Anti-quantum cryptography algorithm parameter dynamic optimization method and system based on artificial intelligence

Abstract

The invention discloses an anti-quantum cryptography algorithm parameter dynamic optimization method and system based on artificial intelligence, and belongs to the technical field of information security and artificial intelligence crossing. The method comprises the following steps: acquiring environment state data including hardware performance, network state, security context and PQC algorithm performance in real time; constructing a multi-dimensional state feature vector; inputting the feature vector into a pre-trained deep reinforcement learning agent, and outputting a parameter optimization strategy by the pre-trained deep reinforcement learning agent; the optimization strategy is applied to the running PQC algorithm instance in real time; and collecting performance indexes under the new parameters, calculating a reward value, and feeding back the reward value to the intelligent agent for online learning. The system comprises an environment perception module, a feature processing module, an AI decision engine module, a dynamic execution module and a feedback learning module. According to the method, the adaptive dynamic optimization of the PQC algorithm parameters is realized, the problem of performance waste or insufficiency caused by static configuration is effectively solved, and the algorithm execution efficiency is remarkably improved and the system power consumption is reduced while the security is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security and artificial intelligence, in particular to an optimization method of post-quantum cryptography (PQC) technology. Specifically, it is a method and system based on artificial intelligence (AI) technology, which dynamically optimizes the running parameters of PQC algorithm according to real-time hardware performance and security threat situation, to realize the optimal balance between security and performance. BACKGROUND

[0002] With the rapid development of quantum computing technology, the existing public key cryptography system based on large number decomposition and discrete logarithm problem (such as RSA, ECC) is facing a serious threat of being broken by quantum computers. As a response, post-quantum cryptography (PQC) has become a global research hotspot and gradually entered the standardization and application stage (such as NIST recommended CRYSTALS-Kyber, CRYSTALS-Dilithium algorithm).

[0003] However, PQC algorithms face a significant technical bottleneck in the process of practical application: their computational complexity, communication overhead and energy consumption are usually much higher than traditional cryptographic algorithms. For example, lattice-based algorithms involve high-dimensional matrix operations, which pose a huge challenge to computing resource-constrained environments (such as Internet of Things terminals, edge computing devices) or high-performance scenarios (such as data center encryption communication).

[0004] Currently, PQC is mostly implemented with static and conservative parameter configuration to ensure the security strength in the worst case. This approach cannot adapt to diverse hardware platforms and dynamically changing application scenarios, resulting in performance waste in most cases or insufficient performance in high-pressure scenarios. Although there are a few manual or traditional optimization algorithm-based tuning solutions, they are inefficient, lack scalability, and are difficult to handle high-dimensional parameter space and multi-objective (security vs. performance) optimization problems.

[0005] Therefore, there is an urgent need in the field for a technology that can automatically, intelligently and dynamically adjust PQC algorithm parameters, maximize system performance while ensuring the security target. SUMMARY

[0006] (I) Invention purpose The purpose of the present application is to overcome the defects of the existing static parameter configuration technology and provide a PQC parameter dynamic optimization method and system based on AI. This method can real-time perceive the state of the running environment and dynamically select and configure the optimal parameter set of PQC algorithm using the trained AI model, thereby realizing the optimal trade-off between security strength and system performance.

[0007] (II) Technical Solution To achieve the above object, the present application adopts the following technical solution:

[0008] S1: Environment state perception and data collection: real-time collection of environment state data of the target system; the environment state data includes hardware performance indicators, network state indicators, security context indicators, and PQC algorithm performance indicators; S2: State feature vector construction: pre-processing and normalization of the collected environment state data to construct a multi-dimensional state feature vector S that can comprehensively describe the current system running environment and task demand; S3: Dynamic decision based on AI model: input the state feature vector S into a pre-trained deep reinforcement learning (DRL) agent; the DRL agent outputs an action A based on the input state S, which is the recommended PQC algorithm parameter combination optimization strategy; S4: Parameter dynamic deployment and execution: apply the optimal parameter combination decided by the DRL agent to the currently running PQC algorithm instance in real time; S5: Performance feedback and online learning of the model: collect the performance indicators of the system under the new parameter configuration, and calculate a reward value R based on the satisfaction of the security target; feedback the reward value R to the DRL agent for online fine-tuning and optimization of its policy network.

[0009] Further, the deep reinforcement learning (DRL) agent is trained by the following methods: State space: the multi-dimensional state feature vector S; Action space: combination of all adjustable PQC algorithm parameters and their effective value ranges; Reward function: designed as a multi-objective weighted function, with the formula: R = α * Performance_Gain + β * Security_Score + γ * (-Power_Consumption); where Performance_Gain is the performance gain, Security_Score is the security score, Power_Consumption is the power consumption; α, β, γ are weight coefficients.

[0010] An artificial intelligence-based anti-quantum cryptographic algorithm parameter dynamic optimization system for implementing the above method, comprising: An environment perception module for real-time collection of hardware, network, security, and algorithm performance indicators; A feature processing module for pre-processing the collected data to construct the state feature vector S; An AI decision engine module, which is built-in the pre-trained and online-learnable deep reinforcement learning (DRL) agent, is used to receive the state feature vector S and output a parameter-optimized strategy. A dynamic execution module is used to receive decision instructions and call an application programming interface (API) of the PQC algorithm to realize hot updating of running parameters. A feedback learning module is used to evaluate the system performance under new parameters, calculate the reward value R, and drive online learning of the AI decision engine module.

[0011] (Three) beneficial effects Compared with the prior art, the present application has the following remarkable advantages: 1. Dynamic self-adaptation: The present application realizes a leap from "static configuration" to "dynamic optimization", enabling the PQC algorithm to adapt to different hardware platforms and real-time changes in business load and security environment.

[0012] 2. Intelligent decision-making: The present application utilizes the ability of deep reinforcement learning to process high-dimensional state and action spaces, finds complex nonlinear relationships that are difficult for humans to discover, and makes more accurate and efficient decisions.

[0013] 3. Significant performance improvement: Under the premise of meeting the same security target, the present application can significantly improve the execution efficiency of the PQC algorithm, resulting in higher throughput, lower latency, and lower system power consumption.

[0014] 4. Continuous evolution: Through the online learning mechanism, the system can continuously optimize its decision-making strategy, adapt to new hardware and software environments and new threats, and has long-term applicability.

[0015] 5. Flexible deployment: The present application can be integrated into existing systems as an independent optimization middleware or embedded in a cryptographic security chip or smart network card as a core function. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is the overall flowchart of the method provided by the embodiment of the present application; Figure 2 is the system module architecture diagram provided by the embodiment of the present application; Figure 3 is the decision-making and training diagram of the deep reinforcement learning agent in the embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application will be described in detail below in conjunction with the drawings and embodiments.

[0018] Embodiment: This embodiment takes optimizing the performance of the CRYSTALS-Kyber algorithm on an edge server as an example.

[0019] Referring to Figure 1 , the implementation steps of the method of the present application are as follows: 1. Environment state perception and data collection (S1): The environment perception module continuously collects the utilization rate, average frequency, L2 cache usage of the 4 cores of the server CPU (ARM Cortex-A76), and the bandwidth (about 50 Mbps) and delay (about 20 ms) of the current network connection. The security context is set to "medium security level".

[0020] 2. State feature vector construction (S2): After normalizing the above data, the feature processing module constructs a 15-dimensional state feature vector S.

[0021] 3. Dynamic decision based on AI model (S3): Input the vector into the DRL agent (using PPO algorithm) that has been trained offline. After reasoning, the agent outputs the action A: it is recommended to adjust the dimension parameter of Kyber algorithm from the default k=4 to k=3, and enable the built-in AVX2 vector instruction optimization library.

[0022] 4. Parameter dynamic deployment and execution (S4): The dynamic execution module writes the new parameters k=3 and the flag of the application instruction set into the initialization function of Kyber through the Hook mechanism or direct calling, completing the dynamic reconfiguration of the algorithm.

[0023] 5. Performance feedback and model online learning (S5): After running for 5 minutes, the feedback learning module measures that the throughput is increased by 40%, the delay is reduced by 35%, and the power consumption is slightly decreased, and the adjusted parameters still meet the "medium security level" requirement. A positive reward value R is calculated and used to fine-tune the policy network of the DRL agent.

[0024] Referring to Figure 2 , the system of the present application comprises an environment perception module, a feature processing module, an AI decision engine module, a dynamic execution module and a feedback learning module connected to each other through a data bus. These modules can be realized in the form of software programs, hardware circuits or a combination of software and hardware.

[0025] The DRL agent is initially trained in a simulation environment containing multiple hardware platforms (x86, ARM) and simulated network conditions through millions of trial and error iterations. After training is completed, its policy network is solidified and deployed to the target system.

Claims

1. An artificial intelligence-based anti-quantum cryptographic algorithm parameter dynamic optimization method, characterized in that, The method comprises the following steps: S1: collecting environment state data of a target system in real time; the environment state data comprises hardware performance indicators, network state indicators, security context indicators, and PQC algorithm performance indicators; S2: preprocessing and normalizing the collected environment state data to construct a multi-dimensional state feature vector S; S3: inputting the state feature vector S into a pre-trained deep reinforcement learning (DRL) agent; the DRL agent outputs a PQC algorithm parameter combination optimization strategy A according to the input state S; S4: applying the parameter combination optimization strategy A to a currently running PQC algorithm instance in real time; S5: collecting performance indicators of the system under the new parameters, and calculating a reward value R in combination with the satisfaction of the security target; feeding back the reward value R to the DRL agent for online learning.

2. The method of claim 1, wherein, The hardware performance indicators comprise one or more of the utilization rate, main frequency, cache hit rate, memory occupancy rate, and power consumption of a CPU / GPU / NPU.

3. The method of claim 1, wherein, The security context indicators comprise one or more of the security level requirement of a current communication link, the threat level of real-time threat intelligence feedback, and the sensitivity level of protected data.

4. The method of claim 1, wherein, The reward function R of the DRL agent is a multi-objective weighted function, and its calculation formula is: R = α * Performance_Gain + β * Security_Score + γ * (-Power_Consumption); wherein, α, β, and γ are weight coefficients.

5. An artificial intelligence-based anti-quantum cryptographic algorithm parameter dynamic optimization system for implementing the method of any one of claims 1-4, characterized by, The system comprises: an environment perception module for collecting the environment state data in real time; a feature processing module for constructing the state feature vector S; an AI decision engine module in which the DRL agent is built in; a dynamic execution module for realizing hot updating of PQC algorithm running parameters; a feedback learning module for calculating the reward value R and driving online learning.

6. The system of claim 5, wherein, The system is integrated in the form of software middleware in an operating system or a PQC algorithm library.

7. The system of claim 5, wherein, The system is integrated in the form of a hardware IP core in a cryptographic security chip or a smart network card.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor to implement the steps of the method according to any one of claims 1-4.

Images