475 results about "Adaptive compression" patented technology

The present invention relates to the technical field of data transmission. Disclosed is an application-driven three-dimensional spatial data transmission method. The method comprises: determining system key performance indicators (KPIs) by means of qualitative and quantitative analysis; constructing an AI-driven adaptive three-dimensional data transmission mechanism, and dynamically adjusting a transmission strategy on the basis of a real-time network state, a device capability, an application scenario and the KPIs; developing an adaptive compression algorithm set oriented to three-dimensional data, so as to meet differentiated compression requirements of different application scenarios; performing loop execution of a test, and adjusting and optimizing the data transmission mechanism and the compression algorithm set on the basis of a test feedback result and the real-time network state; and deploying an optimized transmission method to a production environment, and collecting field data to optimize the system performance and verify the achievement of the KPIs. By constructing a qualitative and quantitative analysis framework based on machine learning, the present invention quantifies differentiated transmission requirements of different application scenarios and formulates transmission strategies meeting the scenario requirements.

The invention relates to the technical field of data monitoring, in particular to a complex equipment data monitoring method and system based on a knowledge graph and edge calculation. The method comprises the steps of multi-modal data acquisition and edge intelligent preprocessing; performing feature extraction on the preprocessed data; fusing multi-source heterogeneous data and constructing a dynamic knowledge graph; performing deep knowledge reasoning and layered fault early warning decision making; and performing self-optimization and knowledge closed-loop management based on early warning decision. According to the differential coding and self-adaptive compression algorithm of the edge layer, the data transmission quantity is reduced by 75%, the network bandwidth requirement is greatly reduced, and the system operation efficiency is improved.

The invention relates to the field of data backup, in particular to an intelligent data backup method and system based on an AI large model. The method comprises the following steps: acquiring an enterprise global data list, performing intelligent data structure deconstruction and dynamic attribute mapping modeling, and constructing a holographic data semantic perception model; performing real-time transient risk mutation detection on the holographic data semantic perception model, and constructing an intelligent backup triggering mechanism; carrying out storage resource demand prediction based on an intelligent backup trigger mechanism, carrying out multi-storage cloud environment resource dynamic scheduling, and constructing an elastic backup storage resource pool; carrying out incremental backup analysis and self-adaptive compression coding to obtain an incremental backup coding packet; and performing dynamic backup sequence adjustment and intelligent incremental backup decision on the incremental backup coding packet based on the elastic backup storage resource pool, and constructing an intelligent incremental backup execution engine. According to the method, the reliability, the accuracy and the traceability of a backup result are improved through self-adaptive intelligent incremental backup.

The embodiment of the invention provides a kafka storage and calculation separated data processing method and device, and the method comprises the steps: constructing a multi-layer data processing architecture which comprises four layers: network communication, task distribution, processing scheduling and storage. Cold and hot data separation is realized through hierarchical design of a log pre-writing area and a main storage area, and intelligent scheduling of system resources is realized in combination with a copy manager, a coordinator and a load predictor. An adaptive compression algorithm and a verification mechanism are adopted to ensure the data transmission efficiency and integrity, a message cache region is created to store hotspot data, and flexible access of cold and hot data is realized based on message sites. According to the method, the defects of the traditional technology in storage and calculation coupling, storage efficiency, load balancing and the like are effectively overcome, and the performance and expandability of a message queue system are remarkably improved.

The invention discloses a data compression transmission method and system applied to a ferry inspection image, and the method comprises the steps: collecting and obtaining the ferry inspection image in real time, recognizing a key inspection target region in the image, and carrying out the segmentation and partitioning of the image; compressing the key inspection target area based on lossless compression coding; the quantization step size is dynamically adjusted by comparing statistical variances of background pixels between continuous frames, and lossy compression coding is carried out on a background area; based on the boundary distance between the key inspection target area and the background area, adaptive compression coding is carried out on the transition area; constructing a hierarchical data packet; and constructing a data transmission optimization model, dynamically allocating data transmission links, and obtaining a transmission scheme with the highest total transmission. The method has the advantages that efficient data compression transmission is realized by accurately segmenting the image area and adopting a lossless, lossy and adaptive compression technology, the overall transmission efficiency is improved through intelligent transmission optimization, and the definition and real-time performance of the inspection image are ensured.

The invention discloses a remote centralized control distributed data processing and analysis method based on a JMS, and the method comprises the steps: achieving the message routing and resource scheduling through the real-time monitoring of the state of a system, reinforcement of learning decision, deep learning optimization and multi-dimensional message distribution; dynamically adjusting message flow rate and queue length based on real-time network conditions, node load and flow fluctuation; through edge node intelligent task scheduling, distributed collaboration of calculation and storage is realized; adaptive compression algorithm selection is carried out by using deep learning, and the consumption of message transmission bandwidth is reduced; non-tampering storage of messages is realized through a block chain technology; based on a container arrangement technology, dynamic expansion and contraction of computing resources are realized; and dynamic resource scheduling and optimization decision are realized by utilizing reinforcement learning and multi-dimensional monitoring data. According to the invention, the bottleneck problem of the traditional message transmission system is overcome, and the system can efficiently and reliably process mass data in a high-concurrency and high-throughput scene.

The invention relates to the technical field of force feedback handles, in particular to a remote surgical robot control system based on a force feedback handle and a control method thereof. The method comprises the following steps: carrying out multi-dimensional force data acquisition on an end effector of the surgical robot, and carrying out contact force size, direction and time sequence feature extraction to obtain multi-dimensional force feature data; performing surgical operation safety baseline analysis according to the multi-dimensional force feature data to generate surgical operation safety baseline data; performing dynamic force feedback threshold calculation according to the surgical safety operation baseline data to obtain force feedback threshold range data; and performing adaptive compression processing on different force feedback signals according to the force feedback threshold range data to obtain dynamic force feedback control parameters. Through real-time control instruction optimization and a closed-loop feedback strategy, the system can quickly respond to an operation intention, early warning is carried out on potential risks, and it is ensured that the operation process is safer and more reliable.

The invention relates to the technical field of computer vision and image recognition, in particular to an image recognition method based on edge computing, which comprises the following steps: dynamically capturing an original image through a plurality of edge nodes, rejecting redundant regions through a multi-modal perception triggering mechanism, and establishing a cooperative processing group. Illumination equalization, noise filtering and resolution self-adaptive compression tasks are distributed according to dynamic role election, a standardized preprocessed image is generated, a lightweight convolutional neural network is operated in parallel to extract a dual-channel feature vector, and after entropy coding lossless compression and equipment identity tag and time sequence stamp attachment, the dual-channel feature vector is transmitted to a cloud end by adopting a lightweight encryption protocol. The cloud end analyzes the data packet, reconstructs a feature topological graph based on space-time relevance, loads a depth residual error recognition model to execute feature fusion and classification decision, feeds back and updates the weight of an edge node model, solves the problems of low collaborative efficiency and feature distortion, and improves the efficiency and precision of image recognition.

The invention relates to the technical field of data sharing, and discloses a multi-department data sharing method and device, equipment and a storage medium. The method comprises the following steps: carrying out acquisition and standardized preprocessing on multi-department data required by enterprise project application to generate an intermediate data packet; establishing an event trigger distributed data control mechanism based on the intermediate data packet, calculating a dynamic trigger threshold, monitoring data change, and generating a target control data stream; performing template-free adaptive learning to generate an optimal data processing flow definition file; executing data classification, self-adaptive compression and increment processing at the edge computing node to generate an optimized data packet and metadata thereof; and inputting the optimized data packet and the metadata thereof into a SaaS cloud server central data warehouse to generate a project declaration data packet, thereby realizing consistency coordination and conflict automatic resolution of data of multiple departments, and realizing unified access and standardized processing of heterogeneous data sources.

The invention discloses a communication method for broadband communication and satellite fusion, which comprises the following steps of: classifying data streams, ensuring priority transmission of high-priority data streams by adopting a self-adaptive flow control protocol, carrying out load balancing according to link quality in combination with a satellite and a ground communication link, and when the communication link fluctuates, outputting the data streams. The method can automatically switch to a backup link or adjust a transmission mode, guarantees the communication stability, predicts the satellite communication time delay through deep learning, carries out the time delay compensation, optimizes the communication efficiency, intelligently allocates bandwidth resources according to the user demands and geographic positions, optimizes the bandwidth utilization rate, reduces the data transmission amount through a self-adaptive compression algorithm, and improves the bandwidth utilization rate. Therefore, the problems of non-uniform bandwidth utilization, unstable link, delay fluctuation and insufficient flow control in the prior art are solved, and through intelligent prediction and real-time adjustment, the system can efficiently utilize bandwidth resources, ensure communication stability, reduce delay fluctuation and improve overall communication quality.

The invention relates to the technical field of electric digital data processing, and particularly discloses a hierarchical federated learning method for adaptive compression coding, which comprises the steps of hierarchical federated learning architecture construction, edge server matching, model generation performance judgment, compression ratio matching and hierarchical federated learning architecture judgment, and specifically comprises the following steps: firstly, constructing a hierarchical federated learning architecture; each layer can make full use of computing resources thereof, edge servers corresponding to the terminal devices are obtained through matching, the terminal devices are used for generating environment data training models, corresponding compression rates are obtained according to network transmission performance of the edge servers, and the compression rates are used for data compression. The edge server receives the environment data training model of each terminal device and carries out adaptive compression coding and edge aggregation in sequence according to a corresponding compression ratio, and the cloud server receives an edge aggregation result and carries out global aggregation; and finally, judging whether the cloud server to which the hierarchical federated learning architecture belongs uploads a global aggregation corresponding result to an environment resource scheduling cloud platform or not.

The invention provides a power big data-based sparsity feature adaptive compression method, which comprises the steps of adaptively optimizing a sensing matrix through a dictionary learning method aiming at a screened multi-granularity feature subset, introducing a sparse regularization item to enhance the sparsity of data reconstruction, and according to a missing mode of a missing value and context information, carrying out adaptive compression on the sparse feature of the power big data. Establishing a mapping relationship among the compression ratio, the reconstruction error and the feature subset, and determining a self-adaptive compression strategy; based on a self-adaptive compression strategy, a compressed sensing method is adopted to carry out self-adaptive compression on the multi-time-granularity power data, the compression ratio is dynamically adjusted according to reconstruction error feedback, the data compression and reconstruction quality is balanced, and the reconstructed multi-granularity power data is synchronized according to timestamp precision.

A system and method for processing genomic data using dynamic latent manifolds that transforms multi-modal genomic datasets into geometric representations within a curved manifold space. The system receives genomic datasets including DNA sequences, genetic variants, and expression data, then extracts biological features and assesses importance using trained neural networks. Manifold curvature values are computed based on biological significance, and genomic data is embedded as geometric structures where semantic relationships are represented through distance and curvature properties. The system generates compression pressure fields that influence processing decisions and computes optimal geodesic paths through the manifold to minimize cognitive action functionals. Adaptive compression rates are determined for different genomic regions based on geometric properties and biological importance. The manifold structure evolves through use, strengthening frequently accessed pathways while applying thermodynamic decay to unused concepts. The system supports hierarchical organization across biological scales, reversible navigation, and federated learning capabilities that enable privacy-preserving collaboration.

The invention discloses a self-adaptive pruning method based on priority experience playback. The method comprises the steps of training a to-be-pruned image classification model, converting a pruning process of a neural network into a Markov decision process, and designing a corresponding state space, an action space and a reward function; constructing and initializing a value network, a strategy network, a target value network and a target strategy network of a depth deterministic strategy gradient algorithm; pruning the to-be-pruned image classification model; storing the pruning process of the to-be-pruned image classification model into an experience playback pool, and setting a priority for the pruning process; updating the sampling probability of each sample according to the time difference error; updating the value network and the strategy network, updating the target value network and the target strategy network, and carrying out loop training until a set maximum number of iterations is reached; and carrying out fine tuning on the image classification model to be pruned of image classification. According to the method, the pruning proportion of each layer of the network can be adaptively generated, and adaptive compression of the network to-be-pruned image classification model is realized.

The invention relates to the technical field of data compression, in particular to an intelligent data compression method and system based on multi-protocol heterogeneous equipment interconnection. The method comprises the following steps: when it is detected that a new heterogeneous device applies to join a network, carrying out communication protocol real-time analysis and protocol conversion adaptation to obtain a heterogeneous device adaptation protocol; carrying out multi-device data stream receiving according to a heterogeneous device adaptation protocol, and generating data streams of a plurality of time windows; adaptive coding and multi-level compression are carried out on the data streams of the multiple time windows, and a multi-device adaptive compression mechanism is constructed; identifying compression load characteristics of each heterogeneous device, and performing dynamic task allocation to obtain an intelligent task allocation strategy; and driving real-time data intelligent compression operation based on an intelligent task allocation strategy and a multi-device adaptive compression mechanism. According to the invention, efficient and adaptive data compression of different devices is realized, and the data transmission efficiency and quality are improved.

The invention provides an ocean buoy data real-time acquisition method based on low earth orbit satellite communication, and belongs to the technical field of electrical digital data processing.The ocean buoy data real-time acquisition method comprises the steps that firstly, a low earth orbit satellite network communication link is built, data characteristics are analyzed through a compression conversion gain index model, and a low earth orbit satellite network communication link is established; and establishing a multi-stage compression mechanism. The system adopts an optimization equation set to dynamically adjust compression parameters, the optimization equation set comprises a resource consumption equation, a compression quality equation and a transmission efficiency equation, and self-adaptive compression of data is achieved. And meanwhile, through a hierarchical algorithm selection mechanism and a parameter optimization strategy and in combination with real-time feedback control, continuous optimization of the compression efficiency is ensured. According to the compression scheme based on multi-level optimization, the technical problem of low compression efficiency caused by poor algorithm adaptability in the ocean buoy data compression process in the prior art is effectively solved.

A trained model is retrained for video quality assessment and used to identify sets of adaptive compression parameters for transcoding user generated video content. Using transfer learning, the model, which is initially trained for image object detection, is retrained for technical content assessment and then again retrained for video quality assessment. The model is then deployed into a transcoding pipeline and used for transcoding an input video stream of user generated content. The transcoding pipeline may be structured in one of several ways. In one example, a secondary pathway for video content analysis using the model is introduced into the pipeline, which does not interfere with the ultimate output of the transcoding should there be a network or other issue. In another example, the model is introduced as a library within the existing pipeline, which would maintain a single pathway, but ultimately is not expected to introduce significant latency.

The invention discloses a time synchronization method suitable for a satellite communication system, and the method comprises the steps: analyzing a forward broadcast signal, and obtaining the reference time of a network clock; constructing a geometric prediction handle based on satellite orbit data, and predicting one-way propagation time delay between a terminal and a satellite by using the geometric prediction handle; estimating the processing delay of the terminal according to the symbol rate and the equipment working condition of the terminal and the delay calibration data; fusing the network clock reference time with the processing delay estimate to synthesize a terminal local time; and the sending time of the return link is calculated based on the local time and the propagation delay prediction. According to the method, the synchronization precision can be improved to a microsecond level, the convergence speed in a dynamic scene is accelerated, adaptive compression of the guard interval is supported, and the stability and spectrum efficiency of a satellite communication system are effectively improved.

The invention provides a self-adaptive data extraction method and device based on multi-level dynamic sampling, and relates to the technical field of data extraction analysis, and the key points of the technical scheme are that the method comprises the following steps: obtaining operation parameters and system resource use state information of a plurality of blast furnaces; the blast furnace monitoring parameters are divided into a high-frequency sampling layer, a conventional sampling layer and a low-frequency sampling layer according to the operation parameters; calculating a blast furnace abnormal state index and a resource utilization rate based on the operation parameters; executing cross-furnace data association analysis, and calculating a cross-furnace association degree; dynamically adjusting the sampling frequency of each sampling layer according to the abnormal state index, the resource utilization rate and the cross-furnace correlation degree; based on the resource use state information and the data importance, carrying out adaptive compression processing on the collected data; and converting the compressed heterogeneous data into a uniform data format. The adaptive data extraction method and device based on multi-level dynamic sampling provided by the invention have the advantage of improving the extraction efficiency and quality.

The invention discloses a customized production-oriented edge node lightweight AI model adaptive compression method, which belongs to the technical field of intelligent manufacturing and edge computing, and comprises the following steps of: dynamically integrating compression strategies such as pruning, quantification and knowledge distillation by analyzing demand constraints and edge node hardware resources of customized production tasks; constructing an adaptive decision engine by utilizing reinforcement learning and Bayesian optimization, and generating an optimal compression scheme; in the deployment stage, compression parameters are dynamically adjusted through real-time monitoring and a closed-loop feedback mechanism, and the balance of model precision, reasoning efficiency and resource occupation is achieved. According to the method, the adaptability of the model in a heterogeneous edge environment can be remarkably improved, the deployment cost is reduced, and the small-batch and multi-task quick response requirement in a customized production scene is met.

The invention provides a communication segmentation learning system and method for adaptive channel compression, and a medium, and relates to the technical field of segmentation learning and communication compression. The system comprises a plurality of clients, a server side and a channel compression device arranged between the clients and the server side, the channel compression device comprises a channel sensitivity modeling module, a rate distortion adaptive compression module and a cross-client fair coordination module. In a channel sensitivity modeling module, channel importance is dynamically evaluated through intermediate layer activation value fusion; differentiated quantization and compression strategies are designed on the basis of sensitivity scores in a rate-distortion self-adaptive compression module, so that key information is reserved while communication overhead is reduced; and meanwhile, the cross-client fair coordination module realizes balanced distribution of communication resources among multiple clients through a fairness regularization and dual optimization mechanism, so that the influence of excessive compression on global convergence is avoided. According to the method, the communication efficiency and the training stability of segmentation learning in a complex heterogeneous environment are remarkably improved.

The invention relates to the technical field of Internet of Things data processing, and discloses an efficient data processing system for Internet of Things equipment. According to the method, multi-node cooperative training and reasoning are supported through the federated learning driven edge computing cluster, and localized efficient processing is realized while data privacy is protected; the cloud edge collaborative intelligent center solves the compatibility problem of heterogeneous equipment and various scenes through a lightweight model dynamic adaptation technology, the communication overhead is reduced, and the model generalization ability is improved; the cross-layer dynamic scheduling subsystem is combined with space-time flow prediction and a global optimization algorithm to realize elastic allocation of computing resources, so that burst loads are effectively dealt with and node overload is avoided; the full-link trusted execution system integrates data desensitization, security aggregation and block chain evidence storage, constructs hierarchical protection from equipment to the cloud, resists hostile attacks and ensures operation traceability; the AI-driven data processing assembly line greatly reduces the redundant data transmission and storage cost through adaptive compression and exception filtering.

The invention relates to the technical field of power systems, in particular to a power system broadband oscillation disturbance source grading positioning method and system, and the method comprises the steps: calculating the compression ratio of a measurement signal according to the sampling resolution of an original measurement signal and a real-time transmission interval, and obtaining an initial compression coefficient; performing frequency domain decomposition on the original measurement signal to obtain frequency spectrum characteristic data, and compressing the original measurement signal based on the frequency spectrum characteristic data and the initial compression coefficient to obtain a measurement compression signal; embedding a time sequence position code in the measurement compression signal to generate time sequence input data; analyzing the time sequence input data by using a multi-head attention mechanism to obtain an attention weighted signal; and generating a reconstruction signal according to the attention weighted signal, and obtaining a broadband oscillation disturbance source grading positioning result based on the reconstruction signal and the disturbance propagation path. According to the method, rapid and accurate grading and positioning of the broadband oscillation disturbance source are realized through technologies of self-adaptive compression, broadband oscillation characteristic reservation, topology propagation path analysis and the like.

The invention relates to a touch all-in-one machine interaction control method and device, computer equipment and a medium, and the method comprises the steps: receiving image data through a preset video stream protocol, and obtaining a corresponding image data stream; acquiring a real-time network bandwidth condition, and then performing adaptive compression processing on the image data stream based on the real-time network bandwidth condition, and dynamically adjusting image transmission parameters; comprehensively analyzing the image data stream by using a deep learning algorithm, extracting operation characteristics of a user, generating an output control instruction based on the operation characteristics, and further controlling and adjusting image display and / or an equipment state; content information is extracted from the image data stream, and content analysis is carried out by adopting a corresponding content analysis strategy according to the content information, so that real-time display and / or storage are / is carried out according to the analyzed content; and obtaining a user interaction behavior, generating a corresponding adjustment feedback instruction based on the user interaction behavior, and performing personalized adjustment on the user interface. The method has the effect of improving the image transmission quality.

The invention discloses a distributed single-mode small model data synchronization method applied to an intelligent electric energy meter, and the method comprises the steps: S1, building a high-fidelity training environment based on a programmable power supply, dynamically optimizing simulation parameters in combination with end-side feedback, and training an initial power utilization anomaly detection model; s2, modeling the trained model according to an equipment portrait, and generating a lightweight model through bidirectional adaptive compression; s3, deploying a lightweight model at an electric energy meter end and monitoring performance attenuation characteristics; s4, the environment offset is reversely fed back to the cloud through a block gradient return mechanism, and the model parameters are iteratively corrected at the cloud; and S5, triggering cloud secondary training by using feedback data, and circularly updating the model until the precision and power consumption dual-index convergence is achieved. According to the method, a closed-loop optimization link is formed through the end cloud bidirectional data flow, the static defect of traditional one-way deployment is broken through, and continuous self-adaption of the model to the dynamic environment is achieved.

The invention provides an image data processing and storing method based on a meta universe game, which comprises the following steps of: integrating and primarily processing image data to obtain an original image data set; a graphic database is designed based on a distributed storage system and by adopting a cold and hot data hierarchical storage strategy; processing the original image data set by utilizing a self-adaptive compression algorithm, a multi-thread asynchronous loading mode, GPU (Graphics Processing Unit) parallel computing, a progressive rendering technology, a physical basic rendering technology and illumination computing to obtain a standard image data set; a virtualization technology is utilized to simulate various hardware configurations, and a plurality of APIs are combined to carry out cross-platform compatible management of image data; and automatically updating and optimizing the content of the image data according to the monitoring result of the image data. According to the method, efficient storage and processing of the image data can be realized, the quality and compatibility of the image data are enhanced, the game performance and the user experience are remarkably improved, and development and popularization of the virtual world are promoted.

The invention discloses a medical device and rehabilitation software information system based on the Internet of Things, and particularly relates to the technical field of communication transmission management, which constructs event-anchored linear alignment on an edge side, takes a real-time channel quality vector as input, and generates a compression strategy through a Pareto multi-arm bandit, so that the compression strategy is compressed, and the communication transmission efficiency is improved. Meanwhile, a dynamic compression offset index is used as a safety gate, so that a self-adaptive compression decision considering distortion and time delay is realized, and the technical problem that the time sequence consistency and the diagnosis integrity of the multi-modal physiological signal are difficult to guarantee simultaneously under the conditions of clock drift and bandwidth limitation is solved; a cross-modal alignment quality monitoring module is arranged and used for recognizing and restraining cross-modal dislocation false correlation in real time, so that it is ensured that rehabilitation evaluation and short-period diagnosis are established on a reliable time sequence corresponding relation, and the dislocation problem existing in electrocardiosignals and action signals is solved.

The invention belongs to the field of natural gas pipeline operation and maintenance, and particularly discloses an intelligent monitor for multi-parameter fusion analysis in natural gas pipeline operation. The sensor module is arranged on a natural gas pipeline and is used for acquiring original data and signal intensity; the preprocessing module preprocesses the original data to obtain preprocessed data, and then adopts a self-adaptive compression algorithm to obtain compressed data; the fusion module adopts a multi-source data fusion algorithm for the compressed data to obtain fusion features; the diagnosis module inputs the fusion features into a fault diagnosis model based on a machine learning algorithm, and when it is diagnosed that a fault exists, the fault type is output; the positioning module obtains the position, geographic information data and historical operation data of the sensor module, and obtains a fault position by adopting a positioning algorithm according to the signal intensity, the position, geographic information data and historical operation data of the sensor module and a fault type; and the early warning module generates a fault report according to the fault position and the fault type, sends out an early warning signal and uploads the fault report.

A technology for the adaptive compression of time series of complex monitoring data records is disclosed. Variants of complex monitoring data may describe multiple observations of the same type and from the same source from different timestamps, or they may describe multiple, related observations from the same source from the same timestamp. Meta-data for individual complex monitoring data records is analyzed to identify portions of the monitoring data record that can be omitted without loss of data to improve the compression rate. Different base compression methods may be applied for individual entities of monitoring data, depending on the data type used to represent the monitoring data entity, like XOR compression for floating point types. Compressed data is stored to a storage stream, where meta-data that is required to determine the omitted portions of complex data records is stored before corresponding observation payload data.

The invention relates to the technical field of artificial intelligence and big language model reasoning optimization, and discloses a big language model-oriented adaptive KV cache compression method and system, and the method comprises the steps: constructing a lexical element importance measurement mechanism; analyzing an attention head distribution structure in large language model reasoning, and constructing a plurality of pruning strategies; based on a lexical element importance measurement mechanism and the attention head distribution structure, designing a self-adaptive key value cache compression hybrid strategy set based on a pruning strategy; constructing a static self-adaptive key value cache compression method, and automatically distributing a key value cache compression strategy in a pre-filling stage of large language model reasoning; and in a decoding stage, performing adaptive compression on the key value cache based on the allocated key value cache compression strategy. According to the method, the key value cache can be efficiently compressed on the premise of not depending on explicit attention score calculation, a system-level reasoning optimization framework is compatible, the generation performance is kept, meanwhile, the video memory consumption is remarkably reduced, and the long context reasoning capability is enhanced.