A training method and system for digital identity encryption model based on federated learning

Abstract

The present invention belongs to the field of information security technology. A training method and system for a digital identity encryption model based on federated learning are provided. The method comprises: a local node constructs a global model of a digital identity encryption model; a large language model is used to analyze the data to be encrypted, and the data sensitivity level of the data to be encrypted is obtained, and a number of target participants located on the training network are determined according to the data sensitivity level; the global model is sent to each target participant through the network, and each target participant trains the global model, and feeds back the first model parameters obtained by the training to the local node; the local node aggregates all the first model parameters fed back to obtain the second model parameters, and uses the second model parameters to update the global model, and determines the global model after multiple iterations as the digital identity encryption model. The present invention uses a dedicated model to tailor the combined encryption method of the data to be encrypted, thereby improving the security of the combined encryption.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a training method and system for a digital identity encryption model based on federated learning. Background Art

[0002] Digital identity encryption is a technology that protects the identity information of individuals or entities on the Internet, which ensures the confidentiality, integrity and reliability of data. Traditional digital identity encryption is mainly implemented using a single encryption algorithm, such as symmetric encryption algorithms (such as AES, DES) and asymmetric encryption algorithms (such as RSA, ECC). However, with the gradual spread of computer computing power, a single encryption algorithm can no longer ensure data security, and the probability of being cracked is increasing. Therefore, combined encryption methods are gradually being applied to digital identity encryption.

[0003] Combined encryption is an encryption strategy that combines symmetric encryption and asymmetric encryption technologies, which can take advantage of the advantages of both to improve data security and encryption efficiency. However, existing combined encryption methods basically adopt fixed strategies, such as using symmetric encryption algorithms (such as AES) to encrypt actual data and using asymmetric encryption algorithms (such as RSA) to encrypt the keys used by symmetric encryption algorithms. Practice has shown that the encryption performance of this combined encryption method is gradually unable to meet the actual data security needs.

[0004] Therefore, a new combined encryption method is needed to improve the security performance of combined encryption. Summary of the invention

[0005] In this regard, the present invention provides a training method, system, electronic device, computer storage medium and computer program product for a digital identity encryption model based on federated learning to solve the above-mentioned technical problems.

[0006] The present invention discloses a training method for a digital identity encryption model based on federated learning, the method comprising the following steps: a local node constructs a global model of the digital identity encryption model, the global model is used to determine a combined encryption method of data to be encrypted; a large language model is used to analyze the data to be encrypted, and a data sensitivity level of the data to be encrypted is obtained, and a number of target participants located on a training network are determined according to the data sensitivity level; the global model is sent to each of the target participants through the training network, and each of the target participants trains the global model and feeds back the first model parameters obtained by the training to the local node; the local node aggregates all the first model parameters fed back to obtain second model parameters, and uses the second model parameters to update the global model, and determines the global model that has been updated for multiple iterations as the digital identity encryption model.

[0007] In some embodiments, the use of a large language model to analyze the data to be encrypted to obtain the data sensitivity level of the data to be encrypted includes: parsing the data to be encrypted into a plurality of parsed data, inputting the data to be encrypted and each of the parsed data into the large language model at the same time, the large language model obtaining associated information of the data to be encrypted based on the Internet and the training network, performing semantic analysis on the associated information, and deriving the data sensitivity level.

[0008] In some embodiments, determining a number of target participants on the network according to the data sensitivity level includes: calculating a screening number of the target participants based on the data sensitivity level and the following calculation formula: ;in, is the number of screenings of the target party, is the comparison number of the data sensitivity level, is the confidence probability value of the data sensitivity level predicted by the large language model, The average confidence probability value of all the data sensitivity levels predicted by the large language model.

[0009] In some embodiments, each of the target participants trains the global model and feeds back the first model parameters obtained from the training to the local node, including: each of the target participants determines the amount of training data based on the data sensitivity level, collects training data according to the training data amount, and constructs a first training data set based on the amount of training data; wherein the amount of training data is positively correlated with the data sensitivity level; uses the first training data set to train the global model until a training termination condition is reached, and feeds back the first model parameters of the trained global model to the local node; and, also accepts an instruction from the local node to continue the next stage of training, re-collects training data according to the instruction and constructs a second training data set, uses the second training data set to retrain the trained global model, and feeds back the first model parameters of the retrained global model to the local node.

[0010] In some embodiments, the local node aggregates all the first model parameters fed back to obtain the second model parameters, including: the local node calculates the average value or median value of all the first model parameters fed back, and uses the average value or the median value as the second model parameter.

[0011] In some embodiments, before the local node aggregates all the fed-back first model parameters, the method further includes: the local node receives the first model parameters fed back by each of the target participants within a set receiving time window, and deletes each of the first model parameters outside the receiving time window; wherein the time length of the receiving time window is determined according to the number of the above-mentioned retrainings, and specifically, the time length of the receiving time window is positively correlated with the number of the above-mentioned retrainings.

[0012] The present invention also discloses a training system for a digital identity encryption model based on federated learning, the system comprising a local node, a training network and target participants; the local node is used to: construct a global model of the digital identity encryption model, the global model is used to determine the combined encryption method of the data to be encrypted; use a large language model to analyze the data to be encrypted, obtain the data sensitivity level of the data to be encrypted, and determine a number of target participants on the training network according to the data sensitivity level; send the global model to each target participant through the network; each target participant is used to: train the global model, and feed back the first model parameters obtained from the training to the local node; the local node is also used to: aggregate all the fed back first model parameters to obtain second model parameters, use the second model parameters to update the global model, and determine the global model that has been updated for multiple iterations as the digital identity encryption model.

[0013] The present invention also discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in any of the preceding items.

[0014] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0015] The present invention also discloses a computer program product. When the computer program product is run on a terminal, the terminal implements any of the above methods when executing the computer program product.

[0016] The beneficial effects of the present invention are as follows: the present invention uses a combined encryption algorithm to encrypt the data to be encrypted, and the present invention improves the traditional combined encryption algorithm and uses a specially designed digital identity encryption model to generate a tailored combined encryption method for the data to be encrypted, thereby improving the security of the combined encryption. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a training method for a digital identity encryption model based on federated learning disclosed in an embodiment of the present invention.

[0019] Figure 2 It is a structural diagram of a training system for a digital identity encryption model based on federated learning disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0021] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0022] like Figure 1 As shown, an embodiment of the present invention discloses a training method for a digital identity encryption model based on federated learning, and the method includes the following steps: a local node constructs a global model of the digital identity encryption model, and the global model is used to determine a combined encryption method for data to be encrypted; a large language model is used to analyze the data to be encrypted to obtain a data sensitivity level of the data to be encrypted, and a number of target participants located on a training network are determined according to the data sensitivity level; the global model is sent to each of the target participants through the training network, and each of the target participants trains the global model and feeds back the first model parameters obtained from the training to the local node; the local node aggregates all the first model parameters fed back to obtain second model parameters, uses the second model parameters to update the global model, and determines the global model that has been iteratively updated for multiple times as the digital identity encryption model.

[0023] The present invention uses a combined encryption algorithm to encrypt the data to be encrypted, and the present invention improves the traditional combined encryption algorithm and uses a specially designed digital identity encryption model to generate a tailored combined encryption method for the data to be encrypted. The combined encryption method is mainly used to determine which data parts of the data to be encrypted are encrypted by different encryption algorithms.

[0024] At the same time, the present invention also adopts a distributed training method, namely federated learning, to train the digital identity encryption model, as follows: the local node first constructs a global model of the digital identity encryption model, which is the initialization model of the digital identity encryption model, and the model parameters are all preset initial values. Use a large language model (such as BERT, GPT, etc.) to analyze the data sensitivity level of the data to be encrypted, and determine from the network a number of target participants who need to participate in the distributed training of the global model. The global model is sent to each target participant through the network, and each target participant trains the global model by itself. After the training meets the standard, the first model parameter obtained by the training is fed back to the local node. The local node aggregates the first model parameters fed back by each target participant to obtain the second model parameter, and uses the second model parameter to update the global model. After multiple iterative updates, the global model reaches the termination condition of the distributed training, and at this time it is the final usable digital identity encryption model. By using the trained digital identity encryption model to conduct in-depth analysis of the encrypted data, it is possible to determine which encryption algorithms to use and which parts of the encrypted data (data header, data carrier, key, etc.) each encryption algorithm should be used to encrypt in order to achieve the lowest probability of being cracked.

[0025] The digital identity encryption model can be constructed based on a neural network or other machine learning models suitable for the prediction task, such as CNN, RNN, etc. The machine learning model is a reinforcement learning model, an ensemble learning model, etc. The training network can be a blockchain network or a conventional Internet network, and there is no specific limitation.

[0026] In some embodiments, the use of a large language model to analyze the data to be encrypted to obtain the data sensitivity level of the data to be encrypted includes: parsing the data to be encrypted into a plurality of parsed data, inputting the data to be encrypted and each of the parsed data into the large language model at the same time, the large language model obtaining associated information of the data to be encrypted based on the Internet and the training network, performing semantic analysis on the associated information, and deriving the data sensitivity level.

[0027] In an embodiment of the present invention, a large language model such as a BERT model, a GPT model, etc. is a type of deep learning model that is trained on large-scale text data and contains a large number of parameters, and it has shown excellent performance in the field of natural language processing (NLP). By learning the patterns and structures in the text data, the large language model can understand and generate natural language and perform a variety of language-related tasks, such as text generation, text classification, machine translation, sentiment analysis, etc. The present invention uses an online large language model to analyze the data sensitivity level of the data to be encrypted. Specifically, the data to be encrypted is first parsed into a number of parsed data, and the parsed data includes a data header, a data carrier, and a data tail. These parsed data are combined to form a complete data. The meaning of each part and the data content contained are as follows: Data header: The data header is the beginning part of the data to be encrypted (which can be in the form of a data packet), which usually contains control information of the data packet, such as source address, destination address, data packet length, protocol type, etc. In an Ethernet frame, the data header also includes frame synchronization information, address information, and error checking. The function of the data header is to guide the correct transmission and identification of the data packet, and it tells the receiving device how to process the next data.

[0028] Data carrier: The data carrier is the part of the data packet that carries the actual data. During data transmission, the data carrier includes the payload that needs to be transmitted, such as file content, messages, pictures, etc. It is the core part of data transmission and contains the actual information that the sender wants the receiver to obtain.

[0029] Trailer: The trailer is the end of the data packet. It usually contains a checksum or other form of error checking code for error detection. The purpose of the trailer is to ensure that no errors occurred during the data transmission. If an error is detected, the receiver can request that the data packet be resent. In some protocols, the trailer may also contain other protocol-specific control information.

[0030] The large language model performs semantic understanding on the encrypted data and each parsed data, and obtains several retrieval factors from them. Based on these retrieval factors, the related information of the data to be encrypted is searched from the Internet and the aforementioned training network (the training network contains training history data, and the training history data contains various types of data to be encrypted and the data sensitivity level determined by the large language model). The large language model then performs semantic analysis on the related information to derive the data sensitivity level. The data sensitivity level is, for example, 1-10, and the smaller the number, the higher the corresponding encryption security requirement.

[0031] The associated information of the data to be encrypted refers to additional descriptive information of the data to be encrypted, such as secret, confidential, etc., as well as descriptive information of the sensitivity level of the data to be encrypted of the corresponding type existing on the Internet.

[0032] In some embodiments, determining a number of target participants on the network according to the data sensitivity level includes: calculating a screening number of the target participants based on the data sensitivity level and the following calculation formula: ;in, is the number of screenings of the target party, is the comparison number of the data sensitivity level, is the confidence probability value of the data sensitivity level predicted by the large language model, The average confidence probability value of all the data sensitivity levels predicted by the large language model.

[0033] In the embodiment of the present invention, a corresponding comparison number is established in advance for each data sensitivity level, such as shown in the following table:

[0034]

[0035] The higher the data sensitivity level (i.e. the smaller the sensitivity level value), the larger the corresponding control number.

[0036] At the same time, when the large language model predicts the data sensitivity level of the data to be encrypted, it also synchronously determines the confidence probability value of the data sensitivity level of the data to be encrypted. The confidence probability value is used to characterize the accuracy of the data sensitivity level predicted by the large language model. The higher the confidence probability value, the higher the prediction accuracy, and vice versa. In addition, the confidence probability values ​​of all data sensitivity levels predicted by the large language model within a certain period are sorted and calculated to obtain their average confidence probability value.

[0037] By inputting the above-mentioned control number, confidence probability value, and average confidence probability value into the above-mentioned calculation formula at the same time, the number of target participants who need to be screened from the training network to participate in the distributed training can be calculated. Among them, the higher the data sensitivity level (that is, the larger the control number), the higher the encryption security requirements for the encrypted data. At this time, the screening number is increased to improve the security of the combined encryption method generated by the digital identity encryption model through distributed training based on more participants; and the deviation between the confidence probability value of the data sensitivity level predicted in this prediction and the average confidence probability value. The larger it is, the greater the probability that the large language model will make a wrong prediction. Correspondingly, the number of screens will be increased, and the security of the combined encryption method generated by the digital identity encryption model will be improved through possible "over-optimization".

[0038] Combining the above two factors can ensure that the number of target participants screened is within an appropriate range.

[0039] In some embodiments, each of the target participants trains the global model and feeds back the first model parameters obtained from the training to the local node, including: each of the target participants determines the amount of training data based on the data sensitivity level, collects training data according to the training data amount, and constructs a first training data set based on the amount of training data; wherein the amount of training data is positively correlated with the data sensitivity level; uses the first training data set to train the global model until a training termination condition is reached, and feeds back the first model parameters of the trained global model to the local node; and, also accepts an instruction from the local node to continue the next stage of training, re-collects training data according to the instruction and constructs a second training data set, uses the second training data set to retrain the trained global model, and feeds back the first model parameters of the retrained global model to the local node.

[0040] In an embodiment of the present invention, the target participant receives the global model fed back by the local node and the data sensitivity level determined by the aforementioned large language model at the same time, and determines the corresponding amount of training data according to the positive correlation relationship based on the level of the data sensitivity level. Based on this, the corresponding amount of training data is collected by the target participant, and a first training data set is constructed. The global model is trained using the first training data set until the training termination condition is reached, and the first model parameters in the global model are fed back to the local node. The local node can aggregate all the first model parameters fed back into the second model parameters, put the second model parameters into its local untrained initialized global model, and use the test data set to test the initialized global model. If the test result meets the standard, the distributed training is terminated, otherwise, the instruction to continue training is issued to each target participant. After receiving the instruction, each target participant re-collects the training data and constructs a second training data set (which can be a mixture of the first training data set and the newly collected training data), uses the second training data set to retrain the trained global model, and feeds back the first model parameters of the retrained global model to the local node, and the local node performs another test in the aforementioned manner (also for the initial global model). This iteration is repeated until the test result meets the standard. Of course, the number of retraining iterations can be capped, for example, 10 times. If the test result still does not meet the standard when the cap is reached, human intervention is performed.

[0041] The training data included in the above training data set refers to the specific method of using the combined encryption method to encrypt data of different data sensitivity levels, that is, which encryption algorithms are used, and which parts of the data to be encrypted (data header, data carrier, key, etc.) are used for each encryption algorithm. The training data also includes corresponding security labels (such as "high security", "medium security", "low security"). Correspondingly, the training data can be expressed as [data sensitivity level, combined encryption method matrix, security label].

[0042] In some embodiments, the local node aggregates all the first model parameters fed back to obtain the second model parameters, including: the local node calculates the average value or median value of all the first model parameters fed back, and uses the average value or the median value as the second model parameter.

[0043] In an embodiment of the present invention, the aggregation processing of the first model parameters fed back by each target participant may be to obtain the average value or median value thereof. For example, each first model parameter = [A, B, C, D], parameters AD are different types of model parameters, and the corresponding average value or median value is calculated based on parameter A among all the first model parameters to obtain parameter A1, and parameters B1, C1, and D1 are calculated similarly, that is, the second model parameter = [A1, B1, C1, D1] is obtained.

[0044] The first model parameter and the second model parameter are the key model parameters of the digital identity encryption model. The purpose of training the digital identity encryption model is to determine the optimal combination of these key model parameters. Taking the digital identity encryption model as a CNN model as an example, its key model parameters include convolution kernel weights, bias terms, pooling window size and type (maximum pooling or average pooling), learning parameters γ (scale) and β (shift), etc.

[0045] In some embodiments, before the local node aggregates all the fed-back first model parameters, the method further includes: the local node receives the first model parameters fed back by each of the target participants within a set receiving time window, and deletes each of the first model parameters outside the receiving time window; wherein the time length of the receiving time window is determined according to the number of the above-mentioned retrainings, and specifically, the time length of the receiving time window is positively correlated with the number of the above-mentioned retrainings.

[0046] In an embodiment of the present invention, the local node can perform aggregation processing on all the first model parameters fed back by calculating the average value or the median value, but the number of first model parameters corresponding to the aggregation processing needs to be determined. Also, since the efficiency of each target participant in training the global model is different, the feedback time nodes of each first model parameter are different. In this regard, the present invention sets a receiving time window, through which the number of first model parameters involved in each aggregation processing is adjusted, that is, the local node only receives the first model parameters fed back by each target participant within the set receiving time window, and deletes each first model parameter outside the receiving time window. Among them, the time length of the receiving time window is positively correlated with the number of the above-mentioned retraining.

[0047] In the early stage of distributed training of the digital identity encryption model, the length of the receiving time window is set to be relatively small, so that the first model parameters fed back can be aggregated at a higher frequency to obtain the second model parameters, and then the second model parameters can be tested and verified. If the verification fails, the target participant can be instructed to replace the training data in time for retraining. This setting has higher training efficiency, especially for effectively dealing with training deviations in the early stage of training.

[0048] As the number of retraining times increases, the distributed training of the digital identity encryption model enters the middle and late stages. At this time, the length of the receiving time window is gradually increased, that is, the number of first model parameters involved in each aggregation processing increases. This can reduce the interference of individual abnormal first model parameters (mainly first model parameters that fall into the local optimal state) on the second model parameters, so that the final second model parameters are optimal.

[0049] In addition, after the local node deletes each first model parameter outside the receiving time window, it can also send an instruction to the corresponding target participant, instructing it to feed back the first model parameter again in the next receiving time window.

[0050] like Figure 2As shown, an embodiment of the present invention also discloses a training system for a digital identity encryption model based on federated learning, the system comprising a local node, a training network and target participants; the local node is used to: construct a global model of the digital identity encryption model, the global model is used to determine the combined encryption method of the data to be encrypted; use a large language model to analyze the data to be encrypted, obtain the data sensitivity level of the data to be encrypted, and determine a number of target participants located on the training network according to the data sensitivity level; send the global model to each target participant through the network; each target participant is used to: train the global model, and feed back the first model parameters obtained from the training to the local node; the local node is also used to: aggregate all the fed back first model parameters to obtain second model parameters, use the second model parameters to update the global model, and determine the global model that has been updated for multiple iterations as the digital identity encryption model.

[0051] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, characterized in that: the processor executes the computer program to implement the method described in the above embodiment.

[0052] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and is characterized in that: the computer program is executed by a processor to implement the method described in the above embodiment.

[0053] The embodiment of the present invention further discloses a computer program product. When the computer program product is run on a terminal, the terminal implements the method described in the above embodiment.

[0054] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0055] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A training method for a digital identity encryption model based on federated learning, characterized in that The method comprises the following steps: the local node constructs a global model of the digital identity encryption model, the global model is used to determine the combined encryption method of the data to be encrypted; the data to be encrypted is analyzed using a large language model to obtain the data sensitivity level of the data to be encrypted, and a number of target participants located on the training network are determined according to the data sensitivity level; the global model is sent to each of the target participants through the training network, each of the target participants trains the global model, and feeds back the first model parameters obtained by the training to the local node; the local node aggregates all the first model parameters fed back to obtain the second model parameters, uses the second model parameters to update the global model, and determines the global model after multiple iterations as the digital identity encryption model; Determining a number of target participants on the network according to the data sensitivity level includes: calculating the screening number of the target participants based on the data sensitivity level and the following calculation formula: ;in, is the number of screenings of the target party, is the comparison number of the data sensitivity level, is the confidence probability value of the data sensitivity level predicted by the large language model, It is the average confidence probability value of all the data sensitivity levels predicted by the large language model; wherein, the higher the data sensitivity level, the larger the corresponding comparison number.

2. The training method of a digital identity encryption model based on federated learning according to claim 1, characterized in that: The data to be encrypted is analyzed using a large language model to obtain a data sensitivity level of the data to be encrypted, including: parsing the data to be encrypted into a plurality of parsed data, inputting the data to be encrypted and each of the parsed data into the large language model at the same time, the large language model obtaining associated information of the data to be encrypted based on the Internet and the training network, performing semantic analysis on the associated information, and obtaining the data sensitivity level.

3. The training method of a digital identity encryption model based on federated learning according to claim 1, characterized in that: Each of the target participants trains the global model and feeds back the first model parameters obtained from the training to the local node, including: each of the target participants determines the amount of training data based on the data sensitivity level, collects training data according to the training data amount, and constructs a first training data set based on the amount of training data; wherein the amount of training data is positively correlated with the data sensitivity level; uses the first training data set to train the global model until a training termination condition is reached, and feeds back the first model parameters of the trained global model to the local node; and also accepts an instruction from the local node to continue the next stage of training, re-collects training data according to the instruction and constructs a second training data set, uses the second training data set to retrain the trained global model, and feeds back the first model parameters of the retrained global model to the local node.

4. The training method of a digital identity encryption model based on federated learning according to claim 3 is characterized in that: The local node aggregates all the first model parameters fed back to obtain the second model parameters, including: the local node calculates the average value or median value of all the first model parameters fed back, and uses the average value or the median value as the second model parameter.

5. The training method of a digital identity encryption model based on federated learning according to claim 3 is characterized in that: Before the local node aggregates all the fed-back first model parameters, the method further includes: the local node receives the first model parameters fed back by each of the target participants within a set receiving time window, and deletes each of the first model parameters outside the receiving time window; wherein the time length of the receiving time window is determined according to the number of the above-mentioned retrainings, and specifically, the time length of the receiving time window is positively correlated with the number of the above-mentioned retrainings.

6. A training system for a digital identity encryption model based on federated learning, the system comprising a local node, a training network and target participants; characterized in that: The local node is used to: construct a global model of the digital identity encryption model, the global model is used to determine the combined encryption method of the data to be encrypted; use the large language model to analyze the data to be encrypted, obtain the data sensitivity level of the data to be encrypted, and determine a number of target participants on the training network according to the data sensitivity level; send the global model to each target participant through the network; each target participant is used to: train the global model, and feed back the first model parameters obtained by the training to the local node; the local node is also used to: aggregate all the first model parameters fed back to obtain the second model parameters, use the second model parameters to update the global model, and determine the global model that has been updated for multiple iterations as the digital identity encryption model; Wherein, determining a number of target participants on the network according to the data sensitivity level includes: calculating the screening number of the target participants based on the data sensitivity level and the following calculation formula: ;in, is the number of screenings of the target party, is the comparison number of the data sensitivity level, is the confidence probability value of the data sensitivity level predicted by the large language model, It is the average confidence probability value of all the data sensitivity levels predicted by the large language model; wherein, the higher the data sensitivity level, the larger the corresponding comparison number.

7. An electronic device comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 5.

8. A computer storage medium storing a computer program, characterized in that: The computer program is executed by a processor to implement the method according to any one of claims 1 to 5.

9. A computer program product comprising a computer program stored on a non-transitory computer readable medium, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images