175 results about "Bidirectional map" patented technology

The invention relates to the technical field of intelligent power grids, in particular to a power distribution network battery digital dynamic management system based on digital twinning. Comprising a data acquisition unit; the digital twinborn modeling unit is used for constructing a battery-power grid-environment multi-dimensional dynamic twinborn body and realizing virtual-real bidirectional mapping and adaptive updating by combining a multi-physics field coupling model and a long-short-term memory network time sequence prediction algorithm; a dynamic optimization unit; and executing the feedback unit. Through a distributed heterogeneous sensing network of a data acquisition unit, multi-dimensional operation data of a battery pack and a key node of a power distribution network are acquired, and a high-fidelity data set containing four-dimensional labels of a battery state, a power grid parameter, time and a position is generated in combination with a spatial-temporal feature extraction technology; the deep fusion of the full life cycle state of the battery and the global operation data of the power distribution network is realized, and the comprehensive data support covering the global is provided for the optimization decision.

The invention provides a method for generating an SQL (Structured Query Language) by a natural language based on bidirectional mapping and semantic parsing, which relates to the technical field of database query and comprises the following steps of: extracting natural language query elements and packaging the natural language query elements into structured data, and establishing a mapping relationship from a query field to a service attribute and a physical data table by adopting a bidirectional Hash index technology; and automatically identifying multi-table association keys, performing semantic extension and compliance verification, generating an abstract syntax tree, performing processing according to user permission, and finally converting the abstract syntax tree into an SQL statement conforming to a target database syntax specification. According to the method, the accuracy and the efficiency of converting the natural language into the SQL are improved, and the flexibility and the safety of the system are enhanced.

The invention discloses a multi-welding-robot collaborative operation system based on digital twinning, and belongs to the technical field of intelligent manufacturing and robot control, the system comprises a physical space module, a virtual space module, a twinning data layer and a control module, and communication connection is established among the physical space module, the virtual space module, the twinning data layer and the control module; the physical space module comprises a plurality of welding robots, a sensor group and a data transmission network; the virtual space module comprises a multi-welding robot twinning body and a workpiece twinning model; the twin data layer is used for connecting the physical space module and the virtual space module; and the control module generates a welding path optimization scheme, a multi-welding-robot collaborative collision avoidance strategy and a welding quality feedback control instruction based on the twin data layer. According to the method, a closed-loop control framework of'physical-virtual 'bidirectional mapping is constructed, so that dynamic collaborative optimization of multi-robot motion trails, welding seam forming and thermal deformation is realized.

The invention discloses a digital production plan scheduling method and system, and belongs to the technical field of optimal scheduling, and the method comprises the steps: constructing a distributed storage architecture based on edge computing nodes; a central coordinator is adopted to realize cross-node data synchronization through an improved Raft consensus algorithm, multi-version concurrency control is realized based on a vector clock, and a global consistent data view is established; a visual scheduling platform is built based on a Vue3 framework, and man-machine interaction is realized by adopting a Canvas and WebGL collaborative rendering framework; establishing a dynamic coordinate conversion model based on bilinear interpolation, designing a space mapping function containing distortion compensation, establishing a multi-thread coordinate service based on WebWorker, and realizing submillimeter-level bidirectional mapping of pixel coordinates and physical coordinates; constructing a three-dimensional space-time analysis model fused with the multi-dimensional features; and all the units are subjected to feature fusion through residual connection, and finally a scheduling scheme with a confidence coefficient weight is output. The method and the device have the effect of meeting various scheduling requirements.

The invention discloses a remote sensing image open vocabulary segmentation method and system based on vision-language interaction. According to the method, the advantages of a multi-modal large language model and a semantic segmentation network based on a visual basic model are cooperatively utilized, language-pixel two-way mapping is taken as a core, five types of marks of images, texts, categories, objects and segmentation are introduced as carriers of cross-modal information, and three types of cross-modal fine-grained information interaction modules are taken as bridges, so that the cross-modal information interaction is realized. Bidirectional mapping and alignment of fine-grained information of the multi-modal large language model and the semantic segmentation network are realized, the open vocabulary segmentation capability of the semantic segmentation model is improved, and the method can adapt to different remote sensing scenes and category definitions. The method has the following advantages: the method has high performance, strong generalization ability and good expansibility, can provide any category of semantic segmentation maps for unlabeled target domain images based on instructions, and has high application value in the aspects of urban planning, map making, disaster response and the like.

The invention relates to the technical field of database management, in particular to a two-way linkage database table and supervision submission form field synchronous construction method which is applied to a supervision data submission scene. According to the scheme, the method comprises the steps that physical structure metadata of a database table and definition metadata of a supervision submission form are obtained, a semantic vector set is formed by combining metadata semantic analysis and semantic coding, and a bidirectional mapping relation graph is constructed through relation weaving; constructing a linkage propagation path according to a field change event, realizing adaptive bidirectional structure adjustment, generating a field synchronous construction scheme, and driving bidirectional mapping relation graph evolution optimization through feedback learning; according to the method, the bidirectional mapping relation graph and self-adaptive bidirectional structure adjustment are creatively combined, and efficient and self-adaptive bidirectional field synchronous construction is realized.

The invention discloses a method and a system for supporting visual and AI bidirectional intercommunication editing views. The method comprises the following steps: establishing a bidirectional mapping rule between a rendering tree protocol and an intercommunication protocol; in response to an operation of a user on the visual editor, generating a first atomic operation instruction, applying the first atomic operation instruction to the current rendering tree and triggering local redrawing of a view; serializing the updated rendering tree into context segments in an intercommunication protocol format according to the mapping rule, and pushing the context segments to an AI model; responding to a natural language editing request initiated by a user to the AI model, receiving streaming data in an intercommunication protocol format returned by the user based on the request, analyzing the streaming data according to the mapping rule to obtain a second atomic operation instruction, applying the second atomic operation instruction to the rendering tree and triggering local redrawing of a view, and obtaining a second atomic operation instruction. The view is edited by the AI, so that real-time two-way intercommunication between visual editing and AI editing is realized, and the user experience is remarkably improved.

The invention relates to the technical field of hydrological survey skill training, in particular to a construction method and system of a hydrological survey knowledge model based on a large model. The method comprises the following steps: acquiring multi-source hydrological data, and constructing a hydrological knowledge graph; generating a confusion sample corresponding to the hydrological knowledge graph, and performing adversarial training on the confusion sample; according to the confrontation training result and historical test questions in the multi-source hydrological data, obtaining bidirectional mapping test questions; inputting the bidirectional mapping test questions into a question setting interface of a hydrological survey learning platform; collecting answer data and student physiological data corresponding to the bidirectional mapping test questions in real time, and updating a student knowledge state according to the answer data and the student physiological data; constructing a hydrological survey knowledge model based on the updated student knowledge state; and generating a student recommendation learning scheme by using the hydrological survey knowledge model. According to the invention, the intellectualization and practicability of hydrological survey education can be improved.

The invention discloses a physical field reconstruction method fusing a reduced-order model and a multi-fidelity model, and belongs to the technical field of data-driven physical field reconstruction. The method comprises the following steps of: firstly, carrying out nonlinear dimension reduction on high-dimensional simulation data by utilizing a deep auto-encoder, extracting a low-dimensional feature vector, training an encoder-decoder network, and establishing bidirectional mapping between high-dimensional data and low-dimensional data; then, a mapping model from the working condition parameters to the low-dimensional features is constructed and used for predicting feature vectors under the new working condition; the prediction features are then reconstructed by a decoder into preliminary physical field data as a low fidelity trend. And finally, on the basis of the trend, in combination with sparse high-fidelity measured data, correction is carried out through a multi-fidelity fusion model, and a high-precision reconstructed physical field is obtained. According to the method, the precision and credibility of physical field prediction are effectively improved by fusing multi-source data, and the method is suitable for the fields of structural health monitoring, digital twinning and the like.

The invention discloses a low-illumination image enhancement method suitable for a complex night operation scene. The method comprises the following steps: preprocessing a collected low-illumination RGB image to obtain a preprocessed image; multi-scale features are extracted from the shallow convolution projection and the multilayer axial converter unit, and cross-layer attention is used for weighted aggregation to form enhanced features; forward noise addition is executed according to noise scheduling to obtain potential representation representing low-illumination noise distribution, and deep semantic representation is extracted in combination with a U-Net encoder embedded in CBAM; in the decoding stage, details are reconstructed through jump connection and a multi-layer converter, meanwhile, reverse denoising of a diffusion model is introduced to gradually remove noise, and an enhanced image with high brightness and low noise is output; in the training process, bidirectional mapping from low illumination to normal illumination is constructed, a weighted target of self-encoding loss and enhanced loss is introduced to carry out joint optimization on a converter and diffusion parameters, and a low-illumination scene is monitored on a general GPU / edge device.

The invention relates to a fabricated building management method and system based on BIM and digital twinning. The method comprises the steps that a BIM project model is constructed; generating project plan information; building a project digital twinborn model based on the BIM project model and the project plan information; driving a project flow process to be executed; receiving feedback information of the Internet of Things to perform space-time constraint rule verification, and driving state evolution update of the digital twin model; when the risk information is recognized, the risk type is judged, and an adaptive regulation and control strategy is executed; according to the method, a bidirectional mapping mechanism of the BIM project model and the digital twinborn model is constructed, time-space constraint verification and state evolution updating are carried out on the construction process in combination with real-time data of the Internet of Things, an adaptive regulation and control strategy is triggered when the risk is identified, real-time monitoring and dynamic optimization of the whole process of the fabricated building are realized, and the construction efficiency is improved. The method has the advantages of improving the collaboration of the construction process, reducing the rework rate and guaranteeing the project progress and quality.

The invention provides a vision-based cross-network interaction method and system, and relates to the technical field of intelligent interaction, and the method comprises the steps: obtaining a visual interaction sequence, constructing multi-modal feature representation, achieving cross-domain semantic alignment, deconstructing visual information into a hierarchical control instruction set, and transmitting the hierarchical control instruction set to a target network environment for execution after security classification. And a bidirectional mapping relation graph is constructed to realize incremental optimization. According to the method, semantic bridging between heterogeneous networks can be established, the cross-domain control precision is improved, and meanwhile safe interaction in a network isolation environment is guaranteed.

The invention discloses a context-aware-based gap containerization TLS compatible method, which comprises the following steps of: distinguishing a simple TLS variable and a complex TLS variable in a source gap system, and establishing a TLS variable source feature; the method comprises the following steps: constructing a dynamic telescopic TLS memory pool matched with a variable access permission in a target system, expanding a kernel state TLS segment descriptor, constructing a bidirectional mapping table of musl TLS offset and a memory pool address, and optimizing and constructing a clustering structure aligned with a CPU cache line for a hotspot variable to improve the access efficiency; tLS variable access tracks are recorded through a dynamic binary translation engine, TLS segment descriptors are updated, the descriptors are read during thread switching to recover TLS context and refresh cache to guarantee consistency, address and offset conversion is completed according to target system alignment requirements by combining source features and a bidirectional mapping table for TLS variables of nested structures, pointers and array types, and the TLS variables of the nesting structures, the pointers and the array types are aligned. And completing dynamic conversion execution of the source system application program in the target system.

The invention provides an authentication document form generation method and system, and relates to the technical field of data processing, and the method comprises the steps: obtaining an authentication document; converting the authentication document into an XML (Extensible Markup Language) file so as to map authentication document data to an XML tree node of the XML file; standardizing the XML file through the global unique identifier and the tag system to obtain a standardized XML file; in combination with the standardized XML file, template candidate items conforming to different manageable data templates in the authentication document are extracted respectively; in combination with the template candidate item, mapping the authentication document to a target form; and establishing a bidirectional mapping relationship between the target form and the authentication document to realize bidirectional conversion between the target form and the authentication document and complete the form processing of the authentication document. Through establishment of the bidirectional mapping relation, synchronous updating of the form and the document content can be achieved, errors caused by manual operation are avoided, and automation, accuracy and practicability of document conversion are remarkably improved.

The invention discloses an operating system containerization kernel compatibility method based on shadow structure mapping, which comprises the following steps of: constructing a multi-dimensional mapping rule base by analyzing the difference of data structures between a container system and a standard system, and creating a shadow structure comprising a data area, a metadata area and an expansion area when the container system runs on a host system; the host system allocates a shadow memory space and a memory pool, establishes mapping between a shadow structure and a physical memory page and bidirectional mapping between the shadow structure and a host structure, intercepts system call of a container application, obtains a target structure and allocates a shadow structure and a host structure memory, and sends the shadow structure and the host structure memory to the host system; according to the method, conversion from the target structure to the host structure is completed according to the multi-dimensional mapping rule base, the host machine completes calling through the host structure, then the calling result is reversely converted into the target structure and returned to the container application, and seamless compatibility of the container system and the host system is achieved under the condition that a user mode program of the container system is not modified.

The embodiment of the invention discloses a visual fine adjustment method and device of an interface, a storage medium and a terminal. Firstly, a source code and a running state preview interface are synchronously obtained, and it is ensured that operation is based on a real rendering state. Then, when the user edits the interface elements, the system accurately positions the visual elements to the corresponding positions in the source codes during operation through a pre-established bidirectional mapping mechanism. Based on the accurate positioning, the system generates a code modification instruction and updates the target code. And finally, generating a new preview by running the updated code in real time, completing an operation feedback closed loop, and realizing a real-time editing preview effect. According to the method, manual modification of complex codes is avoided through the complete process, the modification precision is ensured while the low operation threshold is kept through intuitive visual operation, and the operation simplicity and control precision are effectively balanced.

The invention discloses a regulation-work package bidirectional association and dynamic mapping method and system, and relates to the technical field of enterprise compliance management and task management informatization. The method comprises the following steps: constructing a standardized law and regulation library and a work package attribute library, disassembling a law and regulation file into compliance requirement points, and determining multi-dimensional attributes of a work package; forward mapping and reverse tracing binding of the compliance requirement point and the work package are realized through a mapping matching model; performing integrity, consistency and rationality verification on the mapping relation; monitoring the regulation change and automatically updating the mapping relation; and visually displaying the mapping relationship and monitoring compliance execution. The system correspondingly comprises a regulation management module, a work package management module, a bidirectional mapping module and the like. According to the method, closed-loop management and control of law and regulation requirements, task execution and compliance evidence are realized, the problems of unidirectional and static mapping and rough granularity in the prior art are solved, the compliance management and control accuracy and the dynamic adaptation capability are improved, and the compliance risk of an enterprise is reduced.

The invention belongs to the technical field of semantic retrieval, and particularly relates to an intelligent customer service information retrieval method and system based on a large model. Performing retrieval judgment on a differential hierarchical index tree-shaped standard semantic question and answer information base of each main body node in the distributed isolation mapping security chain, returning retrieved standard question and answer information according to key entity frequency sorting, and generating an answer by utilizing a pre-training reasoning generation model and combining with the information base when the retrieved standard question and answer information is not retrieved; when the retrieval semantics comprise a reservation demand, generating a cross-system reservation parameter set by means of a distributed reservation decision tree and standard question and answer information, outputting recommended reservation information through a bidirectional mapping dictionary and a matching recommendation algorithm, and dynamically updating the bidirectional mapping dictionary and retrieval index information by collecting the consistency frequency ratio of recommendation to real reservation in real time. And the semantic retrieval precision and the cross-system reservation accuracy are effectively improved.

The invention discloses a photovoltaic construction professional knowledge base construction and updating method based on an NLP technology. The method comprises the following steps: firstly, acquiring a multi-source text from multiple channels; photovoltaic domain entity recognition is completed by adopting a named entity recognition technology after preprocessing, an entity relationship is extracted through dependency syntax and semantic role labeling, and a knowledge triple is generated; and then performing unified modeling on entity category levels and relationship types based on the ontology, storing a triple into a graph database, storing detailed attributes and business process data of entities and relationships into a relational database, and realizing interconnection of the two databases by a bidirectional URI mapping mechanism. In the updating stage, minute-level knowledge updating is achieved through increment monitoring, entity alignment, conflict resolution and version control. And finally, performing quantitative evaluation on knowledge quality, and performing feedback by a visual instrument panel to form closed-loop optimization. Real-time and reliable knowledge support can be provided for planning, design, construction and operation and maintenance of a photovoltaic project through knowledge base entity recognition query constructed by the method.

A multi-granularity semantic alignment method oriented to cross-modal image-text retrieval comprises the steps that multi-layer feature representation of an image and a text is established, and a global vector and a local mark sequence are obtained respectively; capturing fine-grained features with higher information amount by using a mark screening mechanism, and inhibiting interference of redundant and invalid features on representation; under a unified semantic alignment framework, a bidirectional mapping relationship among multi-granularity semantics is constructed, and coarse-granularity, fine-granularity and cross-granularity semantic similarity alignment is comprehensively realized so as to strengthen multilayer semantic consistency constraints; and adaptively integrating the granularity similarities through a weighted fusion module, generating single instance-level similarities, and carrying out descending sorting on candidates. According to the method, the multi-granularity characteristics of the image and the text are comprehensively considered, the image-text retrieval precision in an actual scene can be improved, and the method has good application value in the fields of image-text retrieval, multi-modal content analysis and the like.

The invention relates to the technical field of ink proportioning, and discloses an ink proportioning optimization method and system based on machine learning, and the method comprises the steps: constructing a multi-mode ink characteristic database; designing a multi-modal fusion model based on a visual language pre-training technology; a three-mode feature alignment mechanism is realized; establishing a condition generation framework to realize proportion optimization; and deploying a feedback iteration system. Through the multi-modal fusion and visual language pre-training technology, accurate bidirectional mapping between the sensory characteristics and the chemical proportion of the ink is realized, direct generation of an accurate proportion scheme from unstructured requirements of a user is supported, the experiment cost is reduced, the optimization time of the ink proportion is shortened from several weeks to several days, and the optimization efficiency of the ink proportion is improved. And a reverse proportioning design which is oriented by a result is supported, so that powerful technical support is provided for innovation of ink products.

The invention belongs to the technical field of path planning, and discloses an underground carry-scraper autonomous navigation path generation system based on front-end and rear-end interaction, which comprises the following steps: fusing multi-source map data, and constructing a space coordinate bidirectional mapping engine coupling physical space and pixel space association; detecting space conflicts between the user-labeled path points and the roadway structure in real time and generating a dynamic path skeleton; road section types are divided, then paths are dynamically optimized, redundant data are synchronously compressed, and an optimized path packet is generated; constructing a route version control tree, dynamically allocating storage resources according to roadway topology, and generating a simulation ready path file package; the transverse error and the minimum roadway wall distance are solved in real time, a verification report is generated, then an overrun event in the simulation trajectory tracking process is captured, a double-channel tracing report is generated, and a path section needing to be corrected is reversely marked; and performing hot update on the optimization rule, quantifying the path adaptability and performing reverse iteration on the core parameter library.

The invention provides a battery health detection method and system based on vehicle type-charging curve double-library mapping, and the method comprises the steps: constructing a bidirectional mapping relation table of vehicle types and current flow features according to a feature mode boundary and a vehicle type feature library, and calling corresponding feature mode parameters if an input vehicle type has a matching record in the feature library, establishing a specific health state judgment benchmark of the vehicle type; training feature data in the bidirectional mapping relation table by adopting a neural network algorithm, and learning current flow modes of different vehicle types in each health state through a multi-layer perceptron to obtain a battery health detection model capable of identifying vehicle type differences; and acquiring real-time charging current data of the to-be-detected electric vehicle, matching the corresponding characteristic mode parameters from the bidirectional mapping relation table according to the vehicle type identifier, if the matching is successful, extracting the characteristic vector of the current current, and judging the similarity between the characteristic vector and the preset health state mode.

The invention belongs to the technical field of industrial control, and discloses an industrial control virtual equipment resource pool construction method, system, equipment and medium, and the method comprises the steps: optimizing parameters of a transition probability and an observation probability in a time sequence data statistical algorithm according to time sequence characteristics of an industrial control scene, and extracting resource characteristics; generating a virtual resource model, and constructing a bidirectional mapping table of physical resources and virtual resources; optimizing a clustering algorithm by using a time weight factor, dynamically aggregating the time sequence metadata in combination with a streaming computing framework, and performing anomaly detection on a resource state in combination with the aggregated time sequence metadata based on an anomaly detection technology; based on a layered decoupling flattening architecture, a resource pool is divided into a resource layer, a data layer, a link layer, a technical support layer and an interface layer. According to the method, logic-physical layered decoupling of the industrial control resource pool is realized, seamless access and dynamic optimal distribution of multi-protocol equipment are supported, and high-reliability and low-delay bottom support is provided for users.

The present application relates to water level data processing technical field, specifically to water level data acquisition and processing system based on internet of things, system includes trend identification module, sampling frequency modulation module, data encapsulation module, abnormal elimination module, instruction linkage module.In the present application, through analyzing continuous water level change trend, trend identification is realized and sampling frequency is dynamically adjusted, so that sampling interval and actual water level fluctuation state are matched, response sensitivity of water level data and timeliness of data acquisition are improved, time stamp index mechanism and bidirectional mapping structure are used to construct unified encapsulation format, integrity of water level data structure and convenience of index management are enhanced, sudden break point is identified through water level amplitude difference and data is stored in segments, data stability and sudden situation identification ability are improved, and adaptive ability of water level information processing, maintaining level of data continuity and response efficiency to sudden abnormality are overall improved.

The invention discloses an interface data migration method, device and equipment based on bidirectional mapping processing. The method comprises the following steps: receiving an initial processing request corresponding to a first service system and obtaining a corresponding to-be-processed document; performing bidirectional mapping processing on the to-be-processed document to obtain corresponding mapping document information; mapping the document information to perform service logic decomposition on the initial processing request to obtain an interface data processing request corresponding to the second service system; and synchronously storing the initial processing request and the interface data processing request into a preset message middleware so as to perform synchronous execution through the message middleware. According to the interface data migration method based on the bidirectional mapping processing, the bidirectional mapping processing is performed on the incoming parameter and the outgoing parameter of the to-be-processed document through the bidirectional field mapping table, so that smooth migration of the interface data between the first business system and the second business system is realized, and the situation that the maintenance cost is increased due to code redundancy is avoided; and the reconfiguration reliability of the financial transaction system is greatly improved.

The application discloses an operating system containerization kernel compatibility method based on shadow structure mapping, constructs a multi-dimensional mapping rule library by analyzing the difference between the data structure of a container system and a standard system, creates a shadow structure containing a data area, a metadata area and an extension area when the container system runs on a host system, allocates shadow memory space and a memory pool by the host system, establishes the mapping between the shadow structure and a physical memory page and the bidirectional mapping between the shadow structure and a host structure, intercepts the system call of a container application by the host system, obtains a target structure and allocates the shadow structure and the host structure memory, completes the conversion of the target structure to the host structure according to the multi-dimensional mapping rule library, completes the call by the host computer using the host structure, reversely converts the call result into the target structure and returns the call result to the container application, and realizes the seamless compatibility between the container system and the host system without modifying the user state program of the container system.

The invention relates to a trade rule reasoning method and system based on bidirectional mapping and a storage medium, and relates to the technical field of trade management.The trade rule reasoning method comprises the steps that a trade rule text is collected and split, and a plurality of rule terms are obtained; analyzing the rule terms according to preset key entity elements to obtain rule condition segments, and combining the rule feature weight values to obtain trade rule feature vectors; identifying and converting trade rule feature vectors according to the rule attributes and the dependency relationship between the rules, and constructing a rule dependency graph; converting the trade rule text into formalized rule elements according to the semantic framework and the logic structure in combination with the rule dependency graph; generating an optimal rule reasoning path according to the rule query request in combination with the formalized rule elements; aggregating all the optimal rule reasoning paths to form a trade rule corresponding to the rule query request; and establishing a mapping relationship between the semantic structure and the formalized representation of the rule text through a bidirectional mapping mechanism.

The application provides an entity network modeling method and device based on an agent system, including a first agent and a second agent in communication with each other, the first agent is used for constructing a data relationship graph of the entity network, and the second agent is used for generating a data model of the entity network according to the data relationship graph. In the application, the problems that a traditional data modeling method needs to master a large amount of business field and data statistical knowledge, has poor generalization ability, and is not timely, etc. are solved, the global data is automatically and timely preprocessed, relationship is recognized, and data modeling is performed through the agent and the prompt word, obvious advantages are obtained in operability, automation, construction cost, generalization, and real-time performance, and the bidirectional mapping, dynamic interaction, and real-time connection of the 6G physical network and the digital twin network can be effectively supported.

The application discloses a prefix caching method and device suitable for a mixed attention model architecture. In view of the heterogeneous management problem of key-value caching and state snapshot, a unified prefix index mechanism based on chain hashing is proposed, so that the two types of caches share a hash space. The method determines the boundary of the two types of caches consistent and effective through the longest prefix matching and forward traversal verification, combines local snapshot query and remote confirmation, realizes the collaborative recovery and reuse of key-value caching and linear attention snapshot, and reduces the heterogeneous state step-out risk. In the reasoning process, the two types of states are captured synchronously and unified hash index is adopted, and a bidirectional mapping between cache block identification and snapshot hash is established through a double cache coordination manager; when the expulsion condition is met, the linkage cleaning of the two types of caches is realized based on the mapping. The application realizes the unified storage and life cycle collaborative management of heterogeneous caches under the mixed attention architecture, guarantees the whole process consistency, and improves the reasoning performance and system robustness.