1506 results about "Bottleneck" patented technology

The embodiment of the invention provides a chip platform resource dynamic scheduling method, which comprises the following steps of: deploying a resource monitoring probe to acquire a chip platform running state, constructing a resource use index database to record CPU utilization rate and memory bandwidth data, analyzing a time sequence characteristic and a change rule of the running state, establishing a load model, setting a resource threshold value based on the system bottleneck identification model, formulating a scheduling rule through a strategy generator, and creating a resource configuration scheme according to the resource threshold value and the scheduling rule; analyzing a system state through a system bottleneck identification model, selecting a corresponding optimization strategy according to a scheduling rule, constructing a feedback acquisition queue, evaluating a scheduling effect of an execution instruction by using a data analysis technology, inputting the scheduling effect into an evaluation model for verification, and updating the scheduling rule based on a verification result of the evaluation model; according to the invention, the resource management efficiency and performance of the system can be effectively improved, and more flexible cross-platform support is realized.

The invention provides a production scheduling calculation method of a production plan scheduling platform. The method comprises the steps of data collection, demand prediction intelligent analysis, productivity planning and bottleneck analysis, production scheduling plan making, real-time monitoring and dynamic adjustment, plan execution and feedback analysis and report generation and decision support. Through distributed scheduling and real-time dynamic adjustment, the production plan can respond to emergencies such as equipment failure and order change in production in time, the continuity and stability of the production process can be ensured, production stagnation or delay caused by the emergencies is avoided, meanwhile, the production scheduling plan can be rapidly adjusted through real-time data feedback, and the production efficiency is improved. Resource utilization is optimized, waste is reduced, production efficiency is improved, intelligent productivity analysis and bottleneck identification are utilized, an enterprise can comprehensively know the use condition of each production resource, and possible bottleneck problems are found and solved in advance.

The invention provides an intelligent factory production scheduling method and system based on digital twinning, and the method comprises the steps: firstly obtaining a real-time monitoring data set, including equipment operation states, material flow tracks and environment sensing data, of a physical workshop; generating a three-dimensional digital twinborn model containing an equipment dynamic operation topology, a material real-time distribution thermodynamic diagram and an environment state simulation layer, then performing operation state correlation analysis on the three-dimensional digital twinborn model, and generating a feature set of equipment operation efficiency, material flow bottleneck, environment interference and the like; generating a dynamic scheduling parameter set containing equipment load balancing, material priority distribution and environment adaptation compensation parameters, and finally calling a preset scheduling strategy optimization algorithm to perform multi-objective optimization processing on the dynamic scheduling parameter set to generate an equipment scheduling, material distribution and environment regulation and control instruction set. And synchronizing to a physical workshop to execute real-time scheduling operation, thereby realizing intelligent and efficient production scheduling.

The invention belongs to the technical field of multi-target global dynamic production scheduling, and discloses a manufacturing resource dynamic production scheduling data comprehensive processing method, which comprises the following steps: acquiring multi-source heterogeneous data, constructing a field mapping table, constructing an equipment fingerprint database in combination with historical equipment performance data, dynamically adjusting confidence coefficient weight, and further judging an equipment state. Based on an equipment state judgment result, integrating operator report data and multi-source heterogeneous data through a preset rule base trigger event, and generating an abnormal event flow with a context tag; based on the abnormal event type and the equipment state judgment result, a recommendation scheme is constructed through a production scheduling knowledge graph, and a production scheduling instruction set is generated; according to the production scheduling instruction set, generating a local optimization scheme and a global optimization strategy, and forming a final execution scheme; calculating a resource utilization rate and a bottleneck process index through the 3D virtual workshop model, and generating a difference analysis report; according to the method, global resource scheduling and closed-loop iteration are realized, and the production efficiency and the system robustness are improved.

The invention relates to the technical field of project management, in particular to an intelligent management method and system for a civil engineering project, and the method comprises the following steps: obtaining personnel distribution, equipment state and material inventory data, extracting a key task node and an execution state, recognizing a progress lagging task, screening a key path node, analyzing a time difference, and marking a progress trend. And identifying an equipment influence section, judging resource allocation imbalance, and generating a progress monitoring index. According to the invention, through multi-source data acquisition of personnel distribution, equipment state and material inventory, difference identification between a task state and a plan node is enhanced, progress lag early warning efficiency is improved, progress fluctuation dynamic classification and trend identification are realized, construction rhythm optimization is assisted, and through coincidence analysis of equipment operation and a task fluctuation area, the progress lag early warning efficiency is improved. The construction efficiency bottleneck is identified, the scheduling accuracy is improved, fine progress monitoring is realized through investment and period difference comparison, and the progress control capability and the collaboration of on-site resource allocation are comprehensively enhanced.

The invention discloses a game theory improved model-based cloud computing node storage calculation training resource competition scheduling method. The method comprises the following steps of S1, constructing a global resource scheduling system; s2, carrying out initial division on the cloud computing nodes in each logic partition; s3, executing a local resource competition game model; s4, the global resource scheduling system collects a local resource competition result of each logic partition and residual available resources in each logic partition; s5, executing the global collaborative game model; s6, dynamically updating the resource state information of the cloud computing node; s7, the global resource scheduling system performs adaptive adjustment on resource demand strategies in the local game model and the global cooperative game model; and S8, the global resource scheduling system stores and analyzes past resource competition results and task execution conditions, and each cloud computing node updates a resource demand strategy according to the data. According to the invention, the problems of resource bottleneck and overload in the prior art are solved, and the resource utilization efficiency and the scheduling flexibility are greatly improved.

The invention discloses a production process state monitoring scheduling optimization method based on real-time data acquisition, relates to the technical field of manufacturing process scheduling, and is used for solving the problem of insufficient real-time performance and stability of process scheduling. According to the method, a process monitoring mechanism based on real-time acquisition and closed-loop scheduling is constructed, a time sequence structured data frame is formed under a unified time reference, operation stability and quality offset characteristics are extracted, a data credible label and a current process state factor vector are generated, an optimized scheduling model is input, and task conflicts and resource bottlenecks are identified according to the data credible label and the current process state factor vector. According to the method, path compression and sequence adjustment are implemented in combination with scheduling priority mapping and a resource path diagram, scheduling deviation vectors are constructed through task response time delay and process blocking in operation, rules and parameters are triggered to be updated online and locally rearranged, and solidification is performed after verification in a prediction window, so that equipment idling and switching fragmentization in the production process are reduced, and the production efficiency is improved. And the real-time performance of the production process, the resource utilization rate and the system stability are improved.

The invention relates to the technical field of intelligent scheduling, in particular to an enterprise production real-time monitoring and intelligent scheduling system based on artificial intelligence, which comprises a multi-source heterogeneous data fusion unit, a priority resource coupling decision unit, a bottleneck prediction and tracing unit and a scheduling instruction generation unit, the multi-source heterogeneous data fusion unit collects multi-dimensional data such as equipment vibration, temperature, order delivery time and the like in real time and constructs a joint feature vector, and the priority resource coupling decision unit dynamically adjusts task priority and resource allocation through a dual-channel depth Q network to cope with order insertion tasks and equipment health degree fluctuation. The bottleneck prediction and tracing unit predicts production bottlenecks and traces root causes by using a process dependency graph, a multi-modal fusion model and a causal discovery algorithm, and supports preventive maintenance and dynamic scheduling, and the scheduling instruction generation unit synthesizes a preorder result to generate an adaptive scheduling instruction. And enterprise production equipment utilization rate and production efficiency are improved.

The invention relates to the technical field of concrete detection, in particular to a dynamic detection and analysis method and system for early performance of ultra-high performance concrete. According to the technical scheme, the method comprises the steps of establishing a raw material characteristic spectrum library based on multi-physics field coupling, constructing a dynamic detection array based on a time sequence, developing a multi-modal data fusion algorithm, establishing a nonlinear coupling effect decomposition model and implementing dynamic parameter inversion. According to the method, a technical chain of four-dimensional monitoring network-multi-physics field coupling analysis-cross-scale prediction-adaptive regulation and control is constructed, so that dynamic accurate detection and intelligent regulation and control of the early performance of the ultra-high performance concrete are realized, and the technical bottlenecks of a traditional method in the aspects of parameter coupling, cross-scale association and real-time regulation and control are solved; the early performance prediction precision and the maintenance efficiency are remarkably improved, the cracking risk is reduced, and reliable technical support is provided for application of the ultra-high performance concrete in major engineering.

The invention discloses a big data-based shipping center area coordination optimization method and system, and the method comprises the steps: dividing all-day working time into a plurality of decision periods, building a state transition model, maximizing the utilization rate of port resources through an iteration model under constraint conditions, obtaining a resource distribution strategy of each decision period, and obtaining a chain reaction chain. Identifying logistics bottlenecks, adding queuing plans, regarding ships, berths, storage yards and loading and unloading equipment as different agents, dynamically adjusting the queuing plans by using a multi-agent collaborative algorithm, obtaining a final collaborative scheduling method, setting a reward function, and updating a state transition model by using reinforcement learning. And continuous optimization of the scheduling method is realized. According to the method, efficient utilization of port resources is achieved through the multi-agent collaborative algorithm and the reinforcement learning technology, the port operation efficiency is improved, the logistics cost is reduced, the adaptability of a port to uncertainty is enhanced, and meanwhile good interpretability is achieved.

The invention discloses an artificial intelligence-based industrial production line dynamic scheduling method and system, and relates to the technical field of intelligent manufacturing and industrial intelligent scheduling, and the method comprises the steps: collecting real-time state data, carrying out the preprocessing, taking the maximization of productivity, the shortest delivery time and the lowest energy consumption as optimization objectives, constructing a multi-objective reinforcement learning model, and carrying out the optimization of the multi-objective reinforcement learning model; scheduling priority data is generated, and operation distribution of each process node is adjusted in combination with a current equipment load threshold value and production bottleneck node information; when an abnormal condition is detected, triggering a rescheduling mechanism according to scheduling priority data, and updating an operation sequence and a resource allocation result; and synchronously feeding back the updated job allocation result and execution effect to the multi-target reinforcement learning model, and carrying out iterative optimization on the multi-target reinforcement learning model through a priority experience playback mechanism to realize continuous optimization of a scheduling strategy. According to the method, through a mode of combining multi-target reinforcement learning and dynamic scheduling, the learning efficiency and the optimization effect are improved.

The invention relates to the technical field of automatic deployment of cloud computing platforms, in particular to a self-adaptive deployment method and system oriented to a credential heterogeneous environment. Aiming at three technical bottlenecks of low multi-CPU architecture adaptation efficiency, frequent software dependence conflicts and complex security baseline configuration in a localization process, a heterogeneous computing resource intelligent scheduling engine and a dynamic security policy generation mechanism are innovatively provided. The method comprises the following steps: constructing a heterogeneous resource portrait through a hardware feature automatic identification technology, and realizing component installation sequence optimization based on a DAG dependency relationship analysis algorithm and Kahn topological sorting; and creating an adaptive network security policy, and sensing the firewall state of the target system in real time through a probe. Compared with a traditional deployment mode, the method supports cross-architecture compatibility, solves the problem of dependency conflicts, improves deployment efficiency, and guarantees consistency and safety of the system.

The invention provides an edge node scheduling method and system based on a sinking distributed network. The method comprises the following steps: determining a scheduling strategy of an edge node based on a preset load balancing algorithm; receiving a service request from the terminal equipment; identifying a proper service node based on the data type of the service request; distributing the service node to process the service request; and adjusting the scheduling strategy based on the real-time monitored data feedback. Through the scheme of the embodiment of the invention, the problem of data consistency caused by insufficient data synchronization mechanism under the condition of high concurrent requests can be solved. And the problem of communication delay increase caused by lack of an adaptive routing protocol under the condition of dynamically changing network topology is solved. And the problem of computing resource waste caused by an unreasonable task allocation strategy under the condition of resource-limited edge nodes is solved. And under the condition of data intensive application, the problem of data processing bottleneck caused by an imperfect data caching mechanism is solved.

The invention provides a network equipment monitoring time sequence data processing method based on artificial intelligence, which comprises the following steps: acquiring network topology structure data and historical load data, constructing a topological graph model containing node importance indexes and connection strength parameters, and determining a key equipment candidate set; according to real-time load data and the topological graph model, calculating a dynamic importance score of each device in the current network structure, and generating a key device list; based on an equipment dependency relationship of the topological graph model, extracting a time sequence feature vector of collaborative operation between equipment from the time sequence data set; when the fluctuation amplitude of the time sequence feature vector exceeds a preset threshold value, a group cooperation abnormal state is judged, and a performance bottleneck early warning signal is generated; analyzing the high-frequency time sequence data of the key equipment, and identifying potential fault equipment; and updating the topological graph model according to the fault probability, adjusting node weight parameters, and regenerating a key equipment candidate set.

The embodiment of the invention provides a distributed training method and system, electronic equipment and a storage medium. The distributed training method is applied to a distributed training system, in a current round of iterative training, training of at least one stage of a target model is cooperatively executed by at least one corresponding device and at least one corresponding host, and the distributed training method comprises the following steps: for each stage of the target model, the device executes a forward calculation task, obtaining a forward calculation result; the equipment executes a reverse calculation task based on the forward calculation result to obtain gradient data; and in response to the completion of the reverse calculation task, the equipment triggers to execute the optimization update task and obtains the updated parameters for the use of the next round of iterative training. According to the distributed training method, waiting time can be reduced, computing resources of equipment can be saved, efficient overlapping of communication and computing can be realized, assembly line efficiency and parallelism can be effectively improved, and thus the communication bottleneck problem in large-scale distributed deep learning training can be relieved.

The invention relates to the technical field of production scheduling of manufacturing workshops, in particular to a distributed intelligent production scheduling decision-making system based on cloud computing. The system is composed of a cloud computing platform, a data acquisition module, a data processing module, an intelligent production scheduling engine and a distributed execution control module. The problems of performance bottleneck, data acquisition limitation, lack of flexibility and adaptability of a production scheduling scheme and the like in a traditional production scheduling system are solved, intelligent distribution and scheduling of production tasks can be realized through fusion application of advanced technologies such as cloud computing, big data and artificial intelligence, the production efficiency and the product quality are improved, and the production cost is reduced. And greater economic benefits and social benefits are created for enterprises.

The invention discloses a large model training fault recovery method based on distributed memory management, and the method comprises the steps: obtaining a model state at a current moment when judging that a check point storage service needs to be carried out in a current training round, serializing the obtained model state into a storable format, and generating check point data; then transmitting the check point data to a host side memory from a device side memory in the node; after the transmission is finished, continuing to train the task, and storing asynchronous check points at the same time; the asynchronous check point storage comprises data distributed memory backup and data persistence to a disk; and when the training task goes wrong, task recovery is carried out through check point data. When the fault is recovered, the backup data is preferentially read from the memory, and the delay caused by the I / O bottleneck of the disk can be obviously reduced, so that the training pause or the efficiency reduction caused by the operation of storing or loading the check point data is reduced, the efficient utilization of the GPU or other computing resources is kept, and the continuity and the stability of large-scale model training are improved.

The invention relates to the technical field of artificial intelligence, in particular to a cloud edge collaborative production manufacturing management system based on artificial intelligence, which comprises a task scheduling optimization module, a resource scheduling and load balancing module, a production efficiency evaluation module, a dynamic load adjustment module and a fault tracing analysis module. According to the invention, by accurately analyzing the task dependency relationship, optimizing the task scheduling strategy and improving the response sequence and delay control of task execution, the processing bottleneck problem caused by data transmission delay is avoided, the task execution period is flexibly adjusted, the high-frequency tasks in the production process are effectively managed, and the load unevenness caused by fluctuation is reduced; the method can accurately identify and regulate the production bottleneck, improve the adaptive capability of the production process, accurately identify the potential fault source, shorten the fault diagnosis time, reduce the complexity, more rapidly position the problem, improve the stability, efficiency and fault tolerance of the production system, and integrally enhance the self-optimization capability of the production system.

The invention discloses an enterprise digital management method based on data mining. The method comprises the following steps: acquiring information flow data and resource allocation data; analyzing according to the information flow data and the resource allocation data to obtain an initial network diagram; performing clustering analysis according to the initial network diagram, and performing path optimization to obtain a dynamic network diagram; performing feature extraction on the dynamic network diagram to obtain prediction analysis data; according to the dynamic network diagram and the prediction analysis data, network nodes are identified, and key nodes and high-risk nodes are obtained; in combination with the key nodes and the high-risk nodes, identifying a key propagation path to obtain the key propagation path; and based on the key nodes, the high-risk nodes and the key propagation paths, predicting key positions where resource bottlenecks and conflicts occur, and generating resource optimization suggestions. According to the method, the dynamic change of the implicit dependency relationship of the enterprise can be identified, high-risk nodes and key propagation paths are positioned, and resource configuration is optimized.

The invention discloses a high-availability testing method and system for task scheduling of a computing power host management platform, belongs to the technical field of high-availability testing of computers, and aims to solve the technical problems that in the prior art, high-concurrency scene testing coverage rate is insufficient, cross-domain scheduling, failover and self-healing scene verification are imperfect, and testing efficiency is high. According to the technical scheme, the method structurally comprises the steps of hybrid load testing, wherein the elastic capacity expansion and contraction capacity and the performance stability of a system are verified by simulating a long-time stable load and burst flow scene; cross-regional intelligent scheduling: tasks are dynamically scheduled based on cost, time delay and regional strategies, and optimal resource allocation and service continuity are ensured; resource fault migration recovery: simulating node-level or region-level faults, and verifying the automatic detection, migration and recovery capabilities of the system; monitoring and data collection: collecting affairs, resources and stability indexes in real time, and providing multi-dimensional data support for verification; and analysis and optimization: identifying a bottleneck based on test data to generate optimization suggestions, and continuously improving the high availability of the system.

The invention discloses an AI-driven industrial automatic production line intelligent scheduling method and system, and relates to the technical field of production line intelligent scheduling, and the method comprises the steps: carrying out the multi-source real-time collection based on an industrial automatic production line, obtaining a multi-source heterogeneous data set, carrying out the deep analysis and prediction, and recognizing a production dynamic prediction data set; positioning and identifying production equipment, determining production bottleneck position information, and drawing a demand change trend chart; carrying out production operation analysis, generating production bottleneck early warning information, carrying out control scheduling, and formulating a production scheduling strategy; and scheduling monitoring is carried out on the industrial automatic production line, continuous optimization response is carried out, and a real-time production scheduling optimization strategy is obtained. The technical problems of low scheduling efficiency and insufficient adaptability caused by lack of dynamic prediction capability of the industrial automatic production line in the prior art are solved, and the technical effects of realizing dynamic prediction of the industrial automatic production line and improving the scheduling efficiency and adaptability of the production line are achieved.

The invention relates to the technical field of aviation composite material hot press molding scheduling, solves the technical problem that an existing method does not fully consider spatial constraint and time window limitation, and particularly relates to an aviation composite material hot press molding single-machine batch scheduling method considering two-dimensional spatial characteristics. Two-dimensional space characteristics of an autoclave and to-be-processed operation, release time of the operation and time window constraints are comprehensively considered, and a single-machine batch scheduling model with the purpose of minimizing completion time is constructed. And on the basis, solving the scheduling optimization model by adopting an extended whale optimization algorithm to obtain an optimal scheduling scheme. According to the scheme, the processing sequence and the processing time of batches and the placing positions of operation in the batches can be effectively arranged, and the utilization rate of the autoclave is improved to the maximum extent on the premise of meeting space and time constraints, so that the production efficiency is remarkably improved.

The invention discloses a computing resource management scheduling method and system. The method comprises the following steps: carrying out resource demand prediction based on a machine learning algorithm; a main node main scheduler performs an allocation task by using the resource demand prediction result and the resource demand information provided by the execution node in combination with multiple computing resource allocation algorithms to realize multi-resource collaborative scheduling; the main scheduler realizes distributed scheduling of task, node and cluster levels based on a self-adaptive distributed multi-level scheduling architecture; and the main scheduler realizes monitoring data return through a feedback loop and carries out real-time distribution strategy adjustment. Aiming at the bottleneck problem in the resource management process, the invention provides a brand-new and dynamically adaptive computing resource management scheduling method and system, and the computing resource management scheduling performance is remarkably improved.

The invention discloses an intelligent file service system generation method based on a low-code development mode. The method comprises the following steps: constructing a file service core data model, defining file metadata fields, a storage structure and an association relationship, and forming a standardized data architecture. According to the method, the core data model, the visual page component library and the drag-and-drop workflow template are constructed, linkage binding of data-component-process is achieved in combination with a low-code engine, and the efficiency bottleneck of a traditional development mode is broken through; through a three-way real-time synchronization mechanism, real-time hot repair and dynamic branch judgment, page layout and flow rule adjustment can take effect immediately through visual configuration, and the problem of dynamic demand response lag is solved; through the intelligent resource scheduling logic compiled through environment adaptation, a differentiation scheme is generated for different deployment environments, the configuration difference influence under a private cloud and server mixed deployment scene is remarkably reduced, the deployment failure rate is greatly reduced, and the system online stability is improved.

The invention provides a defective part reprocessing decision-making method based on multi-target balance, which comprises the following steps: acquiring real-time defect rate data and part distribution information from production, analyzing a defect rate change trend and a part quantity fluctuation condition, and setting a standard defect threshold according to an industrial standard and historical data; if the defect rate exceeds a standard defect threshold value, identifying and fitting part performance parameters to obtain an average performance curve, calculating slope change and abrupt change amplitude of the curve, and determining time and influence range of a variation point; acquiring a factory processing load, combining the priority distribution sequence with the target part grouping scheme and the factory processing load to obtain a reprocessing flow configuration scheme, and identifying a balance point of resource consumption and a reprocessing coverage rate through simulation; and acquiring a real-time ratio of the reprocessing coverage rate and the resource consumption corresponding to the balance point, comparing the real-time ratio with a historical optimal ratio, and if the real-time ratio is not smaller than the historical optimal value, determining that a performance bottleneck risk exists, and analyzing a risk cause.

The invention discloses an enterprise production management method based on digital twinning and workflow simulation, and particularly relates to the technical field of industrial internet and intelligent manufacturing, and the method comprises the steps: constructing a physical production system digital twinning body and business process workflow model, and carrying out the dynamic association through a model fusion engine to form an integrated digital twinning model; real-time event driving is used for deducing and simulating a future production process, and bottleneck and conflict prediction is output; based on the prediction result, utilizing a multi-objective optimization engine to generate a plurality of alternative scheduling schemes; performing parallel simulation quantitative evaluation on the KPI of each scheme, selecting an optimal scheme, analyzing the optimal scheme into a control instruction, and issuing and executing the control instruction; and model self-correction and closed-loop optimization are realized through real-time monitoring and feedback. According to the method, the problem of service and physical state disjunction caused by digital twinning and workflow independence is solved, the whole-process closed-loop management from prediction to execution is realized, and the production self-adaption and intelligent level is improved.

The invention relates to the technical field of task scheduling, in particular to a container computing-oriented dynamic scheduling system and method, and the system comprises a resource state modeling module which obtains CPU, memory and network bandwidth parameters and analyzes multi-node resource characteristics based on resource use data of nodes in a container cluster. According to the invention, through dynamic resource modeling, analysis of probability change of multi-node resources and accurate prediction of resource fluctuation, a scheduling system can rapidly deal with dynamic resource adjustment, real-time resource comparison is combined, refined load optimization and reallocation are executed, global resource balance is enhanced, resource waste and node overload conditions are avoided, and the scheduling efficiency is improved. In the task scheduling process, matching is carried out based on the priority of tasks and resource multi-dimensional data, scheduling intelligence and high efficiency are ensured, loads are monitored in real time, task distribution is flexibly adjusted, the loads are prevented from being concentrated in a few nodes, task migration among the nodes is optimized, the bottleneck is eliminated, and the system stability and the overall performance are remarkably improved.

The invention belongs to the technical field of industrial control, and discloses a bottleneck prediction optimization method, device and equipment for a battery positive electrode material production line, and the method comprises the steps: collecting the spatial data and time series data of the production line in real time; synchronizing the spatial data and the time sequence data to a digital twinborn model, wherein the digital twinborn model is pre-constructed based on the production line; converting the spatial data and the time sequence data into spatio-temporal data; inputting the spatio-temporal data into a pre-trained spatio-temporal diagram convolutional network to obtain a prediction result of the bottleneck of the production line; when it is judged that a production line bottleneck exists according to the prediction result, based on the prediction result, an optimization strategy is generated according to a priority rule, the optimization strategy is operated on the digital twin model in a simulation mode, a simulation result is obtained, and if it is judged that the simulation result meets the expectation, the production line is controlled according to the optimization strategy. Efficient and stable operation of a production line can be ensured, and production efficiency and stability are improved.

The invention relates to the technical field of semiconductor manufacturing, in particular to an intelligent manufacturing process optimization method and system of an integrated industrial large model. Dynamic optimal combination solution of technological parameters such as photoetching dose, etching rate and film thickness is realized; an equipment operation instruction and an abnormal intervention strategy are generated in real time, and the accuracy and timeliness of operation are improved; and cross-process collaborative scheduling and bottleneck prediction are carried out, the whole production process is optimized, and finally the yield, efficiency and flexibility of semiconductor manufacturing are improved.

The invention relates to the technical field of project management, in particular to a large-model batch reasoning and data flow optimization system oriented to an MOE architecture. According to the method, a collaborative architecture of the request access module, the environment sensing module, the expert routing engine, the resource scheduling module and the dynamic optimization control module is set, the text length and the subject type are extracted by using the request access module, a basis is provided for accurate routing, and the GPU video memory, the I / O bandwidth and the request queue depth are acquired in real time through the environment sensing module, so that the real-time routing is realized. The system load is comprehensively monitored, meanwhile, an expert sub-network is activated through an expert routing engine according to request features, invalid calculation is avoided, weight loading and resource allocation are managed through a resource scheduling module, the I / O bottleneck is reduced, and finally an optimization strategy is intelligently triggered through a dynamic optimization control module based on routing conflict factors. The problems of large reasoning delay fluctuation and unbalanced resource utilization rate mentioned in the background technology are solved, and stable low-delay response and resource collaborative optimization in a high-concurrency scene is realized.