311 results about "Preemption" patented technology

The invention discloses a multi-modal fusion AGV dynamic path planning and cluster scheduling system, and relates to the technical field of multi-modal perception and data fusion, and the system comprises a multi-modal perception module which generates a dynamic obstacle confidence map through multi-source data fusion in combination with a hardware-level time synchronization and Transform feature fusion network; the dynamic path planning module adopts an improved rolling window algorithm, integrates an LSTM space-time conflict prediction model and an adaptive weight cost function, and realizes dynamic obstacle trajectory prediction and non-oscillation global path generation; the cluster scheduling control module is used for optimizing multi-AGV task allocation and conflict resolution in combination with a dynamic priority preemption mechanism and digital twinborn simulation rehearsal based on a distributed contract network protocol of edge computing; and the data conflict resolution module is used for triggering a multi-modal re-calibration process through confidence weighting and sliding window time sequence verification. According to the system, in logistics storage and intelligent manufacturing scenes, the dynamic obstacle avoidance success rate and the robustness and operation efficiency of an AGV cluster are improved.

The invention discloses a digital twinborn model real-time linkage method driven by information physics fusion, particularly relates to the technical field of digital twinborn cooperative control of an industrial automation production line, and is used for solving the problems of instruction conflict and control failure caused by mismatching of virtual model parameters and dynamic capability of physical equipment in the prior art. According to the method, equipment state and material flow data are collected in real time, equipment dynamic degradation parameters are extracted to generate capability attenuation feature vectors, hidden process conflict path detection and resource preemption probability simulation are combined, a collaborative optimization model of equipment health and process scheduling is constructed, and a control instruction set containing dynamic capability constraints is generated. After the screening instruction is verified through physical constraint matching, parameters are corrected in a closed loop mode based on an execution result, a model is iterated, and self-adaptive matching of the virtual instruction and the physical equipment capacity is achieved; the reliability of the digital twinning control instruction and the stability of a production line are remarkably improved, and the risk of abnormal shutdown caused by an overrun instruction is avoided.

The invention discloses a heterogeneous AI computing power resource scheduling method and system, and the method comprises the steps: constructing a heterogeneous AI computing power resource pool, wherein the heterogeneous AI computing power resource pool integrates the computing resources of a plurality of heterogeneous AI acceleration chips; obtaining a scheduling demand of the AI task, wherein the scheduling demand comprises a task type, a resource request quantity, a priority identifier and a task group association relationship; generating a multi-dimensional scheduling strategy according to task requirements, wherein the scheduling strategy comprises a priority scheduling rule, an affinity scheduling rule and a resource preemption rule; based on a multi-dimensional scheduling strategy, the AI tasks are dynamically allocated to target computing power nodes of the heterogeneous AI computing power resource pool, and the task execution state and the resource utilization rate are monitored in real time; and dynamically adjusting computing resource allocation according to the resource utilization rate. Through the heterogeneous AI computing power resource pool, the resource utilization rate is remarkably improved, dynamic resource allocation is realized through a multi-dimensional scheduling strategy, and meanwhile, a communication path is optimized through an affinity scheduling strategy, so that the problem of task starvation caused by resource fragmentation is avoided.

The invention relates to a network real-time synchronization communication system, and relates to the technical field of computer network communication. The system comprises four core modules: a hardware optical IO synchronization module which integrates a multi-wavelength optical transceiver array and an anti-jitter circuit and provides a nanosecond global clock signal, and the synchronization precision is less than or equal to 15ns; the time sequence arrangement module is used for dividing a fixed time window and a dynamic buffer window based on a time sensitive network, supporting a priority preemption mechanism of the SRIO bus and realizing deterministic data transmission; the distributed RTDATA nodes replace a traditional single-board computer, a heterogeneous computing unit and an intelligent storage controller are integrated, and protocol stack processing delay is reduced through the zero copy technology; according to the dynamic multicast routing system, a safely isolated multicast tree is constructed as required, and link load distribution is optimized in combination with machine learning. Through collaborative design of hardware optical synchronization and dynamic data arrangement, the problem of real-time performance reduction caused by multi-node communication concurrence is solved.

The invention relates to the technical field of computing power resources, and discloses a flexible allocation monitoring system for computing power resources, which defines a clear resource use boundary for different types of tasks through a container-level QoS strategy of a resource isolation unit, and adjusts resource allocation in real time in combination with a dynamic partition management module, thereby avoiding resource waste in a traditional fixed allocation mode, and improving the resource allocation efficiency. The dynamic preemption unit accurately screens low-priority tasks for resource recovery through a multi-factor decision engine and a preemption cost evaluation algorithm, in 50 concurrent task scenes, the resource preemption response time is shortened to 20-35 ms and is improved by 50%-70% compared with 70-100 ms of Docker / K8s, the blocking duration of high-priority tasks is reduced to 15-25 ms from 80-120 ms, and the core service interruption risk is greatly reduced.

The invention relates to the technical field of electric power data analysis, and discloses a concentrator collection data analysis method and system based on big data, and the method comprises the steps: capturing a high-priority task in real time, and generating a concentrator collection task flow sequence arranged according to an initial scheduling sequence; performing cross-task dependency chain logic association tracking on the concentrator collection task flow sequence, and constructing a multi-dimensional scheduling feature vector reflecting the importance gradient of the task; based on the multi-dimensional scheduling feature vector, a hierarchical priority aggregation algorithm is utilized to extract a preemption evolution track of a high-priority task to the bandwidth and the acquisition time slot, and potential scheduling error sequence candidate tasks are detected; performing priority grading delimitation on the scheduling error sequence candidate tasks and the associated low-priority task interval, and constructing a local task logic graph with a priority weight gradient; and performing dynamic intervention judgment on the scheduling evolution path based on the local task logic graph. The method has the advantage of improving the overall scheduling efficiency.

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 discloses a modular reclosing circuit breaker protection system based on intelligent electrical cloud coordination, particularly relates to the technical field of intelligent power distribution network relay protection, and is used for solving the problems that fault diagnosis is delayed and a protection strategy cannot be generated in a fault clearing time window due to resource competition in a cloud coordination protection system. A dynamic preemption priority is generated based on power grid topology vulnerability and associated circuit breaker collaborative requirements by monitoring the load state of cloud computing node resources in real time; credibility attenuation control is carried out in combination with a fault current waveform distortion rate; when the resource preemption necessity exceeds a threshold value, freezing a periodic task thread and releasing computing resources, and opening up a high-priority communication channel at the same time; sending a fault recording compression instruction to a target circuit breaker terminal to obtain an edge preprocessing feature vector; and protection strategy optimization is executed in the released resources to generate reclosing parameters, and the reclosing parameters are issued and executed through a high-priority channel, so that quick response of a fault diagnosis task is realized.

The invention belongs to the technical field of multi-modal intelligent data management, and discloses a multi-modal business data co-processing and intelligent conversion method, which comprises the steps of fusing multi-source data, coupling cross-modal time sequence association, detecting image-text semantic conflicts, and generating an RAG semantic unit and a space-time alignment semantic packet. Constructing a four-dimensional priority label to divide resource pool categories, dynamically allocating resource pool computing resources, generating a low-redundancy feature set, triggering a forced preemption strategy during resource competition, and synchronously generating a resource arbitration log; injecting a dynamic rule node according to a data type, executing confidence arbitration in combination with a conflict arbitration identifier, generating business decision content and generating a rule correction decision packet; constructing a full-link semantic thinking chain, synchronously capturing a system operation track, and aligning to generate a dual-channel tracing report; identifying residual sensitive fields, triggering hot update to load corresponding rules, quantizing effectiveness to generate compliance reports, reversely updating an industry rule base, and optimizing a resource allocation strategy.

The invention relates to the technical field of computing resource scheduling, and discloses a heterogeneous multi-task computing power dynamic scheduling method and system.The method comprises the steps that in response to multiple computing task requirements of a user, the priority of computing tasks is analyzed and sorted, and a task priority sequence is obtained; classifying the heterogeneous computing resources according to the resource states of the heterogeneous computing resources, and respectively placing different categories of computing resources in corresponding resource pools to obtain multiple categories of resource pools; in combination with the task priority sequence and the multi-class resource pool, allocating computing resources of a corresponding class to each task through a preset resource matching mechanism; in the task calculation process, when the calculation resources of the first task are insufficient, the calculation resources of the second task are preempted through a preset resource preemption mechanism, the priority of the second task is improved after resource preemption, and the calculation resources are redistributed; according to the method and the device, the corresponding computing resources can be allocated in combination with the task priorities, and the resources are dynamically scheduled when the computing resource gap appears.

The invention discloses a PLC scheduling algorithm based on a multi-core soft real-time operating system, and relates to the technical field of voltage control, and the algorithm comprises the following four steps: 1, starting a target industrial control system, completing hardware self-inspection, memory allocation and communication module starting, and sending an initialization ready signal after confirming that an assembly is normal; 2, according to PLC task functions and real-time performance, four types of standardized tasks are divided, priority levels are defined, sub-priorities are configured, a dependency relationship is identified, and a task chain is established; 3, counting the number of tasks and task chains, distributing resources by using multiple binding cores, and constructing a global ready queue to adapt to dynamic scheduling; and 4, monitoring an interrupt event through an interrupt manager, realizing high-priority task preemption by semaphores, and allocating tasks to corresponding CPU cores, so that the problems of soft real-time multi-core task preemption and data conflict are solved, the scheduling real-time performance and the data security are improved, the system failure rate is reduced, and the industrial control efficiency and stability are improved.

The invention relates to the field of intelligent control, in particular to an intelligent washing control method and system based on multi-modal data, and the method comprises the steps: carrying out the time sequence alignment and feature fusion of a running state parameter set and a communication demand parameter set, and obtaining a multi-modal communication feature matrix; analyzing the coupling relationship between the dehydration stage identifier and the real-time energy consumption value of the washing equipment through a preset conflict prediction model, dynamically calculating the channel preemption priority of each washing equipment according to an instruction execution time window, and generating a clustering type communication resource scheduling sequence in combination with the priority of a control instruction; obtaining the number of instruction conflicts identified by each washing device in the dehydration stage, and constructing a load balancing optimization function with the instantaneous circuit load; and when the output value of the load balancing optimization function is minimized, obtaining a dehydration peak shifting control instruction. The problems of communication quality reduction and control delay during cooperative control of a plurality of washing devices are solved, and the timeliness and safety of control of the washing devices are improved.

The invention discloses a multi-task scheduling method, system and device based on resource queue management and a medium, and relates to the technical field of chip design simulation. The method mainly comprises the following steps: receiving a task creation request, generating a unique task identifier for a task according to a resource type identifier and task validity period information contained in the task creation request, and adding the task into a resource queue of a corresponding resource type; sorting and scheduling the tasks in the resource queue according to the task creation time, the task state and the task validity period; in a task execution process, according to a resource type required by the task, adopting a preemptive or polling resource allocation mode to obtain a target resource; after the task is executed, the task state is updated, the state of the completed task is allowed to roll back to the to-be-scheduled state under the preset condition, and the task enters the corresponding resource queue again to wait for scheduling. According to the method, the fairness of task scheduling, the resource utilization rate and the system robustness can be effectively improved.

Techniques are provided to enable mid-thread (instruction level) preemption in a graphics processor without requiring software intervention by a graphics driver associated with the graphics processor. Hardware based mid-thread preemption is facilitated using the thread dispatch hardware of the graphics processor to trigger execution of a kernel program by the graphics processor that saves the thread state of preempted processing resources and facilitates the subsequent restoration of that thread state to the same or a different set of processing resources.

The invention relates to the technical field of intelligent manufacturing and production optimization scheduling, and particularly discloses a thermal fuse flexible production-oriented order scheduling resource scheduling optimization method, which comprises the following steps of: generating a structured vector by extracting inherent and dynamic characteristics of an order, and constructing a production scheduling dynamic environment model by fusing a device resource state time sequence; the order comprehensive process switching cost and the resource constraint tightness are calculated, order clusters are divided through multi-target constraint clustering, and a pre-arrangement sequence is generated; key resource conflict early warning is completed through time sequence prediction; constructing a multi-target scheduling model by taking the order cluster as a unit, and generating a look-ahead window scheme by combining a rolling optimization solidification plan baseline; and finally, on the basis of a real-time execution data updating model, through dynamic priority preemption and local rescheduling adjustment when abnormity occurs, the problems of high cost of small-batch mixed arrangement, conflict identification lagging, easy damage of a thermosensitive process steady state and poor compliance constraint adaptation are solved, and collaborative optimization of the scheduling efficiency, the production yield and the compliance is realized.

The invention provides a mechanical arm control system and control method based on an embedded virtual machine architecture. A real-time operating system and a non-real-time operating system are deployed on the same embedded board card; the non-real-time operating system transmits control words of all the drivers to a driver communication module deployed on the real-time operating system through a control queue based on the received control request, and the driver communication module transmits state data of all the drivers to the non-real-time operating system through a state queue; and the motion planning module deployed on the non-real-time operating system transmits the generated position data to the driver communication module through the position queue according to a preset control period, and issues the control word and the position data to the driver through the driver communication module. By the adoption of the method, the problems that a non-real-time operating system is interrupted, communication stability is poor due to multi-task resource preemption, control response is slow, state coverage is caused by jitter of the non-real-time operating system, and a mechanical arm is stuck due to position transmission jitter can be solved.

The invention discloses a task scheduling method based on dynamic weight, and belongs to the technical field of workflow engines and distributed computing. The method comprises the steps of establishing a to-be-allocated task set; calculating a dynamic priority weight; establishing a task queue; sequentially processing the tasks according to a task queue sequence; establishing a calculation node set of all available processing tasks, and screening out an optimal calculation node corresponding to a to-be-processed task in the task queue; judging whether the to-be-processed task is partitioned or not; establishing a task preemption condition until the to-be-processed task is processed; and repeating the steps until all tasks are processed. The method can be applied to cloud computing, intelligent manufacturing and distributed system task scheduling scenes, and the resource utilization rate and task completion timeliness are remarkably improved.

The embodiment of the invention discloses an interrupt processing method, a processor, a chip, a storage medium and a program product, and the method comprises the steps: determining the remaining processing time of an assembly line of a target task based on the state of the assembly line of the target task being processed in response to a to-be-processed interrupt signal; determining the real-time weight of the interrupt signal to be processed based on the interrupt type of the interrupt signal to be processed and the current working mode of a processor; the real-time weight is used for representing the urgency of the interrupt signal to be processed; determining a response score of the interrupt signal to be processed based on the remaining processing time of the pipeline and the real-time weight; the response score represents the response priority of the interrupt signal to be processed; and in response to the response score reaching a first threshold value, triggering interrupt preemption, and processing the interrupt signal to be processed.

The invention provides a multi-factor collaborative dynamic priority task scheduling method based on waiting time feedback, which is characterized by comprising the following steps of: initializing waiting time, task criticality and weight values of queue length, initializing a mixed task set and a task waiting queue, and initializing the task waiting queue into an empty set; entering a task waiting queue; updating the priorities of all tasks in the task waiting queue; sorting the tasks in the task waiting queue according to the priority; monitoring whether a new task arrives in real time; judging whether the newly arrived task meets a preemption condition or not; updating the historical average waiting time of the task; updating weight values of task waiting time and queue length; and judging whether all tasks in the mixed task set are scheduled. According to the method, long-term hunger of low-priority tasks is avoided, and the average waiting time of the tasks is shortened; the preemption threshold value is dynamically adjusted, and the context switching overhead is reduced; a dynamic weight mechanism is designed, and the self-adaptive capacity of the system is improved.

The invention provides a computing core preemption method and device, equipment, a storage medium and a program product. The computing core preemption method applied to the graphics processor comprises the following steps: receiving a preemption request for a computing core; for each thread executing the first calculation task in the calculation core, in response to the preemption request, storing task data corresponding to sub-tasks which are not executed by the thread in the process of executing the first calculation task to obtain an idle thread; under the condition that all the threads in the computing core are idle threads, executing a second computing task through the computing core; the priority of the second computing task is higher than that of the first computing task. In the calculation core preemption method, the task data corresponding to the unexecuted subtasks of the first calculation task is unloaded out of the graphics processor, and the obtained idle thread executes the second calculation task, so that the amount of data needing to be unloaded can be reduced, and the unloading efficiency and the execution efficiency of execution preemption can be improved.

The invention discloses an unmanned aerial vehicle real-time task scheduling and resource management system, which belongs to the technical field of unmanned aerial vehicle computing resource management, and comprises a task load prediction module, a priority dynamic evaluation module, a resource elastic distribution module and a closed-loop feedback optimization module, the task load prediction module adopts an LSTM-GRU hybrid neural network to predict a resource demand peak value in advance and triggers a pre-scheduling signal; the priority dynamic evaluation module dynamically adjusts the task priority based on the multi-dimensional evaluation matrix; the resource elastic allocation module realizes elastic allocation through a reserved resource pool and a multi-level preemption mechanism; the closed-loop feedback optimization module monitors performance deviation and reversely adjusts parameters of each module, the four modules are deeply coupled to form a closed-loop cooperation mechanism, and intelligent task scheduling and dynamic resource optimization are achieved.

The invention provides an FPGA (Field Programmable Gate Array) multi-modal control method based on dynamic partial reconfiguration, which comprises the following steps that: a main body adopts a layered hardware architecture to realize multi-modal control, firstly, a system module of a whole system is divided, so that FPGA logic resources are divided into a static region and a reconfigurable region; wherein the static area is responsible for global control and a communication interface part, the reconfigurable area realizes a multi-mode function by loading different configurations, and the middle layer is used for deploying the dynamic part to reconfigure a controller and is responsible for time sequence coordination and bit stream management. Compared with the prior art, the method has the advantages that the mapping relation between the modal feature library and the hardware module is established, the priority preemptive reconfiguration strategy and the modal switching cost model are adopted, the real-time matching of the sensor data flow and the hardware architecture is achieved, the dynamic data path reconstruction technology supports multi-modal parallel processing, and the real-time matching of the sensor data flow and the hardware architecture is achieved. Meanwhile, time sequence constraint, resource occupancy rate and power consumption budget are considered.

The invention relates to the technical field of database monitoring, in particular to a database interaction-oriented software performance bottleneck monitoring system, which comprises a three-dimensional tensor construction module for acquiring a transaction operation chain of each node and a latch contention waveform and an IO bandwidth pulse sequence associated with the transaction operation chain in real time through a distributed transaction probe to generate a three-dimensional tensor; the tensor decomposition module is used for inputting the three-dimensional tensor into a decomposition model to generate a competitive intensity spectrum; the dynamic back pressure regulation and control module is used for identifying competition hotspots according to the competition intensity spectrum, and injecting back pressure pulses with phase difference into a transaction scheduler when detecting that a latch contention waveform and IO bandwidth pulsation form resonance; according to the method, active feedback intervention of a database kernel layer on competitive behaviors is realized, the problem of centralized preemption of lock resources caused by frequency resonance can be effectively broken through especially through scheduling offset and a progressive exit mechanism, and the instantaneous congestion probability of the system and the transaction-level delay fluctuation amplitude are reduced on the premise of not influencing the transaction consistency.

The invention belongs to the technical field of computers, and particularly relates to a resource allocation method and device, equipment and a storage medium. The method comprises the following steps: S1, acquiring a target resource occupation task and real-time state data of a plurality of clusters; s2, calculating a cluster resource utilization rate based on the total resource configuration amount and the resource usage amount; calculating a residual resource proportion of the cluster based on the resource residual amount and the resource demand parameter; calculating a load sensitivity coefficient of the target resource occupation task based on the task priority, the resource demand parameter, an execution duration threshold and a task dependency identifier; and S3, screening the clusters according to preset screening conditions corresponding to different task guarantee levels, determining the clusters meeting the preset screening conditions as candidate clusters, monitoring in real time, triggering resource migration based on quantitative migration conditions, and forming a closed-loop process of data acquisition-index calculation-screening distribution-preemption coordination-monitoring migration.

The invention relates to the technical field of task scheduling, and discloses a scheduling control method and device for a multi-core real-time processing system and a computer storage medium, and the method comprises the steps: when a target process has a high-bandwidth operation request, executing a first bandwidth preemption operation based on a bandwidth lock mechanism, scheduling a target processor core corresponding to the target process to preempt a first shared resource required by the high-bandwidth operation request; the first shared resource comprises a CPU shared resource; based on a dynamic bandwidth preemption priority mechanism, executing a second bandwidth preemption operation according to the process priority and the task information of the target process so as to schedule a task execution unit of the target process to preempt a second shared resource; the second shared resources at least comprise global shared resources. Visibly, the scheduling reliability and accuracy between the processor cores and the main devices of the multi-core CPU in the multi-core real-time processing system can be improved, the problems of inner competition and outer competition are solved at the same time, and therefore the real-time performance of the system is improved.

The invention discloses a task preemption scheduling method, which relates to the field of task preemption scheduling, and comprises the following steps of: performing dynamic task priority analysis and queue topology reconstruction by acquiring task basic data to obtain a coupling task group set, performing resource satisfaction degree decision by acquiring node resource data to obtain a global resource satisfaction degree mark, and performing task preemption scheduling. Performing resource matching degree space-time prediction on the basis of the global resource satisfaction degree mark to obtain optimal preemption time, performing preemption decision on the optimal preemption time and the coupling task group set to obtain a total preemption loss value of the node, and performing resource fragment integration scheduling on the basis of the total preemption loss value of the node to obtain a final scheduling scheme. The node direct scheduling model is created based on the global resource satisfaction degree mark, the optimal node identifier is obtained, the resource fragments are integrated to construct the comprehensive fragment index of the multi-dimensional resources, the overall utilization rate of the resources is improved, the overall income of the system is improved, and the recovery cost and the time overhead of the system can be reduced.

The invention belongs to the technical field of cache prefetching optimization of a firmware layer of a computer storage system, and particularly provides a memory control method, a storage device, a medium and equipment.The instantaneous flow is effectively filtered and persistent hotspot data is accurately recognized through a three-layer structure of combining a first-stage hotspot filtering module with a window cache, a view queue and a protection queue; the second-stage continuous access filtering module dynamically judges a continuous access mode by means of mean-variance statistics of an annular buffer area, and accidental continuous triggering of prefetching is avoided; and the third-stage random access judgment module prohibits invalid prefetching and effectively distinguishes access modes. The pre-fetching bandwidth is limited through the token bucket algorithm, the global priority queue of the skip list is combined, it is ensured that the priority of the write task is higher than that of the pre-fetching task, pre-fetching and foreground I / O resources are isolated, delay jitter control over foreground key I / O can be achieved during full-load pre-fetching, and the problem that delay soaring is caused when pre-fetching preempts the resources is solved.

Embodiments of the present application provide a communication method, a device, a storage medium, and a computer program product. In the embodiments of the present application, a storage device can be integrally configured for a distributed cluster, and in a main node competition link of the distributed cluster, a node which successfully preempts an exclusive write permission of the storage device can become a main node in the distributed cluster. When the node which achieves successful preemption is used as the main node, a working task generated by the node can be written into the storage device, and is broadcasted to slave nodes in the distributed cluster, and the slave nodes do not need to perform non-volatile storage on the working task. In this way, the distributed cluster only needs to be configured with a time-division multiplexed storage device, without needing to configure persistent storage resources for each node in the distributed cluster any more.

The invention relates to an automatic driving simulation test system based on a local cluster. According to the system, a local structured file is used as a database, a plurality of common computers are built into a distributed computing cluster through an SSH protocol, and a Web front end is integrated to provide a user interface. The core executes a priority scheduling algorithm fused with a dynamic adjustment and preemption mechanism through an intelligent scheduling and control module, and drives a simulation software API (Application Program Interface) to realize full-process automation from task submission, scheduling, distributed parallel execution, state monitoring to result collection and report generation. The invention further relates to a corresponding method and device, a processor and a computer readable storage medium. According to the automatic driving simulation test system, method and device based on the local cluster, the processor and the storage medium, the problems that an existing simulation test cloud platform is high in cost and complex in deployment are effectively solved, and an efficient, low-cost and high-flexibility test solution is provided for automatic driving research and development.

The invention discloses a financial scene-oriented model reasoning method and device, equipment and a storage medium, and relates to the field of model reasoning. According to the extracted semantic information, the model structure and the system bandwidth resources, determining a delay estimated value for reasoning the current task; performing score evaluation on the current task based on the task priority information and the delay estimated value, and updating the task queue and the KV cache sharing degree of the queue in combination with the SLO information; based on the KV cache sharing degree and the queue task number, the task queues form task batches, and the task batches are sent into a GPU video memory for model calculation; and when the GPU computing resources are occupied by the low-priority tasks in the task batch, acquiring SLO information of not executing the high-priority tasks, and calculating the preemption state of executing the high-priority tasks through a profit-cost quantification mechanism. According to the method and the device, the task delay, the priority and the KV cache sharing degree are dynamically evaluated, and the benefit-cost quantitative preemption mechanism is combined, so that the method and the device have the advantages of improving the high-priority task response speed and the overall efficiency of the system.