354 results about "Global scheduling" patented technology

The invention provides an emergency command converged communication system resource dynamic scheduling method based on artificial intelligence. The method comprises the following steps: S1, acquiring a plurality of sensors and log data of an emergency command system; constructing a dynamic graph neural network optimization communication topology, and selecting an optimal communication path; combining resource prediction and the optimized communication topology to formulate a global optimal resource scheduling strategy, and generating an optimal resource allocation scheme for each task; the resource pool comprises available computing resources and communication bandwidth; dynamically adjusting resource allocation and a communication path in a task execution process, and generating task execution feedback of a corresponding task; the system state is monitored in real time, and resource prediction, topology adjustment and scheduling strategies are optimized based on execution error feedback of task execution feedback. According to the method, efficient and intelligent dynamic resource scheduling is realized in an emergency command converged communication system through resource prediction of self-supervised learning, self-adaptive topological optimization of a variable topological graph neural network and global scheduling of reinforcement learning driving.

The invention discloses a flexible load multi-target collaborative scheduling system and method, and relates to the technical field of collaborative optimization of power systems. The method is used for solving the problem of lack of accurate prediction and multi-target coordination of agricultural electricity and water utilization regulation and control. Firstly, based on meteorological data, soil moisture content and crop growth characteristics, an irrigation demand prediction model is constructed, irrigation water demand is predicted, and a water pump load power baseline is generated; then, a dynamic baseline constraint condition is generated in combination with historical behavior data and water pump start-stop logic; establishing a power grid side objective function, a user side objective function and a water affair side objective function, and introducing an underground water and carbon emission punishment mechanism; thirdly, dividing the power distribution network into sub-regions, adopting an alternating direction multiplier method to solve a region regulation and control strategy in parallel, coordinating water resource distribution conflicts through virtual interactive variables, and outputting a global scheduling instruction; and finally, collecting real-time response data and correcting the prediction model on line to realize closed-loop adaptive optimization.

The invention discloses a carbon neutralization target-oriented electricity-carbon-green certificate multi-target scheduling optimization method and system, and relates to the technical field of carbon neutralization control, and the method comprises the following steps: obtaining first data of a target energy storage unit, constructing a constraint generation network, and dynamically updating an energy storage operation constraint domain based on a dynamic time warping algorithm; based on the updated operation constraint domain, establishing an electricity-carbon coupling model, extracting a nonlinear mapping relation between an electric energy output behavior and carbon emission intensity, and outputting a dynamic incidence matrix through a multi-dimensional time sequence predictor; constructing a carbon flow distribution topological graph according to the dynamic incidence matrix, and obtaining an optimal carbon flow path through a Dijkstra algorithm; based on the operation constraint domain and the optimal carbon flow path, establishing a multi-subject collaborative decision-making mechanism, solving a multi-target unit equilibrium solution by adopting a distributed alternating direction multiplier method, and generating a global scheduling instruction set; according to the method, a carbon flow topology and dynamic constraint domain cooperation mechanism is constructed, so that efficient solving of a multi-main-body carbon response optimization scheduling problem is realized.

The invention discloses a source load storage dynamic strategy verification method based on double-layer reinforcement learning, and the method comprises the following steps: S1, collecting the operation data of a source load storage system, and constructing a standardized operation data set; s2, constructing a double-layer reinforcement learning model, generating a global scheduling strategy by an upper-layer strategy network, and outputting an action decision strategy by a lower-layer strategy network; s3, performing joint training on the double-layer reinforcement learning model by adopting a strategy gradient optimization method, and outputting a scheduling strategy; s4, introducing an integral gradient method to analyze a scheduling strategy, and constructing a key scheduling state node set; s5, optimizing the generation logic of the global scheduling strategy to obtain an optimized double-layer reinforcement learning model; s6, constructing a plurality of source-load-storage system operation scenes to form a typical operation scene set; and S7, deploying the optimized double-layer reinforcement learning model in the typical operation scene set, and outputting a strategy verification result. According to the invention, through combination of double-layer reinforcement learning and an integral gradient method, source-load-storage dynamic strategy verification is realized.

The invention relates to the technical field of vehicle scheduling, in particular to a multi-terminal vehicle scheduling system based on reinforcement learning. The system comprises a heterogeneous data fusion module, a resource allocation module, a hierarchical reinforcement learning module, an optimization feedback module and a man-machine cooperative control module. Data of vehicle operation, operation tasks, environment monitoring and the like are collected and uniformly packaged into a structured data set, an upper-layer manager model generates a global scheduling instruction set based on a PPO algorithm, and a lower-layer worker model outputs a specific vehicle control instruction based on multi-agent reinforcement learning. The system also evaluates and optimizes a historical scheduling execution effect through an NSGA-II algorithm, selects a Pareto optimal solution set, and realizes continuous iteration of a scheduling strategy. The man-machine cooperative control module supports visual display and manual intervention operation, and improves the adaptability and controllability of the system in a complex operation scene.

The invention discloses an agent collaborative optimization data center management system based on knowledge graph driving, and relates to the technical field of data center management. The system specifically comprises the following modules: a modeling and entity management module, a cross-regional global scheduling optimization module, an agent game negotiation optimization module, an agent trust management module, a game negotiation conflict identification module, a reasoning evidence management module and a cross-regional consistency verification module. By establishing a complete knowledge graph model, systematic modeling of resource attributes, constraints, historical decisions and strategy preferences is realized, a unified data basis is provided for negotiation among multiple agents, the problem of a suboptimal solution caused by information asymmetry is effectively solved, a hierarchical reasoning mechanism is adopted, and the probability of resource disruption is reduced. In combination with global-region-node three-level reasoning and multi-round game negotiation, recursive optimization from global to local is realized, the reasoning complexity is effectively reduced, and the negotiation efficiency is improved.

One or more embodiments of the invention provide a load-aware scheduling method of an inference system and the inference system. The method is applied to a global scheduler in the inference system. The inference system further comprises an inference engine. The inference engine comprises at least one computing instance deployed on each computing node in a computing cluster; computing resources of the computing examples comprise GPUs carried on the computing nodes where the computing examples are located; the global scheduler maintains dynamically updated GPU load information of each calculation instance; the method comprises the steps of obtaining a to-be-executed target reasoning request; based on the maintained GPU load information of each calculation instance, determining a target calculation instance of which the GPU load meets a preset condition; and sending the target reasoning request to the target calculation instance, so that the target calculation instance executes the target reasoning request.

The invention discloses a vehicle-mounted cross-protocol data time sequence cooperative processing method, which comprises the following steps: establishing a unified time reference, and providing an accurate space-time coordinate system for heterogeneous network nodes by electing a global master clock and constructing a clock relation model and a periodic correction mechanism; systematic cognition and management of communication resources are realized, and a verified protocol capability table is generated by automatically identifying protocol types and variants; constructing an automatic mapping system from demand to execution, performing formalized packaging on a target scene by introducing a unified intermediate model, generating a cross-protocol mapping rule based on the model and a protocol capability table, and then forming a global scheduling table in combination with a unified time reference; the cooperative communication of all nodes is finally realized by using the global scheduling table, and the end-to-end delay is constrained within the preset range, so that the problem that the end-to-end time sequence in the vehicle-mounted heterogeneous network is uncontrollable is solved, and the controllability and reliability of the time sequence of the vehicle-mounted network are improved.

One or more embodiments of the invention provide a load-aware scheduling method of an inference system and the inference system. The method is applied to the inference system comprising a global scheduler and an inference engine. The inference engine comprises a Prefill engine and a Decode engine; the Prefill engine and the Decode engine respectively comprise at least one calculation instance; computing resources of the computing instances comprise GPUs carried on the computing nodes where the computing instances are located; the method comprises the following steps: a global scheduler obtains a target reasoning request to be executed, determines a first calculation instance in a Prefill engine of which the GPU load meets a first preset condition based on maintained GPU load information, and sends the target reasoning request to the first calculation instance to execute reasoning calculation in a Prefill stage; and a first local scheduler corresponding to the first calculation instance in the Prefill engine determines a second calculation instance in the Decode engine with the GPU load meeting a second preset condition based on the maintained GPU load information, and sends an inference calculation result of the Prefill stage to the second calculation instance to execute inference calculation of the Decode stage.

The invention provides a large model reasoning acceleration method, device and equipment, a computing service device is configured with a plurality of computing nodes, stores a prefix index structure and is used for indicating a mapping relation between a token sequence prefix and the computing nodes storing cache computing results of the token sequence prefix; the method comprises the following steps: executing a global scheduling mechanism, querying the prefix index structure based on a token sequence of a reasoning request to perform prefix matching so as to determine one or more candidate computing nodes, and selecting a target computing node from the candidate computing nodes according to a real-time load state; executing a local scheduling mechanism, and allocating an execution priority for the reasoning request to perform scheduling processing according to the prefix matching degree of the reasoning request on the target computing node; and loading a cache calculation result corresponding to the matched prefix, and only calling a large model for a non-prefix part of the reasoning request to execute reasoning calculation.

The invention relates to the technical field of power distribution network operation optimization, and discloses an intelligent power distribution network distributed power scheduling system based on edge computing, which comprises a data acquisition and preprocessing module, an edge computing and local scheduling module, an SDN network control module, an application layer global scheduling module and a scheduling execution and control module, according to the invention, by introducing the edge computing node, local preprocessing of data and generation of a preliminary scheduling strategy are realized, delay of data transmission to the central server is significantly reduced, and the system can dynamically adjust the scheduling priority in combination with the improved particle swarm optimization algorithm, so that the scheduling efficiency is improved. Three core indexes of renewable energy utilization rate, power distribution network loss and voltage stability are comprehensively considered, and a more accurate and efficient local scheduling strategy is generated. And meanwhile, the SDN network control module ensures that high-priority data can still be efficiently transmitted when the network is congested by calculating a link priority coefficient, so that the real-time performance and the reliability of the system are further improved.

The invention discloses a Beidou satellite positioning platform remote monitoring and scheduling system and method. The method comprises the following steps: S1, dividing a scheduling task of Beidou satellite positioning data into a first layer, a second layer and a third layer; s2, the first layer generates a preliminary scheduling scheme; s3, a second layer utilizes a graph neural network to process dependency and resource conflicts among tasks, and generates an optimized scheduling scheme in combination with a near-end strategy optimization algorithm based on strategy gradient; s4, the third layer adopts a global graph neural network model to process the global dependency relationship of the tasks, and generates an execution instruction of a global scheduling scheme in combination with a soft actor commentator algorithm; s5, modeling task information into a directed graph, propagating messages layer by layer through a graph neural network, and generating an optimal scheduling scheme; s6, setting a feedback mechanism between the layers; and S7, carrying out graphical display on the finally generated optimal scheduling scheme through a visual tool. According to the invention, accurate dynamic scheduling of Beidou satellite data is realized through hierarchical scheduling and graph neural network optimization.

The invention relates to the technical field of power distribution network dispatching. By providing a power grid load dynamic prediction and optimal scheduling method, device, equipment and medium, the method comprises the following steps: performing multi-source heterogeneous fusion processing on meteorological parameters, historical load curves and new energy output data to generate a dynamic load prediction map; constructing a dynamic network model, and performing power flow distribution simulation processing based on the dynamic network model to obtain a stability margin calculation result and a preset safety threshold boundary; performing stage decomposition processing on the global scheduling target to generate a progressive scheduling stage sequence; performing matching processing on the response characteristics of the power generation equipment to generate a self-adaptive progressive scheduling instruction sequence; and executing an adaptive progressive scheduling instruction sequence, performing feedback processing on the real-time state of the power grid, and generating a dynamic adjustment instruction so as to realize multi-dimensional data association modeling, stability margin quantitative analysis and dynamic instruction optimization, thereby improving the load prediction precision and reducing fault diffusion and transient oscillation.

The invention relates to the technical field of new energy, and particularly discloses a new energy short-term intelligent optimization scheduling method under high-proportion new energy grid connection, and the method comprises the steps: data collection and preprocessing, multi-source data fusion prediction, hierarchical supply and demand balance, distributed collaborative optimization, and real-time regulation and feedback. According to the scheme, a multi-time-scale hierarchical supply and demand balance mechanism is adopted, fine management of different scheduling targets is achieved, the supply and demand relation is coordinated in combination with a dynamic weight adjustment mechanism, key load requirements are met preferentially, economy and stability are considered, and meanwhile the real-time performance and adaptability of scheduling decisions are improved; a distributed collaborative optimization structure is designed, the adaptability and robustness of a scheduling system to operation environment changes are effectively enhanced, a global scheduling problem is decomposed into a plurality of regional sub-problems, parallel solution and asynchronous update are achieved, the scheduling efficiency and the computing resource utilization rate are improved, and the requirements for real-time performance and high concurrency of large-scale new energy grid-connected scheduling are met.

The invention belongs to the technical field of power system optimization scheduling, and provides a source network load storage intelligent collaborative optimization method, which comprises the following steps of: deploying sensors at four ends of a source network load storage respectively, collecting in real time by utilizing a cloud data center, enabling data of the four ends to be consistent in time sequence through a PTP protocol, constructing a topological graph according to parameters and data, and establishing a source network load storage intelligent collaborative optimization system. Selecting a model in a digital twinning environment for simulation; dividing independent agents at four ends of a source network load storage, setting observation data, an execution space and excitation feedback, forming an excitation item by economy, stability and environmental protection, interactively circulating actual data, a prediction instruction and an excitation value, recording into a sequence, inputting the sequence into a strategy network, and calculating and outputting logarithmic probability gradient to update the parameters of the strategy network; and the intelligent agent completes interactive circulation according to the strategy network, generates a local scheduling instruction, aggregates the instruction to perform weighted calculation, generates a global scheduling scheme, issues the global scheduling scheme to execution equipment, updates parameters by using an average deviation calculated by a deviation vector, resolves the global scheduling scheme and issues the global scheduling scheme to form a closed-loop mechanism.

The invention discloses a parallel task scheduling method based on node dynamic load and task relevance measurement. The parallel task scheduling method comprises the following steps: constructing a global scheduler for computing resource virtualization, sample allocation based on reinforcement learning and a local scheduler based on hybrid drive model scheduling on a server; in the preprocessing stage, computing resources are counted, and computing resource virtualization, GPU virtualization multiplexing technology, GPU acceleration and the like are carried out; a global scheduling stage oriented to test task parallel optimization: performing sample scheduling feasibility analysis containing fuzzy time, computational node selection based on an EXP3 algorithm, and sample scheduling oriented to node load; and a multilevel model parallelization stage based on domain knowledge and data hybrid driving, wherein a model cohesion coupling degree measurement mechanism, a model coupling mode knowledge base and a multilevel model parallelization method are included. According to the method, the global scheduler and the local scheduler are comprehensively utilized, the parallelism degree between samples and parallelism between models in the samples are achieved, and the method has wide application value and use prospects in all fields.

The invention discloses a multi-edge device collaborative reasoning method and system for a hybrid expert large model, and the method comprises the steps: collecting and analyzing system data, and adjusting the expert layout according to the system data; in the first stage, the number of experts required by each layer of a server is dynamically determined by balancing activation diversity and memory resource limitation; in the second stage, according to the expert number and the activation mode of each layer obtained in the first stage, low-time-delay reasoning is achieved by minimizing the calling times of remote experts, and the experts are distributed to all the servers. The system includes a global scheduler and an edge server participating in the system. By using the method, edge multi-machine joint reasoning is achieved, the deployment method is optimized, online adjustment deployment can be performed according to data changes, and the reasoning speed is increased compared with other deployment technologies. The method can be widely applied to the technical field of distributed machine learning.

The invention discloses a supply chain performance path optimization and collaborative decision-making method based on an AI algorithm, and particularly relates to the field of supply chain performance path optimization, and the method comprises the steps: firstly obtaining order goods information and warehouse resource data, and judging whether a single warehouse can independently perform a performance or not through matching; if yes, constructing a comprehensive evaluation function including total transportation time, time redundancy, path complexity and resource utilization rate, and generating an optimal path from the warehouse to the delivery place; if multi-warehouse collaboration is needed, dynamically selecting a centralized point warehouse, generating a comprehensive early warning score and selecting an optimal centralized point by comprehensively calculating the cargo transfer cost and the historical performance risk value of each warehouse, and finally planning a transfer path from a supply warehouse to the centralized point and a performance path from the centralized point to a terminal to form a full-link collaboration scheme; according to the method, a traditional single-warehouse decision mode is broken through, global resource scheduling is realized through a double-path optimization mechanism, the transportation delay risk is remarkably reduced, and the supply chain operation efficiency is improved.

The invention provides an intelligent scheduling control method for an electric energy meter verification assembly line, and belongs to the technical field of electric energy meters. S1, features extracted from an original data set are expressed as a feature matrix, and a multi-level collaborative algorithm is designed; s2, the global scheduling layer is responsible for distribution of overall resources and macroscopic scheduling decision making, the input is a feature matrix, and the output is a task priority matrix; s3, the local scheduling layer is responsible for distribution of specific tasks and management of execution details, the input is a task priority matrix, and the output is a specific task distribution matrix; and S4, the equipment control layer is responsible for operation and state monitoring of verification equipment, inputting the specific task distribution matrix, outputting the specific task distribution matrix and the equipment control instruction set, and dynamically adjusting the specific task distribution matrix and the equipment control instruction set according to feedback. According to the invention, the Internet of Things and the big data technology are fully utilized, and the intelligent algorithm is combined, so that efficient, intelligent and automatic control of the electric energy meter verification assembly line is realized.

The invention discloses a full-equipment energy consumption optimization management system in a building, and relates to the technical field of intelligent control, and the system is provided with a multi-modal data collection module which collects building environment parameters and equipment operation data through a distributed sensor network, and generates a multi-modal sensing matrix; an environment state prediction module is set to construct a dynamic knowledge graph based on a multi-modal sensing matrix, a graph attention network is used, an energy flow matrix of equipment topology is output, micro-environment state prediction data is generated through prediction of a finite element solver, a scheduling scheme solving module is set, a mixed integer programming model is constructed based on the micro-environment state prediction data, and a scheduling scheme is set. And solving the mixed integer programming model to obtain a scheduling scheme of the energy consumption equipment, setting a control instruction optimization module, and driving the edge execution equipment to operate based on the scheduling scheme. The global scheduling optimization of the whole building energy consumption equipment is realized, the energy consumption is effectively reduced, and the environmental comfort is improved.

The invention discloses a programmable logic controller programming system with an internet of things function, which relates to the technical field of industrial automation control and comprises a main control module based on a heterogeneous multi-core processor architecture and used for executing user program logic and coordinating operation of a communication module, a task distribution module, a dynamic loading module and a safety verification unit; according to the programmable logic controller (PLC) programming system of the internet of things, the real-time performance and the resource efficiency of an industrial control system are remarkably improved through heterogeneous multi-core collaboration, dynamic resource scheduling and multi-protocol compatibility technologies; on the basis of differential division of labor of the ARM and the FPGA, master core global scheduling and slave core high-real-time control are combined, and millisecond-level response of key tasks is guaranteed; compared with a traditional PLC, the system breaks through the limits of protocol closure, resource rigid distribution, shutdown dependence and the like, provides a flexible, stable and safe intelligent control core for the industrial Internet of Things, and promotes industrial automation to evolve towards high efficiency, networking and high availability.

Disclose are a task optimization method and apparatus, and a non-transitory computer-readable storage medium, the method may include: evaluating tasks to identify a target task to be optimized; adjusting a target task parameter corresponding to the target task; executing the target task parameter by virtual scheduling to perform global scheduling speculation; and determining the target task parameter in response to a speculation effect of the global scheduling speculation meeting an expectation, and executing the target task based on the target task parameter.

The invention belongs to the technical field of semiconductors, and discloses a multi-core GPU interconnection architecture and a self-adaptive cache allocation method thereof.The multi-core GPU interconnection architecture comprises a plurality of GPU clusters and a global cluster, the global cluster comprises a global router, a global scheduler, a directory, a memory and a plurality of GPU clusters, each GPU cluster is composed of a plurality of GPU cores, and the GPU cores are arranged in the global router. Each GPU core grain comprises a plurality of independent primary data caches, secondary data caches, a local router and a network interface, and the network interface judges whether a current request is processed by the GPU core grains in a GPU cluster or not according to whether a target address in a target node belongs to the GPU cluster or not, so that a high-level multi-core grain interconnection architecture supporting cross-cluster communication is formed. According to the method, performance is optimized by adapting to different access modes, private and shared cache modes are switched in real time, single data access delay is reduced, the overall system efficiency is improved, and the method is used for a multi-core-particle heterogeneous system with a plurality of GPUs.

The invention discloses an intelligent construction scheduling method based on the Internet of Things, and relates to the field of intelligent construction, and the method comprises the steps: obtaining call extension data of each floor of a construction hoist, the call extension data comprising landing door state data and call data; converting the landing door state data by using a binarization processing algorithm to obtain landing door state parameters of each floor; calling data are counted through a counter, calling state parameters of all floors are obtained, landing door state parameters and the calling state parameters serve as environment states, a building hoist scheduling problem is modeled as a Markov decision process, a scheduling strategy serves as action, energy consumption of the building hoist serves as a reward function, and a building hoist scheduling strategy is established. Reinforcement learning is carried out through a Q-learning algorithm, and a scheduling strategy is obtained; in order to solve the problem of low intelligent degree of construction hoist scheduling in the prior art, a global scheduling strategy and a local scheduling strategy are obtained by adopting a hierarchical reinforcement learning algorithm, and collaborative optimization is performed by adopting a genetic algorithm.

The invention discloses a manned transport vehicle scheduling method and device, and relates to the manned transport vehicle scheduling technology, and the method comprises the steps: obtaining the vehicle information and map data of a target mining area, and generating a global scheduling scheme through the vehicle information and map data, the global scheduling scheme comprises the step of carrying out vehicle scheduling according to the determined number of vehicles to be allocated at each loading point and the determined number of vehicles to be allocated at each heavy-load transportation path; according to the method, the manned transport vehicles in the target mining area are scheduled according to the global scheduling scheme, the mining area vehicles can be accurately scheduled, accurate and balanced scheduling of the manned transport vehicles is realized, the overall efficiency of mining area operation is improved, and the transportation cost is reduced.

The invention relates to the field of three-dimensional warehouse automatic sowing, in particular to a multi-layer three-dimensional warehouse automatic sowing path optimization scheduling method and system, and the method comprises the following steps: obtaining sowing association data and three-dimensional warehouse association data through a data interaction module, and constructing a multi-dimensional mapping table; calling a preset material characteristic and path constraint mapping library, screening paths conforming to constraints through an improved A algorithm, calculating path cost, and outputting an initial path; the real-time load state of the AGV is input, a regional task allocation plan is output through the global dispatching center, and an optimized task allocation table is output through local obstacle avoidance optimization in combination with multi-source sensing data; and calculating task priorities, allocating resources according to the priorities, processing path conflicts, and outputting a final task scheduling instruction. According to the method, the task priority is calculated through the multi-dimensional parameters, and the AGV load is balanced in combination with distributed global computing power scheduling, so that the problem of coexistence of AGV task overload and idle is effectively solved, and operation interruption caused by hardware faults is avoided.

The invention belongs to the field of power system operation and control, and particularly relates to an operation control method for a wind-light hydrogen storage multi-energy complementary system, which comprises the following steps: performing empirical mode decomposition on an original output sequence to obtain an IMF component, training parameters by combining a hidden Markov chain model and a forward-backward algorithm and solving an optimal state sequence, and fusing a predicted value by using Bayesian estimation to obtain an optimal state sequence; outputting a wind and light output and load demand global prediction value; by taking the minimum comprehensive operation cost, the minimum carbon emission intensity and the minimum power supply shortage rate as targets, obtaining equipment, power balance and hydrogen storage capacity constraints through a constraint verification algorithm, and constructing a multi-target optimization model; a scene trigger function identifies four types of typical scenes, a chaos particle swarm optimization algorithm is improved to solve a model, and density peak clustering is carried out to obtain a global scheduling scheme containing output of an electrolytic cell and a fuel cell and energy storage charge and discharge power. According to the invention, through scene adaptive optimization and intelligent clustering screening, the randomness and volatility of wind and light output are dealt with.

The invention discloses an intelligent manufacturing collaborative decision-making method based on digital twinning. The method comprises the following steps: establishing a global digital twinning model and a local digital twinning model according to an intelligent manufacturing process; the method comprises the following steps of: acquiring production data in an intelligent manufacturing process in real time; determining a regulation and control index data set under sampling times; constructing a time sequence augmented matrix; constructing a dynamic weight matrix by utilizing a Pearson correlation coefficient; carrying out dimensionality reduction on the weighted time sequence augmentation matrix by utilizing the rank of the corrected singular value matrix; and standardizing the dimensionality-reduced weighted time sequence augmentation matrix, dividing the standardized matrix by using a dynamic sliding window, inputting the divided matrix into a double-branch attention network in sequence, and predicting a production decision scheme in combination with a current actual production decision scheme. Through cooperation of local self-cooperation and global scheduling, the response speed of intelligent manufacturing is improved.

The invention discloses a power grid loss intelligent optimization system and method based on a heterogeneous computing architecture, and relates to the technical field of power grid loss optimization. A five-level closed-loop control process taking a disturbance density feature vector group Oflow as a perception basis, a resource mapping feature tensor Omap as scheduling input, a hot spot risk vector Orisk as a state constraint, a scheduling decision vector Odispatch as dynamic control execution and a global scheduling performance value Q as a feedback basis is constructed; a whole-process adaptive scheduling optimization mechanism aiming at heterogeneous task complexity and a power grid disturbance evolution trend in power grid operation is realized, and a local anomaly situation can be perceived in advance by constructing a disturbance density feature vector group Oflow before a clear anomaly is not formed in a task disturbance trend; cross-architecture granularity scheduling can be realized among multiple architectures through a resource mapping feature tensor Omap; risk diffusion constraints in spatial topology can be provided based on the hot spot risk vector Orisk.