Dynamic reconfiguration post-quantum cryptography multi-algorithm collaborative acceleration system

Abstract

The application relates to the technical field of post-quantum cryptography, and discloses a dynamic reconstruction post-quantum cryptography multi-algorithm cooperative acceleration system. The system comprises a dynamic parameter library, a cryptographic graph indexer, a vectorization scheduling engine and a cooperative execution unit. The dynamic parameter library stores a set of real-time running parameters of post-quantum cryptography algorithms, and is updated according to feedback from the cooperative execution unit. The cryptographic graph indexer constructs a dynamic cryptographic algorithm index graph according to the parameters, wherein nodes are algorithm instances, edges are cooperative relationships, and edge weights are adjusted according to path weight feedback from the vectorization scheduling engine. The vectorization scheduling engine vectorization-encodes a user cryptographic task request to generate a task semantic vector, calculates the similarity between the task semantic vector and node embedding vectors of the index graph, and generates an algorithm selection instruction. The cooperative execution unit receives the instruction, calls corresponding algorithm instances to operate, and feeds back parameter changes to the dynamic parameter library. The system realizes dynamic optimization and efficient cooperation of post-quantum cryptography algorithms, and enhances the adaptability to complex scenarios.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of post-quantum cryptography, in particular to a dynamic reconstruction post-quantum cryptography multi-algorithm cooperative acceleration system. BACKGROUND

[0002] With the rapid development of quantum computing technology, the traditional public key cryptography system based on large number factorization and discrete logarithm problem faces the risk of being cracked by quantum algorithms, and post-quantum cryptography algorithm has thus become an important research direction to ensure information security. At present, there are many types of post-quantum cryptography algorithms, and different algorithms have significant differences in security, computational efficiency, resource consumption, etc. For example, lattice-based cryptography algorithms have outstanding performance in resisting quantum attacks, but have high computational complexity; hash-based cryptography algorithms have fast operation speed, but have limitations in key length.

[0003] In practical applications, a single post-quantum cryptography algorithm often cannot meet the needs of complex scenarios. For example, in the financial transaction scenario, the algorithm not only requires high security to protect user funds, but also needs to respond quickly to improve transaction efficiency; Internet of Things devices are limited by hardware resources, and have strict limitations on the amount of calculation and storage requirements of the algorithm. However, there is a lack of effective mechanism in the prior art to cooperatively schedule multiple post-quantum cryptography algorithms, and most systems can only fixedly use a certain algorithm, or realize algorithm switching through simple manual configuration, and cannot dynamically adjust according to real-time task requirements and running environment.

[0004] The running parameters of post-quantum cryptography algorithms fluctuate with changes in application scenarios, such as key update frequency and data transmission volume. In existing systems, parameter management is mostly static configuration, which is difficult to optimize in real time according to the actual running state of the algorithm, resulting in that the performance of the algorithm cannot be fully utilized, and even security risks may be caused due to parameter mismatch. At the same time, the cooperative relationship between different algorithms lacks quantitative analysis, and it is difficult to determine the optimal algorithm combination strategy, further limiting the application effect of post-quantum cryptography technology in complex scenarios. SUMMARY

[0005] The purpose of the present application is to provide a dynamic reconstruction post-quantum cryptography multi-algorithm cooperative acceleration system to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides a dynamic reconstruction post-quantum cryptography multi-algorithm cooperative acceleration system, which comprises:

[0007] a dynamic parameter library, a cryptography graph indexer, a vectorization scheduling engine, and a cooperative execution unit;

[0008] The dynamic parameter library is configured to store a real-time running parameter set of the post-quantum cryptographic algorithm, and dynamically update the real-time running parameter set according to feedback of the cooperative execution unit, the real-time running parameter set including algorithm calculation complexity, resource occupancy rate and security strength index;

[0009] The password graph indexer is configured to obtain the real-time running parameter set and construct a dynamic cryptographic algorithm index graph; wherein the password graph indexer includes an index constructor and a weight optimizer; the index constructor is configured to obtain the post-quantum cryptographic algorithm instances in the algorithm instance library, take each algorithm instance as a node, and create cooperative relationship edges between nodes based on algorithm protocol compatibility and calculation dependency; the weight optimizer is configured to calculate strength change values of each cooperative relationship edge according to path weight data fed back by the vectorization scheduling engine, and dynamically adjust edge weights of the cooperative relationship edges; wherein the path weight data contains node paths triggered in an algorithm selection instruction execution process, and the dynamic adjustment includes: counting trigger frequencies and edge weight correction amounts of each node path, generating strength change values of the cooperative relationship edges, and dynamically superimposing and updating edge weights of corresponding cooperative relationship edges in the dynamic cryptographic algorithm index graph according to the strength change values;

[0010] The vectorization scheduling engine is configured to generate a task semantic vector by vectorizing and encoding a password task request input by a user, calculate similarity between the task semantic vector and embedding vectors of nodes in the dynamic cryptographic algorithm index graph based on the task semantic vector, and generate an algorithm selection instruction; wherein the vectorization scheduling engine includes a semantic encoder and a priority scheduler; the semantic encoder is configured to extract multi-modal features of the password task request input by the user and generate the task semantic vector; and the priority scheduler is configured to calculate initial similarity between the task semantic vector and the embedding vectors of the nodes in the dynamic cryptographic algorithm index graph, and locate a target node set with the highest similarity in an embedding space through multi-layer projection calculation.

[0011] The cooperative execution unit is configured to receive the algorithm selection instruction, call a corresponding post-quantum cryptographic algorithm instance to perform cryptographic operation, and feed back parameter changes in the operation process to the dynamic parameter library.

[0012] Preferably, the dynamic parameter library includes an algorithm instance library and a parameter tracker.

[0013] The algorithm instance library is configured to store initialization configurations and real-time state data of the post-quantum cryptographic algorithm instances.

[0014] The parameter tracker is configured to monitor a running process of the post-quantum cryptographic algorithm instance, and extract a real-time running parameter set from the running process, the real-time running parameter set including algorithm calculation complexity, resource occupancy rate and security strength index.

[0015] The parameter tracker further performs incremental update on real-time state data in the algorithm instance library according to parameter change data fed back by the cooperative execution unit.

[0016] Preferably, the cryptogram index constructor comprises an index constructor and a weight optimizer.

[0017] The index constructor is configured to obtain post-quantum cryptographic algorithm instances in the algorithm instance library, take each algorithm instance as a node, and create cooperative relationship edges between nodes based on algorithm protocol compatibility and calculation dependency relationship; wherein the protocol compatibility is verified by detecting whether the algorithm instances share the same basic mathematical difficult problem structure, and the calculation dependency relationship is verified by analyzing algorithm input and output interfaces and confirming that the output format of the predecessor algorithm matches the input specification of the successor algorithm.

[0018] The weight optimizer is configured to calculate the strength change value of each cooperative relationship edge according to the path weight data fed back by the vectorization scheduling engine, and dynamically adjust the edge weight of the cooperative relationship edge.

[0019] Preferably, the vectorization scheduling engine comprises a semantic encoder and a priority scheduler.

[0020] The semantic encoder is configured to perform multi-modal feature extraction on the user input cryptographic task request to generate a task semantic vector; wherein the multi-modal feature extraction comprises processing and fusing structured fields and unstructured text in the cryptographic task request respectively.

[0021] The priority scheduler is configured to calculate the initial similarity between the task semantic vector and the embedding vector of each node in the dynamic cryptographic algorithm index graph, generate an adjustment weight coefficient in combination with the edge weight of the cryptographic graph index constructor, correct the initial similarity according to the adjustment weight coefficient, and generate an algorithm selection instruction; wherein the adjustment weight coefficient is generated according to the edge weight of the cooperative relationship edge between target nodes, and is corrected by multiplying the initial similarity.

[0022] Preferably, the cooperative execution unit comprises an algorithm executor and a parameter feedback unit.

[0023] The algorithm executor is configured to activate the corresponding post-quantum cryptographic algorithm instance according to the algorithm selection instruction, perform cryptographic operation, and output the encryption result.

[0024] The parameter feedback unit is configured to collect resource consumption trajectories and protocol interaction states during the running of the post-quantum cryptographic algorithm instance, generate parameter change data, and send the parameter change data to the dynamic parameter library.

[0025] Preferably, the index constructor performs the following operations:

[0026] obtain an algorithm computation complexity and a security strength indicator in the real-time running parameter set, and calculate protocol compatibility between algorithm instances; wherein the protocol compatibility is obtained by comparing protocol metadata with the algorithm to generate a protocol compatibility score;

[0027] generate an initial weight of a collaborative relationship edge according to the protocol compatibility and a computation dependency relationship;

[0028] When a new post-quantum cryptographic algorithm instance is added, a new node is created based on the real-time state data of the parameter tracker, and a collaborative relationship edge is connected to an existing node.

[0029] Preferably, the weight optimizer performs the following operations:

[0030] Receive the path weight data fed back by the vectorization scheduling engine, which includes the node path triggered in the algorithm selection instruction execution process and the execution time consumption;

[0031] Statistical frequency of triggering each node path, and generate a strength change value of the collaborative relationship edge according to the triggering frequency and the execution time consumption;

[0032] According to the strength change value, the edge weight of the corresponding collaborative relationship edge in the dynamic cryptographic algorithm index graph is dynamically superimposed and updated.

[0033] Preferably, the priority scheduler performs the following operations:

[0034] Calculate the initial similarity between the task semantic vector and the embedding vector of each node in the dynamic cryptographic algorithm index graph;

[0035] Map the task semantic vector to the embedding space of the dynamic cryptographic algorithm index graph, and calculate the positioning of the target node set with the highest similarity in the embedding space through multi-layer projection;

[0036] Calculate the positioning of the target node set with the highest similarity in the embedding space through multi-layer projection;

[0037] According to the edge weight of the collaborative relationship edge between the target nodes fed back by the cryptographic graph indexer, generate an adjustment weight coefficient;

[0038] Multiply the initial similarity of the target node set by the adjustment weight coefficient to generate a final similarity ranking and output an algorithm selection instruction.

[0039] Preferably, the dynamic parameter library further comprises a reconstruction executor;

[0040] The reconstruction executor is used to restructure the post-quantum cryptographic algorithm instances in the algorithm instance library when the parameter tracker detects that the security strength indicator is lower than a preset threshold.

[0041] The structural reorganization comprises: decomposing component dependency relationship of algorithm instance, recombining algorithm modules according to the cooperative relationship edge of the cryptogram indexer, and updating the real-time running parameter set; Wherein, the structural reorganization is executed in stages, including: dependency analysis stage, for parsing the call graph of algorithm instance to identify the data flow and control dependency between modules;Component extraction stage, for splitting the binary code of algorithm instance into logical units;And reorganization stage, for querying the cryptogram indexer and recombining algorithm modules based on the cooperative relationship graph.

[0042] Preferably, the cooperative execution unit further comprises a cooperative controller;

[0043] The cooperative controller is used to establish data pipeline between instances according to the cooperative relationship edge of the dynamic cryptogram algorithm index graph when the algorithm executor calls multiple post-quantum cryptography algorithm instances;

[0044] The intermediate calculation results and protocol state between algorithm instances are synchronized through the data pipeline, and the synchronized data stream is input into the parameter feedback device to generate parameter change data; Wherein, the protocol state comprises protocol phase identifier and error code.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] Through the cooperative work of the dynamic parameter library, the cryptogram indexer, the vectorization scheduling engine and the cooperative execution unit, the dynamic optimization and efficient cooperation of post-quantum cryptography algorithm are realized. The dynamic parameter library can store and update the running parameter set of post-quantum cryptography algorithm in real time, so that the system can adjust the parameters according to the actual running state of the algorithm, adapt to the demand change in different scenarios, and keep the algorithm in good running state in various application environments.

[0047] The dynamic cryptogram algorithm index graph constructed by the cryptogram indexer takes post-quantum cryptography algorithm instances as nodes, the cooperative relationship between algorithm instances as edges, and dynamically adjusts the edge weight according to the path weight feedback of the vectorization scheduling engine. This way quantifies the cooperative relationship between different algorithms, provides a basis for optimization of algorithm combination strategy, so that the system can more clearly grasp the correlation between algorithms, and thus select a more appropriate algorithm combination in complex scenarios.

[0048] The vectorization scheduling engine performs vectorization encoding on the user input cryptography task request, generates a task semantic vector, and generates an algorithm selection instruction by calculating the similarity between the vector and the embedding vector of each node in the dynamic cryptogram algorithm index graph. This algorithm selection method based on vector similarity can accurately match the task demand and algorithm characteristics, so that the system can select the most suitable algorithm for each specific task, improving the pertinence and effectiveness of task processing.

[0049] The cooperative execution unit, after receiving the algorithm selection instruction, calls the corresponding post-quantum cryptography algorithm instance to perform cryptographic operations, and feeds back the parameter changes in the operation process to the dynamic parameter library, forming a closed-loop optimization mechanism, so that the dynamic parameter library can obtain real-time information in the algorithm operation in a timely manner, constantly improve the parameter set, and at the same time, the entire system forms the ability of continuous iterative optimization, gradually improves the performance in the long-term operation, and adapts to the changing task requirements and environmental conditions.

[0050] Through the organic cooperation of each unit, the system breaks the limitations of a single algorithm, realizes the dynamic reconstruction and cooperative operation of multiple post-quantum cryptography algorithms, can cope with complex demands in different scenarios, enhances the flexibility and adaptability of post-quantum cryptography technology in practical application while ensuring security, expands its application range, and makes post-quantum cryptography technology better serve multiple fields such as finance and the Internet of Things. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a timing diagram of the dynamic reconstruction post-quantum cryptography multi-algorithm cooperative acceleration system described in the present application;

[0052] Figure 2 is a flowchart of the dynamic parameter library;

[0053] Figure 3 is a comprehensive diagram of the dynamic parameter library;

[0054] Figure 4 is a flowchart of the password graph indexer;

[0055] Figure 5 is a comprehensive diagram of the password graph indexer;

[0056] Figure 6 is a detailed flowchart of the weight optimizer;

[0057] Figure 7 is a flowchart of the vectorization scheduling engine. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0059] Please refer to Figure 1The application provides a dynamic reconfiguration post-quantum cryptography multi-algorithm collaborative acceleration system, which comprises four core components: a dynamic parameter library, a cryptography graph indexer, a vectorization scheduling engine and a collaborative execution unit.

[0060] The dynamic parameter library serves as a parameter storage center of the system and continuously maintains a real-time running parameter set of post-quantum cryptography algorithm instances, including key data such as algorithm calculation complexity, resource occupancy and security strength indicators. The cryptography graph indexer constructs a dynamic cryptography algorithm index graph based on the real-time parameters provided by the dynamic parameter library. The graph uses a graph structure to represent the collaborative relationship between algorithm instances, wherein a node represents a specific algorithm instance and an edge represents the protocol compatibility and calculation dependency relationship between instances. The vectorization scheduling engine converts the cryptography task request input by a user into a high-dimensional semantic vector, generates an optimal algorithm selection instruction through similarity calculation with the embedding vectors of the nodes in the index graph, and calls the corresponding algorithm instance to perform operation according to the instruction. Meanwhile, the collaborative execution unit feeds back the parameter changes in real time to the dynamic parameter library, forming a closed-loop control system. During system operation, the components realize real-time interaction through a data flow pipeline. The parameter update of the dynamic parameter library triggers the graph structure adjustment of the cryptography graph indexer. The vectorization scheduling engine optimizes the scheduling strategy based on the latest graph data. The execution feedback of the collaborative execution unit drives the parameter library update, forming a continuously optimized adaptive cycle.

[0061] Embodiment 1: refer to Figure 2 , which relates to the detailed architecture and operation mechanism of the dynamic parameter library. The system component comprises an algorithm instance library and a parameter tracker, which constitute the core storage and monitoring system. The algorithm instance library adopts a distributed key-value storage structure. Each post-quantum cryptography algorithm instance corresponds to an independent partition storage space. The partition is divided into a static configuration area and a dynamic state area. The static configuration area stores algorithm binary code, pre-computed parameters and protocol configuration files, and adopts a read-only mode to prevent runtime tampering. The dynamic state area maintains the runtime snapshot of the instance, including volatile data such as memory image, register value and stack pointer, and guarantees state consistency through copy-on-write technology. The storage engine realizes multi-version concurrent control and supports parallel loading and updating operations of algorithm instances.

[0062] The parameter tracker adopts a microservice architecture design and includes three functional layers of data collection, real-time calculation, and persistence. The data collection layer deploys a lightweight kernel module to capture the instruction execution flow of the algorithm instance by interrupt hijacking, and the sampling interval is dynamically adjusted according to system load, with a basic sampling period of 5 milliseconds in idle state and automatically shortened to 1 millisecond in high load state. The real-time calculation layer runs a stream processing engine to perform window aggregation operations on raw sampling data, calculates the instruction throughput per unit time for the complexity index, and normalizes the result combined with the processor pipeline state; the resource occupancy monitoring integrates memory allocator hook functions to track page table entry modification records and distinguish the usage proportion of permanent memory and temporary cache; the security strength evaluation module runs in a trusted execution environment to detect the actual attack resistance of the algorithm through simulated side-channel attacks. The persistence layer implements an incremental storage strategy, only recording the parameter change amount in adjacent sampling periods, and uses columnar compression format to reduce storage overhead.

[0063] The parameter update pipeline establishes an asynchronous processing channel, and the parameter change data fed back by the execution unit are first entered into a verification queue to verify the legality of the data source through digital signature. The data packet that passes the verification triggers a version merging operation, and the state data in the algorithm instance library is updated according to the modification granularity: register-level changes directly overwrite the original values, memory region changes perform differential comparison for local update, and complete state snapshots trigger atomic replacement. Version logs are maintained during the update process to support state rollback and operation auditing. A bidirectional data channel is established between the parameter tracker and the algorithm instance library, and real-time parameter collection does not affect the normal operation of the instance, and the state update operation is guaranteed to be visible through memory barriers.

[0064] The re-executor runs as an independent security component in privileged mode and continuously monitors the time series changes of security strength indicators. Threshold detection uses a sliding window comparison algorithm, and the window size is adaptively adjusted according to the algorithm type, with a shorter detection window for signature-based algorithms and a longer observation period for encryption-based algorithms. When an indicator anomaly is detected, the reconstruction process is executed in stages: the dependency analysis stage uses static binary analysis tools to parse the call graph of the algorithm instance, identifies the data flow and control dependencies between modules, and extracts components by function boundary detection technology to segment binary code into logical units while preserving debugging symbols and relocation information; the recombination stage queries the password graph indexer to obtain the latest collaboration graph, and preferentially selects adjacent nodes with high security ratings to establish connections, and the newly combined algorithm module is optimized by the linker to generate a complete instance.

[0065] The newly generated algorithm instance undergoes an integrity verification process, including control flow integrity checks, memory safety verification, and protocol compliance testing. The verified instance is assigned a unique identifier and registered in the activation area of the algorithm instance library. The instance activation process adopts a gradual strategy, initially routing a small number of tasks to the new instance while running the old instance as a backup. After a stability observation period, the load proportion of the new instance is gradually expanded. The retirement process of the old instance retains a recovery mechanism, maintaining its state data for a preset time to facilitate quick fallback.

[0066] The fault-tolerant mechanism of the dynamic parameter library implements multi-level protection. The data collection process adopts a redundant sampling design, with key indicators collected by independent threads in parallel. The single-point error is eliminated through a voting mechanism. The real-time calculation layer deploys a checksum mechanism, and the window aggregation result is only adopted after cross-validation. Persistent storage implements erasure coding, storing data in blocks on different physical devices, so that single-point failure does not affect overall availability. Each operation step in the reconstruction process generates a transaction log, allowing recovery to a consistent state after an abnormal interruption.

[0067] The storage optimization of the algorithm instance library adopts a hot and cold data separation strategy. High-frequency access active instances reside in non-volatile memory, and low-frequency usage instances are stored in compressed disk images. The loading manager predicts the calling mode of algorithm instances and decompresses instances that may be needed in advance to the memory buffer. The resource monitoring of the parameter tracker implements dynamic precision adjustment, increasing the sampling frequency for key algorithm instances and using statistical sampling methods for ordinary instances. The transmission of monitoring data uses differential encoding, sending only the change quantity to reduce network overhead.

[0068] The security strength evaluation module implements a multi-dimensional detection system. The side-channel analysis unit simulates timing attacks and power analysis attacks to detect whether the algorithm implementation has information leakage vulnerabilities. The mathematical security evaluation unit runs in a sandbox environment, attempting known algebraic attacks and lattice reduction attacks. The protocol consistency verification unit checks the deviation of algorithm instances from the standard protocol. The evaluation results are weighted to generate a comprehensive security score, which serves as an important basis for reconstruction triggering.

[0069] The parameter synchronization mechanism solves the data consistency problem in a distributed environment. When the system is deployed in a multi-node cluster, the dynamic parameter library adopts a master-slave replication architecture, with the master node receiving all parameter update operations and propagating changes to slave nodes through an optimistic replication protocol. The conflict detection algorithm identifies data races caused by concurrent modifications and resolves conflicts using timestamp ordering. The password graph indexer subscribes to parameter change events and performs incremental graph refresh operations after receiving update notifications to maintain real-time synchronization with the dynamic parameter library.

[0070] Referring to Figure 3, which shows the running state and performance characteristics of the dynamic parameter library. The upper left subgraph compares the performance of different algorithm instances. The three key indicators of computational complexity, resource occupancy, and security strength are compared using a bar chart. As can be seen from the figure, different algorithm instances have significant differences in various indicators. The upper right subgraph shows the trend of security strength over time, with five curves representing the security strength changes of the five algorithm instances, and the marked asterisk indicating the system triggered reconstruction event. When the security strength is close to or below the 0.7 threshold, the system triggers reconstruction operations such as security reconstruction, performance reconstruction, or protocol upgrade. The lower left subgraph reveals the relationship between resource occupancy and security strength, with the size of the dot representing the computational complexity and the color depth representing the security strength. The right lower subgraph shows the call distribution of each algorithm instance, reflecting the system's preference for different algorithm instances.

[0071] Example 2: see Figure 4 , focusing on the detailed operation mechanism of the cryptographic graph indexer, which is implemented by the index builder and weight optimizer to build and optimize the dynamic cryptographic algorithm index graph. The index builder uses the attribute graph model to store the algorithm instance relationship network, with each node corresponding to an independent post-quantum cryptographic algorithm instance. The node attributes include algorithm identifier, protocol version, hardware acceleration flag, and 128-dimensional feature vector. The creation of edges follows a strict double verification process: the protocol compatibility verification phase detects whether algorithm instances share the same underlying mathematical difficult problem structure, such as lattice encryption or hash signature system; the calculation dependency verification phase analyzes the algorithm input and output interface to confirm whether the ciphertext output format of the predecessor algorithm matches the ciphertext input specification requirements of the successor algorithm. The edge weight calculation uses a multi-factor scoring matrix, and the protocol compatibility score is generated by comparing the protocol metadata of the algorithm. The calculation dependency strength is evaluated by a prediction model trained from historical call data, and the two are combined in a predetermined proportion and normalized to convert to a weight value in the range of 0 to 1.

[0072] The dynamic weight adjustment mechanism uses an event-driven architecture. The weight optimizer continuously listens to the path weight data packets pushed by the vectorized scheduling engine, which contains path identifiers, access node sequences, and execution time. The pre-processing module performs three layers of filtering: the format verification layer discards abnormal structure packets, the service verification layer removes protocol version mismatch records, and the topology verification layer filters illegal node jump paths. Effective packets enter the frequency statistics module, which uses a dynamic sliding window counting mechanism, with the window length growing synchronously with the system scheduling period. Recent event weights show an exponential decay distribution over time. The edge weight correction amount calculation introduces an adaptive learning rate adjustment strategy, which dynamically scales the learning rate based on historical correction amplitude fluctuations to avoid weight value oscillation.

[0073] The integration process of the new algorithm instance establishes a standardized hot plug interface. When the parameter tracker detects a new instance registration event, it immediately sends a timestamped notification signal to the index builder. The node creation process is executed in three steps: the node initialization phase assigns a unique graph identifier and writes the basic metadata; the attribute filling phase calls the feature extraction service to generate embedding vectors; the relationship building phase starts the real-time compatibility test tool set. Compatibility testing is run in a secure sandbox environment, including seven dimensions such as protocol handshake testing, data format conversion testing, and resource contention testing. The test results automatically generate a collaborative relationship evaluation report. The new node relationship connection adopts a conservative strategy, only connecting with existing nodes that have passed all compatibility tests. The initial edge weight is assigned based on the quantitative score in the test report.

[0074] The edge weight update operation implements distributed transaction management. The optimizer maintains an edge weight version history log, creating a transaction snapshot before each update operation. The strength change value is checked and entered into the weighted update queue. The update execution stage uses an incremental superposition algorithm: new weight = old weight x decay coefficient + change value x gain coefficient. The graph structure locking mechanism implements hierarchical control: read operations apply shared locks to allow concurrent access, and write operations apply exclusive locks and set a timeout retry strategy. When multiple updates cause the weight of a certain edge to oscillate beyond the stability threshold, the self-healing module triggers a weight rollback operation, selecting a stable value from the historical version that meets the recent average access frequency requirements, and sending a weight abnormality notification to associated components.

[0075] The real-time graph optimization service runs continuously in the background. The defragmentation process periodically reorganizes the physical storage locations of nodes, placing high-frequency collaborative node clusters in adjacent storage areas. The index compression service analyzes the edge weight distribution characteristics and implements marked cleanup for long-term low-weight edges, reducing graph traversal overhead. The memory hotspot monitoring module identifies high-access-frequency subgraphs and caches them in non-volatile memory to speed up access. The graph partitioning service implements vertical splitting based on the collaborative density between nodes, and cross-partition access is achieved through relay nodes.

[0076] The collaborative relationship prediction model establishes an offline training mechanism. Historical path weight data is collected to build a training data set, and twenty-seven-dimensional features such as node attributes, path length, and execution delay are extracted through feature engineering. The graph neural network model learns the implicit association rules between nodes and generates relationship prediction heat maps regularly. The prediction results are converted into relationship strength estimates, which are used for the initialization of edge connections for new nodes. Model updates use a shadow deployment mode, with the new and old models running in parallel for a period of time to verify the prediction accuracy before performing a hot replacement.

[0077] The abnormal synergy link detector maintains system stability. When detecting a sudden drop in the weight of a high-frequency connected node pair, an automatic diagnosis process is triggered: the first round checks the protocol compatibility state, the second round reviews the calculation dependency interface, and the third round verifies the resource competition situation. The diagnosis report generates an abnormal identifier code to drive the weight optimizer to perform targeted correction operations. For persistent abnormal links, the system generates a relationship reconstruction proposal for the index builder to decide.

[0078] Referring to Figure 5 , the structural features and dynamic changes of the password graph indexer are shown. The upper left subgraph is a comprehensive weight distribution histogram, reflecting the overall strength distribution of the synergy relationship between algorithm instances. The upper right subgraph is a weight component analysis, showing the average contribution proportion of protocol compatibility, calculation dependency, and security similarity in the comprehensive weight. The lower left subgraph presents the weight changes of three typical edges over time, revealing the mechanism of the system dynamically adjusting algorithm synergy relationships. As can be seen from the figure, the edge weight fluctuates over time, and the system continuously optimizes the synergy relationship according to the running feedback. The lower right subgraph shows the relationship between protocol compatibility and calculation dependency, with the size of the point representing the comprehensive weight and the color depth representing the security similarity. The chart shows that high protocol compatibility often accompanies high calculation dependency, forming the synergy advantage area of the system.

[0079] Example 3: Referring to Figure 6 , it involves the deep operation details of the weight optimizer, which realizes the intelligent adjustment function of the edge weight in the dynamic password algorithm index graph. The path weight data processing adopts a streaming analysis architecture, and the input data packets are transmitted through an encrypted channel. The packet structure includes a header verification segment and a payload segment. The header verification segment uses CRC-32 cyclic redundancy check code to verify data integrity, and the payload segment uses ASN.1 encoding format to encapsulate path access information. The parsing engine establishes a three-stage processing pipeline: the first-stage unpacker separates the verification header and the payload; the second-stage decoder restores structured path data; and the third-stage cleaner filters illegal node jumps and timeout path records. Valid path data enters a distributed message queue for temporary storage, and the queue partition strategy is determined based on the hash value of the path starting node to ensure that access paths with the same source node are processed in a centralized manner.

[0080] The edge weight adjustment amount calculation introduces a dynamic learning rate mechanism, and the weight update formula is defined as:

[0081]

[0082] Where: represents the edge weight adjustment amount from node i to node j, is the baseline learning rate parameter, is the trigger frequency of the edge within the statistical period, is the total number of outgoing edges of node i, It is a smoothing factor to prevent division by zero errors. It is the execution time of the path. This is the time-normalization constant. This formula achieves a balance in three aspects: the frequency statistics term maintains the weight advantage of popular paths, the normalized denominator prevents a few edges from monopolizing resource adjustments, and the time-logarithmic term optimizes the execution efficiency of long paths. Learning rate. The system dynamically adjusts its load, automatically reducing the load during high loads to avoid drastic fluctuations and appropriately increasing the load during idle periods to accelerate convergence.

[0083] The frequency statistics module implements a time-decay counting method. Each path-triggered event is assigned a time-sensitive tag, and the counting weight decays exponentially over time. Events within the most recent hour retain their full weight, events between 1 and 3 hours ago have their weight halved, and events older than 3 hours are counted only with their base weight. The sliding window statistician maintains a circular buffer for each edge, recording the timestamps and additional weights of the most recent 1024 triggered events. When a new event causes the buffer to fill up, the oldest event record is automatically discarded. Statistical results generate a summary report every hour, including the standardized trigger frequency, average execution time, and weight change trends for each edge.

[0084] Intensity change values ​​are calculated to incorporate path topology features. In addition to basic frequency and time-consuming factors, a path length compensation factor is introduced, granting edges in short paths an additional gain coefficient. A critical path identifier analyzes the betweenness centrality of nodes in the graph, assigning adjustment priority to edges connected to nodes with high betweenness. Weight adjustment requests enter a priority queue, with a queue sorting strategy that comprehensively considers adjustment urgency and impact scope; adjustment requests for critical paths can be processed in advance.

[0085] Edge weight update operations are subject to transactional control. An operation snapshot is created before each adjustment, recording the current weight value and related statistics. Update execution employs a two-phase commit protocol: the preparation phase verifies whether the target edge has been concurrently modified; the commit phase atomically updates edge attributes and related statistics. The version controller maintains the modification history of each edge, retaining the 16 most recent versions for rollback selection. When three consecutive large adjustments in the same direction are detected, a weight stabilization check is automatically triggered, pausing subsequent adjustments until the fluctuation range returns to normal.

[0086] An anomaly handling mechanism establishes a multi-level defense system. The input data validation layer rejects path records containing illegal node references; the frequency statistics module sets an upper limit threshold to prevent malicious data manipulation; and weight adjustments are subject to amplitude limits, with each adjustment not exceeding 30% of the current value. When continuous abnormal fluctuations in the weight of an edge are detected, the isolation detector initiates a special analysis: first, it verifies the liveness status of associated nodes; second, it checks for protocol compatibility changes; and finally, it audits historical adjustment records. Upon confirmation of the anomaly, a diagnostic report is generated, and depending on the severity, options include weight reset, temporary edge disabling, or triggering node reconstruction.

[0087] The adaptive parameter adjustment system runs continuously in the background. Baseline learning rate Dynamic adjustment according to the overall stability of the system, the stability index calculates the variance coefficient of the last 100 weight updates. Time-consuming normalization constant Automatic calibration with hardware performance changes, periodically sample the execution time distribution of typical paths to recalculate the baseline value. Smoothing factor Adaptive adjustment according to the dispersion degree of the edge weight in the graph, to prevent numerical instability in the normalization process.

[0088] The interaction between the weight optimizer and the graph database adopts a batch processing mode. After accumulating enough adjustment requests, start a batch update transaction, package multiple edge weight modifications and execute them, reduce database lock competition. Update results are broadcast through the publish-subscribe mode, and related components subscribe to the edge change events they are interested in. For high-frequency modified edges, implement an incremental merging strategy to combine continuous small adjustments into a single meaningful update operation.

[0089] The visualization monitoring interface displays the weight dynamics in real time. The edge weight heat map uses gradient colors to represent the weight value size, and the dynamic change uses flow animation to illustrate. The topology change monitor highlights the recently adjusted edges and nodes, making it easy for administrators to grasp the system evolution trend. The history backtracking function can view the graph state snapshot at any time point, and supports weight change record retrieval by time range. The alarm panel displays abnormal edge states, including oscillation over-limit, frequent disconnection and other conditions that require manual intervention.

[0090] The version compatibility checker maintains long-term stability. When the algorithm instance library upgrades the protocol version, it automatically scans all related edges in the graph and marks the connections that may be affected. The compatibility risk assessment model predicts the impact of version upgrade on edge weights and generates a pre-adjustment proposal. Before major version changes, perform shadow testing to verify the effectiveness of the weight adjustment strategy in an isolated environment, and confirm that it is correct before applying it to the production system.

[0091] Embodiment 4: refer to Figure 7 , around the operation process of the vectorization scheduling engine, this component converts the password task request into a computable semantic vector, and selects the optimal algorithm instance through a multi-level matching mechanism. The original task request received by the semantic encoder contains structured fields and unstructured text, the structured fields include operation type enumeration value, data size classification and security level identification; unstructured text allows free-form task description, such as "need to quickly process a large amount of sensitive data for security signature". The feature extraction pipeline is divided into three processing stages: the syntax parser analyzes the structured fields to generate the basic feature vector, the natural language understanding model processes the description text to generate a 128-dimensional semantic embedding, and the feature fusion layer concatenates the two vectors and reduces them to a 64-dimensional final representation.

[0092] The priority scheduler maintains a copy of the embedding vectors of each node in the dynamic cryptographic algorithm index graph, which are obtained through offline training, and each vector captures the functional and performance characteristics of the corresponding algorithm instance. The similarity calculation adopts a hierarchical filtering strategy, with the first round of coarse screening based on local sensitive hashing to quickly narrow down the candidate range, and the second round of fine calculation using an improved similarity measurement method that considers the combined influence of node attributes and edge weights. The weight coefficient generation process refers to the topological characteristics of the edges in the graph, analyzing not only the strength of directly connected edges but also the conductive influence of indirect paths. Intermediate data generated during the task request processing process is recorded in the scheduling decision table, as shown in Table 1.

[0093] Table 1: Scheduling decision table.

[0094] ;

[0095] Multi-layer projection calculation is performed on a heterogeneous computing architecture, with CPUs handling control logic and lightweight operations, and GPUs accelerating dense vector operations. The projection space is divided into multiple sub-regions, each corresponding to a specific type of cryptographic operation requirement, such as digital signature area, key exchange area, etc. The region boundaries are dynamically adjusted according to historical task distribution, and the space reorganization process is triggered when new algorithm instances are added. During the candidate node screening process, the region division information is used to quickly locate the relevant node cluster, reducing unnecessary global calculations.

[0096] The cache management module implements intelligent result reuse. The task request feature vector is standardized to generate a cache key, which includes the operation type, data size level, and semantic vector hash value. The cache lookup uses the least recently used strategy, and when a hit occurs, the pre-stored algorithm selection instruction is directly returned, while the access timestamp is updated. The cache invalidation mechanism listens to changes in the algorithm instance library, and when relevant algorithm instances are modified or deleted, the corresponding cache entries are automatically cleared.

[0097] The fault tolerance processing flow guarantees the stability of scheduling. When numerical anomalies occur during similarity calculation, the backup calculation strategy is automatically switched; if the target instance is found to be unavailable during node selection, the backup path search is immediately started. The fault recovery log records abnormal situations and processing measures for subsequent system optimization. The resource monitoring module tracks CPU and memory usage in real time during scheduling, and automatically triggers the load balancing strategy when the threshold is exceeded.

[0098] Task priority management implements multi-queue scheduling. Emergency tasks enter the fast lane, skipping some preprocessing steps; ordinary tasks follow the standard process; and background batch tasks are processed in energy-saving mode. Resource isolation is implemented between queues to prevent low-priority tasks from blocking critical operations. The dynamic priority adjustment mechanism automatically raises the priority based on the task waiting time to avoid starvation.

[0099] The algorithm selection instruction generates a complete execution plan. In addition to the preferred algorithm instance, the instruction also lists a sequence of alternative solutions and failover conditions. Execution parameter tuning suggestions are generated based on historical run data, including recommended parallelism, memory allocation strategies, etc. The instruction code uses a platform-independent format to support reliable transmission across heterogeneous environments.

[0100] Version compatibility checks ensure that tasks match algorithms. Before finalizing the instruction, the checker verifies the task requirements against the protocol version of the target instance and automatically triggers rescheduling if incompatible. The version adaptation layer can handle minor version differences by converting parameters to achieve a smooth transition. When there is a major version mismatch, a detailed error report is generated to guide users to adjust task parameters.

[0101] The visual monitoring interface displays the scheduling decision process. The dynamic topology graph presents candidate nodes and their connection relationships, with color depth representing similarity scores. The task flow animation visually displays processing stages, including feature extraction, candidate selection, and final selection. Historical decision records support multi-dimensional search and analysis, helping to understand system behavior patterns.

[0102] The real-time debugging interface allows authorized users to intervene in the scheduling process. Debugging mode can manually specify algorithm instances for functional verification or inject simulated tasks to test system response. Performance profiling tools record the time distribution of each processing stage to identify potential bottlenecks. Debugging operations are executed in a sandbox environment and do not affect normal task flow.

[0103] The dynamic learning mechanism continuously optimizes the scheduling strategy. Collecting deviation data between actual execution results and expected performance is used to adjust similarity calculation parameters. The feedback loop is sensitive to delays, with recent task data receiving higher weights. Model updates use gradual deployment, first verifying effectiveness in a shadow system before gradually rolling out. Anomaly detectors monitor scheduling quality indicators and trigger alarms and recommend manual review when there is persistent deviation.

[0104] Embodiment 5: Covers the core operation process of the collaborative execution unit, which realizes the efficient collaboration of cryptographic operations through three functional modules: algorithm executor, parameter feedback, and collaborative controller. The algorithm executor deploys a containerized running platform, and each post-quantum cryptographic algorithm instance is allocated an independent container environment. Container resource quotas are dynamically configured according to instance types. Signature algorithm instances are allocated higher CPU quotas but limited to memory upper limits, while encryption instances are allocated larger memory spaces. Instance loading uses a double-layer cache strategy, where the first call loads the complete configuration image from the algorithm instance library and stores it in local cache, and subsequent calls directly reuse cached instances. The hot instance pool maintains the preloaded state of commonly used algorithm instances and dynamically adjusts the pool size based on call frequency.

[0105] The parameter feedbacker constructs a distributed monitoring network, and the probe agent is deployed in three types: a container-level agent binds to each algorithm instance container to collect process resource occupation indicators; a host-level agent monitors the physical resource allocation state; and a protocol-level agent is embedded in the algorithm execution process to capture protocol interaction events. Multi-granularity sampling is implemented for data collection, and adaptive sampling intervals are used for resource monitoring class data, with an initial sampling period of 50 milliseconds. When resource fluctuations intensify, the period is automatically compressed. Event-driven mode is used for protocol state capture, and snapshot recording is triggered when key protocol state transitions occur. The original monitoring data stream is filtered in two stages: the first stage removes abnormal burrs generated by system noise, and the second stage eliminates transient fluctuations through smoothing processing.

[0106] The collaborative controller implements topology management between algorithm instances. When task requirements trigger multi-algorithm collaboration, the controller constructs inter-instance data flow paths based on the latest collaboration relationship data provided by the dynamic cryptographic algorithm index map. The connection path is established using the optimal path algorithm, which preferentially selects low-latency and high-bandwidth node combinations. The data pipeline uses zero-copy shared memory technology, and the transmission control block pre-allocates a ring buffer. Producers and consumers exchange pointers to synchronize progress. The pipeline type is divided into calculation data pipelines and protocol state pipelines, the former transmits intermediate calculation results, and the latter synchronizes session state identifiers between algorithms.

[0107] The protocol state synchronization mechanism is designed as a hierarchical structure. The basic synchronization layer maintains protocol phase identifiers and error codes, and broadcasts change events through a lightweight message queue. The enhanced synchronization layer can optionally transmit detailed protocol parameter history, and the receiving instance can selectively subscribe as needed. The state consistency guarantee uses a version vector mechanism, each algorithm instance participating in collaboration maintains a local version number, and the state update carries version information. The conflict detection algorithm resolves state differences based on the time sequence relationship of version numbers.

[0108] The data processing pipeline of the parameter feedbacker includes a feature extraction module. After the input monitoring data stream is divided by the window, the feature extractor identifies key pattern points: peak duration and rising slope in resource consumption trajectory, and state transition delay and abnormal retry events in protocol interaction. The pattern features and original data are combined to generate differential data packets, and the data packet structure includes a standard header and a variable payload. The header records the data packet type label and time reference, and the payload stores the parameter change sequence after feature labeling. Time series difference encoding is used for data compression, which replaces the absolute value with the change amount between consecutive sampling points. Each data packet is attached with a redundancy check code, and data retransmission is triggered when the check fails.

[0109] The anomaly diagnosis module implements closed-loop monitoring. The algorithm instance container has a built-in health check endpoint that sends a running status heartbeat periodically. The coordination controller maintains an instance health state table, and instances that lose three consecutive heartbeats are marked as suspicious. The data pipeline deploys a traffic monitoring thread that, when it detects that no data has been transmitted for more than 200 milliseconds consecutively, initiates a pipeline integrity test: it sends a test data packet to verify channel functionality, and if a response is not received within a timeout, it determines that the pipeline has failed. Error handling uses a step-by-step response strategy: first, it attempts to restart the affected data pipeline, second, it reloads the algorithm instance container, and finally, it notifies the vectorization scheduling engine to reschedule the task.

[0110] The state recovery mechanism ensures business continuity. Key execution points of each algorithm instance set soft state checkpoints, storing memory snapshots to persistent storage. When the instance recovery process is triggered, the state is reloaded from the most recent valid checkpoint. The distributed checkpoint coordinator ensures state synchronization between multiple cooperating instances by pausing protocol execution for all instances and then creating a checkpoint uniformly, and aligning the state by timestamp when recovering. Incremental recovery mode allows some instances to continue running, and only the instances that have failed are reset and resynchronized with the current state.

[0111] The resource recycling system enables fine-grained management. Algorithm instances that have completed tasks enter a sleep countdown and are automatically unloaded from container resources after the idle time exceeds a specified threshold. The shared memory area implements reference counting cleanup, and the space is immediately recycled when all associated instances are unmapped. Long-period tasks implement a memory block rotation mechanism, where used memory blocks are actively released and then re-allocated to prevent memory fragmentation. Network connections use a connection multiplexing pool, and unused connections are safely closed after a predetermined time.

[0112] The container image update process seamlessly integrates with the running system. When the algorithm instance library pushes a new version of the image, the coordination controller coordinates the rolling upgrade strategy. Instances in an idle state are immediately updated to the new version, running instances are marked as pending update, and version switching is performed after the task is completed. The update process maintains the ability to run two versions in parallel, allowing tasks to roll back to the old version. The upgrade verification unit performs a smoke test to confirm that the new version is functioning properly, and then releases resources for the old version.

[0113] Event trace logs record the complete execution trajectory. Monitoring data is stored in a structured log format, with each record containing an exact timestamp, event type code, and related instance identifier. The log indexing service supports multi-dimensional queries, allowing historical analysis by task sequence, algorithm type, or resource metrics. Log storage implements a hierarchical strategy, with hot data stored in high-speed storage devices and cold data migrated to compressed long-term archives. Log secure storage uses an append-only design, with encryption and signature to prevent illegal tampering.

[0114] The security audit module implements runtime policies. All inter-container communications are forced to two-way certificate authentication, and protocol-level probes verify algorithm outputs against security specifications. Critical parameter modifications trigger an approval process, requiring dual authorization from the management side. Sensitive operations implement multi-factor audit trails, recording operator identity, time stamp, and operation content. The audit rule base is automatically updated on a regular basis, capturing new attack signatures and translating them into monitoring policies in a timely manner.

[0115] It is to be understood that the terminology used herein such as first and second, and the like, is only used to distinguish one entity or action from another entity or action, and does not necessarily require or imply that there is any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0116] While embodiments of the present application have been shown and described, it is to be understood that the embodiments described are merely divergences, modifications, substitutions and variations of the embodiments and are capable of numerous rearrangements, modifications and substitutions of parts and elements without departing from the spirit and scope of the present application, which are defined by the following claims and their equivalents.

Claims

1. A dynamic reconfigurable post-quantum cryptography multi-algorithm collaborative acceleration system, characterized in that, The dynamic parameter library, the password graph indexer, the vectorization scheduling engine and the cooperative execution unit are included. The dynamic parameter library is used for storing a real-time running parameter set of a post-quantum cryptography algorithm and dynamically updating the real-time running parameter set according to feedback of the cooperative execution unit, and the real-time running parameter set includes algorithm calculation complexity, resource occupation rate and security strength index. The password graph indexer is used for obtaining the real-time running parameter set and constructing a dynamic cryptography algorithm index graph; wherein the password graph indexer includes an index constructor and a weight optimizer; the index constructor is used for obtaining post-quantum cryptography algorithm instances in an algorithm instance library, taking each algorithm instance as a node, and creating cooperative relationship edges between nodes based on algorithm protocol compatibility and calculation dependency; the weight optimizer is used for calculating strength change values of each cooperative relationship edge according to path weight data fed back by the vectorization scheduling engine, and dynamically adjusting edge weights of the cooperative relationship edges; wherein the path weight data contains node paths triggered in an algorithm selection instruction execution process, and the dynamic adjustment includes: counting trigger frequencies and edge weight correction amounts of each node path, generating strength change values of the cooperative relationship edges, and dynamically superimposing and updating edge weights of corresponding cooperative relationship edges in the dynamic cryptography algorithm index graph according to the strength change values. The vectorization scheduling engine is used for generating a task semantic vector by vectorizing and coding a password task request input by a user, calculating similarity based on the task semantic vector and embedding vectors of each node in the dynamic cryptography algorithm index graph, and generating an algorithm selection instruction; wherein the vectorization scheduling engine includes a semantic encoder and a priority scheduler; the semantic encoder is used for performing multi-modal feature extraction on the password task request input by the user to generate the task semantic vector; and the priority scheduler is used for calculating initial similarity of the task semantic vector and embedding vectors of each node in the dynamic cryptography algorithm index graph, and positioning a target node set with the highest similarity in an embedding space through multi-layer projection calculation. The cooperative execution unit is used for receiving the algorithm selection instruction, calling a corresponding post-quantum cryptography algorithm instance to perform a password operation, and feeding back parameter changes in the operation process to the dynamic parameter library.

2. The dynamic reconfigurable post-quantum cryptography multi-algorithm cooperative acceleration system of claim 1, wherein, The dynamic parameter library includes an algorithm instance library and a parameter tracker. The algorithm instance library is used for storing initialization configurations and real-time state data of post-quantum cryptography algorithm instances. The parameter tracker is used for monitoring a running process of the post-quantum cryptography algorithm instance, and extracting a real-time running parameter set from the running process. The parameter tracker is also used for incrementally updating real-time state data in the algorithm instance library according to parameter change data fed back by the cooperative execution unit.

3. The dynamic reconfigurable post-quantum cryptography multi-algorithm co- acceleration system of claim 2, wherein, The password graph indexer includes an index constructor and a weight optimizer. The index constructor is configured to obtain post-quantum cryptographic algorithm instances in the algorithm instance library, take each algorithm instance as a node, and create a collaborative relationship edge between the nodes based on algorithm protocol compatibility and calculation dependency relationship; wherein the protocol compatibility is verified by detecting whether the algorithm instances share the same basic mathematical difficult problem structure, and the calculation dependency relationship is verified by analyzing algorithm input and output interfaces and confirming that the output format of the predecessor algorithm matches the input specification of the successor algorithm; The weight optimizer is configured to calculate the strength change value of each collaborative relationship edge according to the path weight data fed back by the vectorization scheduling engine, and dynamically adjust the edge weight of the collaborative relationship edge.

4. The dynamic reconfigurable post-quantum cryptography multi-algorithm co- acceleration system of claim 3, wherein, The vectorization scheduling engine includes a semantic encoder and a priority scheduler; The semantic encoder is configured to perform multi-modal feature extraction on the password task request input by the user to generate a task semantic vector; wherein the multi-modal feature extraction includes processing and fusing the structured fields and unstructured texts in the password task request respectively; The priority scheduler is configured to calculate the initial similarity between the task semantic vector and the embedding vector of each node in the dynamic cryptographic algorithm index graph, generate an adjustment weight coefficient in combination with the edge weight of the cryptographic graph indexer, correct the initial similarity according to the adjustment weight coefficient, and generate an algorithm selection instruction; wherein the adjustment weight coefficient is generated according to the edge weight of the collaborative relationship edge between the target nodes, and is corrected by multiplying the initial similarity.

5. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 4, wherein, The collaborative execution unit includes an algorithm executor and a parameter feedback device; The algorithm executor is configured to activate the corresponding post-quantum cryptographic algorithm instance according to the algorithm selection instruction, perform cryptographic operations, and output an encrypted result; The parameter feedback device is configured to collect resource consumption trajectories and protocol interaction states during the running of the post-quantum cryptographic algorithm instance, generate parameter change data, and send the parameter change data to the dynamic parameter library.

6. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 3, wherein, The index constructor performs the following operations: Obtain the algorithm calculation complexity and security strength indicators in the real-time running parameter set, and calculate the protocol compatibility degree between algorithm instances; wherein the protocol compatibility degree is obtained by comparing the protocol metadata with the algorithm to generate a protocol compatibility degree score; Generate an initial weight of the collaborative relationship edge according to the protocol compatibility degree and the calculation dependency relationship; When a new post-quantum cryptographic algorithm instance is added, create a new node based on the real-time state data of the parameter tracker, and connect the collaborative relationship edge with the existing node.

7. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 6, wherein, The weight optimizer performs the following operations: Receive the path weight data fed back by the vectorization scheduling engine, wherein the path weight data includes the node path triggered during the execution of the algorithm selection instruction and the execution time; Statistically count the triggering frequency of each node path, and generate a strength change value of the collaborative relationship edge according to the triggering frequency and the execution time; According to the strength change value, the edge weight of the corresponding collaborative relationship edge in the dynamic cryptographic algorithm index graph is dynamically updated.

8. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 4, wherein, The priority scheduler performs the following operations: Calculate the initial similarity between the task semantic vector and the embedding vector of each node in the dynamic cryptographic algorithm index graph; mapping the task semantic vector to an embedding space of the dynamic cryptographic algorithm index graph, locating a target node set with the highest similarity in the embedding space through multi-layer projection calculation; locating a target node set with the highest similarity in the embedding space through multi-layer projection calculation; generating an adjustment weight coefficient according to the edge weight of the cooperative relationship edge between the target nodes fed back by the cryptographic graph indexer; multiplying the initial similarity of the target node set by the adjustment weight coefficient to generate a final similarity ranking and output an algorithm selection instruction.

9. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 2, wherein, The dynamic parameter library further comprises a reconstruction executor; The reconstruction executor is configured to perform structural reorganization on the post-quantum cryptographic algorithm instance in the algorithm instance library when the parameter tracker monitors that the security strength indicator is lower than a preset threshold; The structural reorganization comprises: decomposing the component dependency relationship of the algorithm instance, recombining the algorithm module according to the cooperative relationship edge of the cryptographic graph indexer, and updating the real-time running parameter set; wherein the structural reorganization is executed in stages, comprising: a dependency analysis stage for parsing the call graph of the algorithm instance to identify the data flow and control dependency between modules; a component extraction stage for segmenting the binary code of the algorithm instance into logical units; and a recombination stage for querying the cryptographic graph indexer and recombining the algorithm module based on the cooperative relationship graph.

10. The dynamic reconfigurable post-quantum cryptography multi- algorithm co-acceleration system of claim 5, wherein, The cooperative execution unit further comprises a cooperative controller; The cooperative controller is configured to establish an instance-to-instance data pipeline according to the cooperative relationship edge of the dynamic cryptographic algorithm index graph when the algorithm executor calls multiple post-quantum cryptographic algorithm instances; synchronizing the intermediate calculation results and protocol states between algorithm instances through the data pipeline, and inputting the synchronized data stream into the parameter feedback device to generate parameter change data; wherein the protocol state comprises a protocol phase identifier and an error code.

Images