3624 results about "Throughput" patented technology

The invention relates to a risk management and control method and system based on real-time behavior analysis, and the method comprises the steps: carrying out the structural processing of multi-source behavior data through lightweight protocol decoding and behavior label embedding, and constructing an original behavior data set of a user and an entity; extracting multi-dimensional behavior characteristics by using a sliding window analysis and sparse representation mechanism, and constructing a user behavior graph by combining graph embedding learning; constructing a time-sensitive behavior trend model through streaming modeling and an incremental learning strategy, identifying an abnormal evolution trajectory in real time, and introducing a dynamic risk threshold regulation and control mechanism; adopting a high-throughput flow data processing and fast similarity matching algorithm to construct a fusion discrimination model, giving risk levels to abnormal behaviors and classifying the abnormal behaviors; and finally, performing closed-loop optimization in combination with a historical treatment effect. The system has the advantages of high real-time performance, high calculation efficiency, adaptability to complex network environments and the like, and the network security protection capability can be effectively improved.

The invention relates to a deterministic network congestion avoidance flow routing scheduling method, which belongs to the technical field of industrial internet, and comprises the following steps: S1, sensing network congestion based on IFIT flow detection and queue state, and calculating a congestion coefficient; s2, performing planning decision on a traffic routing path based on deep reinforcement learning to avoid congestion; and S3, sinking congestion coefficient calculation and strategy mapping calculation logic to switch hardware by using a P4 programmable data plane to realize localized closed-loop control. Through dynamic path optimization and localization execution, the throughput is remarkably improved, the time delay and jitter are reduced, the intelligent routing mechanism effectively balances the load, and a high-reliability dynamic scheduling solution is provided for a large-scale deterministic network in combination with the deterministic guarantee capability.

The invention relates to the technical field of cloud computing storage, in particular to a cloud storage dynamic optimization method and system, and the method comprises the following steps: based on a cloud storage data access record, recognizing an access time interval sequence, analyzing the variation amplitude of adjacent time intervals, and calculating the standard deviation of the access time intervals; and determining the access period stability according to the standard deviation and the change amplitude, and classifying the data objects to obtain a data access period stability classification value. According to the method, task scheduling is more accurate by identifying a data access time interval sequence and changes thereof and refining data classification and storage resource allocation, a high-throughput and low-delay storage path is matched, the health condition of a storage node is evaluated, a fault node is identified in advance, data stability is guaranteed, and data copy distribution is adjusted; the storage reliability and stability are improved, the storage path for accessing the intensive data is optimized, the dynamic load balance of the storage resources is realized, and the storage efficiency is improved.

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 discloses an SDR-oriented heterogeneous task scheduling and transmission system and method.The system comprises a heterogeneous platform composed of an ARM and an FPGA, a dynamic scheduling decision engine is arranged in the ARM and receives a task feature vector and a platform state parameter set as input, and the dynamic scheduling decision engine evaluates income and cost of task allocation to the ARM or the FPGA for execution to make a task allocation decision; the data generated in the task execution process or the execution completion result are interacted and synchronized between the ARM and the FPGA through the heterogeneous shared memory area HSM. The method realizes reduction of signal processing pipeline delay, improvement of data throughput and optimization of system power consumption, and is suitable for communication, electronic countermeasure and other scenes needing high-performance real-time signal processing.

The present invention relates to adaptive multi-state encoding and processing in virtualized computing environments. The system includes a virtualized state selection module that dynamically transitions between binary, ternary, quaternary, and higher-order encoding states based on workload, bandwidth, security posture, and compliance requirements. A virtual encoding engine utilizes hardware-accelerated components, such as vFPGAs, vGPUs, and cTPUs, to enhance encoding throughput. A compliance-driven feedback controller continuously monitors encoding efficiency, threat levels, and adherence to mandates such as GDPR, HIPAA, and FIPS 140-3. Additional features include AI-based anomaly detection, federated model refinement, distributed ledger-backed audit trails, and quantum-resistant encoding techniques. By integrating intelligent encoding decisions with scalable compliance enforcement, the system enables high-performance, secure, and regulation-ready data processing across distributed cloud and edge environments, delivering measurable improvements in system responsiveness, data integrity, and operational trust.

The invention relates to the technical field of storage optimization, in particular to a data storage management system and method based on big data, and the system comprises an efficiency monitoring module, a medium evaluation module, a decision generation module, a strategy execution module and an effect feedback module. According to the method, input and output delay, a medium health attenuation curve and a bandwidth fluctuation map are collected, a dynamic feature space is constructed, the stability of the system is judged in combination with a Lyapunov index, early warning of performance abnormity is achieved, health indexes such as an erasing cycle, a seeking error rate and a charge retention rate are fused, and the access frequency and the load pressure are combined. Calculating a compensation coefficient, improving heterogeneous medium cooperation efficiency, pre-migration evaluation cost and topology adjustment income, improving resource configuration accuracy, dynamically adjusting hot and cold data distribution and node load, enhancing system scheduling flexibility, comparing throughput, medium wear rate and synchronous delay, and quantifying stability gain brought by configuration change; and a closed-loop process from identification to feedback is constructed.

A low earth orbit satellite (LEO)-based communication system and method. The system comprises a low-orbit satellite constellation, a ground station, a user terminal and a network management and control center. The network management and control center comprises a track and link prediction module, a link quality prediction and evaluation module, a predictive resource pre-allocation and dynamic adjustment module and a space-ground cooperative control module. According to the communication system, active resource optimization and dynamic adjustment before actual switching are realized through track and link prediction, link quality prediction, predictive resource pre-allocation and dynamic adjustment and air-ground cooperative control. According to the method, the switching time delay can be greatly reduced, the packet loss rate is remarkably reduced, and the system throughput and the access success rate are also remarkably improved. Through predictive management, the performance, reliability and user experience of LEO communication are comprehensively improved.

The invention provides an end-side multi-modal large model accelerated reasoning method and system, and the method comprises the steps: carrying out the two-stage screening and rearrangement of visual tokens based on the CLS attention and text-to-visual attention in a visual encoder and pre-filling stage, and constructing a sparse attention and sparse key value cache; in a decoding stage, an important neuron set is judged according to activation gating or historical statistics, only a corresponding feedforward network weight is pulled and calculated, missed weights are loaded on demand through asynchronous I / O, and hot neurons are maintained in a high-speed memory to utilize model sparsity, so that video memory / memory occupancy and calculation overhead are remarkably reduced on an end side; throughput and time delay performance are improved. According to the method, the internal memory and computing resources required by reasoning of the multi-modal large language model are reduced from two dimensions by utilizing the endogenous sparsity of the end-side large language model in input and the model, so that a higher reasoning speed is achieved by utilizing fewer resources on the premise of keeping the size of the model unchanged, and the performance of the whole system is improved.

The present invention relates to a system and computer implemented method for cost-aware autoscaling of artificial intelligence workloads using predictive queueing models, designed to achieve proactive and economically optimized scaling of computational resources across cloud and edge environments. The invention introduces a predictive queueing-based technique that anticipates future workload congestion by modeling dynamic task arrivals and service times using a stochastic queueing process. A cost estimation unit computes the total projected operational cost of potential scaling actions by integrating real-time infrastructure pricing data, predicted delay penalties derived from service-level objectives, and estimated energy consumption. A scaling decision unit applies reinforcement learning-based optimization to select the scaling action that minimizes total cost while ensuring compliance with latency and throughput constraints. The system includes a hardware-integrated autoscaling controller device comprising a predictive computation processor, cost-decision processor, and scaling actuation interface configured for real-time execution of predictive and scaling operations.

The invention provides a bus network data transmission control method based on priority scheduling, which relates to the technical field of network data transmission, and comprises the following steps: dividing a bus network into an edge layer, a core layer and a terminal layer; the terminal layer comprises a plurality of actuator nodes; performing priority classification on the terminal data set in the edge layer; distributing a transmission time slot of priority rating data on the core layer based on a preemptive scheduling algorithm; a priority data transmission strategy is dynamically adjusted in combination with the dynamic time slot allocation table and the local load state; and when a plurality of actuator nodes initiate transmission requests at the same time, outputting a final transmission sequence according to a priority data transmission strategy. Through a static rule base and a dynamic correction mechanism, key data features are accurately identified, the real-time performance and reliability of high-priority data are ensured, and the key business delay risk is reduced. Intelligent allocation of bandwidth resources is achieved through a preemptive scheduling algorithm, real-time service requirements are guaranteed to the maximum extent, and the network throughput efficiency is remarkably improved.

The invention relates to a network slice optimization method for a 5G private network, and the technical scheme comprises the steps: constructing a time sequence performance model in a slice operation process, and monitoring tiny changes including end-to-end time delay, throughput and packet loss rate; then extracting a short-period fluctuation mode of the index, and comparing the gradient with a historical stability threshold value; judging whether the index is in a performance critical state or not by identifying an index drifting behavior which continuously jitters but does not default yet; non-equal-ratio responses including jitter improvement and delay decline after each time of resource adjustment are recorded; then trend fitting is carried out on the data, and whether a performance platform period or a critical jump point exists or not is judged; delimiting three intervals of resource investment in the model: an inactive area with invariable investment, a high-sensitivity area with rapid performance change and a rising saturation area with stable performance and continuous resources, wherein the three intervals include an inactive area with invariable investment, a high-sensitivity area with rapid performance change and a rising saturation area with stable performance; and identifying the current response area of the slice through the model, determining that the regulation strategy does not take equal proportion input as the core, and taking the optimal response area as a resource allocation interval to make a choice.

The invention relates to the technical field of communication, in particular to a dynamic priority dual-mode communication fault-tolerant switching method and equipment, and aims to solve the problems of communication interruption and service quality reduction when a main link is degraded in a complex electromagnetic environment. According to the method, a dynamic priority evaluation model and a dual-mode protocol cooperation mechanism are constructed, throughput, bit error rate, delay jitter, task semantic priority and aging constraint data of a main link and a standby link are collected in real time, a priority weight and a communication mode decision matrix are generated through combined modeling, an atomization switching instruction is executed according to the priority weight and the communication mode decision matrix, and a communication mode is switched. And completing path reconstruction and resource redistribution. The system further introduces link health prediction, a resource reservation pool and an online learning feedback loop, and supports fault pre-judgment, key task preemption and model adaptive optimization. According to the method, the limitation of a static threshold is broken through, the task semantics and channel state joint scheduling is realized, and the communication reliability, the service continuity and the intelligent fault-tolerant capability of a distributed system in a high-risk scene are remarkably improved.

The invention relates to the technical field of cooperative control, in particular to a full-intelligent simulation load distributed cooperative control method, which comprises the following steps of: acquiring node load data, dynamically predicting, adjusting and distributing, performing cooperative scheduling optimization control, and generating an intelligent load control scheme. According to the method, the load state information is extracted in real time and standardized verification is carried out, so that the running state of each node has a unified measurement basis, task allocation is carried out in combination with node processing capacity and throughput performance, and task scheduling and node performance dynamic matching are realized; prejudgment type load regulation and control are achieved by combining historical data trend prediction with a current state, the risk of task delay and node overload is avoided, a task allocation strategy is continuously optimized by utilizing real-time state feedback, the resource use efficiency is improved, the cooperative relation between nodes is strengthened, and the sensitivity and consistency of system scheduling are guaranteed in a dynamic load change environment. And the full-process adaptivity and collaborative stability of task scheduling in a multi-node system are supported.

The present invention provides systems and methods for cellular network performance monitoring and optimization, enabling SIM-based devices to dynamically adapt to changing network conditions for improved connectivity. The invention introduces a process that includes determining baseline path performance through detailed probing of network metrics, continuously assessing current path performance via real-time monitoring, and instructing the SIM to switch from its current connected mobile network carrier to an alternate carrier when predefined performance thresholds are not met. Switching instructions are securely delivered Over-The-Air (OTA) to the SIM, ensuring seamless transitions to the most efficient and reliable network path. The system leverages both active and passive application layer observations to optimize latency, throughput, and reliability while supporting diverse applications, including IoT devices, industrial systems, and consumer devices.

An improved system implements Fitness Gradient Consensus including hash distance and bucket consensus variations within a digital blockchain by calculating the highest fitness value competing blocks to resolve conflicts and allocate the rewards associated with building new blocks. The consensus system applies conflict resolution formulas to incentivize block-building nodes to share blocks generated, as it completes construction, to improve chances of a reward, resulting in enhanced speed and security of blockchain.The hash distance consensus utilizes a hash distance scalar value as part of its fitness metric, and the bucket consensus assigns tokens to buckets and calculates an aggregate value of the assigned tokens. A trust-but-verify variant increases transactional throughput and reduces linearity and computational constraints.The system also utilizes novel record types, such as token genesis, transfer, transaction, trade order, settlement, proposition, determination, and pattern linkage records to facilitate the automation of financial, commercial and legal processes.

The invention discloses a power grid real-time data reasoning analysis system and method based on Rete algorithm optimization, and belongs to the technical field of smart power grids. In order to solve the problems of low rule matching efficiency and insufficient event triggering and operation optimization capability, a data acquisition module efficiently accesses a multi-source heterogeneous data stream, a streaming data processing module realizes high-throughput real-time data processing based on DAG task fragmentation and parallel computing, a rule engine module optimizes a dynamic rule network through a Rete algorithm, and a data processing module realizes real-time data processing based on DAG task fragmentation and parallel computing. The reasoning analysis module is combined with rule matching and event triggering, power grid faults are rapidly positioned, operation state evaluation and dynamic optimization suggestions are provided, and the graphical rule construction module reduces rule design complexity and enhances rule adaptability through visual rule editing and verification tools; the optimization monitoring module guarantees the system stability and the resource utilization efficiency through real-time performance monitoring, load prediction and automatic tuning.

Low Latency, Low Loss, Scalable Throughput (L4S) latency improvements may be provided. Improving L4S can include determining traffic in a L4S queue has crossed above a first threshold. In response to determining the traffic in the L4S queue has crossed above the first threshold, a priority of the L4S queue is increased. When it is determined the traffic in the L4S queue has crossed above a second threshold, a congestion notification is sent.

The invention relates to the technical field of Internet of Things routing management, in particular to a big data-based routing optimization method, which comprises the following steps of: acquiring spectrum data of an Internet of Things routing management area, performing correlation analysis on a spectrum state corresponding to a historical success rate and retransmission times of communication between nodes, and quantifying a probability of successful communication at a future target time. According to the method, the multi-band historical communication success rate and retransmission frequency data are dynamically acquired, the interference risk of the target time period is analyzed and predicted in combination with the time sequence, an interference avoidance strategy based on static channel assessment is upgraded to a probability-driven pre-judgment mechanism, and the anti-interference capability of a transmission path in a complex electromagnetic environment is improved. According to multi-protocol link characteristics of Bluetooth, WiFi, 2.4 G and the like, parameters such as delay, throughput, packet loss rate, flow cost and the like of a single path are synchronously quantified, a unified performance cost file of a heterogeneous network link is established, and end-to-end performance modeling of cross-protocol path combination is realized.

The invention belongs to the technical field of data transmission and processing, and discloses a zero-copy-based multi-thread high-throughput data stream channel separation method and system, and the method comprises the steps: directly writing a mixed data stream into a front-end buffer region configured as an annular structure through a data receiving module by adopting direct memory access; then, a multi-thread processing module dynamically allocates a plurality of processing threads from a thread pool to separate channel data in parallel, each thread adopts a zero copy algorithm based on pointer offset, positions the channel data in a memory, creates pointer reference and associates the channel data to a corresponding rear-end buffer area, and logic separation is achieved without physical copy; and finally, the data storage module efficiently writes the separated data into persistent storage in an asynchronous I / O mode. According to the method, zero-copy, multi-thread parallel and two-stage dynamic buffering strategies are combined, the data separation efficiency is remarkably improved, CPU occupation and memory bandwidth are greatly reduced, and the real-time performance and stability of high-throughput data processing are guaranteed.

The systems and methods provide a model deployment criterion. The model deployment criterion indicates a difference in a value against which the proxy model may be measured to determine when, if ever, the proxy model should be deployed to replace the existing rule engine. The model deployment criterion may be keyed to the proxy model (e.g., based on a difference in its size, throughput speed, number of changes, etc.), the existing rule engine (e.g., based on a difference in its age, update occurrences to its rule base, etc.), and / or comparisons between models (e.g., based on differences in results, throughput speed, efficiency, etc.).

The invention discloses an asynchronous parallel reasoning method, system and equipment for a hybrid expert model and a medium, which are corresponding schemes: decoupling synchronization of calculation and communication between GPUs (Graphics Processing Unit) caused by all-to-all set communication in expert parallelism, allowing asynchronous parallelism of model calculation and lexical metadata communication, and solving the problem of asynchronous parallelism of the model calculation and lexical metadata communication. Data communication overhead caused by expert parallelization is fully masked, and synchronization waiting overhead is eliminated; aiming at the phenomenon of uneven cold and heat of experts in reasoning, the hot experts are preferentially placed in the GPU, the cold experts are laterally loaded in the CPU so as to release the video memory space of the GPU, and the calculation efficiency of the GPU can be improved by increasing the batch size during reasoning; efficient resource scheduling is realized by dynamically selecting a computing unit which is most suitable for execution and a cold expert which needs to be loaded; generally speaking, the communication overhead and the waiting overhead during parallel reasoning of experts can be remarkably reduced, meanwhile, the calculation efficiency of the GPU is improved, and the overall throughput performance in the reasoning process is optimized.

The invention provides a DCU-based high-performance sparse stiffness matrix vector multiplication method, which comprises the following steps of: according to a sparse stiffness matrix, dividing a non-zero element into a plurality of calculation unit blocks by rows, pre-loading non-zero element data to an L1 shared memory or a register file through an on-chip shared memory controller of the DCU, a high-bandwidth crossbar switch of the DCU is used for realizing data copying and transmission; starting multi-row fusion execution for a short row of which the row non-zero element is lower than a DCU single-instruction multi-data width threshold value; constructing a wavefront scheduler based on a DCU asynchronous computing engine: binding an independent instruction cache region for each wavefront, and loading a multiply-add operation instruction set in advance through a prefetch instruction queue; a calculation unit state register is established, and when a wavefront scheduler ready signal is triggered, a scalar unit of the DCU is activated to execute calculation; and realizing cross-thread block reduction by adopting a DCU atomic operation accelerator. According to the method, the calculation throughput and the memory bandwidth utilization rate of large-scale structural mechanics stiffness matrix vector multiplication are effectively improved.

The invention relates to the technical field of data flow control, in particular to a heterogeneous data flow synchronous control method and system.The method comprises the following steps that multi-source data segment receiving time is obtained, a time interval sequence is constructed, an interval difference value and a synchronous difference quantity are calculated, the sensitivity level is judged, a channel throughput change is extracted, a descending section is marked, and a priority index is generated; and mapping the high-sensitivity stream to a first-stage or second-stage output path to sort and distribute key stream segments, judging that serial numbers are continuously combined or independently output, and generating intersection and sequence information. According to the method, the time interval sequence is constructed based on the receiving time of the multiple data sources, the interval deviation is quantified, the labels are divided through the synchronization difference quantity, the sensitive level structure is established, differential management of the synchronization precision requirement of the heterogeneous data streams is achieved, the response efficiency and the alignment precision of the high-sensitivity data segments are improved in the data distribution process, and the data distribution efficiency is improved. And meanwhile, the competition of a low-sensitivity section for output resources is reduced, and the synchronous time sequence coordination capability and the channel bandwidth utilization rate in a multi-source heterogeneous environment are improved.

The invention relates to a cargo-to-person system multi-AGV path planning method based on deep reinforcement learning in the field of computers, and the method comprises the steps: fusing a double-layer planning architecture of improved A * global guidance and local reinforcement learning, employing a non-uniform penalty heuristic strategy in a global layer, and employing a non-uniform penalty heuristic strategy in a local layer; performing collaborative optimization on the path dispersity through a fixed penalty term calculated by a Dijkstra algorithm and a real-time dynamic penalty term based on a historical path; an asynchronous near-end strategy optimization (APPO) algorithm is designed in a local layer, multi-channel observation states (static obstacles, AGV positions and global path codes) of five continuous time steps are processed in combination with an LSTM network, and a five-degree-of-freedom action strategy is output. A fixed punishment exemption mechanism is set for a storage movable shelf (Pod) area, invalid avoidance is avoided, the task success rate and throughput are improved, the calculation load is reduced, and an efficient and extensible solution is provided for solving the multi-AGV path planning problem of a cargo-to-person system.

The invention belongs to the technical field of computers, particularly relates to a prefetch instruction management method, system and equipment, and aims to solve the problem of low instruction fetch efficiency of an instruction prefetch technology. The method comprises the following steps: receiving branch prediction information from a branch prediction unit, and performing label comparison on table entries in an instruction fetching target queue and the branch prediction information; under the condition that the table item is hit, querying an instruction fetching address corresponding to the branch prediction information in the hit table item; under the condition that the table item is not hit, a new storage table item is allocated to the branch prediction information, and an instruction fetching address carried by the branch prediction information is determined through the storage table item; in response to a received prefetching request sent by the prefetching unit, determining a target table item based on the prefetching request, and returning an instruction fetching address queried in the target table item to the prefetching unit; and storing the stable cache line in the target table item to a flow buffer area. According to the method, multiple prediction requests can be processed in parallel, the front-end throughput is improved, and the instruction fetching efficiency is improved.

A system can configure a first number of demodulation reference signal positions in radio resource control information as part of a connection setup with a base station that is configured to facilitate first broadband cellular communications. The system can, after attaching to the base station, receive a medium access control control element message from the base station indicative of modifying the first number of demodulation reference signal positions to a second number of demodulation reference signal positions. The system can conduct second broadband cellular communications with the base station according to the second number of demodulation reference signal positions, wherein a throughput of the second broadband cellular communications is determined as a function of a size of a transport block set based on the second number of demodulation reference signal positions.

The invention relates to the technical field of data processing, and particularly discloses an improved inter-ERP (Enterprise Resource Planning) system data migration and synchronization method, which comprises the following steps of S1, metadata acquisition and preprocessing; s2, field intelligent matching; S3, machine learning recommendation; s4, mapping confirmation and feedback; s5, converting rule configuration; s6, carrying out total migration; s7, performing increment synchronization; and S8, monitoring and alarming: counting delay, throughput and failure rate of increment synchronization in real time, and automatically sending an alarm notification when an abnormal threshold is triggered. According to the method, a complete technical system which is infinitely infinitely and covers the whole migration and synchronous life cycle is constructed from metadata acquisition to intelligent mapping, rule driving and full / incremental ETL (Extract Transform Load), and then to monitoring alarm and rollback management; the method is suitable for efficient, reliable and extensible data migration and synchronization between heterogeneous ERP systems, and provides solid support for enterprise informatization upgrading, system splitting / merging and big data analysis enabling.

Methods, devices, systems, and means for user equipment slicing assistance information by a user equipment, UE, are described herein. The UE detects a condition of the UE (610) and, based on the detecting, evaluating one or more preferences (612). Based on evaluating the one or more preferences, the UE sends UE Slicing Assistance Information, USAI, to a core network entity (614), the USAI being based on a current network slice configuration. The UE receives, from a base station, a reduced radio resource configuration for operating using the low-throughput network slice (616) and communicates using the low-throughput network slice (618).

The invention belongs to the technical field of artificial intelligence and high-performance computing, and particularly relates to a neural network large model efficient reasoning method based on multiple GPGPUs. The method aims to solve the problems of high communication overhead, non-uniform load, low resource utilization rate, high data transmission delay and the like among multiple processors. Dividing a calculation task into a plurality of sub-graphs through static analysis and mixed granularity partitioning of a model calculation graph; distributing the sub-graphs to the optimal GPGPU based on a weighted cost function in combination with heterogeneous resource perception and a dynamic mapping strategy; a global pipeline scheduling plan is constructed by using communication topology perception, and calculation and communication overlap are maximized; data are loaded in advance through a host side hierarchical caching and asynchronous prefetching mechanism, and transmission delay is hidden; multi-stream concurrent execution and event-based lightweight synchronization are adopted on each GPGPU, so that waiting overhead is reduced. According to the method, the reasoning delay can be remarkably reduced, the throughput and the hardware utilization rate are improved, and the method has good adaptivity and expandability.