Event-triggered encryption method based on reinforcement learning

Abstract

The invention relates to an event triggering type encryption method based on reinforcement learning, and belongs to the field of information security. According to the method, the event triggering type encryption strategy based on reinforcement learning is set, so that the encryption security based on event triggering is ensured, and meanwhile, the flexibility and adjustability of an encryption mode are ensured through adjustment of the encryption strategy based on reinforcement learning. By using the above mode, on the premise of ensuring the security of the to-be-encrypted data, a relatively optimized encryption strategy can be comprehensively determined by combining the security of the to-be-encrypted data and the security demand of the service to which the to-be-encrypted data belongs, so that the flexibility of the encryption strategy is established on the basis of security; and meanwhile, the security of the to-be-encrypted data can be adjusted by setting and adjusting the event triggering threshold.

Description

Technical Field

[0001] The present invention belongs to the field of information security, and in particular relates to an event-triggered encryption method based on reinforcement learning. Background Art

[0002] In response to the increasing number of data encryption requirements and different security encryption requirements for different data in the same business, the existing data security encryption requirements have large design defects for overall security under the premise of meeting security and flexibility, and cannot meet the security and flexibility of encryption while meeting adjustable practical encryption strategies. On the other hand, in the existing designs that use reinforcement learning rules for data security division and encryption strategies, the input of reinforcement learning is incomplete, making it difficult to train a good reinforcement learning model in practice.

[0003] In view of the above problems, the present invention designs an event-triggered encryption method based on reinforcement learning, which solves the problems of the security and flexibility of the encryption strategy and the completeness of the data type in the reinforcement learning model input. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] The technical problem to be solved by the present invention is how to provide an event-triggered encryption method based on reinforcement learning to solve the problems of the security and flexibility of the encryption strategy and the completeness of the data type in the reinforcement learning model input.

[0006] (II) Technical solution

[0007] In order to solve the above technical problems, the present invention proposes an event-triggered encryption method based on reinforcement learning, which comprises the following steps:

[0008] The first step is to design event trigger threshold conditions based on the partition strategy of the data to be encrypted and the encryption strategy;

[0009] Step 2: Determine the metadata partitioning strategy for the data to be encrypted; metadata refers to the basic data unit that uses the same encryption method and encryption strength;

[0010] Step 3: Select an encryption strategy for the divided data to be encrypted;

[0011] Step 4: Evaluation of partitioning strategy and encryption strategy based on event triggering;

[0012] Step 5: Division of data to be encrypted and determination of encryption strategy based on reinforcement learning rules.

[0013] (III) Beneficial effects

[0014] The present invention proposes an event-triggered encryption method based on reinforcement learning. The beneficial effects of the present invention are as follows:

[0015] 1. The event-triggered reinforcement learning encryption method proposed in the present invention uses an event-triggered mechanism to strengthen the security constraints caused by encryption policy changes. By setting a policy change security judgment based on event triggering before making policy changes based on reinforcement learning rules, policies that do not meet security requirements are directly excluded, so that reinforcement learning-based judgments are no longer performed for such policies, saving a lot of computational effort while enhancing the overall security of the data to be encrypted.

[0016] 2. In the division and encryption strategy for the data to be encrypted based on reinforcement learning in the present invention, the input of the strategy in the reinforcement learning rule comprehensively utilizes the historical, current and future predicted security performance requirements and environmental requirements of the data to be processed. The future predicted security performance requirements refer to the reference to the security requirements of the data in the specific business to which it belongs in the reinforcement learning rule, because the security requirements of the specific business to which the data belongs have a type of security requirements that do not change frequently over time in a specific business. At the same time, this type of security requirement is also one of the most important factors in constituting the security requirements of the data itself, so this type of security requirement is to meet the predictability of the data security requirements in the future. Therefore, the input information for the reinforcement learning rules involved in the present invention summarizes all data security requirements on the timeline in the data security requirements.

[0017] 3. The present invention innovatively proposes a data partitioning strategy based on reinforcement learning rules. Considering that different partitions will cause security issues with the encryption strategy, and that too fine a partition will reduce the encryption efficiency, and too coarse a partition will reduce the security of the encryption strategy, this is the trade-off between the security of the encryption strategy and the encryption efficiency caused by the partition of the same data to be encrypted. The present invention proposes a partitioning method that is optimized by comprehensive measurement based on reinforcement learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.

[0020] The present invention ensures the security of encryption based on event triggering by setting an event-triggered encryption strategy based on reinforcement learning, and at the same time, the adjustment of the encryption strategy based on reinforcement learning ensures the flexibility and adjustability of the encryption method. The above method can comprehensively determine the more optimized encryption strategy under the premise of ensuring the security of the data to be encrypted, combining the security of the data to be encrypted itself and the security requirements of the business to which it belongs, so that the flexibility of the encryption strategy is based on security, and at the same time, the setting and adjustment of the event trigger threshold makes the security of the data to be encrypted adjustable.

[0021] The present invention proposes an event-triggered encryption method based on reinforcement learning, which mainly takes the characteristics of the data to be encrypted and the usage characteristics of the data to be encrypted in the business as the input feature data for learning;

[0022] The characteristics of the data to be encrypted include: the structure type of the data or file, the business scope to which the data belongs, the relevance of the data to the business, and other characteristic information;

[0023] The usage characteristics of the data to be encrypted in the business include: the business type to which the data belongs, the historical security policy for data usage, the association between the data to be encrypted and different businesses, and other characteristic information.

[0024] Before encryption, the data to be encrypted needs to be initially divided according to the usage scenarios of the specific business and the security encryption strategy used. That is, the data to be encrypted is divided into different metadata according to different time and space security strengths and business requirements.

[0025] Reinforcement learning rules are used in the process of dividing the data to be encrypted into metadata and using different encryption strategies. By comparing the division strategies and encryption strategies used in the past and the feedback results of the current division and encryption strategies, a comprehensive evaluation is performed to obtain the selection of the division and encryption strategy for the current data to be encrypted.

[0026] The principle of event triggering is to prevent frequent changes in the division and encryption strategies of the data to be encrypted. Only when the current division of the data to be encrypted or the encryption strategy has particularly good performance after the change can the corresponding division or encryption strategy be implemented. The good performance here refers to the quantitative calculation of the variable of the division or strategy to be implemented compared with the original division or strategy. When the value of this quantitative calculation exceeds a certain predetermined threshold, the corresponding division or encryption strategy is implemented. Here, exceeding a certain predetermined threshold represents the occurrence of an event.

[0027] This event-based reinforcement learning encryption strategy for data not only meets the flexibility and security requirements of encryption strategies for different metadata in the data, but also meets the requirements of implementing security strategies based on predetermined security parameters. That is, the encryption strategy based on event-triggered reinforcement learning proposed in the present invention is a flexible encryption method for the data to be encrypted while meeting the security requirements.

[0028] The present invention provides an event-triggered encryption method based on reinforcement learning, the method comprising the following steps:

[0029] Step 1: Design event trigger threshold conditions based on the partition strategy of the data to be encrypted and the encryption strategy.

[0030] Specific policy implementation conditions are set for the partitioning strategy and encryption strategy of the data to be encrypted. The implementation conditions mentioned here are event trigger threshold conditions. Events refer to the implementation actions of partitioning and encryption, that is, when the two specific actions of partitioning or encryption meet the conditions required by each, they can be implemented. They are not simply based on the results of reinforcement learning. The setting of the threshold conditions in the event trigger mentioned here is based on historical experience and the specific business needs of the current data to be encrypted. In particular, it is necessary to pre-determine the impact of this threshold on the overall security of the encryption system. The impact of other performances other than security is not considered here.

[0031] Step 2: Determine the metadata division strategy for the data to be encrypted; metadata refers to the basic data unit that uses the same encryption method and encryption strength.

[0032] Because different businesses and the same business in different time periods have different encryption methods and strength requirements for the same data, the metadata division of the data to be encrypted needs to consider the type of business to which the metadata currently belongs, the security policy of the current business, the security requirements of the metadata itself, and other information, and also needs to refer to the historical division of metadata and other information. Based on the above information, the data to be encrypted is divided to obtain a preliminary division of the data to be encrypted.

[0033] Step 3: Select an encryption strategy for the divided data to be encrypted.

[0034] After the data to be encrypted is divided, the encryption strategy is selected based on the data to be encrypted and the encryption strategy itself. The encryption strategy is selected based on the historical encryption strategy and encryption strength of the data, the type of key used, the specific business security policy requirements and other related information.

[0035] Step 4: Evaluation of partitioning and encryption strategies based on event triggers.

[0036] After the preliminary partitioning strategy and encryption strategy are determined, the security and other performance of the partitioning and encryption strategies are determined based on event triggering conditions after their implementation, and a comprehensive decision is made as to whether the partitioning and encryption strategies can be implemented.

[0037] The event triggering conditions are set by using the partitioning and encryption strategies as inputs and taking the security indicators as the basis for determination. The respective event triggering conditions are determined according to the input partitioning and encryption strategies.

[0038] Taking partitioning as an example, the determination is made based on the quantitative value of the refinement degree of the plaintext partitioning. The refinement of the fixed plaintext can be based on the current partitioning value X h and the value Y of the latest partition that has been performed h The ratio X h / Y h Quantify. The maximum value of the division value is 1, and the minimum value is 0. Here, 1 and 0 represent a practical security strategy. The division of 1 represents a strong security division of one character or one bit at a time, while the division of 0 represents a weak security division of the plaintext as a group. The event trigger condition for the division can be set as the ratio X h / Y h Exceeds a number between [0,1] l , that is, when X h / Y h<l When X is set, the current partition is not updated, and the partition is executed according to the latest partition that has been executed before; when X h / Y h >= l When the current partitioning strategy is updated and executed.

[0039] The setting of event trigger conditions for encryption strategies also follows a similar approach. It should be noted that the setting of event trigger conditions for partitioning and encryption strategies here are mutually influential, rather than independent of each other. The specific setting is determined based on the designer's consideration of the overall security of the encrypted system and the actual security requirements. Here we only provide a design framework and a relatively large degree of design freedom for actual users to design according to specific circumstances.

[0040] This step does not need to consider the impact of reinforcement learning, because the event triggering conditions run independently in the framework based on reinforcement learning strategies. Once a partitioning strategy or encryption strategy does not meet the strategy change conditions based on event triggering conditions, then such corresponding strategies do not need to be determined through subsequent reinforcement learning-based rules to determine whether the strategy needs to be changed.

[0041] The division strategy for the encrypted data and the encryption strategy based on the event trigger threshold are independent and have no correlation with each other. The two strategies are considered together only when judging the quality of the two strategies based on the rules of reinforcement learning.

[0042] Step 5: Division of data to be encrypted and determination of encryption strategy based on reinforcement learning rules.

[0043] After determining the partition and encryption strategy of the data to be encrypted based on event trigger conditions, the partition and encryption strategy needs to be determined based on reinforcement learning rules. Specifically, the partition and encryption strategy needs to go through certain strategy conversion steps to convert the corresponding strategy into a data structure that can be recognized and calculated by the reinforcement learning model. Then, after trying the current partition strategy or encryption strategy, the security performance of the entire data, encryption efficiency, the overall security impact of the business to which the data belongs, and other possible impact results are evaluated and scored to obtain the metadata security probability value, the overall data security probability value, and the overall efficiency impact value after the specific strategy is implemented. Then, according to the specific values ​​fed back in these environments, the strategy is constantly changed so that the above indicators change towards the established goals, and finally a suitable comprehensive strategy is determined. The suitable comprehensive strategy mentioned here is because it is difficult to determine the optimal strategy in theory and examples. Generally, only a relatively optimized strategy can be determined based on the specific situation, and it is difficult to determine whether there is any strategy that is better than the current selection.

[0044] For example, in the encryption strategy mentioned here, after implementing the specific encryption method according to the selected encryption strategy, the encrypted data can be cracked without a key to obtain the cracked plaintext data, and then the security performance of the key used in the strategy is determined by comparing the difference between the cracked plaintext data and the original plaintext data. The repetition probability between the two is an important basis for determining the security of the key used in this strategy. The reinforcement learning rule changes the security strength of the key used in the specific strategy based on this repetition probability, so that this repetition probability changes in a decreasing direction. The factors involved in other encryption strategies such as encryption methods (such as grouping and sequence) and key length in the encryption strategy are also optimized in a similar way.

[0045] Beneficial effects of the present invention:

[0046] 1. The event-triggered reinforcement learning encryption method proposed in the present invention uses an event-triggered mechanism to strengthen the security constraints caused by encryption policy changes. By setting a policy change security judgment based on event triggering before making policy changes based on reinforcement learning rules, policies that do not meet security requirements are directly excluded, so that reinforcement learning-based judgments are no longer performed for such policies, saving a lot of computational effort while enhancing the overall security of the data to be encrypted.

[0047] 2. In the division and encryption strategy for the data to be encrypted based on reinforcement learning in the present invention, the input of the strategy in the reinforcement learning rule comprehensively utilizes the historical, current and future predicted security performance requirements and environmental requirements of the data to be processed. The future predicted security performance requirements refer to the reference to the security requirements of the data in the specific business to which it belongs in the reinforcement learning rule, because the security requirements of the specific business to which the data belongs have a type of security requirements that do not change frequently over time in a specific business. At the same time, this type of security requirement is also one of the most important factors in constituting the security requirements of the data itself, so this type of security requirement is to meet the predictability of the data security requirements in the future. Therefore, the input information for the reinforcement learning rules involved in the present invention summarizes all data security requirements on the timeline in the data security requirements.

[0048] 3. The present invention innovatively proposes a data partitioning strategy based on reinforcement learning rules. Considering that different partitions will cause security issues with the encryption strategy, and that too fine a partition will reduce the encryption efficiency, and too coarse a partition will reduce the security of the encryption strategy, this is the trade-off between the security of the encryption strategy and the encryption efficiency caused by the partition of the same data to be encrypted. The present invention proposes a partitioning method that is optimized by comprehensive measurement based on reinforcement learning.

[0049] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An event-triggered encryption method based on reinforcement learning, characterized in that: The method comprises the following steps: The first step is to design event trigger threshold conditions based on the partition strategy of the data to be encrypted and the encryption strategy; The second step is to determine the metadata division strategy for the data to be encrypted; Metadata refers to the basic data unit that uses the same encryption method and encryption strength; Step 3: Select an encryption strategy for the divided data to be encrypted; Step 4: Evaluation of partitioning strategy and encryption strategy based on event triggering; Step 5: Division of data to be encrypted and determination of encryption strategy based on reinforcement learning rules.

2. The event-triggered encryption method based on reinforcement learning as claimed in claim 1, characterized in that: The method uses the characteristics of the data to be encrypted and the usage characteristics of the data to be encrypted in the business as input feature data for learning; The characteristics of the data to be encrypted include: the structure type of the data or file, the business scope to which the data belongs, and the degree of relevance of the data to the business; the usage characteristics of the data to be encrypted in the business include: the business type to which the data belongs, the historical security strategy for the use of the data, and the relationship between the data to be encrypted and different businesses.

3. The event-triggered encryption method based on reinforcement learning as claimed in claim 1, characterized in that: In the first step, specific policy implementation conditions are set for the partition strategy and encryption strategy of the data to be encrypted. The implementation conditions here are event trigger threshold conditions. Events refer to the implementation actions of partitioning and encryption. That is, the two specific actions of partitioning or encryption can only be implemented when they meet their respective required conditions. It is not a result obtained simply based on reinforcement learning.

4. The event-triggered encryption method based on reinforcement learning as claimed in claim 3, characterized in that: The setting of the threshold conditions in the event trigger is based on historical experience and the specific business needs of the current data to be encrypted. It is necessary to predetermine the impact of this threshold on the overall security of the encryption system.

5. The event-triggered encryption method based on reinforcement learning according to any one of claims 1 to 4, characterized in that: In the second step, the metadata division of the data to be encrypted needs to take into account the type of business to which the metadata currently belongs, the security policy of the current business, and the security requirement information of the metadata itself. At the same time, it is necessary to refer to the historical division information of the metadata. Based on the above information, the data to be encrypted is divided to obtain a preliminary division of the data to be encrypted.

6. The event-triggered encryption method based on reinforcement learning as claimed in claim 5, characterized in that: In the third step, after the data to be encrypted is divided, the encryption strategy is selected for the divided data to be encrypted itself. The encryption strategy is selected based on the historical encryption strategy and encryption strength of the data, the type of key used, and the specific business security policy requirements.

7. The event-triggered encryption method based on reinforcement learning as claimed in claim 6, characterized in that: In the fourth step, after the preliminary partitioning strategy and encryption strategy are determined, the security and other performance of the partitioning strategy and encryption strategy are determined based on the event triggering conditions after the partitioning strategy and encryption strategy are implemented, and it is comprehensively determined whether the partitioning strategy and encryption strategy can be implemented; The event triggering conditions are set by using the partitioning and encryption strategies as inputs and taking the security indicators as the basis for determination. The respective event triggering conditions are determined according to the input partitioning and encryption strategies.

8. The event-triggered encryption method based on reinforcement learning as claimed in claim 7, characterized in that: For the division, the quantized value of the refinement degree of the plaintext division is used for determination. The refinement of the fixed plaintext can be determined based on the value X of the current division. h and the value Y of the latest partition that has been performed h The ratio of X h / Y h Quantify; the maximum value of the division value is 1, and the minimum value is 0. Here, 1 and 0 represent a practical security strategy. The division of 1 represents a strong security division of one character or one bit at a time, while the division of 0 represents a weak security division for the plaintext as a group; the event trigger condition for the division can be set to the ratio X h / Y h Exceeds a number between [0,1] l , that is, when X h / Y h<l When X is set, the current partition is not updated, and the partition is executed according to the latest partition that has been executed before; when X h / Y h >= l When the current partitioning strategy is updated and executed.

9. The event-triggered encryption method based on reinforcement learning as claimed in claim 7, characterized in that: The settings of event trigger conditions for the partitioning and encryption policies affect each other, and the policy change decisions based on event trigger thresholds for the partitioning policy and encryption policy for the data to be encrypted are independent.

10. The event-triggered encryption method based on reinforcement learning as claimed in claim 7, characterized in that: The fifth step comprises: After determining the division and encryption strategy of the data to be encrypted based on event trigger conditions, the division and encryption strategy need to be determined based on reinforcement learning rules; specifically, the division and encryption strategy need to go through certain strategy conversion steps to convert the corresponding strategy into a data structure that can be recognized and calculated by the reinforcement learning model, and then evaluate and score the security performance, encryption efficiency, and overall security impact of the entire data after trying the current division strategy or encryption strategy, and obtain the metadata security probability value, overall data security probability value, and overall efficiency impact value after the implementation of the specific strategy, and then continuously change the strategy based on the specific values ​​fed back in these environments so that the above indicators change toward the established goals, and finally determine the appropriate comprehensive strategy; For the encryption strategy, after implementing the specific encryption method according to the selected encryption strategy, the encrypted data is cracked without a key to obtain the cracked plaintext data, and then the security performance of the key used in the strategy is determined by comparing the difference between the cracked plaintext data and the original plaintext data. The repetition probability between the two is an important basis for determining the security of the key used in this strategy. The reinforcement learning rule changes the security strength of the key used in the specific strategy based on this repetition probability, so that this repetition probability changes in a decreasing direction.

Images