1588 results about "Multiple stages" patented technology

The invention discloses a PLC code generation and verification method based on a large language model, and the method comprises the steps: generating multi-stage multi-mode input data which is used for representing user demands and formulating a system design scheme according to a demand analysis result; the multiple stages comprise a demand analysis stage and a design stage of a complete engineering cycle; the multi-mode input comprises a Pamp (Passive Amplitude Point); an ID graph and a natural language; then, according to the multi-stage multi-mode input data, a PLC code is generated; then, verification testing is carried out on the generated PLC code, and the verification testing comprises code verification and compiler verification; and finally, after the verification test on the PLC code is completed, deploying the PLC code. The PLC code deployment method is high in accuracy and automation degree, an efficient and accurate PLC code deployment method can be provided for network security researchers, and the PLC code deployment method can be widely applied to the technical field of computers.

The invention discloses a Text2SQL intelligent question and answer query method and system based on an LLM and a vector model, and aims to construct an interactive joint model with the LLM as a core and the vector model as an auxiliary, and realize an intelligent Text2SQL question and answer query chain through multi-dimensional data representation, task decoupling, model multi-task synchronous fine tuning and link construction. The method includes collaboration of multiple stages of SQL generation and execution, answer generation and the like, so that query requirements expressed by a user in a natural language are converted into professional MySQL query statements. According to the method, a complex Text2SQL task is disassembled into a plurality of relatively simple tasks through task decoupling design and link construction, task node scheduling is optimized through link construction, and fault-tolerant processing is carried out for model generation abnormity; meanwhile, in order to assist the user in checking, each subtask result is displayed at the front end, and the subtask results are editable, so that the final result is traceable.

Apparatus and methods are disclosed for implementing a copilot as a network of microservices including specialized large language models (LLMs) or other trained machine learning (ML) tools. This architecture supports flexible, customizable, or dynamically determinable dataflow. Compared to much larger competing LLMs, comparable or superior performance is achieved for certain tasks, while significantly reducing computation time and hardware requirements, even to a single compute node with a single GPU. An expansion microservice converts client input tokens into associated tokens to diversify targeted tasks presented to a core microservice. An ML tool in the expansion microservice is pretrained in multiple stages for varying tasks using varying general and target-specific corpora. Synthesized training data derived from a knowledge graph can also be used. The expansion ML tool is subsequently fine-tuned in one or multiple stages. Variations and additional techniques are disclosed.

This disclosure describes techniques relating to development of data flow pipelines. In one example, this disclosure describes a method that includes accessing, by a computing system, metadata identifying characteristics of a data flow pipeline; generating, by the computing system and based on the metadata, the data flow pipeline, wherein the data flow pipeline includes a plurality of stages, and wherein the plurality of stages includes a multi-sourced stage in which data output by each of a subset of stages in the plurality of stages are used as input to the multi-sourced stage; executing, by the computing system, the data flow pipeline; detecting, by the computing system, modifications to the metadata; generating, based on the modifications, an updated data flow pipeline; and executing, by the computing system, the updated data flow pipeline.

The invention relates to the technical field of hydrogen production, in particular to a bipolar plate runner structure-mass transfer efficiency parameter simulation optimization method. Comprising the steps that a flow channel structure is designed, specifically, a bipolar plate flow channel is constructed through a fractal tree-shaped network topological structure, the flow channel is composed of multiple stages of branch channels, the section size of each stage of branch channel is decreased progressively according to a self-similarity proportion, and a three-dimensional flow channel network with fractal dimensions is formed; multi-physics field coupling modeling: establishing a multi-physics field coupling model including fluid flow, electrochemical reaction and heat and mass transfer based on computational fluid mechanics and a finite element method, wherein the model considers a turbulence effect in a flow channel, interface impedance of a porous medium diffusion layer and reaction kinetic parameters at the same time; parameter sensitivity analysis: screening key parameters influencing mass transfer efficiency through orthogonal test design, including flow channel fractal dimension, branch angle, porosity and surface wettability parameters, and constructing a parameter response curved surface; the flow channel structure of the bipolar plate can be accurately optimized, and the mass transfer efficiency is effectively improved.

Provided in embodiments of the present disclosure are a method for scheduling workloads of a data center in an electricity market environment, and a device. In the embodiments of the present disclosure, a multi-stage random optimization solution that allows a data center to participate in a multi-level electricity market is provided. The types and data dependency relationship of workloads are taken into consideration, and a load scheduling model, in a time dimension, for the workloads is constructed; on the basis of electricity source information of the data center, a multi-stage electricity cost optimization model is constructed, so as to comprehensively consider the exogenic uncertainty of an electricity price and carbon emissions and the endogenic uncertainty of the workloads; on the basis of the multi-stage electricity cost optimization model, multi-stage optimization solving is performed on the load scheduling model, so as to obtain a load scheduling scheme and an electricity purchase strategy. Therefore, a more accurate and more reliable estimation is made concerning the electricity flexibility of the data center, such that the electricity cost of the data center can be reduced, support is provided for the reduction of carbon emissions, and the consumption of renewable energy sources is encouraged, thereby achieving the energy use flexibility of the data center in the multi-level electricity market environment.

Provided is an efficient and scalable attention model that can be referred to as multi-axis attention. Example implementations can include two aspects: blocked local and dilated global attention. These design choices allow global-local spatial interactions on arbitrary input resolutions with only linear complexity. The present disclosure also presents a new architectural element by effectively blending the proposed multi-axis attention model with convolutions. In addition, the present disclosure proposes a simple hierarchical vision backbone, example implementations of which can be referred to as MaxViT, by simply repeating the basic building block over multiple stages. Notably, MaxViT is able to “see” globally throughout the entire network, even in earlier, high-resolution stages.

The invention discloses a method for dynamically adjusting the frequency of a resonant cavity of a cyclotron and a tuning circuit, which are applied to the field of cyclotrons and solve the problem that when the frequency of the resonant cavity of the cyclotron is adjusted by judging the magnitude of reflection power, due to the fact that the reflection power can change in multiple stages in a single macro pulse period, the frequency of the resonant cavity of the cyclotron is adjusted. And an accurate numerical value is difficult to obtain through a simple amplitude sampling circuit. According to the invention, through determining the stable resonance time period of the resonant cavity, the phase difference between the incident power signal and the reflection power signal is sampled in the corresponding stable resonance time period of the resonant cavity, and the change value of the phase difference between the incident power signal and the reflection power signal is dynamically compared. According to the conversion relation between the phase difference change value and the frequency change value, the resonant cavity frequency change value is obtained through conversion, then the resonant cavity tuning device is adjusted to enable the resonant cavity frequency to be consistent with the frequency of the incident power signal all the time, and the dynamic tuning method of the resonant cavity is achieved.

Apparatuses, systems, and techniques to perform a cryptographic operation using multiple iterations, wherein each iteration includes two or more stages operating in parallel on inputs derived from a common value, one of the stages computing real data and other stages computing dummy data.

The invention relates to a scissor type unfolding mechanism which comprises a top short rod assembly, scissor type stage group rod piece units, a bottom supporting rod assembly, a driving unit assembly and an unfolding in-place indicating switch, the unfolding mechanism is connected with an unfolded object through the top short rod assembly, and the multiple stages of scissor type stage group rod piece units are connected through hinge shafts to form a multi-stage unfolding mechanism. The long rods of the multiple stages of unfolding mechanisms are connected in sequence; the bottom supporting rod assembly is connected with the driving unit assembly, the driving unit assembly drives the bottom supporting rod assembly to achieve unfolding and folding of the unfolding mechanism, after the unfolding mechanism is unfolded, a stable triangular structure is formed through the bottom supporting rod assembly, the scissor type stage group rod piece units and the limiting devices, the structural rigidity is improved, and the unfolding mechanism is controlled to be unfolded.

The invention relates to a method for improving the calculation speed of a model based on a mercuric chloride AI processor. The method comprises the following steps: deploying the mercuric chloride AI processor and a deep learning model in a server, and carrying out adaptation and optimization on the mercuric chloride AI processor; before the data in the first buffer area is read, predicting and preloading the data to be processed, and loading the data from the global memory to the second buffer area in advance; a double-buffer mechanism is arranged in the mercuration AI processor, interrupt and event trigger points are set, the operation of loading data to a next buffer area is immediately started when a specified calculation stage is finished, and the time sequence of data circulation is accurately controlled; and a control parameter is automatically adjusted based on monitoring data of the real-time monitoring module, the calculation process is decomposed into a plurality of stages, different buffer areas are allocated for each stage, and access to the shared memory is optimized by setting a specified cache replacement strategy and the size of a cache line. According to the process, the calculation speed of the model in the deep learning field is improved, the resource utilization rate is improved, and the defect of manual adjustment and optimization is overcome.

The invention provides a pole piece and a manufacturing method thereof, a rolling system and a secondary battery, and the manufacturing method of the pole piece comprises the following steps: coating slurry on a current collector and drying to obtain a dried pole piece; the difference between the thickness of the dried pole piece and the preset thickness after rolling is used as the total rolling reduction; the dried pole piece is subjected to continuous multi-stage rolling, adjacent two-stage rolling is carried out, the percentage of the rolling reduction of the previous-stage rolling in the total rolling reduction is larger than or equal to the percentage of the rolling reduction of the next-stage rolling in the total rolling reduction, and the sum of the rolling reduction of all stages of rolling is equal to the total rolling reduction; in the multi-stage rolling, the rolling reduction of each stage of rolling is gradually reduced on the whole; and the multi-stage rolling is cold pressing. According to the manufacturing method of the pole piece, the multi-stage rolling mode is adopted, the rolling reduction decreasing and roller gap adjusting mode is adopted, the diameter, roughness and hardness of the roller body are combined to adapt to different rolling reduction, and finally the consistency of stress release and the pole piece is achieved.

The invention relates to a conflict-free AGV scheduling method for punctualized material distribution in a workshop, and the method comprises the following steps: obtaining a material distribution task, and dividing a decision period into a plurality of stages; an independent scheduling algorithm is used in each stage for AGV scheduling, tasks and departure time distributed by each AGV in each trip are optimized, a conflict-free optimal path is planned, an AGV scheduling scheme for minimizing time window penalty and distribution cost is realized, the scheduling algorithm comprises task allocation and path planning, in the task allocation, the path planning is performed, and in the path planning, the time window penalty and the distribution cost are reduced. According to the method, travel division is carried out based on departure time estimation, AGV assignment of each travel is carried out based on a heuristic rule, initial path planning is carried out by adopting an A * algorithm according to a task allocation result in path planning, conflict detection is carried out based on a path time window, and conflict resolution is carried out in combination with waiting and local re-planning. Compared with the prior art, cooperation of task allocation and path planning under the punctualization distribution requirement is achieved, and the calculation efficiency and the resource utilization rate can be improved.

The invention relates to a BIM-based deep foundation pit construction dynamic monitoring method and system, and specifically, the method comprises the steps: dividing a BIM design model into a plurality of stages according to a construction stage, and obtaining deep foundation pit site point cloud data and a current construction stage; constructing a point cloud picture based on the affiliation consistency probability and boundary conformity of the points in the point cloud in the BIM design model at the current construction stage, and segmenting the point cloud picture by adopting minimum segmentation with hierarchical constraint to obtain a plurality of point cloud clusters; calculating the internal geometrical morphology of the point cloud cluster based on the characteristic value of the covariance matrix of the point cloud cluster, and associating the point cloud cluster with the construction member in the BIM design model at the current construction stage by using the position relationship between the geometrical morphology and the point cloud cluster; and performing three-dimensional reconstruction on the point cloud cluster by using construction components associated with the point cloud cluster, and updating and visually displaying the current construction progress state of each component in the BIM design model.

The invention relates to a heterogeneous cluster-oriented resource allocation method and device and a storage medium, and the method comprises the steps: executing a plurality of iterations of an initial model on a single node of each heterogeneous cluster, and obtaining the performance characteristics of the model; generating a plurality of parallel strategy combinations comprising the first parallelism degree and the second parallelism degree; generating a plurality of assembly line parallel combinations based on the first degree of parallelism, calculating the load imbalance rate of each assembly line parallel combination, and screening out N combinations with the minimum load imbalance rate from the assembly line parallel combinations; traversing the second parallelism degree and the micro-processing batch in each stage corresponding to each parallel combination to generate a plurality of stage strategy combinations; and calculating video memory occupation and execution time under each stage strategy combination, screening out K stage strategy combinations of which the video memories are smaller than or equal to a video memory threshold value and the execution time is shortest, and carrying out resource allocation on the heterogeneous cluster based on the stage strategy combinations. According to the invention, the problems of low training efficiency and low resource utilization rate are solved.

The invention discloses a deformable wing with the characteristics of large unfolding-folding ratio and high bearing capacity. The deformable wing comprises a bearing structure, a driving mechanism, a wing surface dimensional structure and an accessory structure, the bearing structure comprises a plurality of groups of telescopic mechanisms, wing ribs and corner boxes; the accessory structure comprises a mounting plate, and the mounting plate is connected with the machine body; each group of telescopic mechanisms comprises a plurality of stages of telescopic rods which are arranged along the chord length direction and are fixed on the mounting plate; the wing ribs are arranged in the wingspan direction and are connected with the multi-stage telescopic rods through the angle boxes to form a wing framework; the driving mechanism penetrates through the hollow parts of the wing ribs, and two ends of the driving mechanism are respectively connected with the mounting plate and the wing rib on the outermost side of the wing; the wing skeleton is externally coated with the wing surface dimensional structure to form a wing shape.

The invention relates to the field of grinding machinery, and discloses a cylinder body casting grinding device and a grinding process thereof.The device comprises a frame body and a controller, a three-jaw chuck is arranged on the frame body, and a to-be-ground cylinder body is clamped through the three-jaw chuck; a grinding mechanism is further arranged on one side of the frame body and comprises a hydraulic telescopic cylinder and a driving part, the hydraulic telescopic cylinder is fixedly installed on the frame body, the driving part is fixedly installed at the end of the hydraulic telescopic cylinder, and the output end of the driving part is in transmission connection with a grinding shaft; a grinding shell is fixedly arranged at the end of the grinding shaft, multiple stages of grinding wheels are arranged in the grinding shell, an alternating assembly is further arranged in the grinding shell, and the alternating assembly is used for pushing the different stages of grinding wheels to sequentially stretch out of the grinding shell to conduct the grinding action of the cylinder body. By means of the mechanism, real-time replacement of different levels of grinding wheels in the working state in the cylinder body is achieved, and the grinding efficiency is higher.

The invention discloses an efficient hybrid parallel training method for a multi-modal model, and belongs to the technical field of computers. The method comprises the steps that firstly, a multi-modal model architecture is divided into a plurality of independent modal sub-modules according to the multi-modal model architecture, and each sub-module processes one kind of modal data; secondly, formulating a computing power distribution scheme by statically analyzing the computing load of each sub-module; calculating time and parameter quantity of each layer of a modal sub-module are analyzed, then the layers are divided into a plurality of stages by using a partition algorithm, and asynchronous pipeline parallel and data parallel are combined to accelerate training; according to the method, the hardware resource utilization rate is remarkably improved through the calculation power distribution and hybrid parallel strategy for the modal calculation load, and the acceleration of the multi-modal model training process is realized.

The invention discloses an anti-explosion pressure relief ball valve which comprises a pressure monitoring system and a control execution system. The pressure monitoring system is used for forming a pressure distribution model; the control execution system comprises a signal processing module used for calculating the pressure change rate; the abnormal pressure identification module is used for judging whether an abnormal pressure event exists or not according to the judgment condition; the multi-stage pressure relief control module is used for controlling the opening number of the micro-channel systems and controlling the pressure relief amount according to the relation between the current pressure and a preset threshold value; the stepped pressure relief structure comprises multiple stages of pressure relief cavities, so that the pressure relief process conforms to a pressure change formula; and the dynamic feedback control module is used for dynamically adjusting the opening amount of the micro-channel by adopting a PID control algorithm. The system has the following beneficial effects that through pressure monitoring, intelligent judgment and the multi-stage controlled pressure relief module, early recognition, accurate intervention and safe disposal of abnormal pressure events are achieved, and safety accidents such as pipeline burst and flange leakage are effectively prevented.

The invention discloses a plant disease identification method and system based on unsupervised category incremental learning, and a storage medium, relates to the technical field of plant disease identification, and aims to solve the problems that an existing deep learning model cannot meet the application requirements of real-time updating and optimization of the model, the timeliness is poor, and the efficiency is high. And the problem of failure of the class incremental learning strategy when the new class data is not labeled by the label exists. Comprising the following steps: S1, dividing a data set into category sets of a plurality of stages according to categories; s2, performing interpolation between the image of the current task and the random image uniformly sampled and selected from the cache region of the old category data by using a hybrid playback algorithm to generate a synthetic sample; s3, constructing a SimSiam model which comprises two identical neural network branches and shared parameters of the two branches, constructing a loss function based on the negative cosine similarity of the two network branches, and training the SimSiam model in multiple stages; s4, evaluating the performance of the model based on a KNN algorithm; and S5, classifying the plant diseases by adopting the trained SimSiam model.

The invention discloses a lifting stage structure control system and method, and relates to the technical field of stage control, and the lifting stage structure control system and method comprises the steps: S1, collecting the multi-source state data of stage equipment in real time, carrying out the data preprocessing, and constructing a multi-source state database; s2, constructing a motion state space model, predicting a time sequence motion sequence at a future moment, judging whether stage equipment is overlapped or not, predicting a collision risk, and taking risk avoiding measures; s3, the obstacle avoidance priority of the stage equipment is judged, a stage equipment movement scheme is generated, and a control instruction is issued; s4, motion operation is executed, the health degree of motion execution of the stage equipment is evaluated in real time, and graded early warning measures are taken; s5, displaying the stage situation and risk based on the collision risk and health degree evaluation result, and generating an analysis report; the problem that potential collision risks cannot be avoided in time due to dynamic path intersection caused by cooperative operation of multiple pieces of stage equipment in a complex stage scene is solved.

The invention relates to an underground engineering construction design technology, discloses a method for analyzing the influence of underground excavation under-crossing on an existing subway station, and improves the efficiency and precision of analyzing the influence of underground excavation under-crossing construction on the existing subway station. The method comprises the steps that by defining a structure and geometrical shapes and material attributes of a stratum, boundary conditions and external loads are set, grid division is carried out, and a finite element three-dimensional basic calculation model is constructed; according to theories of porous media and the like, a multi-physical field coupling model of a seepage-deformation-temperature full coupling effect is constructed; the rigidity of the underground excavation vertical pile foundation is calculated, and the rigidity of the underground excavation vertical pile foundation is corrected based on the influence factors of the soil squeezing effect and the negative friction resistance; subsurface excavation construction is subdivided into a plurality of stages, structural stress, deformation and stratum displacement are calculated through an incremental method, and the timeliness of a supporting structure and the influence of different excavation modes on stratum disturbance are analyzed; and obtaining a structural response through finite element iteration solution, and judging the influence of underground excavation under-crossing on the existing subway station.

The invention discloses a BIM-based building full life cycle compliance inspection system and method, and the system can be integrated into each process of a building project, automatically and accurately extracts key data of a plurality of stages, such as design, construction, operation and maintenance, from a BIM model in real time, and forms a complete information chain of the full life cycle of the building project. The system realizes comprehensive coverage of multiple stages of building design, construction, operation and maintenance and the like; in the design stage, the system can assist a designer in compliance review and discover and correct design defects in advance; in the construction stage, the system can monitor the construction progress and quality in real time, and the compliance of the construction process is ensured; in the operation and maintenance stage, the system can continuously monitor the operation state of the building facilities, timely maintenance suggestions and early warning information are provided, and long-term compliance operation of the building project is guaranteed; the comprehensive, efficient and accurate compliance inspection mode provides powerful guarantee for smooth promotion and successful delivery of the construction project.

In certain embodiments, model unlearning verifiability may be facilitated. In some embodiments, in connection with a request for unlearning of a dataset for a machine learning model, an unlearning process may be performed on the machine learning model, where the unlearning process involves multiple stages, for the unlearning of the dataset, at which model instances of the machine learning model are respectively generated. One or more proofs or hash-based values corresponding to the model instances may be generated, such as first and second hash-based values corresponding to first and second model instances of the machine learning model generated at first and second stages of the unlearning process, respectively (e.g., where the second stage depends on the first stage). The proofs or hash-based values may then be stored at one or more database locations of a database to enable verification of the performance of the unlearning process.

The invention relates to a filling mining method stope structure parameter optimization method and system based on numerical simulation, and the method comprises the steps: obtaining mine engineering geological conditions, building a mine entity model, and determining a stope structure parameter scheme; establishing a stope model according to the determined stope structure parameters; combining the mine entity model and the stope model, re-grouping the combined model, and assigning values to parameters in the combined model by setting mechanical parameters extracted from initial conditions and mine engineering geological conditions; and carrying out excavation and filling simulation on the combined model, respectively determining the stability indexes of the stope in the states of the original rock, the goaf and the filling body, and screening out an optimal stope structure parameter scheme. The problem that the resource development efficiency is affected due to the fact that an existing stope structure parameter optimization scheme is poor in analysis efficiency and accuracy in mine stability analysis under the multi-stage, multi-time-sequence and multi-scale conditions is solved.

The invention provides a three-dimensional point cloud segmentation method and system based on bidirectional fusion of a point cloud and an aerial view, and belongs to the technical field of three-dimensional scene environment perception, and the method comprises the steps: obtaining to-be-segmented original point cloud data; processing the obtained original point cloud data by using a pre-trained segmentation model to obtain a segmentation result; wherein the segmentation model comprises an encoder module, a backbone network and a fusion segmentation head. According to the method, a bidirectional fusion mechanism of the point cloud and the aerial view is provided, through interactive fusion of the point cloud and the BEV features in multiple stages, the expression ability of the model for local and global features is improved, the problems of fuzzy semantic boundary, difficulty in small target recognition and the like are remarkably improved, and the segmentation precision is obviously superior to that of an existing BEV method. Compared with a voxel method and a multi-representation fusion method, the method has the advantages that the weight of a calculation graph is kept light in structural design, compared with the voxel method and the multi-representation fusion method, the method has remarkable advantages in the aspects of parameter quantity and reasoning speed, precision and real-time performance are better considered, and the method is suitable for scenes with extremely high efficiency requirements such as automatic driving.

A bioreactor for suspended cells is provided. The bioreactor for suspended cells comprises: a housing which is formed in the shape of a container having a culture space with a predetermined volume, and which comprises pots for seeding; a support assembly comprising a plurality of supports arranged at intervals in multiple stages in one direction of the housing; at least one gas inlet penetrating the housing by a predetermined area; and a porous member for covering the gas inlet, wherein the culture space comprises a medium storage space filled with a predetermined amount of a medium so that all the plurality of supports are submerged therein, and the medium storage space can comprise a plurality of cell growth spaces divided through the plurality of supports arranged at intervals in multiple stages.

The invention discloses a plane matrix type high folding and unfolding bionic antenna truss structure with a negative Poisson's ratio effect, and relates to the technical field of space folding and unfolding mechanisms, the plane matrix type high folding and unfolding bionic antenna truss structure comprises a plurality of cell elements and a plurality of cell element connecting rods, transverse cell elements are connected through the cell element connecting rods, and longitudinal cell elements are fixedly connected through adjacent cell element cross beams; the cell element connecting rods are hinged to the cell elements on the two sides respectively, and the cell elements are three-level positioning inwards-concave hexagonal cell elements; the folding and unfolding antenna truss structure is in a folded state during launching, transportation and silent latency and is in an unfolded working state during on-orbit task execution, the high folding and unfolding ratio characteristic of the antenna provides bionic improvement inspiration for various urgent properties such as stealth and size reduction of a satellite antenna structure, the shape of the antenna folding and unfolding array can be changed in a multi-stage mode, and the antenna folding and unfolding array is suitable for large-scale popularization and application. Therefore, the functions of optimal frequency band adjustment, optimal area expansion, switching of different scanning ranges and the like are realized.

The invention provides an end-to-end intelligent system development method and system based on an AI large model. The method comprises the following steps: S1, receiving a system demand described by a user through a natural language; s2, analyzing the requirements through an AI large model and converting the requirements into formalized specifications; s3, automatically generating a system architecture and dividing components; s4, generating each component code in a recursive manner; s5, dynamically detecting and repairing the generated code problem; s6, automatically integrating all components to form a complete system; and S7, continuously optimizing the system based on system feedback or user feedback. According to the method, work of multiple stages such as demand analysis, architecture design, module division, code writing and testing is automatically completed, distributed integration and smart contract are performed, distributed integration verification is performed through a block chain smart contract and a graph database, data consistency, interface matching and security among components are ensured, the integration process is optimized, and the integration efficiency is improved. Cross-domain cooperation is achieved, and the development efficiency is high.