122 results about "Adaptive computing" patented technology

The invention discloses an intelligent manufacturing workstation safety control system oriented to man-machine cooperation, and belongs to the technical field of intelligent manufacturing and man-machine cooperation. The system comprises a multi-modal sensor array, a personnel state sensing module, a robot state monitoring module, a dynamic safety area adaptive calculation module, an intention prediction and risk assessment module, a cooperative control decision module and an execution feedback module. The dynamic safety area self-adaptive calculation module constructs geometric representation of a man-machine interaction space based on a topological manifold theory, and calculates a dynamic safety boundary in real time; the intention prediction and risk assessment module is fused with multi-source information to predict a future trajectory of a person and assess a collision risk; the cooperative control decision module performs multi-objective optimization with safety and efficiency as double objectives to generate an optimal adjustment strategy, the system realizes prospective safety control based on intention prediction, the cooperative efficiency is improved by 25%-45% on the premise of guaranteeing safety, and the working safety accident rate is reduced to be close to zero.

The invention discloses an industrial Internet of Things federated learning method based on a federated increment decision tree, and the method comprises the steps: a cloud server deploys and initializes a federated learning global model and the federated increment decision tree, and sets a statistical histogram bucket boundary set of each feature value; in each federated learning iteration, the cloud server broadcasts a federated learning global model parameter, each industrial device adopts a local data set to train and count to obtain a local gradient histogram parameter, the edge server performs local aggregation on the local gradient histogram parameter and the federated learning model parameter, and the edge server performs local aggregation on the local gradient histogram parameter and the federated learning global model parameter; and the cloud server globally aggregates the local aggregation parameters of the gradient histogram and then incrementally trains the federated increment decision tree, simultaneously aggregates the parameters of a federated learning global model, and adaptively calculates an aggregation weight based on a second-order gradient value during local aggregation and global aggregation of the parameters of the federated learning model. The federal learning overhead can be effectively reduced, and the performance of the federal learning model is improved.

The invention discloses an unmanned aerial vehicle performance picture generation method, device and equipment and a medium, and the method comprises the steps: directly generating a line draft image according with an unmanned aerial vehicle performance specification through a multi-modal generation model trained based on a special data set according to text or image input. And carrying out structural analysis on the line draft, and separating a contour skeleton from an internal filling area. In the point distribution stage, key inflection points of the contour are innovatively recognized and locked as fixed anchor points, and it is ensured that geometric features are not lost in follow-up processing; and combining binary search to adaptively calculate an optimal spacing, and applying constraint space relaxation in a filling area. And finally, through local security assessment, minimum addition and deletion are carried out on point locations only on the security line segments, and the target sortie is accurately matched. The problem of low production efficiency caused by excessive manual operation is effectively solved, and the mass production requirement of high-frequency commercial performance is fully met.

The invention relates to the technical field of medical image processing and surgical navigation, in particular to a liver segmentation, modeling and surgical simulation system based on a cascade neural network and a multi-scale reconstruction technology. The multi-scale dynamic registration module is used for carrying out coarse segmentation and fine segmentation on liver and focus areas of the medical image and is used for non-rigid registration and respiratory motion compensation of the multi-modal medical image; the physical enhancement modeling engine is used for constructing a liver vessel tree three-dimensional model containing hemodynamic characteristics; the operation navigation and risk assessment module is used for performing real-time navigation and operation path planning in the operation; and the self-adaptive computing framework provides underlying computing resource support for the system. And high-precision segmentation: through a cascade recursive network and a boundary sensitive loss function, the liver tumor segmentation Dice coefficient reaches 94-96%, which is increased by 6-12% compared with the traditional method.

The invention discloses a secure closed-loop circulation and physical output control method and system for an oversized raster image. The method comprises the following steps: a client adaptively calculates fragmentation granularity according to a network state, fragments an image and encrypts and transmits the image; the server performs secondary encryption and distributed storage on the received data; the preview end constructs a multi-resolution hierarchy based on a viewport mapping technology, only extracts visual slices and superposes watermarks, and transmits the visual slices and the watermarks to a client memory for zero-cache rendering; and the printing end constructs a secure channel bound with a hardware fingerprint, carries out streaming decryption on the data, embeds traceability information, and injects a printing drive in real time. According to the method, the problems of unstable GB-level image transmission and unsmooth preview are solved, and high-efficiency circulation and strict protection of design assets are realized through full-process data non-landing streaming control.

The invention discloses an element size effect elimination method for concrete structure damage finite element simulation, which comprises the following steps of: 1, performing finite element mesh generation on a concrete structure, endowing all elements with a concrete damage zone model based on a power law type tensile damage equation and initial model parameters, and establishing a concrete structure damage finite element model; step 2, obtaining size information of each unit in the concrete structure damage finite element model in the step 1, and adaptively calculating model parameters of each unit related to the size of the unit based on the size information of the unit; and step 3, simulating a concrete structure loading process by using the model parameters calculated in the step 2 to obtain a concrete structure damage finite element simulation result in which the unit size effect is eliminated. The method breaks through the bottleneck that the unit size effect in concrete structure damage finite element simulation cannot be eliminated in the prior art, and has wide application prospects in the process of objectively simulating concrete structure damage evolution.

The invention discloses an optical neural network multi-architecture adaptive computing chip and a computing method thereof. The chip comprises an optical signal input unit, an input data modulation unit, a weight and bias processing unit and a photoelectric detection unit. The optical signal input unit is used for generating and outputting an optical signal; the input data modulation unit is used for modulating and loading information carried by the source data and the bias data to an optical signal output by the optical signal input unit; the weight and bias processing unit comprises a photon weight module and a photon bias module, and the photon weight module is used for carrying out optical coding on weights and realizing optical multiplication and addition calculation of input data of optical signals and corresponding weight parameters; the photon bias module is used for bias coding, summing the bias coding and an optical multiplication and addition operation result, and outputting an operation result containing bias compensation; the photoelectric detection unit is used for converting operation results into electric signals to be output. The method has high universality, and the calculation accuracy of the optical neural network is effectively improved.

The present application relates to the technical field of industrial visual inspection, and specifically relates to a visual inspection system and method with differentiable imaging parameters and adaptive computing power, which mathematically differentiates the physical camera imaging process into a differentiable computation graph node, constructs an end-to-end optimization link of "differentiable imaging-inference", combines multi-modal closed-loop control based on a three-dimensional state space of "speed-temperature-confidence", realizes adaptive allocation of computing power, and compensates for light distribution through ROI partition and adaptive light control; the visual inspection system and method with differentiable imaging parameters and adaptive computing power can automatically search for optimal imaging parameters for a detection task, solve the defects of traditional visual system imaging and algorithm fragmentation and rigid computing power allocation, significantly improve the defect detection rate under complex light scenes, reduce the overkill rate and equipment operating temperature, balance the detection real-time performance and accuracy, and adapt to the dynamic working condition requirements of industrial production lines.

The application discloses an intelligent dynamic grid self-adapting method and equipment for droplet pool simulation of additive manufacturing, and the method comprises the following steps: establishing a simulation model of additive manufacturing by using a pool simulation tool; extracting the simulation real-time gas phase fraction gradient, solid phase fraction gradient, solute concentration gradient and pool temperature gradient of each grid at each time step of simulation as the perceived physical parameters; adaptively calculating the encryption urgency index of each grid according to the perceived physical parameters; obtaining the historical encryption urgency index of each grid, calculating the second-order difference value, and predicting the change trend of the encryption urgency index of each grid at the future time; determining the encryption threshold and the coarsening threshold according to the encryption urgency index of all grids at the current time step, and combining the change trend of the encryption urgency index, the size relationship between the encryption urgency index and the threshold to determine whether the grid needs to be encrypted or coarsened; and iterating until the simulation is completed. The application has high precision, low cost and high automation degree.

The application discloses a charged water body equivalent impedance model construction and parameter acquisition method and related devices, and innovatively establishes a calculation system containing resistance and inductance correction factors, and realizes high-precision characterization of water body impedance characteristics by introducing an electrode spacing function. Specifically, a scalable grid technology is used to construct an adaptive calculation framework, and the calculation accuracy of resistance and inductance parameters is significantly improved through a power function correction algorithm. A reliable numerical calculation is provided for water body impedance network model analysis in complex hydrological environments. The application solves the problems of the prior art, such as complex modeling process, difficulty in accurately describing nonlinear behavior, insufficient precision, and inability to accurately capture the real behavior of water body.

The invention belongs to the technical field of intelligent monitoring, and provides a self-adaptive node addressing method and system suitable for a multi-node intelligent equipment system.The technical scheme includes the steps that an addressing instruction is received, and a search initial address range is obtained; taking the search initial address range and the maximum effective device number as input, and under the constraint of the size of an available stack memory, performing adaptive calculation to obtain a fuzzy dividing and conquering factor for dividing the search range and fine scanning granularity for fine search; the method comprises the following steps: initializing an address stack, pressing a search initial address range into the address stack, and circularly executing the following steps: sending a popped query instruction of a stack top address range, dynamically matching a corresponding branch processing mode according to a received response condition until the address stack is empty, and obtaining an addressing result of each node. The system deployment speed is improved, the communication overhead is reduced, and the stability in a complex environment is enhanced.

The embodiment of the invention provides a communication method, and the method comprises the steps: a first computing node subscribes to a mobility event of a terminal device, and determines that a device providing an access service from a first computing service of the terminal device is switched from a first network device to a second network device through second information, and the first computing node can release the first connection between the terminal equipment and the first computing node by sending the third information, so that the situation that a path is not good or unreachable due to the fact that the terminal equipment still tries to be connected with the original computing node is reduced, and the terminal equipment can be conveniently connected with the adaptive computing node subsequently.

This invention discloses a robot environment perception method and system, belonging to the field of intelligent robot technology. The method and system collect environmental perception data through multiple perception links; establish a dynamic confidence model for each perception link and calculate the dynamic confidence level in real time; construct an environmental complexity assessment module to calculate the environmental complexity index in real time; and dynamically adjust the perception mode and fusion strategy within an adaptive fusion framework based on the dynamic confidence level and environmental complexity index. The adaptive fusion framework includes a predictive mode switching layer, a confidence-weighted fusion layer, and an adaptive computational resource scheduling layer. A navigation module receives the fused data and performs path planning and motion control. This achieves predictive switching of perception modes and confidence-weighted fusion, enhances system robustness, optimizes computational resource utilization, and improves the robot's autonomous navigation capability in complex dynamic environments.

The invention relates to the technical field of storage and calculation integrated chips, and provides a dynamic reconstruction neural network calculation structure and method based on a memristor array. The method aims at solving the problems that an existing memristor neural network computing architecture is rigid in structure and cannot adapt to computing characteristics of different network layers, so that the resource utilization rate is low, invalid computing is frequent, and energy efficiency is reduced. According to the main scheme, the reconfigurable memristor array comprises a core reconfiguration unit composed of a plurality of reconfigurable memristor sub-arrays, each sub-array is integrated with a memristor cross array, a configurable switching network, an input / output buffer and local control logic, and three working modes of high parallelism, high density and sparse calculation are supported; the core units are connected through the configurable internet, so that dynamic switching of parallel, cascade or hybrid topology is realized; and a mode scheduling controller is matched to analyze network layer characteristics and issue a reconstruction instruction, and a data flow control module coordinates data distribution and synchronization, so that efficient self-adaptive calculation of a convolutional layer, a full-connection layer and a sparse layer is realized.

The present application relates to the technical field of GPU computing optimization, and in particular to a sparse attention computing method, device and medium for GPU, wherein the method realizes the high efficiency of long context reasoning through the geometric perception sparse attention framework of ball hashing, and combines a large-scale parallel hashing optimization algorithm and a load adaptive computing kernel. Compared with the existing sparse attention methods based on heuristics or gradient learning, the present application realizes higher retrieval recall rate, lower preprocessing overhead and efficient hardware adaptation to irregular sparse patterns.

The application discloses a long context processing robust enhancement method and system based on neuron dynamic adjustment, and belongs to the technical field of text sequence processing, and comprises the following steps: embedding a nerve regulation unit in each Transform layer of a large language model used for a long context processing task; and the nerve regulation unit sequentially performs the following steps: dynamically activating abnormal perception, that is, adaptively calculating an abnormal detection threshold of a current layer according to activation value statistical characteristics of the current layer; activating a hard-gate truncation, that is, generating a binary hard mask according to the abnormal detection threshold, and the binary hard mask realizes deterministic isolation of abnormal components in the activation value while keeping normal components unchanged; and performing hierarchical response compression, that is, performing learnable dimension-level linear compression on the isolated abnormal components, re-integrating the compressed abnormal components and the normal components through gate fusion, and outputting an activation tensor after suppressing the abnormality. The application can prevent rank collapse and gradient oscillation, and enhance the robustness of long context task execution.

The invention belongs to the technical field of slope engineering and earthwork calculation, and particularly relates to a self-adaptive calculation method and system for external normal vectors in slope earthwork calculation. The method comprises the following steps: obtaining and verifying the effectiveness of a slope closed boundary polyline; adopting a hybrid strategy to adaptively judge the polygon winding direction; calculating a line segment direction vector and an initial normal vector thereof, and performing weighted smoothing processing; obtaining an outer normal vector according to winding adjustment, unitizing the outer normal vector, and verifying correctness; and finally, calculating the earthwork volume by using a curvature-corrected segmented integration method. The system correspondingly comprises a boundary data reading and verification module, a winding judgment module, a direction vector calculation module, an initial normal vector smoothing module, a normal vector adjustment module, a unitization and verification module, an earthwork calculation module and the like. According to the method, the external normal vector can be determined in a self-adaptive and high-precision mode, and the accuracy and reliability of complex slope earth volume calculation are effectively improved.

The application discloses a ground-non-ground fusion network high-speed terminal seamless switching method and system based on trajectory prediction and resource reservation, and belongs to the technical field of space-ground integrated communication. The core of the method is: through bidirectional motion prediction of the terminal and the NTN node, the switching area is accurately judged; the terminal speed, network delay and other factors are combined to adaptively calculate the dynamic switching window; the AI-enabled link scoring model is used to intelligently select the optimal target link; and the resource is cooperatively pre-reserved and uplink and downlink synchronization is completed before the window. The application has the beneficial effects that: the failure problem of traditional RSRP judgment in the NTN dynamic environment is overcome, the switching success rate and prediction accuracy in the high-speed mobile scene are significantly improved, the service interruption time and signaling overhead are greatly reduced, the continuity and reliability of communication are effectively guaranteed, and the application is suitable for intelligent networked vehicles, unmanned aerial vehicles and other high-dynamic service scenarios in 6G networks.

This invention discloses a vibration signal denoising and feature extraction method based on wavelet adaptive thresholding, belonging to the field of vibration sensor signal processing technology. The method includes the following steps: Step 1, performing multi-level decomposition of the signal using a filter bank constructed based on discrete wavelet transform; Step 2, adaptively calculating the noise threshold; Step 3, implementing wavelet coefficient shrinkage denoising based on the SureShrink algorithm; Step 4, reconstructing the vibration signal based on quasi-discrete wavelet transform; Step 5, automatically extracting signal features based on a multi-layer recurrent neural network. This invention, through signal decomposition based on discrete wavelet transform and an adaptive thresholding denoising algorithm, reduces the error impact caused by the vibration sensor's own noise and environmental disturbances, retains most of the features of the original vibration signal, and establishes a multi-layer recurrent neural network for the denoised vibration signal to achieve automatic extraction of time features, providing an important foundation for the subsequent development of fault diagnosis and health monitoring algorithms.

This invention discloses a method for improving the accuracy of rice irrigation forecasts in irrigation districts based on digital twins, belonging to the field of agricultural water conservancy engineering technology. Based on digital twin technology, this invention divides the irrigation district into sub-regions using GIS and constructs a digital mapping model; it establishes a dynamic adaptive calculation system with multi-model selection, combining meteorological data completeness, rice growth period, and historical errors, and calculates the appropriate model using a dynamic weighted adaptation algorithm, obtaining the final value from the input data; it introduces soil moisture and field management correction coefficients to construct a dynamic calculation model; combined with the calculation of real-time water requirements for rice, it generates a refined irrigation scheduling plan for the sub-regions and achieves coordinated irrigation; and it corrects parameters using a gradient descent algorithm to ensure forecast accuracy. This method, through its dynamic adaptive calculation system, solves the accuracy problem in scenarios with missing data and has broad application value.

This invention discloses a resume screening and display method and system for human resource management. It constructs a keyword-dimension mapping table based on job description text using a natural language processing engine; analyzes keyword frequency, syntactic position, modification strength, and semantic relevance to generate a job requirement intensity vector; retrieves historical recruitment data based on job meta-information to generate a historical weight vector; adaptively calculates a smoothing coefficient based on the amount of historical data; and integrates dual-source weights to generate a dynamic weight vector; standardizes the scoring of each dimension of the resume, using a normalized formula based only on valid dimensions to calculate the matching degree, avoiding score distortion caused by missing fields; and finally displays the results sorted by matching degree. This invention abandons the preset weight mode, achieving dynamic screening with a tailored approach for each job, improving matching accuracy, operational efficiency, and the quality of human-machine collaborative decision-making, providing a scientific, efficient, and transparent intelligent solution for human resource management.

The invention provides a computing power matching method and system for large-model cross-modal retrieval, and relates to the field of large-model technology application, and the method comprises the steps: carrying out scene computing power demand mapping processing on a large-model cross-modal retrieval demand, and obtaining a scene computing power demand result containing a basic computing power demand and a dynamic computing power demand; constructing a resource supply capability portrait of each computing power node in a preset computing power pool, executing double-loop adaptation adjustment processing based on the result to obtain an inner-loop local adaptation conclusion and an outer-loop global adaptation conclusion, generating a computing power adaptation judgment conclusion according to a double-loop adaptation result, and determining the computing power adaptation according to the computing power adaptation judgment conclusion. According to the method, a scene-based computing power demand is determined to indicate the adaptation state of the scene-based computing power demand and the resource supply capability, finally, a computing power scheduling instruction is generated according to a determination conclusion, and computing power tasks are accurately allocated to adaptive computing power nodes, so that the efficiency and performance of large-model cross-modal retrieval are effectively improved, and the computing power cost is reduced.

The present application relates to the technical field of automation plate processing and carrying control, and particularly relates to a plate automatic feeding and discharging control system and a control method thereof, which comprises: a task analysis module, which receives an external production task and calculates the external production task into processing procedure parameters containing plate geometric size, thickness and material density characteristics; and a gravity calculation module, which calculates plate mass distribution based on the processing procedure parameters and outputs gravity matrix data containing target gravity coordinates, suction cup matrix gating strategy and plate specification characteristics. In the present application, through automatic detection and calculation of plate geometric characteristics and density information, on-demand dynamic gating of the suction cup array and adaptive calculation of the vacuum negative pressure threshold are realized, frequent parameter adjustment caused by plate specification changes in the traditional manual teaching mode is avoided, and the automatic adaptation capability of the feeding process to multi-specification plates is improved.

The invention discloses a multi-modal event analysis method and system for realizing efficient calculation, and belongs to the field of artificial intelligence and edge calculation. The method comprises the following steps: receiving data streams from a plurality of heterogeneous sensors; calculating a cross-modal confidence score reflecting the consistency between the data streams by a fusion engine; the method is characterized in that a calculation regulator dynamically selects an optimal processing stage from a recognition engine with a plurality of hierarchical processing stages according to the confidence score; and processing the data and generating event classifications using only the selected stages. According to the method, a feedback control loop from the multi-modal confidence coefficient to the calculation depth is established, adaptive calculation of deep analysis of fuzzy and conflict data and rapid processing of clear and consistent data is achieved, the average calculation load is remarkably reduced, and meanwhile the accuracy and reliability of an analysis result are greatly improved.

The invention relates to a fragmented transmission generation method and system for Internet of Things low-orbit satellite terminal data, and the method comprises the steps: actively sensing and quantifying the state of a data queue through an Internet of Things terminal, calculating and reporting a transmission priority and a local resource demand degree, converging all terminal information through a low-orbit satellite node, and transmitting the collected terminal information to the Internet of Things terminal. An algorithm based on cooperative multi-agent reinforcement learning is utilized to carry out joint optimization scheduling, a resource allocation strategy including fragment granularity guidance is generated and issued, and the Internet of Things terminal adaptively calculates the size of an optimal fragment according to the resource allocation strategy, cuts data and packages the data into data fragment units which are provided with priority labels and can be independently transmitted, and sends the data fragment units to the Internet of Things terminal. And finally, sending is completed on the precise resources allocated by the satellite. According to the method and the device, cross-layer accurate matching of data generation and resource scheduling is realized, the system throughput is remarkably improved, the service time delay is reduced, and fairness among massive terminals is guaranteed.

This invention relates to a parametric adaptive density filtering method and apparatus for point cloud data of grain pile volume in silos. The parametric adaptive density filtering method for point cloud data of grain pile volume in silos includes the following steps: S1, acquiring point cloud data of the grain pile volume measured inside the silo; S2, generating random samples from the point cloud data using the Monte Carlo method, and calculating the optimal total number of clusters K based on the random samples. best Then according to K best The adaptive mean Euclidean distance S3 is calculated first based on the calculated global density mean of the point cloud data and the local densities ρ1 to ρ of the N original points. N ; and then based on ρ1~ρ N The adaptive density threshold ρ is calculated. y This invention employs clustering to adaptively calculate the neighborhood radius parameter. Then, by calculating the neighborhood density of each original point and using statistical characteristics to dynamically adjust the adaptive density threshold, low-density noise points are eliminated, thereby ensuring the accuracy of point cloud data filtering and improving the accuracy of subsequent grain storage calculations.

The invention discloses a current transformer adaptive compensation and protection strategy dynamic adjustment method and system, which can accurately identify the saturation state of a current transformer, adaptively calculate a compensation current capable of suppressing the current saturation of the current transformer according to the real-time saturation state of the current transformer, and meanwhile, automatically adjust the protection strategy of the current transformer. A reliable current data source is provided for an external protection device according to the saturation state of the current transformer, and a corresponding protection strategy adjustment instruction sent to the external protection device is dynamically adjusted; therefore, it is ensured that the CT can still keep high measurement precision under the large fault current, the reliability, selectivity and adaptive capacity of an external protection device are comprehensively improved, and the development requirements for safe, flexible and efficient operation of a power grid in the future are met.

The invention relates to the technical field of JESD link dynamic adjustment, in particular to a JESD link dynamic adjustment method and system based on parameter adaptive calculation, and the method comprises the steps: calculating an ADC sampling clock frequency, an SYSREF frequency and a GT reference clock frequency according to a target sampling rate, calculating a target VCO frequency according to the ADC sampling clock frequency, the SYSREF frequency and the GT reference clock frequency, and when the target VCO frequency belongs to a VCO frequency range, determining that the target VCO frequency is within the VCO frequency range; performing frequency division coefficient verification on the target VCO frequency, performing crystal oscillator frequency verification on the target VCO frequency to obtain a crystal oscillator frequency verification result when the frequency division coefficient verification result is that the frequency division coefficient meets the chip specification requirement, and performing feasibility judgment on the QPLL linear rate according to the verification result when the crystal oscillator frequency verification result is that the crystal oscillator frequency verification is passed. And obtaining a feasibility judgment result, and when the feasibility judgment result is passed, carrying out automatic reconfiguration and link reconstruction on the JESD link. According to the invention, the flexibility and reliability of JESD link adjustment can be improved.

The present disclosure sets forth systems, apparatuses, and methods that autonomously adapt networks to acquire additional capabilities and / or optimize current capabilities. An example system comprises processors and memory that stores instructions to be executed to monitor first performance metrics of a first network, determine, based on the monitored first performance metrics, network patterns indicative of network inefficiencies or inabilities, determine, based on the determined network patterns, a network mutation, create, based on the first network and the network mutation, a second network, compare the first performance metrics of the first network to second performance metrics of the second network, and based on determining that the second performance metrics are an improvement over the first performance metrics by a threshold amount, apply the network mutation to the first network.

The invention discloses a monitoring and log integrated fault diagnosis system based on dynamic time window association, which is suitable for operation and maintenance of a data center and a GPU (Graphics Processing Unit) cluster. When the monitoring index is triggered abnormally, the system adaptively calculates a retrieval window according to a sampling period, log delay, network jitter, clock skew and the like, automatically retrieves and aggregates related logs in the window, and generates a structured associated event. Calculating correlation degree through template matching, semantic similarity and time decay, and outputting confidence and candidate root causes; gPU temperature, power consumption, a video memory and PCIe errors are fused to construct a causal chain, early warning and self-healing are realized, a maintenance suppression window is provided for noise reduction, and positioning efficiency and stability are improved.