A secure data transmission system based on dynamic algorithm selection

Abstract

The application provides a kind of safe data transmission system based on dynamic algorithm selection, belong to data transmission security technical field.The system includes: multi-dimensional context perception module, real-time acquisition network environment, equipment resources, data attributes, security threat situation and compliance requirements and other multi-dimensional information;Algorithm capability registry module, store the capability description information of each encryption algorithm;Security assessment engine module, output the security score index of each algorithm;Intelligent decision engine module, through deep reinforcement learning model and index mutual correction mechanism, independently select the optimal encryption algorithm combination;Algorithm switching execution module, complete the seamless switching of encryption algorithm under the condition of uninterrupted communication;Feedback evaluation and self-evolution module, with decision performance index as reward signal to drive model continuous optimization.The application realizes multi-dimensional perception, intelligent decision, seamless switching and self-learning evolution of encryption algorithm, significantly improves the security, adaptability and long-term self-optimization capability of data transmission.

Description

Technical Field

[0001] This invention relates to the field of data transmission security technology, and in particular to a secure data transmission system based on dynamic algorithm selection. Background Technology

[0002] Encrypted transmission is a core technical means to ensure network data security. Current mainstream transport layer security protocols (such as TLS and IPsec) generally employ a fixed cipher suite negotiation mechanism: the two communicating parties negotiate and select an encryption algorithm during the session establishment phase, and then use the same algorithm for data encryption and decryption throughout the entire session. This static algorithm selection mechanism has revealed significant shortcomings in the face of increasingly complex network environments and constantly evolving attack threats. On the one hand, the network state, device resources, and security posture at the time of session establishment may change drastically during communication, while the algorithm, once selected, cannot be adjusted, potentially leading to performance degradation or increased security risks later in the communication process. On the other hand, when a security vulnerability is discovered in a certain algorithm, existing solutions can only interrupt the current session and re-initiate the handshake negotiation, failing to achieve a smooth switch between algorithms.

[0003] Some existing dynamic encryption schemes attempt to address the aforementioned issues. Some schemes introduce context-aware modules to collect device resources or network status and dynamically adjust encryption strategies or key lengths accordingly. However, these schemes still rely on relatively singular decision-making criteria, failing to fully integrate multi-dimensional factors such as data sensitivity levels, real-time security threat intelligence, and industry compliance requirements. This results in algorithm selection outcomes that struggle to simultaneously meet security, performance, and compliance needs. Other schemes employ multi-layered or hybrid encryption to enhance security, but their algorithm sets are fixed, and their selection logic is based on preset rules. They lack the ability to learn from historical decisions and continuously optimize, making it difficult to maintain decision accuracy over long-term operation.

[0004] Furthermore, existing technologies generally lack the ability to seamlessly switch encryption algorithms during communication. Even if the decision module determines that the current algorithm is no longer the optimal choice, the system cannot complete a hot switch of the algorithm without interrupting the data stream or rebuilding the session. This means that any algorithm adjustment is accompanied by the overhead of communication interruption and renegotiation, which is unacceptable for applications with high real-time requirements (such as video conferencing, industrial control, and financial transactions). At the same time, existing solutions also fail to take into account the smoothness of post-quantum cryptography algorithms, lacking a mechanism for mixed deployment and gradual migration with traditional algorithms in the face of the long-term security threats posed by quantum computing.

[0005] Therefore, there is an urgent need for a secure data transmission system that can integrate multi-dimensional contextual information, has the ability to seamlessly switch algorithms during communication, can continuously optimize decision-making strategies through reinforcement learning mechanisms, and supports the hybrid deployment of post-quantum cryptographic algorithms, in order to overcome the above-mentioned shortcomings of existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide a secure data transmission system based on dynamic algorithm selection, in order to solve the problems of existing technologies such as single-dimensional encryption algorithm selection, lack of dynamic switching capability during communication, lack of learning and evolution capability in decision-making mechanism, weak security assessment mechanism, and difficulty in quantum-safe deployment.

[0007] To achieve the above objectives, the present invention provides a secure data transmission system based on dynamic algorithm selection, comprising: The multi-dimensional context-aware module is configured to collect context information from multiple dimensions that affect the selection of encryption algorithms in real time. The algorithm capability registry module is configured to store capability description information for each encryption algorithm supported by the system. The security assessment engine module is configured to perform real-time security assessments on the selected encryption algorithms and output security score indicators for each algorithm. The intelligent decision engine module, coupled with the multi-dimensional context awareness module, the algorithm capability registry module, and the security assessment engine module, is configured to autonomously select the optimal combination of encryption algorithms in the current communication scenario based on the collected multi-dimensional context information, algorithm capability description information, and security scoring indicators through a deep reinforcement learning model. The algorithm switching execution module is configured to perform a seamless switch between the current algorithm and the target algorithm when the intelligent decision engine module determines that the encryption algorithm needs to be switched. The feedback evaluation and self-evolution module is configured to collect the execution performance indicators and security event data of each algorithm decision, and feed them back to the intelligent decision engine module as reward signals to drive the continuous training and policy optimization of the deep reinforcement learning model.

[0008] Preferably, the contextual information includes: network environment status, device computing resource status, attributes of the data to be transmitted, real-time security threat situation, and compliance requirements.

[0009] Preferably, the capability description information stored in the algorithm capability registry module includes: algorithm type, security strength level, computational overhead characteristics, key length, applicable scenario label, and quantum security identifier.

[0010] Preferably, the real-time security assessments performed by the security assessment engine module include: dynamic security score calculation based on current security threat intelligence, detection of known vulnerabilities in the algorithm, assessment of resistance to quantum attacks, and assessment of side-channel attack risks.

[0011] Preferably, the intelligent decision engine module is specifically configured as follows: Calculate the context fitness index using the following formula. : ; in, Normalized available bandwidth, For normalized CPU utilization, The normalized packet loss rate, Sensitivity level of the data to be transmitted To meet compliance requirements and ensure a high degree of matching, to Preset weighting coefficients; The algorithm performance index is calculated according to the following formula. : ; in, For the normalized safety strength level, For normalized computational overhead, The quantized value of the quantum security identifier. to Preset weighting coefficients; Receive security score indicators output by the security assessment engine module ; right , , Perform mutual corrections and calculate the corrected context fitness index according to the following formula. Corrected algorithm performance metrics and cross correction terms : ; ; ; in, , , , , This is the preset correction factor; Calculate the comprehensive decision-making index using the following formula. : ; in, , , , The dynamic weighting coefficients are updated online according to the following formula: ; in, To preset the learning rate, The reward signal value is fed back from the evaluation and self-evolution module. As a comprehensive decision-making indicator for the current moment, The value of the corresponding item; Comprehensive decision-making indicators As part of the state input of the deep reinforcement learning model, the deep reinforcement learning model autonomously selects the optimal combination of encryption algorithms for the current communication scenario.

[0012] Preferably, the multi-dimensional context awareness module is further configured to adopt a heterogeneous sensor fusion architecture to perform time synchronization and confidence-weighted fusion of context data from different sources.

[0013] Preferably, the security assessment engine module is further configured to access external security threat intelligence sources, obtain real-time updates on the security status of algorithms, and dynamically adjust the security scores of each algorithm accordingly; external security threat intelligence sources include national information security vulnerability databases and general vulnerability disclosure systems.

[0014] Preferably, the algorithm switching execution module is configured to support active switching mode, performance optimization switching mode, and strategy scheduling switching mode; The active switching mode is as follows: when the security score index of the current algorithm output by the security assessment engine module is lower than the preset security threshold, the switching is actively triggered. The performance optimization switching mode is triggered when the CPU utilization rate in the device computing resource status collected by the multi-dimensional context awareness module exceeds the first preset threshold for three consecutive samples, or when the available bandwidth in the network environment status is lower than the second preset threshold. The strategy scheduling switching mode is as follows: the algorithm is rotated according to a preset time period, and the length of the time period is dynamically adjusted by the intelligent decision engine module based on the historical switching costs and the rate of change of comprehensive decision indicators.

[0015] Preferably, the deep reinforcement learning model employs an algorithm based on proximal policy optimization. The state space of the deep reinforcement learning model is composed of feature vectors from various dimensions collected by the multi-dimensional context-aware module and a comprehensive decision index. Together, they constitute the action space of a deep reinforcement learning model, which is the selection operation of each algorithm or algorithm combination in the algorithm capability registry module. Reward function of deep reinforcement learning model Calculate according to the following formula: ; in, As a comprehensive decision-making indicator, To account for the execution overhead of switching from the current algorithm to the target algorithm, The preset switching cost coefficient; Weighting coefficient The feedback evaluation and self-evolution module updates online according to the following formula: ; in, To preset the learning rate, For the first The actual cumulative reward obtained after each decision. This is the predicted value of the current reward function. .

[0016] Preferably, the system also includes a quantum-safe hybrid encryption module, configured to perform hybrid encryption of traditional and post-quantum cryptographic algorithms when the intelligent decision engine module selects a post-quantum cryptographic algorithm, so as to achieve a smooth transition of quantum-safe capabilities; the hybrid encryption adopts a key exchange hybrid mode or a signature hybrid mode.

[0017] Therefore, the present invention employs the above-mentioned secure data transmission system based on dynamic algorithm selection, and the beneficial technical effects are as follows: (1) This invention collects contextual information in real time from five dimensions: network environment status, device computing resource status, data attributes to be transmitted, real-time security threat situation, and compliance requirements. The intelligent decision engine module calculates the context fitness index and algorithm performance index, and simultaneously receives the security score index output by the security assessment engine module. The above indicators are mutually corrected through preset correction coefficients to obtain a comprehensive decision index. This mechanism enables algorithm selection to take into account performance, security, compliance, and real-time threat situation, avoiding decision bias caused by relying on only a single dimension (such as device resources or network status), and significantly improving the accuracy and robustness of algorithm selection in complex environments.

[0018] (2) This invention solves the problems of existing technologies being unable to switch algorithms during communication and lacking learning and evolution capabilities by using a seamless switching process for the algorithm switching execution module and an online update mechanism for the deep reinforcement learning model. Specifically, the algorithm switching execution module completes the current session state snapshot saving, parallel establishment of the new algorithm session, secure migration of the encrypted context, atomic switching operation, and smooth release of old session resources without interrupting the communication session, thus realizing hot switching of the algorithm. At the same time, the feedback evaluation and self-evolution module collects the execution performance indicators of each decision, calculates the reward signal, and feeds it back to the deep reinforcement learning model, driving the online optimization of the dynamic weight coefficients in the intelligent decision engine module and the reward function weight coefficients of the deep reinforcement learning model, enabling the system to continuously learn from historical decisions and continuously improve the quality of decisions over a long period of operation.

[0019] (3) This invention addresses the weaknesses of existing security assessment mechanisms and the lack of quantum security considerations by combining real-time security assessment through the security assessment engine module with the collaborative work of the quantum-secure hybrid encryption module. Specifically, the security assessment engine module accesses external security threat intelligence sources (such as the National Information Security Vulnerability Database and the General Vulnerability Disclosure System), dynamically calculates the security scores of each algorithm in real time, and proactively triggers a switch when the security score falls below a preset threshold. When the intelligent decision engine module selects a post-quantum cryptographic algorithm, the quantum-secure hybrid encryption module automatically executes hybrid encryption of traditional and post-quantum cryptographic algorithms (using a key exchange hybrid mode or a signature hybrid mode), achieving a smooth transition to quantum security capabilities without affecting the compatibility of existing systems. The above mechanism enables the system to respond promptly to zero-day vulnerability threats and possesses the security evolution capability for the quantum computing era. Attached Figure Description

[0020] Figure 1 This is an architecture diagram of a secure data transmission system based on dynamic algorithm selection according to the present invention; Figure 2 This is a flowchart of the internal process of the intelligent decision engine module; Figure 3 Flowchart for seamless algorithm switching. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0023] Example 1 This embodiment uses an internal data transmission gateway of an enterprise as an example to illustrate in detail the specific implementation of a secure data transmission system based on dynamic algorithm selection. Figure 1 As shown, the system in this embodiment includes: a multi-dimensional context-aware module, an algorithm capability registry module, a security assessment engine module, an intelligent decision-making engine module, an algorithm switching execution module, a feedback evaluation and self-evolution module, and a quantum-secure hybrid encryption module. The specific workflow of each module is described in detail below (refer to...). Figures 1-3 ).

[0024] (1) Multi-dimensional context awareness module, configured to collect context information from multiple dimensions that affect the selection of encryption algorithms in real time.

[0025] The system is deployed on the gateway of an enterprise data center. This gateway simultaneously carries both the office network (peak traffic 50Mbps, latency sensitive) and the production network (peak traffic 200Mbps, throughput sensitive). The multi-dimensional context-aware module collects the following five types of information once per second, and the specific collected data is shown in Table 1.

[0026] Table 1 Real-time data collected by the multi-dimensional context-aware module

[0027] After being timestamped, the above five types of information are fused by the multi-dimensional context awareness module using a heterogeneous sensor fusion architecture (based on extended Kalman filtering) with confidence weighting to generate a unified multi-dimensional context vector. .

[0028] (2) Algorithm capability registry module, configured to store capability description information of each encryption algorithm supported by the system.

[0029] When the system starts, it first initializes the algorithm capability registry module. In this embodiment, some encryption algorithms are pre-registered, and their capability description information is shown in Table 2.

[0030] Table 2. Partial descriptions of algorithm capabilities stored in the algorithm capability registry module.

[0031] Each algorithm also includes detailed benchmark data on computational overhead characteristics (such as average encryption latency, CPU cycle consumption, etc.), which will not be elaborated here. New algorithms must undergo functional correctness and security pre-verification through the algorithm sandbox verification module before registration.

[0032] (3) Security assessment engine module, configured to perform real-time security assessment of the selected encryption algorithm and output the security score index of each algorithm.

[0033] The security assessment engine module connects to the API interfaces of CNNVD and CVE, updating the algorithm security score every 5 minutes. At the time of this embodiment, the security score index S (out of 100) for each algorithm is shown in Table 3. The score is calculated based on a weighted average of the following dimensions: theoretical security strength (40%), known vulnerability status (30%), quantum attack resistance (20%), and side-channel risk assessment (10%). Security Score Index It is transmitted in real time to the intelligent decision engine module.

[0034] Table 3 Real-time security scores output by the security assessment engine module

[0035] (4) Intelligent decision engine module, coupled with multi-dimensional context perception module, algorithm capability registry module and security assessment engine module, is configured to autonomously select the optimal combination of encryption algorithms in the current communication scenario based on the collected multi-dimensional context information, algorithm capability description information and security scoring indicators through a deep reinforcement learning model.

[0036] The intelligent decision engine module is specifically configured as follows: Calculate the context fitness index using the following formula. : ; in, Normalized available bandwidth, For normalized CPU utilization, The normalized packet loss rate, Sensitivity level of the data to be transmitted To meet compliance requirements and ensure a high degree of matching, to The preset weighting coefficients are set to 0.3, 0.2, 0.1, 0.25, and 0.15 in this embodiment, respectively.

[0037] The algorithm performance index is calculated according to the following formula. : ; in, =0.95 is the normalized security strength level (256-bit security strength normalized value). =0.3 represents the normalized computational cost (normalized value of AES-256-GCM computational cost). =0 represents the quantization value of the quantum security identifier (non-PQC algorithm). to The preset weighting coefficients are set to 0.5, 0.6, and 0.2 in this embodiment, respectively.

[0038] Receive security score indicators output by the security assessment engine module =0.95 (after normalization).

[0039] right , , Perform mutual corrections and calculate the corrected context fitness index according to the following formula. Corrected algorithm performance metrics and cross correction terms : ; ; ; in, , , , , The preset correction coefficients are set to 0.6, 0.4, 0.5, 0.5, and 0.3 respectively.

[0040] Calculate the comprehensive decision-making index using the following formula. : ; in, , , , The dynamic weighting coefficients (0.4, 0.3, 0.2, and 0.1 in this embodiment, respectively) are updated online according to the following formula: ; in, To preset the learning rate, The reward signal value is fed back from the evaluation and self-evolution module. As a comprehensive decision-making indicator for the current moment, This is the value of the corresponding item.

[0041] Similarly, the D-values ​​were calculated for other candidate algorithms (SM4-CBC, Kyber-768, etc.) in the algorithm capability registry, and the results are shown in Table 4.

[0042] Table 4. Calculation results of the comprehensive decision index D for each candidate algorithm.

[0043] As can be seen from Table 4, the comprehensive decision-making indicators of SM4-CBC The highest score was 0.812, followed by Kyber-768 (0.788). The intelligent decision engine module integrates decision-making indicators. As part of the state input of the deep reinforcement learning model (the state space also includes the original feature vector), the optimal algorithm combination for the model output is SM4-CBC (balancing compliance, security and performance).

[0044] Comprehensive decision-making indicators As part of the state input of the deep reinforcement learning model, the deep reinforcement learning model autonomously selects the optimal combination of encryption algorithms for the current communication scenario.

[0045] The security assessment engine module is further configured to access external security threat intelligence sources, obtain real-time updates on the security status of algorithms, and dynamically adjust the security scores of each algorithm accordingly; external security threat intelligence sources include the National Information Security Vulnerability Database and the General Vulnerability Disclosure System.

[0046] (5) Algorithm switching execution module, configured to perform seamless switching between the current algorithm and the target algorithm when the intelligent decision engine module determines that the encryption algorithm needs to be switched.

[0047] The algorithm switching execution module is configured to support active switching mode, performance optimization switching mode, and strategy scheduling switching mode. The active switching mode is as follows: when the security score index of the current algorithm output by the security assessment engine module is lower than the preset security threshold, the switching is actively triggered. The performance optimization switching mode is triggered when the CPU utilization rate in the device computing resource status collected by the multi-dimensional context awareness module exceeds the first preset threshold for three consecutive samples, or when the available bandwidth in the network environment status is lower than the second preset threshold. The strategy scheduling switching mode is as follows: the algorithm is rotated according to a preset time period, and the length of the time period is dynamically adjusted by the intelligent decision engine module based on the historical switching costs and the rate of change of comprehensive decision indicators.

[0048] In this embodiment, the preset thresholds are as follows: The safety scoring index has preset safety thresholds: =80 points (0.80 after normalization); CPU utilization first preset threshold: =70%; Second preset threshold for available bandwidth: =30Mbps; Initial time period for policy scheduling: =600 seconds (10 minutes).

[0049] Currently, the system is using the AES-128-GCM algorithm. The security assessment engine module outputs a security score of 82 (normalized to 0.82) for this algorithm, which is not lower than... Therefore, the active switching mode was not triggered. The CPU utilization rate collected by the multi-dimensional context awareness module was 45% (the three consecutive sampling values ​​were 44%, 45%, and 46%, none of which exceeded 70%), and the available bandwidth was 42Mbps (not lower than 30Mbps), so the performance optimization switching mode was also not triggered.

[0050] However, the intelligent decision engine module uses comprehensive decision indicators Comparison (see Table 4, SM4-CBC) =0.812, higher than the current AES-128-GCM. =0.780), indicating that switching to SM4-CBC would yield higher overall returns. Meanwhile, 580 seconds have passed since the last algorithm rotation, approaching the current policy scheduling cycle. (Initially 600 seconds). Therefore, this switchover is triggered by the strategy scheduling switchover mode and executed according to the dynamically adjusted cycle.

[0051] The method for dynamically adjusting the policy scheduling time period is as follows: After each algorithm switch, the intelligent decision engine module records the cost of that switch. (In milliseconds, including total time spent on key derivation, context transition, atomic switching, etc.) and the rate of change of comprehensive decision-making indicators before and after the switch. Then update the policy scheduling period according to the following formula: in, =0.812 represents the comprehensive decision index value of the new algorithm (the target algorithm after switching); =0.780 represents the comprehensive decision index value of the original algorithm (the current algorithm before the switch); This indicates the updated policy scheduling period (in seconds), which is the time interval for the next algorithm rotation. =600s represents the current policy scheduling cycle (unit: seconds), which is the cycle used when this switch was triggered; =6.8ms represents the actual total cost of this algorithm switch (unit: milliseconds), including the total time spent on session snapshot saving, key derivation, context migration, atomic switch and resource release; =10ms represents the baseline handover cost, which is the standard handover time of the system under ideal conditions and is used to normalize the handover cost. This represents the switching cost adjustment coefficient (0.2 in this embodiment), which controls the sensitivity of the switching cost to periodic adjustments. The larger the value, the more drastic the cycle response to changes in switching overhead. This represents the rate of change in the comprehensive decision-making indicators resulting from this switchover, calculated using the above formula; =0.05 represents the expected rate of change of the comprehensive decision-making index (set to 0.05 in this embodiment, i.e. 5%), which represents the minimum performance improvement that the system expects to bring with each switch; This represents the indicator improvement adjustment coefficient (0.3 in this example), which controls the sensitivity of the deviation between the actual improvement and the expected improvement to periodic adjustments. The larger the cycle, the stronger the penalty for insufficient performance improvement.

[0052] The physical meaning of the formula is: If the actual switching cost Higher than the benchmark ,but ,lead to This means extending the cycle to reduce the overhead of frequent switching; If actual indicators improve Below expectations ,but Two negatives make a positive, which will also make (This indicates insufficient switching benefits, and the switching frequency should be reduced); conversely, if the improvement is significantly higher than expected, the cycle should be shortened to switch more aggressively. The combined effect of these two factors enables the cycle to adaptively converge to the optimal value, balancing switching overhead and performance gains.

[0053] Therefore, calculation =563s, the next strategy scheduling cycle will be adjusted to 563 seconds (approximately 9.4 minutes). If the switching cost increases or the indicator improves more significantly in the future, the cycle will be extended or shortened accordingly to achieve adaptive optimization.

[0054] (6) Feedback evaluation and self-evolution module, configured to collect the execution performance indicators and security event data of each algorithm decision, and feed them back to the intelligent decision engine module as reward signals to drive the continuous training and policy optimization of the deep reinforcement learning model.

[0055] The deep reinforcement learning model employs an algorithm based on proximal policy optimization. The state space of the deep reinforcement learning model is composed of feature vectors from various dimensions collected by the multi-dimensional context-aware module and a comprehensive decision index. Together, they constitute the action space of a deep reinforcement learning model, which is the selection operation of each algorithm or algorithm combination in the algorithm capability registry module. Reward function of deep reinforcement learning model Calculate according to the following formula: ; in, As a comprehensive decision-making indicator, The execution overhead of switching from the current algorithm to the target algorithm includes the total time spent on session snapshot saving, key derivation, context migration, atomic switching, and resource release. Set the preset switching cost coefficient (to 0.1); Weighting coefficient The feedback evaluation and self-evolution module updates online according to the following formula: ; in, To preset the learning rate, For the first The actual cumulative reward obtained after each decision. This is the predicted value of the current reward function. .

[0056] Taking the switch to SM4-CBC in Part 5 as an example, the actual rewards... =0.92, comprehensive decision-making index =0.812, switching overhead =0.0068, current weight =0.5, predicted reward =0.4053, according to The calculation formula updates the weights. Simultaneously, this reward signal is used to update the PPO policy network: gradient updates are performed every 32 samples with a learning rate of 0.0003, and the target network is updated every 100 steps. After 30 days of continuous operation, the mean reward signal increased from 0.78 to 0.91, validating the effectiveness of the self-evolution.

[0057] (7) Quantum-safe hybrid encryption module, configured to perform hybrid encryption of traditional algorithm and post-quantum cryptography algorithm when the intelligent decision engine module selects post-quantum cryptography algorithm, so as to achieve a smooth transition of quantum security capability; the hybrid encryption adopts key exchange hybrid mode or signature hybrid mode.

[0058] When the intelligent decision engine module determines that the current communication scenario requires post-quantum security (e.g., transmitting government documents with long-term confidentiality requirements exceeding 10 years), it will select a PQC algorithm such as Kyber-768. At this time, the quantum-safe hybrid encryption module automatically activates a hybrid encryption mode: it uses a traditional algorithm (such as AES-256-GCM) to encrypt the actual data, while using Kyber-768 for key encapsulation. Specifically, the sender generates a temporary key pair and sends the public key to the receiver; the receiver uses this public key to encapsulate an AES-256 key (Key Encapsulation Mechanism KEM) and returns the ciphertext; the sender decapsulates the key to obtain the shared key. In this way, even if a quantum computer breaks Kyber-768 in the future, the attacker cannot obtain the encrypted session key (because the traditional key has already been used), while also maintaining compatibility with existing intermediate devices that do not support PQC.

[0059] After 30 days of continuous operation, the system in this embodiment showed the following results: the average decision latency of the encryption algorithm was 25ms, and the switching success rate was 99.97%. Compared with the static configuration scheme, the switching was triggered an average of 45 seconds earlier when the security score was below the threshold (avoiding potential vulnerability exploitation). When resources were scarce, the system automatically downgraded to a low-overhead algorithm, reducing peak CPU utilization by 22%. The average reward signal recorded by the feedback evaluation module increased from the initial 0.78 to 0.91, indicating continuous optimization of the decision-making strategy.

[0060] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0061] Therefore, the present invention employs a secure data transmission system based on dynamic algorithm selection, which significantly improves the security, adaptability, and long-term self-optimization capabilities of data transmission.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A secure data transmission system based on dynamic algorithm selection, characterized in that, include: The multi-dimensional context-aware module is configured to collect context information from multiple dimensions that affect the selection of encryption algorithms in real time. The algorithm capability registry module is configured to store capability description information for each encryption algorithm supported by the system. The security assessment engine module is configured to perform real-time security assessments on the selected encryption algorithms and output security score indicators for each algorithm. The intelligent decision engine module, coupled with the multi-dimensional context awareness module, the algorithm capability registry module, and the security assessment engine module, is configured to autonomously select the optimal combination of encryption algorithms in the current communication scenario based on the collected multi-dimensional context information, algorithm capability description information, and security scoring indicators through a deep reinforcement learning model. The algorithm switching execution module is configured to perform a seamless switch between the current algorithm and the target algorithm when the intelligent decision engine module determines that the encryption algorithm needs to be switched. The feedback evaluation and self-evolution module is configured to collect the execution performance indicators and security event data of each algorithm decision, and feed them back to the intelligent decision engine module as reward signals to drive the continuous training and policy optimization of the deep reinforcement learning model. The intelligent decision engine module is specifically configured as follows: Calculate the context fitness index using the following formula. : ； in, Normalized available bandwidth, For normalized CPU utilization, The normalized packet loss rate, Sensitivity level of the data to be transmitted To meet compliance requirements and ensure a high degree of matching, to Preset weighting coefficients; The algorithm performance index is calculated according to the following formula. : ； in, For the normalized safety strength level, For normalized computational overhead, The quantized value of the quantum security identifier. to Preset weighting coefficients; Receive security score indicators output by the security assessment engine module ; right , , Perform mutual corrections and calculate the corrected context fitness index according to the following formula. Corrected algorithm performance metrics and cross correction terms : ； ； ； in, , , , , This is the preset correction factor; Calculate the comprehensive decision-making index using the following formula. : ； in, , , , The dynamic weighting coefficients are updated online according to the following formula: ； in, To preset the learning rate, The reward signal value is fed back from the evaluation and self-evolution module. As a comprehensive decision-making indicator for the current moment, The value of the corresponding item; Comprehensive decision-making indicators As part of the state input of the deep reinforcement learning model, the deep reinforcement learning model autonomously selects the optimal combination of encryption algorithms for the current communication scenario.

2. The secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, Contextual information from multiple dimensions includes: network environment status, device computing resource status, attributes of data to be transmitted, real-time security threat situation, and compliance requirements.

3. The secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The capability description information stored in the algorithm capability registry module includes: algorithm type, security strength level, computational overhead characteristics, key length, applicable scenario tags, and quantum security identifier.

4. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The real-time security assessments performed by the security assessment engine module include: dynamic security score calculation based on current security threat intelligence, detection of known vulnerabilities in the algorithm, assessment of resistance to quantum attacks, and assessment of side-channel attack risks.

5. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The multi-dimensional context awareness module is further configured to adopt a heterogeneous sensor fusion architecture to perform time synchronization and confidence-weighted fusion of context data from different sources.

6. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The security assessment engine module is further configured to access external security threat intelligence sources, obtain real-time updates on the security status of algorithms, and dynamically adjust the security scores of each algorithm accordingly; external security threat intelligence sources include the National Information Security Vulnerability Database and the General Vulnerability Disclosure System.

7. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The algorithm switching execution module is configured to support active switching mode, performance optimization switching mode, and strategy scheduling switching mode. The active switching mode is as follows: when the security score index of the current algorithm output by the security assessment engine module is lower than the preset security threshold, the switching is actively triggered. The performance optimization switching mode is triggered when the CPU utilization rate in the device computing resource status collected by the multi-dimensional context awareness module exceeds the first preset threshold for three consecutive samples, or when the available bandwidth in the network environment status is lower than the second preset threshold. The strategy scheduling switching mode is as follows: the algorithm is rotated according to a preset time period, and the length of the time period is dynamically adjusted by the intelligent decision engine module based on the historical switching costs and the rate of change of comprehensive decision indicators.

8. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The deep reinforcement learning model employs an algorithm based on proximal policy optimization. The state space of the deep reinforcement learning model is composed of feature vectors from various dimensions collected by the multi-dimensional context-aware module and a comprehensive decision index. Together, they constitute the action space of a deep reinforcement learning model, which is the selection operation of each algorithm or algorithm combination in the algorithm capability registry module. Reward function of deep reinforcement learning model Calculate according to the following formula: ； in, As a comprehensive decision-making indicator, To account for the execution overhead of switching from the current algorithm to the target algorithm, The preset switching cost coefficient; Weighting coefficient The feedback evaluation and self-evolution module updates online according to the following formula: ； in, To preset the learning rate, For the first The actual cumulative reward obtained after each decision. This is the predicted value of the current reward function. .

9. A secure data transmission system based on dynamic algorithm selection according to claim 1, characterized in that, The system also includes a quantum-safe hybrid encryption module, which is configured to perform hybrid encryption of traditional and post-quantum cryptographic algorithms when the intelligent decision engine module selects a post-quantum cryptographic algorithm, so as to achieve a smooth transition of quantum security capabilities; the hybrid encryption adopts a key exchange hybrid mode or a signature hybrid mode.

Images