1474 results about "Calculation technique" patented technology

The invention discloses a major hazard source risk dynamic calculation system based on AI large model enabling, and relates to the technical field of risk calculation, and the system comprises a multi-modal data collection and preprocessing module which is used for achieving the synchronous collection of original data through the butt joint of an industrial protocol with a sensor; the heterogeneous data space-time fusion engine module is used for constructing a space-time diagram model and analyzing a nonlinear coupling relationship of multi-source data; the online incremental learning and model fine tuning module is used for triggering dynamic parameter adjustment based on the real-time data flow; a risk conduction probability calculation module; a multi-modal knowledge self-evolution module; and a self-adaptive threshold management and alarm module. According to the method, sliding window dynamic confidence interval calculation is combined with a time-varying confidence coefficient adjustment mechanism, self-adaptive fitting of a threshold value to real environment disturbance is achieved by embedding a periodic correction term, the bidirectional contradiction between detection sensitivity and false alarm suppression is effectively cracked, and a complete technical closed loop from dynamic sensing and cross-domain verification to rapid linkage is formed.

The invention discloses a cost optimization method for resource scheduling management of a cloud data center, and relates to the technical field of cloud computing, and the method comprises the following steps: S1, collecting and modeling a multi-dimensional resource state of the cloud data center, and generating a resource change trend based on a sliding time window and a prediction model; and S2, constructing a multi-target game scheduling model taking calculation, storage, bandwidth and energy consumption as participants, outputting a scheduling game solution in combination with task modal adaptability parameters, and forming task-resource optimal matching. According to the method, through multi-dimensional resource state collection, a sliding time window and an advanced prediction model, resource dynamic changes and future trends can be captured more accurately, more reliable input is provided for scheduling decisions, resource waste or performance bottlenecks caused by information lag are avoided, calculation, storage, bandwidth and energy consumption are modeled as multi-party game participants, and the game efficiency is improved. Nash equilibrium is solved in combination with task modal adaptability parameters, and an optimal scheduling scheme giving consideration to resource utilization rate, performance and cost can be found.

The invention relates to the technical field of high-pressure dredge pump simulation calculation, and discloses a construction and application method of a high-pressure efficient dredge pump simulation calculation model. A model is established based on dynamic fluid-solid coupling parameters (mud pulsating pressure, impeller dynamic stress and the like), soil multiphase rheological parameters (particle phase volume fraction, viscous resistance coefficient and the like) and pipeline transient resistance parameters (transient pressure drop gradient and the like), and factors such as rotating speed-power matching and the like are incorporated to calculate an operation envelope and a critical failure threshold. A multi-physics field coupling analysis system comprising parameter configuration, dynamic simulation (integrating a flow field and a structural vibration solver) and a safety evaluation module is developed, after actually measured data and self-defined parameters are input, dynamic flow velocity distribution, pressure pulsation characteristics and the like are obtained through fluid-solid bidirectional coupling calculation, and an optimal configuration scheme is generated. The system supports database management, multi-rheological model matching and three-dimensional visualization, the simulation precision and the engineering efficiency are improved, and support is provided for performance optimization and safety evaluation of the high-pressure working condition.

The invention relates to the technical field of electrical design and intelligent optimization calculation, and discloses a capacitor structure design optimization method and system and a storage medium. The method comprises the following steps: carrying out modeling processing on capacitor structure parameters, and dividing a parameter space to obtain an initial model set; obtaining a multi-physical field simulation model according to the geometric model and the material attributes; obtaining performance indexes such as capacitance value, heat distribution and stress based on the simulation model; setting a target function and constraint conditions, and operating an optimization algorithm to obtain an optimal solution set; performing feedback control processing on an optimization result, and analyzing a convergence path to obtain a structure parameter; according to the invention, the execution efficiency of capacitor structure design optimization is improved, and the consistency and stability of the optimization process and the performance prediction result are improved.

The invention relates to the technical field of PC (Personal Computer) terminal AI (Artificial Intelligence) computing, and discloses a method and a system for dynamically scheduling generative AI heterogeneous computing resources of a PC terminal. The system comprises a resource state acquisition module, a scheduling graph generation module, a resource fluctuation entropy analysis module and a scheduling decision engine module. The resource state acquisition module captures running state parameters of a GPU kernel, a CPU thread and a memory block in real time, and generates a resource state feature tensor through normalization processing; the scheduling atlas generation module analyzes and computes the node connection topology, extracts the correlation between the devices, and constructs a multi-dimensional scheduling atlas; the resource fluctuation entropy analysis module separates the load feature vectors, calculates the entropy of each calculation unit and generates a heterogeneous resource entropy matrix; and the scheduling decision engine module jointly analyzes the atlas and the matrix, identifies bottleneck node resource competition characteristics, generates a dynamic scheduling instruction set, adapts to generative AI task requirements, and ensures efficient and stable operation of the task.

The invention provides an intelligent computing power scheduling method in a distributed computing environment, and relates to the technical field of distributed computing, and the intelligent scheduling method specifically comprises the steps of collecting resource information data, constructing a resource and task model, evaluating a computing power demand, formulating an adjustment strategy, distributing and scheduling the computing power, monitoring and adjusting in real time, and feeding back and optimizing. Through comprehensive modeling and dynamic computing power evaluation of computing nodes and tasks and in combination with multiple intelligent scheduling strategies, accurate allocation of computing power resources can be realized, the problems of resource waste and node load imbalance are effectively avoided, the overall utilization rate of resources in a distributed computing environment is remarkably improved, and the computing power of the distributed computing environment is improved according to the characteristics and requirements of the tasks. The execution nodes are reasonably selected, the resource allocation is optimized, the task correlation is considered, the data transmission overhead is reduced, the task execution speed can be increased, the task completion time can be shortened, and the processing capacity and response speed of the system are improved.

The invention provides a multi-data fusion method and system based on edge computing. Belongs to the technical field of edge computing. The method comprises the following steps: receiving heterogeneous resource data from different sensors, equipment and sources through an edge task allocation unit; according to the encryption level corresponding to each task in the heterogeneous resource data, dividing each task data into data groups with identifiers according to a preset segmentation strategy; according to the operation condition of the edge task processor, performing distribution calculation on the data set according to a preset distribution strategy; after the distribution is completed, obtaining an identification sequence of the distribution mode; the distributed data set is calculated and processed through the edge task processor, and after data calculation and processing are completed, the result in the data set is sent back to the edge task distribution unit. Through intelligent scheduling of the edge task distribution unit, data from different sensors and equipment can be received and processed in real time, so that the data processing efficiency and response speed are improved.

The invention relates to the technical field of cloud side end computing, and discloses a cloud side end cooperative computing resource dynamic allocation method based on task-aware migration. The method comprises the following steps: acquiring real-time load characteristics and resource demand characteristics of calculation tasks in a cloud side end system, and dividing task priority queues in combination with task type identifiers; extracting historical execution records of tasks at cloud, edges and terminal nodes, constructing a task execution feature library, and generating a resource demand prediction model in combination with real-time load features; analyzing network transmission time delay characteristics of a cloud end and edge nodes, measuring real-time calculation capability fluctuation data of terminal equipment, and establishing an inter-node resource collaboration degree evaluation matrix; generating an initial migration strategy according to the prediction model and the evaluation matrix, monitoring actual resource occupancy deviation of the task, forming a final decision in combination with a node resource state correction strategy, triggering cross-node migration, and synchronously updating the priority queue and the evaluation matrix.

The invention discloses a distribution network low-voltage equipment state intelligent analysis method and system based on edge computing, and relates to the technical field of distribution network equipment state detection.The method comprises the steps that mesh network topology between edge computing nodes is constructed, and task allocation weights between the nodes are set; constructing a fault knowledge graph based on the multi-dimensional state features of the edge computing nodes; collaborative calculation is carried out based on the edge calculation nodes, calculation results of the edge calculation nodes are fused based on a weighted voting mechanism of a consistency algorithm, and a state evaluation result is obtained. According to the method, the multi-dimensional state features and the edge computing technology are combined, and intelligent analysis of the distribution network low-voltage equipment state is achieved. A mesh network topology is constructed based on electrical characteristics and environment characteristics, and efficient allocation of computing resources is realized through comprehensive evaluation of load complementation characteristics and state evaluation. The accuracy of fault propagation path identification is improved through double constraints of a feature association propagation chain and an equipment physical connection relationship.

The invention discloses a diastrophic stratum deformation analysis and calculation method for an active fault zone, and relates to the technical field of geological engineering calculation, and the method comprises the following steps: (1) constructing a three-field dynamic coupling model based on an energy density gradient, and unifying dimensions of physical parameters of a mechanical field, a seepage field and a temperature field as an energy density function; according to the diastrophic stratum deformation analysis and calculation method for the active fault zone, the problems of multi-physical field splitting and time scale solidification in a traditional method are solved by constructing an energy-mediated multi-field dynamic coupling model and a space-time decoupling algorithm. On the basis of an interaction path of energy density gradient tensor unified mechanics, seepage and temperature fields, instantaneous bidirectional feedback of three-field parameters is achieved, compared with a traditional model, the pore pressure prediction error is reduced, and the calculation efficiency is improved; and meanwhile, the joint simulation efficiency of the second-level seismic event and the ten-thousand-year tectonic motion is improved, the interface parameter continuity error is controlled, and the precision and stability of cross-scale modeling are improved.

The invention relates to the technical field of data linkage calculation, and provides a multistage document dependent freight trial calculation data linkage calculation method and system, and the method comprises the steps: expanding the attribute of a document node from a single business field to a composite vector containing a spatial-temporal feature through time information and geographic identification spatial-temporal dimension modeling; the dependency graph can capture the fine-grained dependency relationship of the freight service, and the space-time accuracy of data linkage is improved; when the basic information or the document information changes, the priority and the calculation path of the task queue are updated in real time through deep reinforcement learning, and calculation errors caused by change lag are avoided; through tensor synthesis operation, parallel subtask division, and a dual verification mechanism of a domain knowledge graph and a forced dependency edge, it is ensured that the whole process from dependency graph construction to calculation result output conforms to industry specifications, and the compliance and risk control ability of freight services are significantly improved.

The invention belongs to the technical field of distributed computing, and relates to a distributed computing resource dynamic scheduling method and system suitable for a new energy scene. According to the method, the priorities of the calculation tasks are obtained by carrying out priority division on the calculation tasks according to the task characteristics. And a distributed computing node resource pool is established, so that centralized management and flexible scheduling of computing resources can be conveniently realized. Performance indexes of all nodes in the resource pool are collected and monitored, and reasonable allocation and efficient utilization of resources can be achieved. And scheduling the calculation tasks based on the priorities of the calculation tasks and the performance indexes of the nodes, generating a scheduling strategy, and ensuring that the tasks with high priorities preferentially obtain the calculation resources so as to meet the requirements on the task execution sequence in a new energy scene. And the scheduling strategy is adaptively optimized according to the performance index of each node, so that the utilization rate of resources and the stability of the system are improved. According to the method, the priority of the calculation task can be dynamically adjusted according to the characteristics of a new energy scene, and flexible scheduling of calculation resources is realized.

The invention relates to the technical field of data analysis and calculation, in particular to a distributed scheduling method based on tasks, nodes and link resources, which is applied to a solution for task scheduling and management in a large-scale distributed system. The method is particularly suitable for cloud computing, big data processing and distributed storage systems which need to process mass data and high-concurrency requests. And dynamically adjusting resource allocation by using real-time monitoring data and a prediction algorithm in combination with task priorities and resource requirements. Compared with traditional static resource allocation, a dynamic resource allocation mechanism is adopted, resource allocation can be intelligently adjusted in real time according to task requirements and node resource conditions, and therefore the resource utilization rate and the task execution efficiency are improved.

The invention provides a numerical calculation method for double-explosion-source underwater explosion impact cavitation loads, and belongs to the technical field of computer data processing-based explosion impact wave calculation, and the method comprises the following steps: solving a propagation process of far-field underwater double-explosion-source explosion impact waves in a two-dimensional acoustic flow field by using a local discontinuous Galerkin (LDG) method; and the cavitation characteristics of the flow field under the action of the double-explosion-source underwater explosion shock waves are researched. A flow field dynamic pressure result calculated by the method is compared with a result calculated by an acoustic finite element method. The result shows that the flow field dynamic pressure result calculated by the method is well matched with the calculation result data of the acoustic finite element method, and the effectiveness and the accuracy of the method are verified.

The invention discloses an efficient task scheduling method based on distributed collaboration, and belongs to the technical field of distributed computing. Aiming at the problems of non-uniform resource allocation, insufficient task type difference adaptation, lack of multi-task dependence global optimization and the like existing in the existing scheduling strategy, the invention provides the following technical scheme: firstly, classifying tasks based on calculation characteristics and establishing a multi-dimensional resource index; secondly, performing dynamic weight evaluation on the heterogeneous resources by adopting an optimal worst weighting method (BWM), and constructing a task-resource matching matrix; a task dependency relationship is modeled through a directed acyclic graph (DAG), and a global priority sequence is generated in combination with a list scheduling algorithm; and finally, single-task optimal node matching is realized by adopting an elimination selection method, and cooperative scheduling is performed on multiple tasks by applying an arithmetic optimization algorithm (AOA). According to the method, accurate resource matching of a calculation-intensive task and a data-intensive task is realized through three technical dimensions of task feature perception, resource dynamic adaptation and dependency relationship collaborative optimization.

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.

The invention provides a parallel computing method and device, electronic equipment and a storage medium, and relates to the technical field of parallel computing, and the method comprises the steps: carrying out the first protocol operation of a target tensor based on each computing core in each stream processor cluster, and generating a data block containing the computing result of each computing core; writing a data block generated by each stream processor cluster into a shared cache; under the condition that each stream processor cluster completes the first protocol operation, reading a data block written by each stream processor cluster from the shared cache; and executing a second protocol operation on the data block read from the shared cache to generate a calculation result of the target tensor. According to the method and device provided by the invention, the parallel architecture and memory access characteristics of the artificial intelligence chip can be better matched, the unnecessary calculation delay and synchronization overhead of the cross-flow processor cluster in the parallel calculation process are reduced, the bandwidth utilization rate of the shared cache is improved, and the overall performance and calculation efficiency of parallel calculation are remarkably improved.

The invention discloses a visual setting calculation system and method, and relates to the technical field of power grid setting calculation, and the method comprises the following steps: obtaining operation data of a target system, extracting feature data, and constructing a knowledge graph of entities and relationships, the feature data including control data; constructing a system structure diagram, and generating a semantic enhancement diagram in combination with semantic information in the knowledge graph; inputting the semantic enhancement graph into a pre-trained graph neural network model, performing node feature aggregation and edge weight learning in combination with a prior rule of a knowledge graph, and outputting sensitivity scores of all control data nodes on performance indexes; according to the sensitivity score, constructing a boundary constraint condition, and generating a control data setting candidate solution based on a CSP solver; performing parameter response simulation verification on the control data setting candidate solution, and screening out an optimal control data setting candidate solution; according to the application, accurate sensitivity scoring can be realized by fusing the knowledge graph, the semantic enhancement graph and the graph neural network.

The invention discloses a power network parameter-based power flow system ill-conditioned evaluation method and system, and mainly relates to the technical field of data calculation of a power system. Comprising the following steps: generating a multi-dimensional evaluation index set comprising a network topology structure index, an admittance matrix characteristic index and branch impedance parameter statistics according to basic network parameters of a power system; through comprehensive analysis of the generated multi-dimensional evaluation index set, traditional Jacobian matrix singular value calculation is replaced, and the ill-conditioned condition number of power flow calculation of the power system is evaluated; according to the power grid type and the operation scene, dynamically adjusting the judgment threshold value of each index, and outputting the ill-conditioned risk level and the optimization suggestion. The method has the advantages that calculation independent of Jacobian matrix singular values is achieved, dynamic quantitative evaluation is carried out on bad operation conditions of the system, and meanwhile the load flow calculation stability evaluation precision of a large-scale power grid is remarkably improved.

The invention relates to a final-stage cache design method based on near memory calculation, and belongs to the technical field of near memory calculation. And compared with a near memory processor near a traditional main memory, data access and calculation can be carried out more quickly, and the calculation throughput rate can be improved. According to the near storage calculation, a near storage processor and a last-stage cache storage array are directly designed in an integrated mode, faster memory access is achieved through an internal bus interconnection mode, and the problem of a storage wall between the calculation speed and the memory access speed and the problem of a power consumption wall of data handling are relieved to a certain degree. The last-stage cache controller cooperates with the embedded processor, and can dynamically coordinate between a standard cache function and a calculation mode. Through a synchronization mechanism in a last-stage cache controller, a memory access request from a host CPU and a calculation task executed by an embedded processor in a last-stage cache are transparently balanced, so that efficient near-memory calculation under cache consistency is realized.

The invention discloses a chlorophyll calculation method based on a double-branch multi-scale fusion network model, which belongs to the technical field of chlorophyll calculation, is used for calculating a chlorophyll a concentration value, and comprises the following steps: taking the chlorophyll a concentration value of a measurement site as a label of a water body characteristic sequence data set, dividing the data set into a training set and a verification set, inputting the training set into the double-branch multi-scale fusion network model to obtain a trained double-branch multi-scale fusion network model; inputting the verification set into the double-branch multi-scale fusion network model to obtain a verified double-branch multi-scale fusion network model; and taking the neighborhood spectral feature tensor and the neighborhood inherent optical feature tensor as the input of the verified double-branch multi-scale fusion network model to obtain the chlorophyll a concentration value of the pixel at the corresponding position. According to the method, the calculation result of the chlorophyll a with relatively high precision is obtained through neural network processing, and the complex relationship among the concentration change of the chlorophyll a in the water body, the spectrum and other optical parameters can be captured.

The invention provides a task scheduling method and device, and relates to the technical field of computers, in particular to the technical field of cloud computing. According to the specific implementation scheme, in response to the fact that a task is awakened, it is determined that the task which is awakened is executed in all CPUs of a physical machine; determining a queue corresponding to the task identifier in the virtual scheduling queue and the physical scheduling queue of the determined CPU, and storing the awakened task in the determined queue; determining a target queue from the virtual scheduling queue and the physical scheduling queue of the target CPU in response to an empty queue appearing in the local scheduling queues of the CPUs of the physical machine; and obtaining a to-be-executed task from the target queue, and calling the target CPU to execute the to-be-executed task. By applying the task scheduling scheme provided by the embodiment of the invention, scheduling of various tasks can be realized.

The invention relates to the technical field of accurate calculation of mining quantity, in particular to an intelligent accurate calculation system of mining quantity based on a three-dimensional geologic model, which comprises a multi-source heterogeneous data acquisition module used for acquiring and preprocessing various data such as geologic structure and mineral distribution; the dynamic three-dimensional geologic modeling module constructs a digital geologic model; the precise reserve calculation module calculates exploitable reserves; the intelligent mining planning module is used for generating a mining path optimal solution set by adopting a hybrid taboo genetic optimization algorithm and evaluating the mining efficiency and the safety risk; the resource collaborative scheduling module is used for establishing a multi-objective optimization model of the equipment-personnel-transportation system by using a space-time constraint particle swarm optimization algorithm; and the visual decision support module supports real-time simulation and adjustment of the mining scheme through a WebGL three-dimensional visual engine and an AR interactive interface. According to the intelligent mining method, accurate calculation of the mining quantity and intelligent mining are achieved, the mining efficiency is improved, safety is guaranteed, resource allocation is optimized, and economic benefits are improved.

The invention discloses a quantum approximate optimization method and device based on easy Hamiltonian, and belongs to the technical field of quantum computation.The quantum approximate optimization method comprises the steps that a binary constraint optimization problem in intelligent decision making and resource configuration is converted into a linear constraint equation; preparing the initial state of the quantum circuit into an initial state corresponding to the special solution of the linear constraint equation; a variable component sub-circuit composed of a target Hamiltonian and an easy-to-drive Hamiltonian is designed, so that quantum state evolution is strictly limited in a feasible sub-space; a final quantum state is simulated and measured through Hamiltonian to obtain a candidate solution, and parameters are iteratively adjusted by using a classical optimization algorithm to approach an optimal solution. According to the method, convergence is accelerated through special solution initialization, the number of iterations is reduced, the evolutionary process is guaranteed to always meet constraint conditions through peaceability, and the solving efficiency and the success rate are remarkably improved.

The invention provides a distributed reasoning task allocation method for an edge computing large model, which belongs to the technical field of mobile edge computing and comprises the following steps of: considering a large language model distributed deployment scene, constructing a workflow for cooperatively finishing a reasoning task by a plurality of edge servers, and adopting accurate secondary reconstruction by considering the generation type characteristic of the reasoning task; the uncertainty of reasoning task resource occupancy is solved, mathematical model expression is constructed for the edge reasoning task allocation problem in the scene, the aim is to maximize the service provider income, the constructed mathematical model is reconstructed into a combination selection problem, an approximate solution with theoretical guarantee is given by adopting a dual method, and the problem of edge reasoning task allocation is solved. And the reasoning task distribution decision of the service provider is optimized, so that the income of the service provider is maximized.

The embodiment of the invention discloses a dynamic computing power scheduling method and device based on deep learning and a medium, belongs to the technical field of cloud computing, and solves the problem that cloud computing resources are seriously wasted due to the fact that a cloud computing scheduling method in the prior art is prone to unbalanced resource allocation. Comprising the following steps: preprocessing acquired cloud computing data to generate a structured feature vector; wherein the cloud computing data at least comprises cloud node state data, computing task attributes and environment data; through a multi-head attention mechanism, task resource space-time correlation features corresponding to the structured feature vectors are extracted; the task resource space-time correlation features are input into a weight dynamic prediction module and a resource matching degree scoring module at the same time so as to carry out parallel processing of weight prediction and matching degree scoring calculation; and solving a multi-objective optimization function corresponding to computing power scheduling based on a parallel processing result, and generating a computing power scheduling instruction based on a solving result.

The invention relates to the technical field of quantum computing, and provides a polariton quantum computing simulation method and related device.The method comprises the steps that parameterized quantum circuits are generated based on micromodels used for describing polaritons for all to-be-computed excitation states respectively, and the parameterized quantum circuits are generated based on initial Hamiltonian of polaritons in a ground state; the target Hamiltonian amount of the polariton in the current excitation state is obtained, then the value of each parameter is adjusted at least once, and the target energy of the corresponding excitation state is obtained. In the primary adjustment process, parameter updating is carried out on the quantum circuit, a reference quantum state of the corresponding excitation state is obtained by utilizing the updated quantum circuit, reference energy of the corresponding excitation state is obtained based on the reference quantum state and in combination with the target Hamiltonian, and then when a set convergence condition is met, the target Hamiltonian is obtained. And taking the reference energy as target energy of the corresponding excited state. The polaritons are simulated through the quantum computer, and the simulation efficiency is improved.

The invention relates to the field of intelligent early warning of guaranteed power supply, and provides an abnormal early warning method and system for a guaranteed power supply system, and the method comprises the steps: obtaining the topological structure, real-time operation data and equipment state data of each node in a guaranteed power supply network, employing the conditional entropy and instantaneous causal entropy calculation technology, precisely evaluating the causal intensity between variables, and achieving the early warning of the abnormal state of the guaranteed power supply system. And a causal adjacency matrix dynamically changing along with time is constructed to reflect a complex causal relationship in the network. Then, in combination with an equipment topological structure and dynamic causal information, multi-scale time sequence features are extracted through feature updating and expansion causal convolution, and a comprehensive feature vector is formed; after the vector is input into the Bayesian neural network, a plurality of prediction value samples are generated through sampling weight distribution, a mean value is used as a prediction result, and the prediction uncertainty is measured through variance. Accurate prediction and timely early warning of potential anomalies of the power supply system are realized, and the stability and safety of the system are effectively improved.

The invention discloses a reinforcement learning calculation simulation method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence computing, and the method comprises the steps: inputting the determined current model parameter configuration, current hardware configuration and current working load into a target simulation system to obtain a plurality of parallel grouping combinations, determining a target simulation system according to the current hardware configuration, determining an effective parallel packet combination from the plurality of parallel packet combinations based on a preset Monte Carlo method, inputting the effective parallel packet combination into a simulator of a preset neural network model, and performing delay time calculation according to the effective parallel packet combination through the simulator to obtain a delay time sequence; and the combination corresponding to the shortest delay time is used as a target parallel grouping combination, so that the technical problems of mismatching of simulation scenes, insufficient precision and lack of effective support for heterogeneous clusters are solved, reliable performance prediction and optimal parallel strategy suggestions are provided through high-precision performance modeling and automatic exploration, and the method is suitable for large-scale popularization and application. Therefore, the resource consumption of large-scale GRPO training is reduced.

To accelerate search speeds for approximate nearest neighbor searches of vector databases, compute-in-memory techniques using NAND memory structures are introduced. For each element of the database, a kernel of its M nearest neighbors is determined. For each vector of the database, both the vector and its kernel are programmed in the arrays of a NAND memory based accelerator card, so that the vectors will be written into the memory arrays both as themselves and also in kernels of vectors for which they are a nearest neighbor. Metadata, associating the locations of the kernel members with the correspond vector is also stored in the memory system. After determining the input's nearest neighbor at one level of search, the metadata is then used to locate that nearest neighbor's nearest neighbors and their distances to the input vector are then computed in parallel in a compute-in-memory vector-vector dot product multiplication.