32results about How to "Improve load balancing" patented technology

The invention belongs to the technical field of computing power scheduling, and particularly relates to a computing power distribution scheduling method and system for an edge computing platform, and the method comprises the steps: generating a business pool based on a historical resource request task; constructing a business computing power demand prediction model based on XGBoost, adding the influence factors into the prediction model, generating resource request business prediction output of each node device, updating the business pool according to the prediction output, evaluating the computing power scale required by the resource request business in the updated business pool, and obtaining the computing power demand; according to the computing power demand obtained by each node device, taking energy consumption minimization and response time minimization as targets, carrying out multi-target solution based on an improved firefly algorithm, and obtaining a computing power distribution scheme which enables the energy consumption and the response time to reach the optimal; and allocating schedulable computing power resources in the edge computing platform based on the computing power allocation scheme. According to the method, overload of part of the data centers can be avoided, meanwhile, the overall calculation efficiency is improved, and calculation power resources are reasonably utilized.

The invention discloses a permanent magnet motor performance testing method and device, and particularly relates to the technical field of permanent magnet motor performance testing. The method comprises the following steps: monitoring a large-scale permanent magnet motor model test process, formulating an intelligent task coordination distribution strategy, determining task coordination efficiency between processing units, when the task coordination efficiency between the processing units is reduced, judging whether communication performance between the processing units is abnormal or not, and adjusting the intelligent task coordination distribution strategy; when the communication performance of the processing units is abnormal, the overall working state between the processing units in the adjustment process of the intelligent task coordination distribution strategy is judged, the adjustment execution effect is evaluated and fed back, and the communication efficiency abnormality between the processing units is effectively monitored in time, so that the communication performance between the processing units is improved, and the user experience is improved. And the execution effect of the formulated intelligent task coordination strategy is fed back, so that the stability of the overall working state between the processing units is ensured.

The invention provides a GPU cluster NLP task parallelization system and method based on a particle swarm optimization, and the method comprises the steps: carrying out the analysis of an input NLP model based on a task decomposition module, dividing the NLP model into a plurality of subtasks, and collecting and recording the GPU cluster resource state based on a GPU resource management module, a plurality of sub-tasks are mapped on a GPU cluster based on a task allocation module, a plurality of independent particle swarms are constructed, task allocation is optimized by using a particle swarm algorithm based on a particle swarm optimization module, task execution is scheduled according to an optimized task allocation scheme based on a task scheduling module, and task execution is completed. And based on a result processing module, collecting and integrating sub-task calculation results on each GPU, and outputting the sub-task calculation results. According to the method, the NLP training speed is remarkably increased, the resource utilization efficiency is effectively improved, and the method is suitable for being used in a compute-intensive NLP task and particularly has a remarkable effect in a scene where GPU resources are rich and highly parallel computing is needed.

The invention discloses a geometric multi-grid aggregation method based on array plane propulsion, and belongs to the technical field of aerodynamic numerical calculation. In order to establish an efficient and rapid geometric multi-grid aggregation method, the method comprises the following steps: establishing a fine grid control body based on a generated grid suitable for flow field numerical solution, and carrying out layered marking; selecting a seed unit to be aggregated; constructing a propulsion array plane for the obtained seed units to be aggregated by using non-aggregated peripheral adjacent units, and establishing a list of units to be aggregated; optimizing to-be-aggregated units in the to-be-aggregated unit list to obtain an optimal to-be-aggregated unit and adding the optimal to-be-aggregated unit into a coarse grid control body; calculating an aggregation rate of the coarse grid control volume, judging whether aggregation is completed according to the obtained aggregation rate of the coarse grid control volume, and if so, performing next coarse grid aggregation; if not, continuing polymerization; and detecting whether all seed units of the fine grid control body are aggregated or not, if not, continuing aggregation, and if yes, removing isolated units and ending aggregation.

The invention provides a traffic load balancing method and device, electronic equipment, a storage medium and a product. The method comprises the following steps: receiving a message; the message carries a flow identifier of the AI large model training data flow to which the message belongs; if the flow identifier or the outlet port corresponding to the flow identifier is not queried in the forwarding record table, selecting a target outlet port with the maximum available bandwidth from the outlet ports meeting the transmission condition of the data flow to forward the message so as to realize the flow load balancing of the outlet ports; the aging time of the forwarding record table item is set to cover one iteration period of AI large model training. According to the method, the data flow does not need to be segmented at a time interval, a proper port can be selected to forward the data flow based on the real available bandwidth of the output port, the bandwidth and queue size of the port are not measured in real time, the accuracy of load measurement is improved, the load balancing effect is improved, network congestion caused by flow load imbalance of AI large model training is avoided, and the network performance is improved. And the resource matching efficiency of the whole service flow is improved.

The utility model provides a tower crane jacking oil cylinder extension detection device, the tower crane jacking oil cylinder extension detection device comprises a first oil cylinder, a detection assembly and a stabilizing assembly, a piston rod of the first oil cylinder is fixedly connected with a first connecting rod, and one end of the first connecting rod is provided with a first positioning frame; the detection assembly is arranged on one side of the first positioning frame, the detection assembly comprises a fixing plate, a supporting block, a movement guide plate, a sliding frame and a connecting block, the fixing plate is fixedly connected to one side of the first positioning frame, and the supporting block is fixedly connected to the top of the fixing plate. The tower crane jacking oil cylinder extension detection device solves the problems that when part of existing jacking oil cylinders conduct jacking operation, real-time detection on the extension length is not convenient, workers cannot find the abnormal condition of extension of the oil cylinders in time, equipment faults are likely to be caused, and safety risks are increased. The tower crane jacking oil cylinder extension detection device solves the problems that in the prior art, when part of jacking oil cylinders conduct jacking operation, the extension length is not convenient to detect in real time, and therefore workers cannot find the abnormal condition of extension of the oil cylinders in time.

The invention provides a power distribution network partition optimization method and system. The method comprises the steps that a non-inductive load characteristic index set is constructed, non-inductive load characteristic indexes are constructed into characteristic vectors, electrical attributes of a power supply unit are constructed into attribute vectors, and then a two-dimensional evaluation model is constructed; obtaining a plurality of candidate partitions based on electrical attributes of the power supply units, calculating a comprehensive evaluation value of each load point and each power supply unit through a two-dimensional evaluation model, distributing each load point to the candidate partitions meeting a preset partition requirement through the comprehensive evaluation value, obtaining a preliminary partition scheme, constructing a global optimization model, and obtaining a final partition scheme; and optimizing each candidate partition to obtain a preliminary partition optimization scheme, constructing a corresponding power flow balance equation, and locally adjusting the preliminary partition optimization scheme to obtain a power distribution network partition optimization scheme so as to optimize the power distribution network partition. According to the invention, multi-dimensional evaluation is carried out on the power distribution network to optimize the reasonability of load division of the power distribution network, so that the operation reliability of the power grid is improved.

The utility model relates to a foundation pit supporting structure for hydraulic engineering, and relates to the technical field of foundation pits, the foundation pit supporting structure comprises a reinforcement cage and a fixing column fixedly mounted on the outer side of the reinforcement cage, and the fixing column is provided with a through hole for concrete to pass through. The underground diaphragm wall has the effect of improving the defects that the combination compactness between the concrete and the reinforcement cage is poor and the bearing capacity of the underground diaphragm wall is insufficient.

The invention discloses an Internet of Things node load balancing method and device, electronic equipment and a storage medium. The method comprises the following steps: determining a dynamic load value corresponding to the complex building group Internet of Things; determining a load integration value according to the load data volume corresponding to all non-uniformly distributed Internet of Things nodes in the complex building group Internet of Things; under the condition that the load integration value is greater than the dynamic load value, determining a priority corresponding to each non-uniformly distributed Internet of Things node; according to the priority, a load balancing strategy is determined and executed, and the load balancing strategy is used for scheduling load data in the non-uniformly distributed Internet of Things nodes so as to reduce the overall load degree of the non-uniformly distributed Internet of Things nodes. According to the method and the device, the technical problem that the load of the Internet of Things of the complex building group is unbalanced due to the fact that the management mode in the related technology lacks the consideration of communication and cooperation between the nodes of the Internet of Things in the complex building group is solved.

The invention relates to the technical field of computer science, in particular to a multi-thread concurrent processing method and system based on adaptive entropy segmentation and interactive differential learning, and aims to dynamically divide preprocessed data through an adaptive entropy segmentation algorithm and optimize a data processing path by using an interactive differential learning algorithm. Finally, the system realizes the backtracking and correction of the abnormal data through a multi-level error correction backtracking mechanism, and ensures the consistency and accuracy of the final data. According to the method, the problems of non-uniform load, insufficient data processing path optimization and poor abnormal data processing in the traditional multi-thread concurrent processing are effectively solved, and the processing efficiency and data reliability of the system are greatly improved.

The invention relates to the technical field of bucket wheel machine fault detection, in particular to a bucket wheel machine unmanned collaborative operation fault on-line diagnosis system, which comprises a single machine path and instruction anomaly detection module used for performing path offset and instruction loss diagnosis on the operation state of a single bucket wheel machine; the single-machine fault root cause analysis module is used for performing root cause analysis on the detected single-machine path offset and instruction loss faults; the collaborative task fault influence evaluation module is used for evaluating the dynamic influence of the single-machine fault on the collaborative operation task of the multi-bucket wheel machine; and the cooperative task dynamic adjustment module is used for optimizing cooperative task distribution of the multiple bucket wheel machines according to the single-machine fault influence evaluation result, and redistributing operation tasks of the fault bucket wheel machine to the healthy bucket wheel machine. According to the method, the depth and accuracy of fault diagnosis are improved, whether the fault bucket wheel machine is shut down or not is dynamically judged, the task completion rate and the system efficiency are balanced, unnecessary resource waste is avoided, and meanwhile the emergency processing requirement of major faults is quickly responded.

The present application provides an adaptive load balancing processing method and system based on a blockchain network, which can accurately analyze and pair key and derivative resource usage indicators by collecting the continuous load status characteristics of each blockchain node in the blockchain network, including multiple resource usage indicators. When encountering a unique resource usage tag and the current load balancing strategy does not cover the unique resource usage tag, if the derivative resource indicator is already in the strategy and the resource cluster associated with it contains other unique resource tags, an updated load balancing solution will be generated based on these resource usage indicator pairing sets to ensure that all unique resources are reasonably load balanced and regulated, thereby effectively improving the performance and stability of the blockchain network.

The invention discloses a GPU batch matrix multiplication accelerator, comprising a main branch circuit composed of a workgroup manager, an instruction decoding processor, a high-speed storage unit, a shared memory unit, a computing unit, and a DMA, which is used to perform batch order optimization on matrix data to obtain matrix slices, and load the matrix slices into the computing unit; a side branch circuit composed of an instruction decoding processor, a high-speed storage unit, a shared memory unit, a computing unit, and a DMA, which directly performs kernel function calculation on the matrix data to obtain matrix calculation results when batch order optimization is not performed; the invention can effectively improve the utilization efficiency of the computing unit and achieve load balancing of the task amount of each computing unit, improve the computing density during operation, achieve higher instruction parallelism, thread parallelism, and memory access parallelism, thereby giving full play to the computing power of the hardware and achieving the purpose of computing acceleration.

The invention discloses a method for accelerating complex scene ray tracing rendering, and relates to the technical field of computer graphic analysis. The method comprises the following steps: carrying out rapid ray tracing rendering on a complex scene by utilizing a distributed rendering cluster, and improving the calculation efficiency when the distributed rendering cluster renders a frame of complex scene by matching a load balancing strategy with the distributed rendering cluster; a frame of rendering result image generated by the distributed rendering cluster and corresponding auxiliary information are input into a multi-scale denoising device for post-processing of filtering denoising and time anti-aliasing, the sense of reality of the rendering result image is improved, and the rendering result image output after processing of the multi-scale denoising device is the final high-sense-of-reality rendering result image. The invention aims at effectively carrying out vivid rendering on a complex scene, and improving the rendering efficiency and effectively denoising while keeping the rendering effect.

The invention relates to a split-phase power control method and system suitable for a three-phase four-wire system converter, and belongs to the field of power grid balance control. Subtracting a converter DC side voltage sampling value from a converter DC side neutral-point voltage target value which is set to be 0, and superposing with output of zero-sequence current control after passing through a neutral-point potential controller to obtain a zero-axis voltage control value; the zero-axis voltage control value, a d-axis voltage control value obtained according to the d-axis current instruction value and a q-axis voltage control value obtained according to the q-axis current instruction value are subjected to Park inverse transformation to obtain a three-phase voltage control value, and PWM waves used for controlling a converter switching tube are output according to the three-phase voltage control value; the zero-sequence current control comprises the steps of subtracting a zero-sequence current component sampling value from a zero-sequence current instruction value and then outputting the result through a zero-sequence current loop controller. According to the control values of the zero-axis voltage, the d-axis voltage and the q-axis voltage, the PWM waves used for controlling the switching tubes are obtained, and the load balancing capacity of the transformer area converter is remarkably improved.

The invention relates to a cross-modal semantic alignment method based on a visual semantic shared space. The method comprises the following steps: S1, acquiring an image embedding representation and a text embedding representation; s2, inputting the embedded representations into a routing gate module, and determining a routing probability value of each embedded representation; s3, inputting the embedded representation into an aggregation expert module, and aggregating the embedded representation into an aggregation vector result by the obtained overall vector; and S4, a loss function is calculated based on an aggregation vector result, a routing gate module and an aggregation expert module are trained based on the loss function, a cross-modal semantic alignment model is obtained, and the cross-modal semantic alignment model is used for realizing cross-modal semantic alignment of actual image data and text data in a multi-modal task. Compared with the prior art, the cross-modal semantic alignment method has the advantages of improving the cross-modal semantic alignment performance and the like.

The invention provides an optical network multipath resource allocation method, system and device, and the method comprises the steps: obtaining a current preset routing table of an optical network according to each node pair in a current directed graph of the optical network; obtaining a service set comprising a plurality of service flows; inputting the preset routing table and the service set into a preset optical network multi-path resource allocation model, and solving the optical network multi-path resource allocation model based on a preset constraint condition and a gradient descent method to obtain a continuous solution representing the allocation proportion of each service flow on different paths and an integer allocation matrix corresponding to the continuous solution; and performing conversion processing on the integer distribution matrix to obtain a corresponding flow table used for recording the distribution relationship between each service flow and each path resource, and issuing the flow table to each network device. The network load balancing degree can be improved, so that the network throughput is increased, the network time delay is reduced, and the network load balancing is realized.

The invention belongs to the field of task adjustment, and particularly relates to a task division-based intrusion detection system scheduling method, which comprises the following steps of: obtaining load information of processing cores, and obtaining a processing core with the highest load and a processing core with the lowest load according to the load information; obtaining the number of flows in the processing core with the highest load and the processing core with the lowest load and the load distribution of all flows; obtaining the load difference between the processing core with the highest load and the processing core with the lowest load according to the load distribution of all flows in the processing core with the highest load and the processing core with the lowest load; judging whether elephant flows exist in the processing cores or not according to the flow quantity in the processing core with the highest load and the processing core with the lowest load and the load difference of the processing cores; if the elephant flow exists, migrating the elephant flow after task division; and if the elephant flow does not exist, migrating the flow in the processing core with the highest load to the processing core with the lowest load. The migration flow between the processing cores can be balanced, and the load of the processing cores can be optimized.

The present disclosure relates to a graphics processing method, a graphics processing device and a computing device. The graphics processing method comprises: dividing a frame to be rendered into a plurality of tiles, wherein the frame to be rendered comprises a plurality of primitives, each primitive covering at least one tile; determining a tile-level splitting mode for at least one tile of the plurality of tiles based on a number of primitives covering the tile; obtaining a frame-level splitting mode for the frame to be rendered; enabling tile splitting for at least part of the plurality of tiles based on the tile-level splitting mode and the frame-level splitting mode, wherein the tile splitting comprises splitting one tile into a splitting number of tile regions corresponding to the tile-level splitting mode or the frame-level splitting mode; and assigning the plurality of tile regions obtained by the tile splitting to a plurality of tile processing cores, wherein each tile processing core is configured to perform a tile processing operation for the assigned tile region. The method helps to provide a flexible splitting mode determination manner and improve graphics processing efficiency.

The application discloses a SAS extender intelligent routing method and system based on GNN. The method comprises the following steps: S1, constructing a path link bipartite graph model; S2, performing load-aware graph neural network inference; S3, path allocation proportional decoding and normalization; S4, strategy compilation and hardware delivery; and S5, closed-loop monitoring and incremental learning optimization. The application realizes load balancing, low delay and globally optimal routing through graph neural network modeling; has the advantages of fast fault self-healing, strong generalization and closed-loop adaptive capacity; and is compatible with the existing SAS protocol, has strong real-time performance and low deployment cost.

A geometric multi-grid aggregation method based on array advancement belongs to the technical field of numerical calculation of aerodynamics. In order to establish an efficient and fast geometric multi-grid aggregation method, the present invention includes establishing a fine grid control body based on a generated grid suitable for numerical solution of flow field for hierarchical marking; selecting seed units to be aggregated; constructing an advancement array with unaggregated surrounding adjacent units for the obtained seed units to be aggregated, and establishing a list of units to be aggregated; optimizing the units to be aggregated in the list of units to be aggregated, obtaining the optimal units to be aggregated and adding them to the coarse grid control body; calculating the aggregation rate of the coarse grid control body, judging whether the aggregation is completed according to the obtained aggregation rate of the coarse grid control body, and if it is completed, performing the next coarse grid aggregation; if it is not completed, continuing the aggregation; detecting whether all seed units in the fine grid control body have completed the aggregation, judging if it is not, continuing the aggregation, judging if it is, removing the isolated units, and ending the aggregation.

This application provides a deployment method, apparatus, electronic device, and storage medium for a hybrid expert model. The hybrid expert model includes multiple transformer modules, each transformer module comprising a normalization layer, a compressed multi-head attention mechanism layer, and a multi-level hierarchical hybrid expert layer. The compressed multi-head attention mechanism layer includes an encoder / decoder, a multi-head attention mechanism module, and a position encoder. The multi-level hierarchical hybrid expert layer includes a multi-level hybrid expert architecture. The deployment method includes: acquiring hardware devices of multiple levels to be deployed; deploying the multi-level hybrid expert architecture in the corresponding levels of the hardware devices according to their hierarchy from high to low; the level of the hybrid expert architecture deployed in lower-level hardware devices is no higher than the level of the hybrid expert architecture deployed in higher-level hardware devices. By deploying the hybrid expert architecture according to the hierarchy of the hardware devices and the hierarchy of the hybrid expert architecture, the utilization of hardware bandwidth at different levels in the cluster is optimized, and the computational performance of the hybrid expert model is improved.

The invention relates to an in-core-group multi-thread optimization method and system based on a SW many-core processor, and belongs to the technical field of electronic information. Comprising the following steps: (1) data preprocessing; separating a dependent operation task from a coring calculation task in the task; (2) the master core and the slave core assist in processing; according to the calculation logic of an original calculation task, a dependent operation task and a slave core calculation task are processed by a master core and a slave core respectively, and parallel operation of multiple instruction streams and multiple data streams in a core group is achieved; (3) slave core calculation optimization; applying for an independent area in the slave kernel LDM to store an intermediate result; in addition, a part of main storage space is applied for receiving the calculation result of the slave core; and when all the threads in the core group reach a preset end position, merging operation results. According to the method, calculation interruption and data errors caused by multiple times of data transmission are avoided, the independence of the slave kernel function is effectively enhanced, and the calculation efficiency is improved.

This application provides a message processing method, apparatus, storage medium, and computer device. The method includes: in response to any data core receiving a message to be forwarded, generating a control core scheduling request using the data core and sending it to a target scheduler associated with the data core; using the target scheduler, generating scheduling indication information based on the control core retrieval request and the latest load information fed back by each control core under load feedback conditions, and feeding it back to the data core; using the data core, determining a target control core from multiple control cores based on the scheduling indication information, and distributing the message to be forwarded to a target lock-free queue between the target control core and the data core; and using the target control core, retrieving the message to be forwarded from the target lock-free queue and processing the message.

The invention relates to the field of Internet of Vehicles, in particular to a calculation task allocation method and a control module for a vehicle-road cooperation vehicle formation. The method comprises the following steps: firstly, dynamically updating calculation tasks of a vehicle formation, classifying the calculation tasks and adding label information to form a global task pool; and obtaining residual computing power values of computing power nodes such as a head vehicle, a following vehicle, a road test MEC and a cloud server, and detecting node health degrees to form a multi-source computing power pool. Then predicting a potential task of each calculation task and calculating a calculation power dependency degree and a priority degree; performing basic capability scoring on each computing power node, and screening candidate nodes of which the basic capability scores are higher than a preset value; and finally, performing many-to-many matching on each calculation task and candidate node in the double pools by taking the lowest total execution cost of the calculation task and the overall load balance of the calculation power node as optimization targets. According to the method, the problems of relatively large task delay and unbalanced equipment load caused by unreasonable calculation task allocation in the existing vehicle formation can be solved.

The application discloses a kind of high-low voltage switch cabinet load balancing control method and system, it is related to load control technical field, including: constructing multi-cabinet integrated model, each switch cabinet in switch cabinet group is dynamically monitored, obtains each cabinet state matrix;According to each cabinet state matrix, load prediction equalization adjustment is carried out to switch cabinet group, obtains load adjustment first group;Multi-node load balancing evaluation optimization is carried out to load adjustment first group, obtains load adjustment second group, fault risk prediction is carried out to switch cabinet group, obtains fault risk matrix;According to load adjustment second group, power quality risk prediction is carried out to switch cabinet group, obtains power quality risk matrix;Multi-dimensional optimization is carried out to load adjustment second group, obtains load balancing control strategy.The application solves the technical problem of load imbalance of high-low voltage switch cabinet in the prior art, achieves the technical effect of improving load balancing effect, reducing fault risk and optimizing power quality.

The present application relates to the field of logistics resource management and control, and discloses a logistics resource allocation method, device, equipment and storage medium. The method comprises the following steps: obtaining an initial storage task population corresponding to a to-be-controlled logistics resource; continuously recombining a task individual corresponding to each first storage task by using a preset acceleration genetic strategy until the initial storage task population meets a preset optimization condition, and then stopping the recombination to obtain an optimal storage task population; pre-allocating the to-be-controlled logistics resource to each processing node corresponding to a second storage task by using a preset dynamic allocation strategy to obtain an initial allocation decision sequence of the to-be-controlled logistics resource in each processing node; and stopping the allocation until the initial allocation decision sequence meets an allocation condition, and then taking the initial allocation decision sequence as an optimal allocation decision sequence. The present application realizes cloud storage of logistics resources on multiple general servers, improves the storage utilization of the general servers, and optimizes the system load balancing.

The utility model discloses a balanced thread structure which comprises a base body, a thread and an axial decreasing subtractive material, the thread is formed on the surface of the base body, the axial decreasing subtractive material is at least formed in a screwing area of the thread, and the subtractive quantity of the axial decreasing subtractive material is gradually decreased from the maximum load end of the thread to the minimum load end of the thread in the axial direction of the thread. According to the balanced thread structure, on the premise that basic parameters of standard threads are not changed and axial installation space is not increased, through local axial decreasing and material decreasing, threaded connection load distribution optimization is achieved.

The present invention relates to the technical field of 5G vehicle networking, and in particular to a prediction-based dynamic load balancing method in the vehicle networking CoMP scenario. The present invention includes counting the initial load status of each base station, and using the lightly loaded base station as a cooperative base station for the fully loaded or overloaded base station; for the overloaded base station, the user is first diverted, and some edge users are diverted to the adjacent lightly loaded base station for service, and then the user's location distribution is analyzed; for the fully loaded or lightly loaded base station, the user's location distribution is directly analyzed; the central user is directly served by this base station, and the edge user must first be predicted, and the CoMP technology is selectively used to provide services to the user. The present invention shows through simulation analysis that the base station overload rate is significantly reduced during both non-peak and peak periods, and the user's rate is also improved to a certain extent, achieving the purpose of load balancing between base stations, while ensuring the user's communication quality and facilitating the user's life.