121results about How to "Improve computing power" patented technology

The invention provides a collaborative computing method, system and device based on SIMD and SIMT, and belongs to the technical field of computer processing. The method is applied to a Warp composed of a plurality of SPs, and the Warp comprises an SIMT front end and an SIMD rear end. Each SP corresponds to a vector register combined by a plurality of short registers; generating instruction fetching indication information under the condition that a thread bundle scheduler at the front end of the SIMT determines that a first thread bundle corresponding to the calculation task is in an activated state; obtaining a first cache instruction corresponding to the first thread bundle through an instruction fetching unit, and sending the first cache instruction to a decoding unit at the front end of the SIMT for analysis; obtaining one or more corresponding required operands through an operand collection unit at the rear end of the SIMD; and distributing the required operands to the corresponding calculation management units through the SIMD rear end to execute the corresponding calculation tasks, and writing the calculation results corresponding to the calculation tasks back to the storage unit through the write-back unit.

The invention relates to the field of nonvolatile memories, in particular to an SOT-MRAM circuit unit based on a 2T2M differential structure and a storage array circuit of the SOT-MRAM circuit unit. The SOT-MRAM unit comprises a first magnetic tunnel junction, a second magnetic tunnel junction, a first field effect transistor, a second field effect transistor, a first phase inverter, a second phase inverter, a write line, a read line, an input line and an output line which have a specific connection relation. According to the SOT-MRAM circuit unit, through the special 2T2M differential structure design and in combination with the excellent performance of a spin-orbit moment magnetic tunnel junction, the storage and operation performance and efficiency of a 2T2MSOT-MRAM differential circuit unit array are integrally improved, the dependence on the performance of the MTJ is reduced, the SOT-MRAM circuit unit has the better energy efficiency ratio and the more stable read-write speed and precision, and the SOT-MRAM circuit unit has the advantages of being simple in structure, low in cost and high in reliability. And the reliability and the stability are excellent.

The invention relates to a homology analysis method and device for binary code files and computer equipment. The method comprises the following steps: obtaining and identifying a function call graph corresponding to each binary code file through a security monitoring program; identifying isomorphic sub-graphs among the binary code files through a hybrid search strategy, and performing symbol program slicing processing on respective write function nodes in each isomorphic sub-graph to obtain symbol program slicing codes corresponding to the respective write function nodes of each isomorphic sub-graph; performing escape processing on each symbol program slice code corresponding to the isomorphic sub-graph to obtain each piece of mathematical semantic information corresponding to the isomorphic sub-graph, and performing similarity calculation to obtain a code similarity value between the binary code files; therefore, the homologous binary code file is identified. By adopting the method, the homology analysis accuracy of the highly confused binary code file can be improved.

The invention discloses a power grid supply chain distribution network protocol inventory material demand prediction method based on a time sequence model, and belongs to the field of data processing. Comprising the steps of data collection and preprocessing; determining a dimension establishment model; model prediction; and correcting the model. The method comprises the following steps: introducing a time sequence model into power grid supply chain demand prediction, and establishing a time sequence model suitable for power grid supply chain distribution network protocol inventory material demands; by means of a big data analysis method and an artificial intelligence algorithm, an association rule of project and demand prediction is mined, a material demand characteristic portrait is formed by using a data analysis method, and through historical data cleaning, model establishment and event management, the material characteristic portrait is carded, analyzed, diagnosed and evaluated in an all-around and multi-view manner. The operation capability and the data applicability of an algorithm model library are improved, a material demand prediction mode is optimized and evaluated, and a support is provided for a power grid supply chain material purchase supply decision. The method can be widely applied to the field of power grid supply chain distribution network protocol inventory material demand prediction and management.

The embodiment of the invention provides a hardware processing device for a digital signature algorithm. The device comprises a control logic module, a hash module, a lattice calculation module, a signature logic module and an authentication logic module, the control logic module is used for receiving input data and inputting the input data into the hash module to generate a random polynomial; inputting the random polynomial generated by the hash module into a lattice calculation module for lattice operation; calling the signature logic module to sign the lattice operation result to generate a signature value; and calling the authentication logic module to verify the validity of the signature value. According to the embodiment of the invention, the lattice hash module and the calculation module are independently provided instead of being placed in the signature logic module and the authentication logic module, so that the control logic module can directly call the hash module to generate the random polynomial and call the lattice calculation module to carry out lattice operation, the utilization rate of resources is improved, the operation speed of an algorithm is optimized, and the operation efficiency is improved. And the overall calculation performance is improved.

The invention discloses a three-dimensional heterogeneous integrated optical computing chip, and belongs to the field of optical computing. The computing chip consists of three parts: a spatial diffraction optical neural network core particle, a plane matrix reconfigurable optical neural network component and a plane switchable optical switch module, the single or multiple spatial diffraction optical neural network core particles, the plane matrix reconfigurable optical neural network assembly and the plane switchable optical switch module are combined together in a three-dimensional heterogeneous integration mode, and efficient multi-task optical calculation processing is achieved. The spatial diffraction optical neural network and the plane matrix reconfigurable optical neural network are combined with each other, a traditional optical calculation framework is broken through, the method has wide application prospects in the optical calculation field, and the blank of related technical fields is filled.

The embodiment of the invention provides a secret sharing and packaging method and device. And the plurality of participants respectively encode vectors formed by secret sharing of k pieces of data owned by the participants in a specified domain space to respectively obtain first domain elements owned by the participants, and the first domain elements have data structures meeting addition homomorphism. Then, the multiple participants perform secret sharing addition operation based on first domain elements owned by the multiple participants through first data interaction based on secure multi-party calculation, so that the multiple participants respectively obtain secret sharing of first sum values of the first domain elements owned by the multiple parties; the secret sharing of the first sum value is used as the secret sharing of a second domain element obtained by coding a vector formed by k data in a directional domain space.

The invention discloses a multi-bus decoupling distributed embedded hardware rendering architecture, system and method, and relates to the technical field of ultra-high-definition video hardware processing, and the key points of the technical scheme are that the architecture comprises a multi-bus decoupling module which comprises a plurality of hardware buses which are physically isolated from one another, each hardware bus is an exclusive hardware transmission link bearing a single-type manufacturing task; the modular processing unit is an independent hardware board card and is integrated with an external signal input port, an upstream bus signal input port and a downstream bus signal output port; the plurality of modular processing units are cascaded in sequence through a hardware bus in the multi-bus decoupling module to form a distributed assembly line rendering link; when the modular processing units are newly added, the signal processing capability and the rendering synthesis capability are linearly improved along with the increase of the number of the modular processing units. According to the invention, the problem of performance bottleneck of the traditional architecture in processing 4K / 8K ultra-high-definition and multichannel rendering is fundamentally solved.

The invention provides a DNN model-based division and reasoning acceleration method and system for vehicle-mounted edge intelligence. The method comprises the following steps: constructing a DAG graph based on a vehicle-mounted DNN model; constructing an auxiliary network flow graph of the DAG; adding a head node representing local calculation of a vehicle and a tail node representing calculation of an edge server or an auxiliary vehicle, and setting the weight of an edge as time delay overhead for segmenting the edge; obtaining different layering modes of an auxiliary network flow diagram of the DAG and corresponding maximum tolerance point sets meeting local time delay, memory and energy consumption limitation of the vehicle; correspondingly modifying the weight of the edge in the auxiliary network flow graph of the DAG based on each maximum tolerance point set, and calculating the maximum flow of each modified network flow auxiliary graph; and selecting a division mode corresponding to the maximum flow as a DNN model division mode. According to the method, the vehicle and road cloud cooperative reasoning is realized through a DNN model division mode, so that the DNN task cooperative reasoning completion time delay is reduced, and the vehicle safety and the user experience are improved.

The invention discloses a GPGPU thread block synchronization method and device based on a thread lock and a medium, and relates to the technical field of electric digital data processing. The method comprises the following steps: a thread block scheduling module allocates thread blocks for each stream multiprocessor and calculates a thread start ID (Identity) to maintain a corresponding thread lock table; analyzing whether a dependency relationship exists between the memory access request and a thread in the currently occupied thread block or not under the condition that the memory access requirement exists, and if yes, adding the memory access request with the dependency relationship to a lock queue; querying a thread lock state of a thread having a dependency relationship with the memory access request in a corresponding thread lock table so as to add the thread lock state to an entry state corresponding to the lock queue; and dequeuing the memory access request from the lock queue under the condition that the entry states of all the threads with the entries in the lock queue and the thread bundle having the dependency relationship are unlocked, and performing memory access according to the memory access request to realize synchronization of GPGPU thread blocks.

The invention relates to a remote sensing image transmission tower detection method based on MFF-YOLO, and the method comprises the following steps: S1, obtaining a slice data set of a satellite image of a region, screening slice data containing a transmission tower as a data set, and carrying out the image preprocessing; s2, labeling the power transmission tower by utilizing Labelimg software, and obtaining a data set label in an xml (Extensible Markup Language) format; s3, constructing a power transmission tower detection network model based on MFF-YOLO; s4, training and verifying the transmission tower detection network model based on the constructed data set; and S5, on the basis of the trained transmission tower detection network model, detecting a satellite image of a to-be-detected area. According to the invention, the available computing resources can be utilized more efficiently to detect the transmission tower target, and the efficiency and accuracy of transmission tower identification are improved.

The invention provides a navigation computer for a small inertial navigation system, the size of the navigation computer is only 55mm * 86mm * 16mm, a navigation resolving module, an I / F conversion module and a power supply module are integrated, and the power consumption is less than 3W; the navigation resolving module takes a ZYNQ7000 all-programmable SoC as a core processor, and has the functions of inertial device original data acquisition, navigation information preprocessing, navigation resolving and the like; the I / F conversion module can directly convert a current signal of the accelerometer into a pulse signal, and the pulse signal is accessed to the navigation resolving module for navigation resolving; the power supply module can output + 5V (8W),-5V (3W), + 15V (9W) and-15V (3W) isolated voltage-stabilized power supplies to the outside, and can supply power to other loads of the inertial navigation equipment. The method is suitable for laser inertial navigation equipment with small volume, low power consumption and high reliability.

The invention belongs to the technical field of remote sensing, and particularly relates to a nuclear power plant surrounding ground feature element extraction method based on deep learning, and the method comprises the steps: 1, making a nuclear power plant surrounding ground feature deep learning sample; step 2, based on the HRNet model, constructing a Mama attention residual module, an inner convolution residual module and an asynchronous convolution residual module through Mama, inner convolution and heterogeneous convolution, and constructing an improved HRNet model; 3, training of a U-Net model, a Deeplabv3 model, a HRNet model, a Swin Transformer model and an improved HRNet model is completed through the sample set training data and the verification data; 4, evaluating land element extraction results of different deep learning models; and 5, extracting elements of the surrounding land of the large-scale nuclear power plant by using the trained improved HRNet model. The method can effectively improve the extraction precision of complex ground feature elements around the nuclear power plant, remarkably improves the calculation efficiency of the model, and is suitable for efficient monitoring and classification of multiple ground feature elements around the nuclear power plant in a large range.

The invention provides a wireless power supply mobile edge computing (WPT)-MEC (wireless power supply mobile edge computing) optimization algorithm assisted by a user cooperation (UC) and a non-orthogonal multiple access technology (NOMA) in a non-linear energy harvesting (EH) mode, the transmission power, the user unloading time and the user unloading power among different energy beams of the HAP are jointly optimized, the energy and the computing resources are fully utilized, the energy utilization rate of the HAP is improved, and the energy utilization rate of the HAP is improved. According to the method, the system weight sum calculation rate (WSCR) is maximized, and the system weight sum calculation rate (WSCR) is maximized. The model is non-convex, so that a formulation problem is difficult to solve. In order to solve the problem, an iterative optimization algorithm based on successive convex approximation (SCA) is provided, firstly, a variable substitution method is utilized to re-express a non-convex optimization problem, the non-convex optimization problem is converted into a form which is easier to process, and then the SCA is utilized to solve the non-convex optimization problem. A numerical result shows that compared with a reference scheme, the scheme provided by the invention can generate better system performance.

The invention belongs to the technical field of data center cooling, and relates to a modular self-adaptive immersion jet liquid cooling system for a high-power server, the high-power server comprises a case, the case is provided with a sealed cavity, the sealed cavity is internally provided with a plurality of devices with different heating degrees, the devices are completely immersed in a refrigerant, and the components are arranged in the sealed cavity. A modularized jet flow device is arranged on a device with the relatively high heating degree, the modularized jet flow device independently conducts jet flow cooling on the device with the relatively high heating degree, and the sealing cavity is further provided with a machine box refrigerant inlet and a machine box refrigerant outlet. And the case refrigerant inlet and the case refrigerant outlet penetrate from the inside of the sealing cavity to the outside of the case. The invention aims to solve the problem that high-efficiency and accurate heat dissipation of high-power heating elements such as chips cannot be realized in the prior art.

The application discloses a kind of highway maintenance digital twinborn application system construction method and system, it is related to digital twinborn technical field, the steps of this method include: real-time acquisition of the multi-source perception data of highway section, and introduce constraint engine, to highway three-dimensional entity construction, generate several three-dimensional sub-image model;Among them, multi-source perception data includes highway image, highway laser point cloud and component attribute;Based on three-dimensional sub-image model, abstractly constructed to form topological space, adopt interpolation algorithm to generate datum curve, and combine the grid vertex of three-dimensional sub-image model, identify envelope boundary, to execute cross section scanning and generate target rendering road;Based on target rendering road, identify disease area, and based on disease area introduce preset maintenance rule, through semantic mounting mapping back to topological space, form digital twinborn application system;The application improves the efficiency of highway maintenance management, improves the computing power and rendering performance of three-dimensional rendering.

The invention provides a Raft election strategy-based stable federated learning method and related equipment, and relates to the field of federated learning. The method comprises the following steps: dividing all equipment participating in federated learning into a plurality of clusters, and electing proxy equipment in each cluster based on a Raft algorithm; judging whether the proxy equipment is online or not, if yes, continuing training, and if not, electing the proxy equipment again based on the Raft algorithm; the server issues the global model obtained by the last round of aggregation to the proxy device; the proxy device distributes the received global model to each device in the cluster, and each device in the cluster performs local training and uploads a local training result to the proxy device; the proxy device aggregates local training results in the cluster and uploads a cluster model obtained by aggregation to a server; the server aggregates the cluster models of the clusters to generate a global model; and repeating the training process until a stop condition is reached. The method can effectively improve the stability and efficiency of the federal learning process.

The application provides a kind of deep learning optical imaging system design method based on complex function neural network, comprising the following steps: step one, object distance, system focal length, F number as input, according to the ideal optical system imaging relationship, determine image distance, imaging scale ratio;Step two, according to the imaging scale ratio and the object image of ideal optical system, the image of ideal optical system is calculated;Step three, construct complex function generator network model, step four, construct discriminator network model, step five, after the establishment, training of the above-mentioned complex function neural network deep learning model, step six, use phase.The deep learning optical imaging system design method based on complex function neural network provided by the application solves the correlation between the advantages and disadvantages of the design result and the experience value of the designer, without the designer providing initial structure parameters;Greatly improve the computing ability and optimization speed.

Provided are an optimization method and an optimization system for non-multiplication-addition type computing operations in an FPGA hardware accelerator. The optimization method comprises: generating a first computing graph by decomposing the non-multiplication-addition type computing operations into a plurality of basic operations; obtaining at least one merged operation by merging consecutive basic operations in the first computing graph; generating a second computing graph based on the first computing graph and the at least one merged operation; and executing the operation in the FPGA hardware accelerator according to the second computing graph; wherein the executing the operation in the FPGA hardware accelerator according to the second computing graph comprises: executing each merged operation of the at least one merged operation by pipeline optimization.

According to the top heat dissipation type liquid cooling container and the application control method thereof, the cooling tower container top arrangement design is adopted, a CDU module is omitted, the overall cost is reduced, more Tanks can be deployed in the spare space after the CDU module is omitted, and the computing power of a single liquid cooling container is greatly improved. The cooling tower is compact in structural form, low in space occupancy rate, higher in integration degree and good in universality and reliability, and the ground space of the cooling tower is saved. After a certain cooling tower fails, normal operation of other devices is not affected, and the disaster recovery capacity is improved. The cooling tower and the Tank carry out heat exchange one by one, the heat dissipation efficiency is higher, the number of water cooling pipelines is smaller, and the integration degree is higher.

This invention discloses a liquid-cooled heat dissipation chip structure and its packaging method, relating to the field of chip technology. The liquid-cooled heat dissipation chip structure includes: a chip with accommodating channels formed therein, and coolant filled into the accommodating channels; a substrate, on which the chip is packaged; and a cooling fan mounted on the side of the chip away from the substrate. This invention solves the problem of chip overheating leading to decreased chip computing performance and slowed data processing speed.

The present invention provides a method and system for generating high-frame-rate and high-resolution videos based on synchronous spatiotemporal super-resolution. The method comprises: extracting frames of a video file one by one to form a video frame sequence; extracting features from each pair of adjacent frames to obtain feature vectors, and obtaining inter-frame motion representation vectors through an inter-frame motion alignment network; using RBPN to super-resolve the adjacent frames to obtain high-resolution frame vectors and low-resolution frame vectors corresponding to the adjacent frames; using a spatiotemporal super-resolution network to generate high-resolution frame vectors and low-resolution frame vectors corresponding to inserted frames; optimizing the high-resolution frame vectors corresponding to the adjacent frames and the inserted frames to obtain target high-resolution vectors corresponding to the adjacent frames and the inserted frames and generating three target video frames by upsampling; and re-encoding the generated target video frames into high-frame-rate and high-resolution videos.

The application provides an activation module for data calculation, an AI processor and a data calculation method. When it is determined that a result calculated by a preceding matrix acceleration module meets an output condition, the output calculation data is stored in a cache unit. A read-write control unit reads the output calculation data from the cache unit and distributes the output calculation data to each calculation unit. Each calculation unit performs a softmax calculation and outputs a target calculation result. In this way, the output calculation data output by the preceding matrix acceleration module can be stored in the cache unit of the activation module. After processing by the read-write control unit and data distribution, the softmax calculation is performed in parallel by at least one calculation unit. The calculation and access of all data are performed in the activation module, reducing the overhead of accessing memory and improving overall calculation performance. The parallel calculation processing scheme can also improve calculation efficiency.

The embodiments of the present disclosure provide a data structure conversion method and an electronic device. The method includes: obtaining a first data structure; performing a first structure conversion on the first data structure based on a preset rule to determine a second data structure; obtaining thread information of a hardware execution unit, wherein the thread information is related to the total number of multiple threads; and performing a second structure conversion on the second data structure based on the thread information to determine a third data structure, wherein the third data structure is used for the hardware execution unit to access at least one element in the first data structure through multiple threads. The computing performance of neural network operations can be improved.

The invention provides a blast furnace condition diagnosis method based on a multi-time granularity dynamic reasoning mechanism, which comprises the following steps: S1, setting a plurality of time granularities, and establishing a furnace condition diagnosis rule base according to a preset rule; s2, collecting data in real time, storing the data in a dynamic database, extracting the data of the dynamic database every first time granularity, performing furnace condition diagnosis of the first time granularity according to rules of a furnace condition diagnosis rule base, and storing a diagnosis result in a process database; s3, calling data of the process database and the dynamic database every Nth time granularity, performing furnace condition diagnosis of the Nth time granularity according to rules of the furnace condition diagnosis rule base, and storing a diagnosis result in the process database; and S4, calling diagnosis results of different time granularities. According to the blast furnace condition diagnosis method with multiple time granularities, a dynamic reasoning mechanism can reduce a large amount of database read-write burden, and the calculation performance of a furnace condition diagnosis model is improved.

The application provides a synthetic aperture radar active composite jamming method and system, the method comprising: acquiring intermediate frequency radar signals to be jammed, radar related parameters and control parameters; calling DSP and multiple FPGA calculation boards to calculate modulation coefficients corresponding to each false scattering point according to pre-downloaded deception templates and radar related parameters; generating deception jamming signals corresponding to the intermediate frequency radar signals according to the intermediate frequency radar signals and the modulation coefficients corresponding to each false scattering point; generating suppression jamming signals corresponding to the intermediate frequency radar signals according to the control parameters; and determining composite jamming signals corresponding to the intermediate frequency radar signals according to the deception jamming signals and the suppression jamming signals corresponding to the intermediate frequency radar signals. The application can obtain composite jamming signals, has better jamming effect, is not easy to expose the position of the jammer, and has lower cost of the jammer. The signal processing architecture of DSP and FPGA is adopted, the calculation performance of the system is improved, and the real-time performance of system simulation is ensured.

The invention provides a configuration server. The configuration server comprises a case and a mainboard, a first installation area and a second installation area are arranged at the two ends of the case in the length direction respectively, and a first GPU set and a second GPU set which are electrically connected to the mainboard are arranged in the first installation area and the second installation area respectively. The first GPU group comprises at least eight first GPU cards, and the first GPU cards are horizontally installed in the first installation area; the second GPU group comprises at least eight second GPU cards, the total number of the first GPU cards and the second GPU cards is at least sixteen, and the second GPU cards are vertically installed in the second installation area. The installation areas are arranged at the two ends of the case respectively, the installation mode of horizontal and vertical combination is adopted, integration of sixteen GPU cards is achieved, the number bottleneck of existing GPU cards is broken through, the computing power is greatly improved, scenes such as large-scale model training can be efficiently supported, the processing period can be shortened, and application in the artificial intelligence high-end field can be assisted.

The present invention relates to a manifold modeling method for environmental parameter space, comprising the following steps: pre-processing environmental parameter data of a candidate region to establish a numerical space model of the environmental parameters of the region; using the geometric structure and spatial characteristics of a Riemann manifold to construct a manifold space representation corresponding to the environmental numerical parameters; performing manifold space representation on simulation parameters according to the manifold space positions of the candidate region at different times; performing manifold space representation on simulation parameters and calculating the corresponding geodesic distance, outputting the geodesic distance between the environmental simulation parameters and the historical environmental parameters of the region in the manifold space; combining the geodesic distance of the environmental simulation parameters in the manifold space, and finally evaluating the reliability of the environmental simulation parameters. The present invention targets a large amount of historical data of environmental parameters, effectively utilizes the geometric structure of the data, and performs manifold modeling on the numerical space of environmental parameters.

The present disclosure provides a Kubernetes storage resource binding method, system, device and medium based on location information, so as to at least solve the problem that the current Kubernetes storage resource scheduling process cannot give full play to the maximum computing power of MEC. The method includes: judging whether a request for a persistent volume declaration PVC from a user terminal is received, the request carrying the access mode of the user terminal; if the request is received, obtaining the location information of the user terminal; judging whether the access mode of the user terminal is a static access mode; if it is a static access mode, selecting the first storage resource pool closest to the user terminal based on the location information of the user terminal; and binding the first persistent volume PV1 created in the first storage resource pool to the PVC. The present disclosure takes the location information of PV into consideration in the resource scheduling process, and implements the binding of storage resources to the storage resource pool closest to the user, so that MEC can give full play to its maximum computing power.

The invention relates to a cooperative firepower attack intelligent decision-making system based on a CPS. The cooperative firepower attack intelligent decision-making system comprises an intelligent decision-making center, a fire control system, a weapon system, a situation awareness system, a combat command and control center, a combat data center, reconnaissance equipment and a real-time communication network. The situation awareness system is used for acquiring weapon and target information of a battlefield and transmitting the information to the intelligent decision center; the intelligent decision-making center fuses and processes information from all the systems to obtain a cooperative firepower strike decision-making scheme; the combat command control center is used for receiving the preliminary task decision and sending the final task decision to the intelligent decision center; the fire control system is used for receiving the attack decision task and controlling the weapon system to attack the target; the combat data center is used for providing data support; and the real-time communication network is used for realizing real-time information transmission. According to the method, the advantages of information-physical fusion of the decision making system are fully exerted, the cooperative firepower strike decision is optimized, and the overall firepower strike capability of the combat unit is improved.