately monitor and predict different meteorological events in different regions. Among them, the Federated Avg algorithm provides a new solution for diverse meteorological monitoring tasks. Traditional meteorological monitoring methods may involve centralizing data to a central server for processing. In contrast, the federated averaging algorithm allows data to be stored on local meteorological equipment, reducing reliance on data centers and helping to handle large-scale and distributed meteorological data. Furthermore, the algorithm can adapt to different regional meteorological data distributions because model training takes place on local meteorological equipment, each capable of local training on its own data. However, the federated averaging algorithm is relatively weak at handling heterogeneity between equipment. In some cases, differences in performance and data distribution between equipment can lead to instability in the global model.

[0004] In view of this, a new method is needed to solve at least some of the above problems. Summary of the Invention

[0005] To address one or more problems in existing technologies, this invention proposes a meteorological monitoring method based on GRU neural networks and differential privacy federated learning. By combining personalized differential privacy algorithms with FedProx federated learning algorithms, the LPDP-FedProx federated learning algorithm is proposed, constructing a federated learning framework that can effectively handle heterogeneous data, effectively protect the privacy of local data, and improve the security of federated learning.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A meteorological monitoring method based on GRU neural network and differential privacy federated learning includes:

[0008] S1. Preprocessing: The client collects local meteorological data and performs preprocessing, generating a meteorological time series data sequence based on the preprocessed data;

[0009] S2. Initialization: Each client obtains the global model initialization parameters from the central server, and each device selects a suitable model structure based on its own heterogeneity.

[0010] S3. Local Training: Based on the generated meteorological time-series data, train a local meteorological monitoring model based on a GRU neural network.

[0011] S4. Local Model Update: After local training is completed, the client updates the parameters of the local model using the LPDP-FedProx federated learning algorithm and uploads the updated model parameters to the central server through the communication module.

[0012] S5. Global Model Aggregation: After receiving model parameters from multiple clients, the central server aggregates the models according to the global model weighted average aggregation algorithm.

[0013] S6. Update Feedback: The central server feeds back the updated global model parameters to each client. After receiving the feedback, the client performs the next round of local training until the best monitoring model is obtained.

[0014] S7. Meteorological Monitoring Visualization: Utilize a global model to monitor and issue early warnings for meteorological parameters, and visualize the parameters and early warning information.

[0015] Furthermore, the meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention includes preprocessing local meteorological data in S1, including data acquisition, filtering, cleaning and enhancement. Among them, data filtering includes encrypting the data using homomorphic encryption technology, and data cleaning includes estimating missing values, resolving conflicts and standardizing data in local meteorological equipment.

[0016] Furthermore, in the meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention, S3, the local meteorological monitoring model training bureau based on GRU neural network includes:

[0017] S3-1. Input the preprocessed meteorological data sequence into the GRU neural network. Let x be the input at time step t. t The gating signal r is calculated. t and z t The calculation formula is as follows:

[0018] r t =σ(W r ·[h t-1 ,x t ])

[0019] z t =σ(W z ·[ht-1 ,x t ])

[0020] Where σ(·) represents the sigmoid function, which compresses the result to between 0 and 1, W r W represents the reset gate weight matrix. z This represents the update gate weight matrix. These two weight matrices are used to perform a linear transformation between the input of the current time step and the hidden state of the previous time step. [] indicates that the two vectors are concatenated. h t-1 This indicates the output of the previous time step;

[0021] S3-2, Calculate candidate hidden layers The formula is as follows:

[0022]

[0023] Where tanh(·) represents the tanh activation function, which scales the activation result to the range between -1 and 1 to obtain the candidate hidden layer states. This is the weight matrix, used to linearly transform the output of the reset gate and the input of the current time step; * represents matrix multiplication.

[0024] S3-3、z t Determine the hidden layer h from the previous time step t-1 How much information is forgotten and how much hidden layer information of the current time step is added? h was calculated t The formula is as follows:

[0025]

[0026] Among them, h t This is the final output of the GRU neural network.

[0027] Furthermore, in the meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention, model aggregation in S5 specifically includes:

[0028] S5-1. In the first round of training, the central server selects a subset of clients {C1, C2…C…} k} as a participant and initialize the global model Then the global model This information is sent to each client via the base station, serving as the client's local GRU model. The initialization model, and then each client C k The LPDP-FedProx algorithm is used to guide parameter updates of the local model, utilizing a private dataset D. kThe local meteorological monitoring model is trained to generate an updated local model. The update formula is as follows:

[0029]

[0030] Where α represents the learning rate, This indicates the calculation of the gradient. Describe the loss function for each client k;

[0031] S5-2. The central server calculates the aggregated values ​​of the updated client model parameters to complete model aggregation. The model aggregation formula is as follows:

[0032]

[0033] Where, N k This represents the amount of data (dataset) from the k-th client, and N represents the total amount of data from the k clients.

[0034] S5-3, The central server updates the global model for the next round based on the aggregation results:

[0035]

[0036] S5-4. Perform a performance evaluation on the updated global model, and iterate and update the model based on the evaluation results to continuously optimize the global model.

[0037] Furthermore, in the meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention, the specific steps of updating the parameters of the local model using the LPDP-FedProx federated learning algorithm in S4 include:

[0038] S4-1. Define a sampling threshold to divide the original dataset into an unsampled part and a sampled part;

[0039] Users in the unsampled segment will have their privacy budget wasted because their data is not used, thus leading to sampling errors:

[0040]

[0041] If a user's privacy budget in the sampling phase exceeds a threshold, the user's data will be protected against noise exceeding privacy requirements, resulting in noise error.

[0042]

[0043] in, It's the user's privacy budget. Let τ be the privacy budget for the i-th user, and τ be the sampling threshold.

[0044] S4-2. The sampling threshold is determined by using the minimum value of the loss function, and the weights of sampling error and noise error are adaptively modified. The loss function is defined as:

[0045]

[0046] For a given Scope When the threshold value τ minimizes the waste function and At that time, this value is the optimal sampling threshold.

[0047] S4-3. The client data is sampled using the optimal sampling threshold calculated in each round. This process is repeated multiple times until the remaining value of the privacy budget is less than a given value. This sampling method is used to perform secondary sampling on the original dataset, minimizing the error caused by random sampling and noise to the algorithm.

[0048] S4-4. Train the local GRU model using the subsampled data and update the local model parameters.

[0049]

[0050] Where, h k It is the local objective function, F k (θ) is the local loss function on the k-th device, and μ is the regularization parameter.

[0051] A meteorological monitoring system based on GRU neural networks and differential privacy federated learning includes:

[0052] Local meteorological equipment with built-in GRU neural network is used to collect and preprocess meteorological data locally, train local meteorological monitoring models, and update model parameters based on LPDP-FedProx federated learning algorithm;

[0053] The central server is used to aggregate local meteorological monitoring model parameters and update the global model based on the LPDP-FedProx federated learning algorithm, and then distribute the updated model parameters to each client.

[0054] The communication module is used to transmit model parameters between local meteorological equipment and the central server.

[0055] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0056] 1. The meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention allows local meteorological equipment to train its own model, and only transmit the model and updated parameters to the central server. Differential privacy technology is used to process the original data and model parameters, which ensures that local data can be effectively protected by privacy and improves the security of federated learning.

[0057] 2. The meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention allows model training on local meteorological equipment without transmitting raw data to a central server, enabling faster adaptation to new data and situations and improving the real-time performance and efficiency of the system.

[0058] 3. The meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention only needs to transmit a small number of model parameters to the central server, which reduces data transmission, reduces the burden on central processing, and improves the overall computing speed.

[0059] 4. The meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention introduces GRU neural network when training the model on local equipment, which reduces the gradient vanishing problem generated during model training. At the same time, since GRU has fewer parameters and a simpler structure, it reduces memory occupation and consumption, making the training of local models simpler and more efficient. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and, together with the description, serve to explain embodiments of the invention, but do not constitute a limitation thereof. In the drawings:

[0061] Figure 1 A flowchart of the meteorological monitoring method based on GRU neural network and differential privacy federated learning of the present invention is shown.

[0062] Figure 2 The schematic diagram of the GRU neural network of the present invention is shown.

[0063] Figure 3 A flowchart of the differential privacy federated learning algorithm of the present invention is shown. Detailed Implementation

[0064] To further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, and are not intended to limit the scope of the claims of the present invention.

[0065] The description in this section pertains only to typical embodiments, and the present invention is not limited to the scope of the embodiments described. Combinations of different embodiments, substitution of some technical features in different embodiments, and substitution of similar or identical prior art with some technical features in the embodiments are also within the scope of the description and protection of the present invention.

[0066] Currently, meteorological monitoring typically involves individual meteorological stations collecting local meteorological data to train meteorological monitoring and forecasting models. However, using limited local meteorological data is insufficient to train a high-precision model, and the resulting model often exhibits poor generalization ability and is not applicable to meteorological monitoring at other stations. Therefore, a cloud-edge collaborative training model can be introduced to address this issue. This involves building a central server in the cloud and uploading meteorological data collected by each station. This provides the cloud with a large amount of labeled meteorological data, allowing for centralized training of a high-precision and highly generalizable meteorological monitoring model, leading to more accurate meteorological monitoring. However, this strategy faces data security challenges. Due to the privacy inherent in meteorological data, stations cannot fully trust each other, making data sharing difficult and creating data silos, thus preventing the full utilization of a large amount of data.

[0067] To solve this problem, such as Figure 3 This invention introduces a federated learning framework. Each weather station collects local meteorological data and independently trains a meteorological monitoring model. The model parameters are then uploaded to a central server. The central server aggregates and updates the model using the LPDP-FedProx algorithm and sends the updated parameters back to each weather station. This process iterates several times to obtain an optimal monitoring model. During this process, the client only uploads model parameters, not the raw meteorological data, thus effectively protecting user privacy while ensuring full utilization of the large amount of data.

[0068] Figure 1 This paper presents an overall flowchart of a meteorological monitoring method based on GRU neural network and differential privacy federated learning, according to the present invention. It mainly includes: data preprocessing, local model training based on GRU neural network, and federated learning derivation process based on LPDP-FedProx. Specific steps are as follows:

[0069] S1. Data Acquisition and Preprocessing: This process mainly includes data acquisition, filtering, cleaning, and enhancement operations. The purpose of these operations is to mitigate statistical heterogeneity. When using federated learning for meteorological monitoring, many heterogeneity problems arise, leading to varying degrees of performance degradation in the global model. These heterogeneity problems can be broadly categorized into four types: statistical heterogeneity, model heterogeneity, communication heterogeneity, and device heterogeneity. Statistical heterogeneity refers to the inconsistent data distribution of edge devices in federated learning and the lack of adherence to identical sampling; that is, the data is non-independent and identically distributed (Non-IID). Due to differences in local data distribution at various meteorological stations, for example, coastal areas may focus more on tidal and wave data, while inland areas may focus more on soil moisture data. Furthermore, the local optimization objectives of edge devices are inconsistent with the global optimization objectives. Therefore, statistical heterogeneity may cause local models to converge in different directions, reaching local optima rather than global optima, thus reducing the performance of federated learning.

[0070] In federated learning, the quantity, quality, and diversity of data collection determine how much useful information edge devices provide to the federated learning system. Therefore, local data collection is a crucial part of federated learning that needs optimization. This invention therefore considers various data-related factors (error rate, classification distribution, content diversity, and data size) that affect model performance to measure the quality of the sample data.

[0071] Data filtering is the process of removing or excluding irrelevant or malicious data from a dataset. This invention utilizes homomorphic encryption to encrypt data, allowing computational operations to be performed while the data is encrypted without decrypting the ciphertext. Through homomorphic encryption filtering, various computational operations can be performed on the data while protecting data privacy.

[0072] Data cleaning is the process of correcting or improving incomplete, inaccurate, or inconsistent data in federated learning. It typically involves applying relevant techniques locally on the client side to estimate missing values, resolve conflicts, and standardize data.

[0073] Data augmentation is a technique that artificially expands the training dataset by generating more data from the limited original data. It can effectively alleviate the data shortage problem in federated learning. For example, due to limited technological development, existing meteorological data in polar regions and the deep sea may be insufficient. In such cases, data augmentation methods can be used to increase the amount of data.

[0074] The above data preprocessing operations can improve the quality and usability of the dataset, while reducing statistical heterogeneity, making the dataset more suitable for model training.

[0075] S2. Training of a Local Meteorological Monitoring Model Based on a GRU Neural Network: Considering the strong spatiotemporal correlation and long-term dependencies in meteorological data, and the good performance and effectiveness of GRU neural networks in processing this type of data, this invention uses a GRU neural network to train the local model. Figure 2 The diagram shown illustrates the basic principle of a GRU neural network. In the diagram, z represents the update gate and r represents the reset gate. represents the candidate hidden layer, and h represents the hidden layer information of the final output.

[0076] GRU neural networks are a variant of LSTM (Long Short-Term Memory) networks. GRU combines the input and forget gates of LSTM into an update gate, and the hidden layers and memory units into a reset gate. The update gate determines how much previous memory information is retained in the current time step, while the reset gate determines how current information is combined with previous memory information. While retaining the advantages of LSTM's gating mechanism and memory units, GRU neural networks improve computational efficiency and reduce gradient explosion by simplifying the structure and optimizing parameters, thus improving the training efficiency and performance of the model to some extent.

[0077] The training process of the local meteorological monitoring model based on the GRU neural network is as follows: The preprocessed meteorological data sequence is input into the GRU neural network, and the input at the t-th time step is denoted as x. t The gating signal r is calculated. t and z t The calculation formula is as follows:

[0078] r t =σ(W r ·[h t-1 ,x t ])

[0079] z t =σ(W z ·[h t-1 ,x t ])

[0080] Where σ(·) represents the sigmoid function, which compresses the result to between 0 and 1, W r W represents the reset gate weight matrix. z This represents the update gate weight matrix. These two weight matrices are used to perform a linear transformation between the input of the current time step and the hidden state of the previous time step. [] indicates that the two vectors are concatenated. h t-1 This indicates the output of the previous time step.

[0081] Before computing the candidate hidden layers, this invention designs a shared hidden layer to meet the needs of multi-task learning. The shared hidden layer allows different tasks to share the model's low-level features. This enables the model to learn a general feature representation, thus better adapting to the needs of different tasks and improving data utilization efficiency. Furthermore, the shared hidden layer allows multiple tasks to share the same model parameters, thereby reducing the number of model parameters, lowering model complexity, and enabling better training of the model on limited data, preventing overfitting.

[0082] Then calculate the candidate hidden layers. The formula is as follows:

[0083]

[0084] Where tanh(·) represents the tanh activation function, which scales the activation result to the range between -1 and 1 to obtain the candidate hidden layer states. This is the weight matrix, used to linearly transform the output of the reset gate and the input of the current time step. * represents matrix multiplication.

[0085] Finally z t The decision needs to be made from the hidden layer h of the previous time step. t-1 How much information is forgotten, and how much hidden layer information from the current time step needs to be added? h was calculated t The formula is as follows:

[0086]

[0087] h t This is the final output of the GRU neural network.

[0088] S3. Federated learning deductive process based on LPDP-FedProx: Figure 3 This paper presents a general framework diagram of a federated learning-based meteorological monitoring method. As shown, the architecture consists of three layers: a cloud layer, a fog layer, and an IoT layer. The cloud layer is a central server that performs global model aggregation and updates, possessing high computing and storage capabilities. The fog layer comprises multiple edge servers (base stations, etc.) and is responsible for communication between the cloud layer and the IoT layer, i.e., communication between the central server and meteorological station equipment in various locations. The IoT layer consists of edge devices (sensors and computers at various meteorological stations, etc.), responsible for local model training and transmitting model parameters to the central server via the fog layer. Compared to the traditional two-layer federated learning architecture, this invention introduces an intermediate layer, the fog layer, which performs the model aggregation step, significantly alleviating the communication pressure between the cloud layer and the IoT layer.

[0089] First, in the first round of training, the central server selects the local equipment {C1, C2…C…} of each weather station. k} as a participant and initialize the global model Then the global model The data is transmitted from the base station through the fog layer to each edge device as a local GRU model. Initialization, followed by each local device C k Using private dataset D k Perform the following local model update:

[0090]

[0091] Where α represents the learning rate, This indicates the calculation of the gradient. Let represent the loss function for each local device k.

[0092] Next, the central server calculates the aggregated values ​​of the updated model parameters for the edge devices, thus completing the model aggregation. The model aggregation formula is as follows:

[0093]

[0094] Where, N k This represents the amount of data (dataset) on the k-th local device, and N represents the total amount of data sets on the k local devices.

[0095] Then, the central server updates the global model for the next round based on the aggregation results:

[0096]

[0097] Finally, the updated global model is evaluated for performance, such as the accuracy of meteorological monitoring or forecasting. Based on the evaluation results, the model is iteratively updated to continuously optimize the global model to adapt to complex meteorological events.

[0098] Specifically, during the local model update process, this invention uses the FedProx algorithm as the baseline algorithm and incorporates differential privacy technology into it, proposing the LPDP-FedProx algorithm. FedProx adds a proximal term to FedAvg to address the problems caused by device heterogeneity and data heterogeneity. FedProx considers the differences in communication and computing capabilities among different devices; different devices do not need to be trained for the same number of rounds, and the local model does not need to be trained with extreme precision, but only needs to obtain an imprecise solution, thus effectively mitigating the device heterogeneity problem. To make the FedProx model converge faster for higher efficiency, while better ensuring data security and saving communication overhead, this invention proposes a personalized local differential privacy method, LPDP, and uses LPDP to improve FedProx.

[0099] LPDP is primarily implemented through sampling and the addition of noise. Random sampling and random noise introduce sampling error and noise error, respectively. Sampling error refers to the impact of random sampling on the training performance of the original dataset, while noise error refers to the degree to which different privacy requirements of clients perturb the training performance. By defining a sampling threshold, the original dataset is divided into an unsampled portion and a sampled portion. Users in the unsampled portion waste their privacy budget because their data is not used, thus leading to sampling error. Conversely, if the privacy budget of users in the sampled portion exceeds the threshold, their data will be protected by noise exceeding privacy requirements, resulting in noise error. Sampling error and noise error are defined as follows:

[0100]

[0101] in, It's the user's privacy budget. Let τ be the privacy budget for the i-th user, and τ be the sampling threshold.

[0102] Since the sampling threshold has opposite effects on sampling error and noise error, a larger threshold can reduce the impact of noise error on the algorithm; conversely, a smaller threshold will increase the impact of sampling error. To obtain the optimal sampling threshold in each round of computation, this invention uses the minimum value of the loss function to determine the sampling threshold and allows for adaptive modification of the weights of the two types of errors. The loss function can be defined as:

[0103]

[0104] For a given Scope When the threshold value τ minimizes the waste function and When this value is reached, it is the optimal sampling threshold.

[0105] The client data is sampled using the optimal sampling threshold calculated in each round, iterating multiple times until the remaining value of the privacy budget is less than a given value. This secondary sampling of the original dataset minimizes the error caused by random sampling and noise to the algorithm.

[0106] LPDP sets different privacy budgets on different devices based on users' personalized privacy needs, reducing the waste of privacy budgets and improving the convergence speed of FedProx. At the same time, it adds noise perturbation to the raw data and local model parameters, improving the security of federated learning.

[0107] Next, the local GRU model is trained using the subsampled data, and the local model parameters are updated. This model can be viewed as an imprecise solution. Specifically, this imprecise solution is called... The definition is as follows: For an objective function h to be optimized k (θ;θ t If:

[0108]

[0109] in, Then θ is called * It is h k One Inaccurate solution. Where F k (·) represents the local loss function, and μ represents the regularization parameter.

[0110] So, local model parameters It can be calculated using the following formula:

[0111]

[0112] Where, h k It is the local objective function, F k (θ) is the local loss function on the k-th device, and μ is the regularization parameter.

[0113] After receiving the updated local model parameters Then, the differential privacy method is used to add noise perturbation to improve the confidentiality and privacy security of the data.

[0114] FedProx regular expressions The differences between global model parameters and local model parameters on various devices are considered, which helps to better maintain consistency between the global and local models during optimization. The selection of the hyperparameter μ is crucial for model convergence. When μ = 0, model convergence deteriorates with increasing data heterogeneity; when μ > 0, convergence performance remains relatively good with increasing statistical heterogeneity, but excessively large μ slows down convergence. Therefore, selecting a suitable μ value is particularly important. Considering that this invention only needs to determine the optimal value of one hyperparameter, a grid search method is appropriate. First, the candidate range of the hyperparameter μ is determined to be [0.01, 1], and the parameter values ​​are selected using a mixed interval method, i.e., 0.01, 0.02, ..., 0.1, 0.2, ..., 1 as candidate values. Then, for each μ value, the model is trained using a cross-validation training set, and the model's performance is evaluated on the validation set. Based on the cross-validation results, the μ value with the best performance on the validation set is selected.

[0115] S4. Visualization: Utilizing an optimal global model for meteorological monitoring and forecasting across various regions, the visualization module displays various meteorological data and their changes, providing timely warnings when extreme weather is predicted. Users can access relevant information in real-time on their devices through the visualization module, allowing them to plan their travel time more effectively and avoid losses.

[0116] The description and application of the present invention herein are illustrative and not intended to limit the scope of the invention to the embodiments described above. The effects or advantages described in the specification may not be apparent in actual experimental cases due to uncertainties in specific conditions or other factors, and such descriptions are not intended to limit the scope of the invention. Variations and modifications to the embodiments disclosed herein are possible, and various substitutions and equivalents of the components in the embodiments are well known to those skilled in the art. It should be understood by those skilled in the art that the invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the invention. Other variations and modifications can be made to the embodiments disclosed herein without departing from the scope and spirit of the invention.

Claims

1. A meteorological monitoring method based on GRU neural network and differential privacy federated learning, characterized in that, include: S1. Preprocessing: The client collects local meteorological data and performs preprocessing, generating a meteorological time series data sequence based on the preprocessed data; S2. Initialization: Each client obtains the global model initialization parameters from the central server, and each device selects a suitable model structure based on its own heterogeneity. S3. Local Training: Based on the generated meteorological time-series data, train a local meteorological monitoring model based on a GRU neural network. S4. Local Model Update: After local training is completed, the client updates the parameters of the local model using the LPDP-FedProx federated learning algorithm and uploads the updated model parameters to the central server through the communication module. The specific steps for updating the parameters include: S4-1. Define a sampling threshold to divide the original dataset into an unsampled part and a sampled part; Users in the unsampled segment will have their privacy budget wasted because their data is not used, thus leading to sampling errors: ， If a user's privacy budget in the sampling phase exceeds a threshold, the user's data will be protected against noise exceeding privacy requirements, resulting in noise error. ， in, It's the user's privacy budget. ( ) is the first Individual user privacy budget, It is the sampling threshold; S4-2. The sampling threshold is determined by using the minimum value of the loss function, and the weights of sampling error and noise error are adaptively modified. The loss function is defined as: ， For a given Scope When the threshold value Minimize the waste function and At that time, this value is the optimal sampling threshold. S4-3. The client data is sampled using the optimal sampling threshold calculated in each round. This process is repeated multiple times until the remaining value of the privacy budget is less than a given value. This sampling method is used to perform secondary sampling on the original dataset, minimizing the error caused by random sampling and noise to the algorithm. S4-4. Train the local GRU model using the subsampled data and update the local model parameters. : ， in, It is a local objective function. It is the local loss function on the k-th device. It is a regularization parameter; S5. Global Model Aggregation: After receiving model parameters from multiple clients, the central server aggregates the models according to the global model weighted average aggregation algorithm. S6. Update Feedback: The central server feeds back the updated global model parameters to each client. After receiving the feedback, the client performs the next round of local training until the best monitoring model is obtained. S7. Meteorological Monitoring Visualization: Utilize a global model to monitor and issue early warnings for meteorological parameters, and visualize the parameters and early warning information.

2. The meteorological monitoring method based on GRU neural network and differential privacy federated learning according to claim 1, characterized in that, S1 performs preprocessing on local meteorological data, including data acquisition, filtering, cleaning, and enhancement. Data filtering includes encrypting the data using homomorphic encryption technology, and data cleaning includes estimating missing values, resolving conflicts, and standardizing data in local meteorological equipment.

3. The meteorological monitoring method based on GRU neural network and differential privacy federated learning according to claim 1, characterized in that, The local meteorological monitoring model training bureaus based on GRU neural networks in S3 include: S3-1. Input the preprocessed meteorological data sequence into the GRU neural network. The input at time step t is denoted as... The gating signal is calculated. and The calculation formula is as follows: ， ， in, (·) represents the sigmoid function, which compresses the result to between 0 and 1. This represents the reset gate weight matrix. This represents the update gate weight matrix. These two weight matrices are used to perform a linear transformation between the input of the current time step and the hidden state of the previous time step. [] indicates that the two vectors are concatenated. This indicates the output of the previous time step; S3-2, Calculate candidate hidden layers The formula is as follows: ， Where tanh(·) represents the tanh activation function, which scales the activation result to the range between -1 and 1 to obtain the candidate hidden layer states. This is the weight matrix, used to linearly transform the output of the reset gate and the input of the current time step; * represents matrix multiplication. S3-3 Determine the hidden layer from the previous time step How much information is forgotten and how much hidden layer information of the current time step is added? Calculations yielded The formula is as follows: ， in, This is the final output of the GRU neural network.

4. The meteorological monitoring method based on GRU neural network and differential privacy federated learning according to claim 1, characterized in that, Model aggregation in S5 specifically includes: S5-1, In the first round of training, the central server selects a portion of the clients { } as a participant and initialize the global model Then the global model This information is sent to each client via the base station, serving as the client's local GRU model. The initialization model is then implemented by each client. The LPDP-FedProx algorithm is used to guide parameter updates of the local model, utilizing a private dataset. The local meteorological monitoring model is trained to generate an updated local model. The update formula is as follows: ， Where α represents the learning rate, This indicates the calculation of the gradient. Describe the loss function for each client k; S5-2. The central server calculates the aggregated values ​​of the updated client model parameters to complete model aggregation. The model aggregation formula is as follows: ， in, This represents the number of data sets for the k-th client. This represents the total amount of data from k clients; S5-3, The central server updates the global model for the next round based on the aggregation results: ， S5-4. Perform a performance evaluation on the updated global model, and iterate and update the model based on the evaluation results to continuously optimize the global model.

5. A system for the meteorological monitoring method based on GRU neural network and differential privacy federated learning as described in any one of claims 1-4, characterized in that, include: Local meteorological equipment with built-in GRU neural network is used to collect and preprocess meteorological data locally, train local meteorological monitoring models, and update model parameters based on LPDP-FedProx federated learning algorithm; The central server is used to aggregate local meteorological monitoring model parameters and update the global model based on the LPDP-FedProx federated learning algorithm, and then distribute the updated model parameters to each client. The communication module is used to transmit model parameters between local meteorological equipment and the central server.

Images