Agent dynamic address and identity management system based on cache capability

Abstract

The invention discloses a proxy dynamic address and identity management system based on cache capability, and the system comprises a capability feature extraction module which is used for obtaining a capability attribute set of a proxy in real time, carrying out the denoising and normalization of the capability attribute set, and generating a standardized capability feature set; and a high-dimensional semantic coding module connected with the capability feature extraction module and used for mapping the standardized capability feature set to a high-dimensional feature space by using a preset semantic coding model to generate a high-dimensional feature vector. Perfect balance is achieved between data timeliness and resource efficiency through the system, the updating strategy periodically scans capacity change through a fixed time window, and the passive updating strategy can respond to model upgrading or hardware recovery emergency state events. The combination of the two ensures that the data in the edge node cache is always kept synchronous with the proxy body state of the physical world, and the storage space is prevented from being fully occupied by invalid historical data through a first-in first-out strategy.

Description

A dynamic address and identity management system for proxies based on caching capabilities Technical Field

[0001] This invention belongs to the field of AI agent interconnection technology, specifically a dynamic address and identity management system for agents based on caching capabilities. Background Technology

[0002] With the rapid development of AI agent technology, the agent economy is gradually emerging, and large-scale, cross-domain, and multi-specification collaboration has become the mainstream application scenario, especially in the financial, industrial, and office collaboration fields of large companies, where a large number of agents are needed to achieve accurate interconnection and efficient collaboration. In existing technologies, the identification of agents mainly relies on static IP addresses or Uniform Resource Locators (URLs). This static identification method has many technical bottlenecks and cannot adapt to the needs of dynamically changing agent capabilities.

[0003] First, existing static identification methods can only identify the agent's identity, failing to associate it with dynamic capability attributes (such as model version, processing power, and domain of expertise). This prevents requesting agents from achieving precise capability matching based on their own needs, forcing them to search for suitable collaborating agents through traversal searches. This significantly reduces the search efficiency for large-scale agent collaboration and increases bandwidth consumption of network discovery protocols, resulting in wasted network resources. Second, in cross-domain agent collaboration, static identification suffers from high addressing latency, failing to achieve sub-second node location and impacting the real-time performance of agent collaboration. This problem is particularly pronounced in latency-sensitive scenarios such as financial transactions and industrial scheduling. Third, existing identification methods lack effective caching mechanisms, requiring recalculation and searching for each address, increasing hardware computing power consumption and reducing hardware resource utilization. Furthermore, with the widespread adoption of the domestic HarmonyOS system and domestic large-scale models, existing identification systems lack corresponding interface adaptation capabilities, failing to achieve compatibility and interconnection with agent systems from different vendors. They are also vulnerable to circumvention by adversaries through protocol name changes and interface modifications, hindering the promotion and application of the technology and the effectiveness of legal protection. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a dynamic address and identity management system for proxies based on caching capabilities, so as to at least partially solve the above-mentioned technical problems.

[0005] The technical solution adopted in this invention is as follows: This invention proposes a dynamic address and identity management system for proxies based on caching capabilities, comprising: a capability feature extraction module, used to acquire the capability attribute set of the proxy in real time, and to perform denoising and normalization processing on the capability attribute set to generate a standardized capability feature set; a high-dimensional semantic encoding module, connected to the capability feature extraction module, used to map the standardized capability feature set to a high-dimensional feature space using a preset semantic encoding model to generate a high-dimensional feature vector; and an AID generation module, connected to the high-dimensional semantic encoding module, used to process the high-dimensional feature vector based on a locality-sensitive topology algorithm. The system performs a hash calculation to generate a dynamic proxy identifier containing a capability hash value, a home domain identifier, a timestamp, and a checksum. A caching module, connected to the AID generation module, stores the dynamic proxy identifier and its corresponding high-dimensional feature vector to edge nodes and performs cache update operations. A distributed addressing module, connected to both the caching module and the high-dimensional semantic encoding module, receives addressing requests, generates a required feature vector, and matches the target dynamic proxy identifier in a distributed hash table based on cosine similarity. An interface adaptation module performs format conversion and verification of communication protocols for different operating systems and large model interfaces.

[0006] In one embodiment of the present invention, the denoising process in the capability feature extraction module specifically employs a wavelet threshold denoising algorithm, which removes high-frequency noise data from the capability attribute set by setting a soft threshold function; the normalization process specifically employs a min-max normalization method, which linearly maps the numerical data in the capability attribute set to the closed interval [0,1], and its calculation formula is as follows: ;in Represents the normalized i-th Item ability attribute value, The values ​​represent the original capability attribute values. min(P) and max(P) represent the minimum and maximum values ​​in the capability attribute set P, respectively, in order to eliminate the weight interference of capability attributes with different dimensions on the subsequent encoding process.

[0007] In one embodiment of the present invention, the Locality Sensitive Topology Algorithm used in the AID generation module is specifically configured with a Gaussian hash function, the hash calculation process of which is defined as: h(V)=⌊(w⋅V+b) / r⌋; where V is the input high-dimensional feature vector, is a random projection vector that follows a standard normal distribution, b is an offset that is uniformly distributed in the interval [0,r], and r is a preset hash bucket width parameter; the AID generation module performs multiple independent hash operations, and concatenates and combines the generated multiple hash values ​​to form the capability hash value in the dynamic proxy identifier, so that feature vectors that are close in distance in the high-dimensional space are mapped to the same hash bucket with high probability, thereby realizing the quantitative fingerprint expression of capability similarity.

[0008] In one embodiment of the present invention, the cache update mechanism established by the cache module includes an active update strategy and a passive update strategy. The active update strategy is configured with a fixed time window, the length of which is 1 to 5 minutes. At the end of each time window, the capability feature extraction module re-collects the capability attribute set of the agent. If the Euclidean distance change rate between the newly collected capability attribute set and the historical capability attribute set stored in the cache exceeds 10%, the AID regeneration process is triggered and the old cached data in the edge node is replaced. The passive update strategy listens to the status events of the agent. When it detects that the agent has undergone a model version upgrade, hardware failure recovery, or task type switch event, it immediately receives the update instruction sent by the agent and forcibly triggers the immediate refresh of the cached content.

[0009] In one embodiment of the present invention, the specific adaptation method of the interface adaptation module for the HarmonyOS operating system is as follows: the dynamic proxy body identifier is encapsulated using the HarmonyOS distributed soft bus protocol, the binary data stream in the identifier is converted into a service discovery message format recognized by the HarmonyOS system, and the message header field is optimized to reduce transmission overhead; the specific adaptation method for the domestic large model interface standard is as follows: a two-way handshake verification mechanism is added to the data transmission link, a key token is exchanged before sending the dynamic proxy body identifier, and the integrity of the check code in the identifier is verified at the receiving end. If the verification fails, the data packet is discarded and a security log is recorded to ensure data consistency and anti-tampering capability when communicating across heterogeneous platforms.

[0010] In one embodiment of the present invention, an identity verification module is further included, which is connected in series with the output of the distributed addressing module. When the distributed addressing module returns the target proxy node information, the identity verification module extracts the dynamic proxy identifier fed back by the target node and parses the verification code and home domain identifier fields therein. The identity verification module has a built-in asymmetric encryption verification algorithm, which uses a pre-stored public key to decrypt the verification code and compares the decryption result with the home domain identifier. If the comparison result is inconsistent or the verification code expires, the target node is determined to be a forged node and the connection request is blocked. The final data communication channel is only established when the verification is successful.

[0011] In one embodiment of the present invention, the cache module further includes a historical snapshot storage unit. The historical snapshot storage unit uses a version chain structure to store the identifiers of dynamically generated proxies and their corresponding timestamps. When an address tracing request occurs, the historical snapshot storage unit can retrieve the high-dimensional feature vectors of historical versions according to a specified time range to reproduce the capability status of the proxies within the time period. At the same time, the historical snapshot storage unit is set with a size limit. When the storage capacity reaches a preset threshold, the historical snapshot data with the earliest timestamp is automatically deleted to maintain the storage space balance of the edge nodes and prevent retrieval delays caused by the accumulation of historical data.

[0012] In one embodiment of the present invention, the AID generation module introduces a dynamic weight adjustment mechanism when generating dynamic proxy identifiers. The mechanism automatically adjusts the weight coefficients of each dimension in the high-dimensional feature vector according to the professional field type to which the proxy belongs. For computing power sensitive fields, the weight coefficient of the computing power dimension is increased by 1.5 times; for latency sensitive fields, the weight coefficient of the response latency dimension is increased by 1.5 times. The weight coefficients directly participate in the dot product operation of the Gaussian hash function, so that the generated capability hash value can better reflect the core capability differences in specific scenarios, thereby improving the accuracy of distributed addressing in specific application scenarios.

[0013] In one embodiment of the present invention, the distributed addressing module integrates a load balancing routing strategy during the addressing process. When multiple target agent nodes with cosine similarity values ​​higher than a preset threshold are calculated, the distributed addressing module further obtains the real-time load rate and network round-trip latency data of each target node. The system constructs a comprehensive scoring function, using cosine similarity as the first weight factor, the reciprocal of the real-time load rate as the second weight factor, and the reciprocal of the network round-trip latency as the third weight factor, and calculates the comprehensive score of each candidate node by weighting. Finally, the node with the highest comprehensive score is selected as the addressing result and returned, thereby achieving uniform load distribution and optimal response speed for the collaborative tasks of the entire network agent while ensuring the accuracy of capability matching.

[0014] The beneficial effects of the technical solution of this invention are as follows: This invention achieves a perfect balance between data timeliness and resource efficiency through the system. The update strategy uses a fixed time window to periodically scan for changes in capabilities, preventing data staleness caused by slow drift. The passive update strategy can respond to sudden events such as model upgrades or hardware recovery. The combination of the two ensures that the data in the edge node cache is always synchronized with the agent state in the physical world. At the same time, the version chain structure and automatic cleanup mechanism introduced by the historical snapshot storage unit not only meet the audit traceability requirements for the integrity of historical data, but also avoid the storage space being filled with invalid historical data through the first-in-first-out strategy, effectively preventing the performance degradation of edge devices due to storage overflow, and realizing lightweight and intelligent full lifecycle data management.

[0015] This invention breaks through the limitations of single similarity matching in the decision quality of distributed addressing. The load balancing routing strategy integrated in the distributed addressing module incorporates three dimensions—capacity matching degree, real-time node load rate, and network round-trip latency—into a comprehensive scoring function. The optimal solution with the highest comprehensive score is selected through weighted calculation, thereby achieving uniform load distribution and optimal response speed across the entire network and improving the overall throughput and stability of large-scale clusters in high-concurrency scenarios.

[0016] The interface adaptation module of this invention breaks down the communication barriers between heterogeneous operating systems and the large model ecosystem by encapsulating the HarmonyOS distributed soft bus protocol and the bidirectional handshake verification mechanism of the domestic large model interface. It not only achieves seamless conversion of binary data streams to service discovery messages, but also ensures data consistency in cross-platform transmission through key exchange and integrity verification, enabling the entire system to achieve plug-and-play and secure interconnection in complex and diverse domestic software and hardware environments.

[0017] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of embodiments taken in conjunction with the accompanying drawings, in which: Figure 1 is a system framework diagram of a proxy dynamic address and identity management system based on caching capability proposed in an embodiment of the present invention; Figure 2 is a first system flowchart of a proxy dynamic address and identity management system based on caching capability proposed in an embodiment of the present invention; Figure 3 is a second system flowchart of a proxy dynamic address and identity management system based on caching capability proposed in an embodiment of the present invention; and Figure 4 is a third system flowchart of a proxy dynamic address and identity management system based on caching capability proposed in an embodiment of the present invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0020] The following describes an embodiment of the present invention, a dynamic address and identity management system for proxies based on caching capabilities, with reference to the accompanying drawings.

[0021] As shown in Figures 1 to 4, this embodiment of the invention provides a dynamic address and identity management system for proxies based on caching capabilities, comprising: a capability feature extraction module, used to acquire the capability attribute set of the proxy in real time, and to perform denoising and normalization processing on the capability attribute set to generate a standardized capability feature set; a high-dimensional semantic encoding module, connected to the capability feature extraction module, used to map the standardized capability feature set to a high-dimensional feature space using a preset semantic encoding model to generate a high-dimensional feature vector; an AID generation module, connected to the high-dimensional semantic encoding module, used to perform hash calculation on the high-dimensional feature vector based on the locality-sensitive topology algorithm to generate a dynamic proxy identifier containing a capability hash value, a home domain identifier, a timestamp, and a checksum; a caching module, connected to the AID generation module, used to store the dynamic proxy identifier and its corresponding high-dimensional feature vector to edge nodes and perform cache update operations; a distributed addressing module, connected to the caching module and the high-dimensional semantic encoding module respectively, used to receive addressing requests, generate a required feature vector, and match the target dynamic proxy identifier in a distributed hash table based on cosine similarity; and an interface adaptation module, used to perform format conversion and verification of communication protocols for different operating systems and large model interfaces.

[0022] In a specific application of this invention, after the system starts running, the capability feature extraction module first intervenes in the workflow. The module continuously collects raw capability attribute data from various agent nodes. The raw data is mixed with high-frequency noise caused by network fluctuations or sensor errors. The capability feature extraction module then calls the built-in wavelet threshold denoising algorithm to clean the data. By setting a reasonable soft threshold function, useless noise signals are removed. Then, the module performs min-max normalization calculation, linearly mapping all numerical capability indicators of different dimensions to a closed interval of 0 to 1, eliminating the weight bias caused by different units of computing power, memory, and bandwidth, and finally outputting a standardized and clean capability feature set for use in subsequent stages.

[0023] After the standardized capability feature set is generated, it is immediately transmitted to the high-dimensional semantic encoding module. The high-dimensional semantic encoding module loads the preset deep semantic encoding model and sends the feature set as the input vector into the neural network structure. The multi-layer nonlinear transformation mechanism inside the model starts to operate, mapping the low-dimensional discrete capability indicators to a continuous high-dimensional feature space. Agents with similar semantics or capability structures will naturally be closer in geometric distance, while agents with huge differences will be pushed away. The high-dimensional semantic encoding module finally outputs a dense high-dimensional feature vector that can accurately represent the comprehensive capability of the agent.

[0024] Once the high-dimensional feature vector is generated, it enters the processing flow of the AID generation module. The AID generation module integrates a topology algorithm based on the principle of locality-sensitive hashing. The algorithm uses a set of random projection vectors that follow a standard normal distribution to perform a dot product operation with the input high-dimensional feature vector, and then adds a uniformly distributed offset and divides it by the width of the hash bucket for rounding. This ensures that vectors that are close in distance in the original high-dimensional space have a high probability of being mapped to the same hash bucket. The AID generation module concatenates the results of multiple independent hash operations to form a capability hash value, and adds a domain identifier representing the management affiliation, a timestamp marking the generation time, and a check code for integrity verification. These four parts are assembled into a unique dynamic agent identifier. This identifier not only represents the identity of the agent, but also directly encodes its current capability status.

[0025] The dynamic proxy identifier and its corresponding high-dimensional feature vector are then pushed to the cache module. The cache module writes the data to the storage media distributed across the edge nodes. To improve the timeliness of the data, the cache module runs a dual update mechanism. On the one hand, the module maintains a fixed-length time window. Whenever the time window expires, the module instructs the capability feature extraction module to re-collect the data. If the Euclidean distance between the new and old data exceeds a preset threshold of 10%, the module automatically triggers the AID regeneration process and overwrites the old cache. On the other hand, the cache module constantly monitors the status event signals from the proxy. Once it detects a sudden event such as a model version upgrade or hardware failure recovery, the module will immediately force a refresh of the cache content of the corresponding node to ensure that the edge nodes always store the latest capability profile of the proxy.

[0026] When an external service initiates an addressing request, the distributed addressing module begins to work. The distributed addressing module first parses the intent description in the request, calls the high-dimensional semantic encoding module to convert the intent text into a demand feature vector, and then the module searches in a distributed hash table composed of all edge nodes. The core of the search process is to calculate the cosine similarity between the demand feature vector and all high-dimensional feature vectors in the cache. The module calculates the cosine value of the angle through dot product operation and modulus calculation. If the value exceeds the preset threshold of 0.85, the distributed addressing module determines that it has found a target agent with matching capabilities. If the first round of search is unsuccessful, the module will automatically expand the query radius and try again until the optimal match is locked or the maximum search depth is reached. This process realizes intelligent addressing based on capability semantics rather than fixed IP addresses.

[0027] After the addressing result is determined, the interface adaptation module takes over the subsequent communication establishment task. The interface adaptation module identifies the operating system type and large model interface standard of the target proxy. If the target is in the HarmonyOS environment, the module encapsulates the dynamic proxy identifier into a message format that conforms to the HarmonyOS distributed soft bus protocol and optimizes the header fields to adapt to its service discovery mechanism. If the target is connected to a domestic large model interface, the module first initiates a two-way handshake verification, exchanges key tokens and verifies the checksum in the identifier. Only after the verification is passed will the module perform the communication protocol format conversion, completely eliminating communication barriers between heterogeneous platforms and ensuring that the instructions can be accurately delivered to the target proxy.

[0028] After the distributed addressing module returns the target node information, the identity verification module intercepts the dynamic proxy identifier returned by the target, extracts the checksum and home domain identifier, and uses the pre-stored asymmetric encryption public key to decrypt and compare the checksum. If the decryption result does not match the home domain or the timestamp has expired, the identity verification module immediately determines that the node is a fake node and blocks the connection. Only requests that pass the verification completely will be allowed to establish the final data channel. This mechanism effectively prevents malicious nodes from impersonating legitimate proxies to access the system.

[0029] The historical snapshot storage unit inside the cache module also synchronously records the system status. The unit adopts a version chain structure to store the identifier and timestamp of each generated dynamic agent in sequence. When historical tracing or fault analysis is required, the system can retrieve these historical snapshots according to the specified time range to reproduce the capability status of the agent at a certain point in the past. At the same time, the unit has a capacity limit. Once the storage space reaches the threshold, the unit will automatically delete the oldest historical data to maintain system operating efficiency. This retains the necessary audit clues while avoiding the unlimited occupation of storage resources.

[0030] In specific scenarios, the AID generation module also activates a dynamic weight adjustment mechanism. The mechanism automatically adjusts the weight coefficients of each dimension in the high-dimensional feature vector according to the professional field to which the current task belongs. For example, when processing computing-intensive tasks, the module will increase the weight of the computing power dimension by 1.5 times, while when processing latency-sensitive tasks, it will increase the weight of the response latency dimension. The adjusted weights directly participate in the hash calculation, making the generated capability hash value more reflective of the core capability differences in specific scenarios. This allows the distributed addressing module to maintain extremely high matching accuracy in complex and ever-changing business scenarios.

[0031] Before finally selecting the target node, the distributed addressing module executes a load balancing routing strategy. When the cosine similarity of multiple candidate nodes meets the standard, the module further collects the real-time load rate and network round-trip latency data of these nodes, constructs a comprehensive scoring function, and calculates the comprehensive score of each node using similarity, the reciprocal of load rate, and the reciprocal of latency as weighting factors. Finally, only the node with the highest score is returned as the addressing result, ensuring that the system can achieve accurate matching of capabilities while also evenly distributing tasks across the entire network of agents, avoiding single-point overload and maximizing the overall response speed. All modules work together to form a closed-loop, adaptive, and highly reliable dynamic address and identity management system.

[0032] In one specific implementation, the denoising process in the capability feature extraction module employs a wavelet threshold denoising algorithm, which removes high-frequency noise data from the capability attribute set by setting a soft threshold function; the normalization process specifically employs a min-max normalization method, which linearly maps the numerical data in the capability attribute set to the closed interval [0,1], and its calculation formula is as follows: ;in Represents the normalized i-th Item ability attribute value, The values ​​represent the original capability attribute values, and min(P) and max(P) represent the minimum and maximum values ​​in the capability attribute set P, respectively, to eliminate the weight interference of capability attributes with different dimensions on the subsequent encoding process. The local sensitive topology algorithm used in the AID generation module is specifically configured with a Gaussian hash function, and its hash calculation process is defined as: h(V)=⌊(w⋅V+b) / r⌋; where V is the input high-dimensional feature vector, which is a random projection vector that follows a standard normal distribution, b is the offset that is uniformly distributed in the interval [0,r], and r is the preset hash bucket width parameter; the AID generation module performs multiple independent hash operations, and concatenates and combines the generated hash values ​​to form the capability hash value in the dynamic proxy identifier, so that feature vectors that are close in distance in the high-dimensional space are mapped to the same hash bucket with high probability, thereby realizing the quantitative fingerprint expression of capability similarity.

[0033] In a specific application of this invention, after the system starts running, the capability feature extraction module first intervenes in the workflow. The module continuously collects raw capability attribute data from each agent node. The raw data contains high-frequency noise caused by network fluctuations or sensor errors. The capability feature extraction module immediately calls the built-in wavelet threshold denoising algorithm to perform deep cleaning of the data. The algorithm decomposes the raw signal into sub-bands of different frequencies, and automatically identifies and removes high-frequency noise components representing random interference by setting a reasonable soft threshold function. At the same time, it completely retains the low-frequency effective signal that reflects the true capability level of the agent. After this processing, the signal-to-noise ratio of the data is improved. The module performs min-max normalization calculation, linearly mapping all numerical capability indicators of different dimensions to a closed interval of 0 to 1. The calculation process strictly follows the formula: ;in Represents the normalized i-th Item ability attribute value, The original capability attribute value is represented by min(P) and max(P), which represent the minimum and maximum values ​​in the capability attribute set P, respectively. This eliminates the weight bias caused by different units of computing power, memory, and bandwidth, enabling various indicators with huge numerical differences to be compared fairly on the same scale. The standardized capability feature set is then transmitted to the AID generation module. The AID generation module integrates a Gaussian hash function based on the principle of locality-sensitive topology. The module performs a dot product operation between the input high-dimensional feature vector V and a random projection vector w that follows a standard normal distribution, and then adds an offset b that is uniformly distributed in the interval [0, r]. Finally, it divides by the preset hash bucket width parameter r and rounds down to obtain the single hash value h(V) = ⌊(w⋅V+b) / r⌋.

[0034] The AID generation module independently and repeatedly performs this hash operation multiple times, each time using a different random projection vector and offset. The generated hash values ​​are sequentially concatenated to form the core part of the dynamic proxy identifier, namely the capability hash value. This ensures that feature vectors that are geometrically close in the high-dimensional feature space have a probability of being mapped to the same hash bucket, while vectors that are far apart are distinguished. This transforms the abstract capability similarity into a concrete quantitative fingerprint expression, solving the problems of inconsistent dimensions and noise interference in multi-source heterogeneous data. It also achieves a deep binding between the proxy identity and its dynamic capability state, enabling the system-generated identifier to sensitively perceive subtle changes in the proxy's capabilities. This lays a solid mathematical foundation for subsequent cache-based fast addressing and accurate matching, ensuring that the entire system can efficiently and accurately complete the tasks of proxy identity management and dynamic address resolution in a large-scale distributed environment.

[0035] In one specific implementation, the cache update mechanism established by the cache module includes an active update strategy and a passive update strategy. The active update strategy is configured with a fixed time window, the length of which is 1 to 5 minutes. At the end of each time window, the capability feature extraction module re-collects the capability attribute set of the agent. If the Euclidean distance change rate between the newly collected capability attribute set and the historical capability attribute set stored in the cache exceeds 10%, the AID regeneration process is triggered and the old cached data in the edge nodes is replaced. The passive update strategy listens to the status events of the agent. When it detects that the agent has undergone a model version upgrade, hardware failure recovery, or task type switch event, it immediately receives the update instruction sent by the agent and strongly... To trigger real-time refresh of cached content, the interface adaptation module adapts to the HarmonyOS operating system by encapsulating the dynamic proxy identifier using the HarmonyOS distributed soft bus protocol, converting the binary data stream in the identifier into a service discovery message format recognized by the HarmonyOS system, and optimizing the message header fields to reduce transmission overhead. For the domestic large-scale interface standard, a two-way handshake verification mechanism is added to the data transmission link. Key tokens are exchanged before sending the dynamic proxy identifier, and the integrity of the checksum in the identifier is verified at the receiving end. If verification fails, the data packet is discarded and a security log is recorded to ensure data consistency and tamper-proof capabilities during cross-heterogeneous platform communication.

[0036] In practical applications, this invention first activates its built-in dual update mechanism to ensure continuous synchronization of the latest state of the agent. The mechanism includes two parallel parts: an active update strategy and a passive update strategy. The active update strategy is configured with a fixed time window whose length can be flexibly adjusted between 1 and 5 minutes. Whenever this time window reaches its end, the capability feature extraction module immediately takes action to re-collect the capability attribute sets of all agents. The system compares the newly collected real-time data with the historical capability attribute sets stored in the cache, calculates the Euclidean distance between the two and obtains the rate of change. Once this rate of change exceeds a preset threshold of 10%, the system determines that the agent's capability has drifted and automatically triggers the AID regeneration process. The newly generated dynamic agent identifier completely replaces the old cached data in the edge nodes, thereby ensuring the accuracy of data in the static time dimension.

[0037] Meanwhile, the passive update strategy continuously monitors various status event signals from the agent in the background. When the system detects a sudden critical event such as model version upgrade, hardware failure recovery, or task type switch in the agent, the passive update strategy will immediately receive the update instruction actively sent by the agent. Instead of waiting for the time window to expire, it will directly force the instant refresh operation of the cached content. This two-pronged approach ensures that both slow capability changes and drastic state changes can be captured and reflected in the cache in a timely manner. Then, the interface adaptation module takes over the complex task of cross-platform communication.

[0038] When the target proxy runs on the HarmonyOS operating system, the interface adaptation module automatically uses the HarmonyOS distributed soft bus protocol to deeply encapsulate the dynamic proxy identifier. The module accurately converts the binary data stream inside the identifier into a service discovery message format that the HarmonyOS system can directly recognize, and specifically optimizes the message header field structure to reduce network transmission overhead and improve communication efficiency on resource-constrained edge devices. When the target interface is a domestic large-scale interface standard, the interface adaptation module adds an extra two-way handshake verification mechanism to the data transmission link. Before officially sending the dynamic proxy identifier, the module exchanges key tokens with the receiving end to establish a secure channel, and the receiving end performs strict integrity verification on the checksum in the received identifier. If the verification result shows that the data is inconsistent or the verification fails, the module will immediately discard the data packet and record a detailed security log to prevent malicious tampering. This enables the entire system to maintain the real-time accuracy of the proxy identity information and the consistency and integrity of the communication data in complex network conditions and diverse device ecosystems, effectively solving the technical bottlenecks of traditional static address management being unable to adapt to dynamic capability changes and the difficulty of interoperability between heterogeneous systems.

[0039] In one specific implementation, an identity verification module is also included, which is connected in series with the output of the distributed addressing module. After the distributed addressing module returns the target proxy node information, the identity verification module extracts the dynamic proxy identifier fed back by the target node and parses the verification code and home domain identifier fields. The identity verification module has a built-in asymmetric encryption verification algorithm, which uses a pre-stored public key to decrypt the verification code and compares the decryption result with the home domain identifier. If the comparison result is inconsistent or the verification code expires, the target node is determined to be a forged node and the connection request is blocked. The final data communication channel is only established when the verification is successful. The cache module also has a historical snapshot storage unit, which uses a version chain structure to store the dynamically generated proxy identifiers and their corresponding timestamps. When an address tracing request occurs, the historical snapshot storage unit can retrieve the high-dimensional feature vector of the historical version according to the specified time range to reproduce the proxy's capability status within the time period. At the same time, the historical snapshot storage unit is set with a size limit. When the storage capacity reaches the preset threshold, the historical snapshot data with the earliest timestamp is automatically deleted to maintain the storage space balance of the edge nodes and prevent retrieval delays caused by the accumulation of historical data.

[0040] In specific applications of this invention, after the system completes initial distributed addressing and obtains the target proxy node information, the identity verification module immediately intervenes in the workflow to perform security access control. The module first extracts the dynamic proxy identifier from the data packet returned by the target node and parses out the two security elements contained therein: the checksum and the home domain identifier field. Subsequently, the module calls the built-in asymmetric encryption verification algorithm and uses the public key pre-stored and managed by the system to decrypt the extracted checksum. The original information restored after decryption is compared bit by bit with the home domain identifier field. If the comparison result shows that the two are inconsistent or the system detects that the checksum has exceeded the valid time window, the identity verification module will immediately determine that the target node is a forged node or an illegal node subjected to a replay attack, and decisively block all connection requests to prevent malicious data from flowing in. Only when the verification results are completely consistent and the time is within the validity period will the module allow access and establish the final data communication channel, thereby building a solid trust barrier at the forefront of the communication link.

[0041] Meanwhile, the historical snapshot storage unit in the caching module also runs continuously in the background to support the system's traceability and resource optimization. The unit uses a version chain structure to store the dynamically generated agent identifiers and their corresponding timestamps in an orderly manner, forming a complete capability evolution chain. When the system receives an address tracing request, the historical snapshot storage unit can quickly retrieve the corresponding historical version's high-dimensional feature vector according to the user-specified time range, reproducing the agent's true capability status within the time period for auditing or fault analysis. To prevent the infinitely growing historical data from consuming too many edge node resources, the historical snapshot storage unit also has a size limit mechanism. Once the storage capacity reaches the preset threshold limit, the unit will automatically identify and delete the historical snapshot data with the earliest timestamp, avoiding retrieval delays or system crashes caused by excessive accumulation of historical data. The real-time security protection of the identity verification module and the full lifecycle data management of the historical snapshot storage unit work together to ensure that the entire agent dynamic address and identity management system based on caching capabilities can resist immediate identity forgery attacks in an open network environment, retain necessary historical traces for post-event tracing, and maintain long-term stable operating performance in resource-constrained edge computing scenarios.

[0042] In one specific implementation, the AID generation module introduces a dynamic weight adjustment mechanism when generating dynamic proxy identifiers. This mechanism automatically adjusts the weight coefficients of each dimension in the high-dimensional feature vector based on the professional domain type of the proxy. For computationally sensitive domains, the weight coefficient for the computational power dimension is increased by 1.5 times; for latency-sensitive domains, the weight coefficient for the response latency dimension is increased by 1.5 times. These weight coefficients directly participate in the dot product operation of the Gaussian hash function, making the generated capability hash value more reflective of the core capability differences in specific scenarios. This improves the accuracy of distributed addressing in specific application scenarios. The distributed addressing module performs addressing... During the process, a load balancing routing strategy is also integrated. When multiple target agent nodes with cosine similarity values ​​higher than a preset threshold are calculated, the distributed addressing module further obtains the real-time load rate and network round-trip latency data of each target node. The system constructs a comprehensive scoring function, using cosine similarity as the first weight factor, the reciprocal of the real-time load rate as the second weight factor, and the reciprocal of the network round-trip latency as the third weight factor, and calculates the comprehensive score of each candidate node by weighting. Finally, the node with the highest comprehensive score is selected as the addressing result and returned. In this way, while ensuring the accuracy of capability matching, the load distribution and response speed of the collaborative tasks of the entire network agent are evenly distributed.

[0043] In specific applications, when the AID generation module initiates the generation process of dynamic proxy identifiers, it first activates the built-in dynamic weight adjustment mechanism to perceive the professional field type to which the current proxy belongs. The system automatically modifies the weight coefficients of each dimension in the high-dimensional feature vector based on the identified field characteristics. When the system determines that the proxy belongs to a computing power-sensitive field, the mechanism increases the weight coefficient of the computing power dimension to 1.5 times the original value. When the system identifies that the proxy serves a latency-sensitive field, the mechanism then increases the weight coefficient of the response latency dimension to 1.5 times. The dynamically adjusted weight coefficients directly participate in the dot product operation of the subsequent Gaussian hash function, so that the final calculated capability hash value can more sensitively capture and reflect the core capability differences in specific application scenarios. This targeted weighted processing improves the accuracy of the system in distributed addressing in specific business scenarios, ensuring that the found proxy fully matches the task requirements in the most critical performance indicators.

[0044] Next, the distributed addressing module takes over the specific addressing operation. After calculating multiple target proxy nodes whose cosine similarity is higher than the preset threshold, the module further activates the integrated load balancing routing strategy to obtain the real-time load rate and network round-trip latency data of these candidate target nodes. The system then constructs a comprehensive scoring function to quantitatively evaluate the merits of each candidate node. The function uses the previously calculated cosine similarity as the first weighting factor to ensure the accuracy of capability matching, the reciprocal of the real-time load rate as the second weighting factor to prioritize nodes with more idle resources, and the reciprocal of the network round-trip latency as the third weighting factor to select the node with the fastest network response. The system performs a weighted calculation on these three factors to obtain the comprehensive score of each candidate node. Finally, the distributed addressing module selects the node with the highest comprehensive score as the unique addressing result and returns it to the requester. This effectively avoids local node overload or network congestion caused by solely pursuing capability matching, achieving a perfect balance between uniform load distribution and optimal response speed in the collaborative tasks of the entire proxy network. This ensures that the entire proxy dynamic address and identity management system based on caching capabilities can maintain an efficient, stable, and intelligent operating state under complex and ever-changing business loads.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A proxy dynamic address and identity management system based on caching capabilities, characterized in that, include: The capability feature extraction module is used to acquire the capability attribute set of the agent in real time, and to perform noise reduction and normalization processing on the capability attribute set to generate a standardized capability feature set. A high-dimensional semantic encoding module, connected to the capability feature extraction module, is used to map the standardized capability feature set to a high-dimensional feature space using a preset semantic encoding model to generate a high-dimensional feature vector. The AID generation module, connected to the high-dimensional semantic encoding module, is used to perform hash calculation on the high-dimensional feature vector based on the locality-sensitive topology algorithm to generate a dynamic proxy identifier containing a capability hash value, a home domain identifier, a timestamp, and a check code. The caching module, connected to the AID generation module, is used to store the dynamic proxy identifier and its corresponding high-dimensional feature vector to the edge node and perform a cache update operation. The distributed addressing module, connected to the caching module and the high-dimensional semantic encoding module respectively, is used to receive addressing requests, generate demand feature vectors, and match the target dynamic agent identifier in the distributed hash table based on cosine similarity; the interface adaptation module is used to perform format conversion and verification of communication protocols for different operating systems and large model interfaces.

2. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, The denoising process in the capability feature extraction module specifically employs a wavelet threshold denoising algorithm, which removes high-frequency noise data from the capability attribute set by setting a soft threshold function; the normalization process specifically employs the min-max normalization method, which linearly maps the numerical data in the capability attribute set to the closed interval [0,1], and its calculation formula is as follows: ;in Represents the normalized i-th Item ability attribute value, The values ​​represent the original capability attribute values. min(P) and max(P) represent the minimum and maximum values ​​in the capability attribute set P, respectively, in order to eliminate the weight interference of capability attributes with different dimensions on the subsequent encoding process.

3. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, The Local Sensitive Topology Algorithm used in the AID generation module is specifically configured with a Gaussian hash function, whose hash calculation process is defined as: h(V)=⌊(w⋅V+b) / r⌋; where V is the input high-dimensional feature vector, which is a random projection vector that follows a standard normal distribution, b is the offset that is uniformly distributed in the interval [0,r], and r is a preset hash bucket width parameter; the AID generation module performs multiple independent hash operations, and concatenates and combines the generated hash values ​​to form the capability hash value in the dynamic proxy identifier, so that feature vectors that are close in distance in the high-dimensional space are mapped to the same hash bucket with a high probability, thereby realizing the quantitative fingerprint expression of capability similarity.

4. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, The cache update mechanism established by the cache module includes an active update strategy and a passive update strategy. The active update strategy is configured with a fixed time window, the length of which is 1 to 5 minutes. At the end of each time window, the capability feature extraction module re-collects the capability attribute set of the agent. If the Euclidean distance change rate between the newly collected capability attribute set and the historical capability attribute set stored in the cache exceeds 10%, the AID regeneration process is triggered and the old cached data in the edge node is replaced. The passive update strategy listens to the status events of the agent. When it detects that the agent has undergone a model version upgrade, hardware failure recovery, or task type switch event, it immediately receives the update instruction sent by the agent and forcibly triggers the immediate refresh of the cached content.

5. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, The specific adaptation method of the interface adaptation module for the HarmonyOS operating system is as follows: the dynamic proxy body identifier is encapsulated using the HarmonyOS distributed soft bus protocol, the binary data stream in the identifier is converted into a service discovery message format recognized by the HarmonyOS system, and the message header fields are optimized to reduce transmission overhead. The specific adaptation method for the domestic large model interface standard is as follows: a two-way handshake verification mechanism is added to the data transmission link, a key token is exchanged before sending the dynamic proxy body identifier, and the integrity of the check code in the identifier is verified at the receiving end. If the verification fails, the data packet is discarded and a security log is recorded to ensure data consistency and anti-tampering capability when communicating across heterogeneous platforms.

6. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, It also includes an identity verification module, which is connected in series with the output of the distributed addressing module. When the distributed addressing module returns the target proxy node information, the identity verification module extracts the dynamic proxy identifier fed back by the target node and parses the verification code and home domain identifier fields therein. The identity verification module has a built-in asymmetric encryption verification algorithm, which uses a pre-stored public key to decrypt the verification code and compares the decryption result with the home domain identifier. If the comparison result is inconsistent or the verification code expires, the target node is determined to be a forged node and the connection request is blocked. The final data communication channel is only established when the verification is successful.

7. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, The caching module also includes a historical snapshot storage unit. This unit uses a version chain structure to store the identifiers of dynamically generated proxies and their corresponding timestamps. When an address tracing request occurs, the historical snapshot storage unit can retrieve the high-dimensional feature vectors of historical versions within a specified time range to reproduce the proxies' capability status during that time period. Simultaneously, the historical snapshot storage unit has a size limit. When the storage capacity reaches a preset threshold, it automatically deletes the historical snapshot data with the earliest timestamp to maintain the storage space balance of edge nodes and prevent retrieval delays caused by the accumulation of historical data.

8. The proxy dynamic address and identity management system based on caching capability according to claim 3, characterized in that, When generating dynamic proxy identifiers, the AID generation module also introduces a dynamic weight adjustment mechanism. This mechanism automatically adjusts the weight coefficients of each dimension in the high-dimensional feature vector based on the professional field type to which the proxy belongs. For computing power-sensitive fields, the weight coefficient of the computing power dimension is increased by 1.5 times; for latency-sensitive fields, the weight coefficient of the response latency dimension is increased by 1.5 times. These weight coefficients directly participate in the dot product operation of the Gaussian hash function, making the generated capability hash value more reflective of the core capability differences in specific scenarios, thereby improving the accuracy of distributed addressing in specific application scenarios.

9. The proxy dynamic address and identity management system based on caching capability according to claim 1, characterized in that, During the addressing process, the distributed addressing module also integrates a load balancing routing strategy. When multiple target agent nodes with cosine similarity values ​​higher than a preset threshold are found, the distributed addressing module further obtains the real-time load rate and network round-trip latency data of each target node. The system constructs a comprehensive scoring function, using cosine similarity as the first weight factor, the reciprocal of the real-time load rate as the second weight factor, and the reciprocal of the network round-trip latency as the third weight factor, and calculates the comprehensive score of each candidate node by weighting. Finally, the node with the highest comprehensive score is selected as the addressing result and returned. In this way, while ensuring the accuracy of capability matching, the load distribution and response speed of the collaborative tasks of the entire network agent are evenly distributed.

Images