25results about How to "Improve parallel processing capabilities" patented technology

The invention belongs to the technical field of distributed data access, discloses an RDMA (Remote Direct Memory Access) distributed data transmission method supporting high-concurrency message transmission, and aims to improve system performance in a high-concurrency message transmission environment. The method comprises the following core steps: 1) peer-to-peer server design: each device is an independent server, and connection is established by a connection channel monitored at any time; 2) establishing a connection resource pool in advance; 3) creating and registering an RPC message; 4) transmitting the RDMA to a memory mailbox pool of a target server; 5) the client and the server perform high-concurrency processing; 6) dynamically polling the server; (7) zero-copy transmission of data is achieved through the RDMA technology. According to the invention, the dynamic expansion capability of the architecture is improved by adopting a connection resource decoupling mode, QP jitter can be reduced and the stability of the system can be enhanced by registering a memory of a continuous large page at one time, and the message transmission efficiency under the distributed network architecture is effectively improved by adopting a combined sending strategy.

The invention relates to the technical field of intelligent Internet of Vehicles, in particular to a task unloading method in a vehicle-mounted edge computing environment based on deep reinforcement learning, which comprises the following steps: acquiring a task vehicle to generate a to-be-executed task, uploading the task to an RSU, generating a scheduling decision according to the current state of the task and available resources, and sending the scheduling decision to the RSU. The task is allocated to a service vehicle or locally executed; performing priority division on the tasks by utilizing an analytic hierarchy process, and obtaining the current state of each to-be-executed task; obtaining service vehicle numbers and task vehicle numbers of the tasks based on task states and priority level utilization model scheduling, and calculating unloading and scheduling strategies; obtaining available computing resources of the current service vehicle; and training the sequence to the sequence model by using an Actor-Cr-it ic algorithm to obtain an optimal task part unloading and scheduling strategy. The computing resources of the service vehicle and the edge server can be fully utilized, the execution time delay of all tasks in the whole time period is smaller, and the task execution success rate is higher.

The invention discloses a method and device for accelerating a homomorphic encryption algorithm by using fast number theory transformation and a medium. The method comprises the following steps: initializing homomorphic encryption parameters and generating a public key, a private key and a relinearization key; encrypting a plaintext to be operated by using the public key to generate a corresponding ciphertext polynomial; performing fast number theory transformation on the ciphertext polynomial to transform the ciphertext polynomial into a point value form; performing homomorphic multiplication on the two point value form ciphertexts to obtain three ciphertexts; using a relinearization key to transform and decompose one ciphertext by using a fast number theory and accumulate the decomposed ciphertext into the other two ciphertexts to obtain two similar primary ciphertexts; using fast number theory inverse transformation to transform the first-order ciphertext from a point value form into a polynomial form; and decrypting the ciphertext polynomial by using the private key to obtain a plaintext multiplication result. The method adopts fast number theory transformation and inverse transformation, improves the polynomial multiplication calculation efficiency through the processes of primitive root sampling, bit inverse sequence and the like, is suitable for the field of polynomial operation, and improves the polynomial operation efficiency.

The invention discloses an automatic code vectorization optimization method and device and a medium, belongs to the technical field of high-performance computing, and aims at solving the problems that existing automatic code vectorization is inconvenient, the manual operation proportion is large, vectorization of irregular codes or codes with a load dependency relationship is difficult to support, and the vectorization efficiency is high. And improvement of calculation performance and reduction of energy consumption are not facilitated. The method comprises the steps of obtaining a scalar instruction diagram of a calculation task according to a calculation kernel of parallel calculation; performing protocol conversion processing on the scalar instruction diagram to obtain a protocol instruction diagram; according to the high-level organization structure of the instruction, grouping processing related to mapping vectors is carried out on the protocol instruction graph, and a grouping graph is obtained; based on the limited vector length of hardware, performing splitting matching processing on instruction vectors in the grouping graph to obtain a vector matching grouping graph; and executing configuration on each group of elements in the vector matching grouping graph to obtain a vector operation graph.

The invention discloses a global memory allocation method and device for a GPGPU (General Purpose Graphics Processing Unit) and a medium, belongs to the technical field of GPGPU memory allocation, and is used for solving the technical problem that in the data transmission cooperative work process of an existing host and a GPGPU device, the real-time performance of a task is easily influenced, and a large device memory space is possibly occupied, so that the memory space is wasted. The method comprises the following steps: performing data block division processing on each piece of input data required by an input task to obtain input data blocks; performing operation processing of the device end on the first group of input data to obtain an operation result of the first group of input data; if the operation state of the device end is completed before the host end finishes sending the second group of input data, performing first operation completion control on the device end to obtain a first operation result of the second group of input data; and if the operation state of the device end is completed after the host end finishes sending the second group of input data, obtaining a post-operation result of the second group of input data.

The invention provides a low-delay wave beam direction finding technology realized based on an FPGA. The low-delay wave beam direction finding technology comprises the following steps: establishing a wave beam direction finding simulation mathematical model according to a wave beam direction finding algorithm in combination with specific hardware parameters and indexes and feature information of a to-be-detected flight target; converting the established wave beam direction finding simulation mathematical model into an FPGA program; the method comprises the following steps: establishing a hardware test platform, and programming an FPGA (Field Programmable Gate Array) program into an FPGA baseband board; and performing operation according to the direction testing process to complete direction testing. Compared with the prior art, the delay caused by direction finding is greatly reduced by virtue of the strong parallel processing capability of the FPGA chip and the high-speed clock, and meanwhile, the resolution of direction discrimination is improved. Through application of the FPGA and accurate mathematical modeling, the method has significant advantages in the aspects of direction finding precision, real-time performance, system expansibility, anti-interference capability and the like, and can effectively cope with complex flight target direction finding tasks. Therefore, the technology has wide application prospects in the fields of aerospace, military reconnaissance, unmanned aerial vehicles and the like.

The embodiment of the invention provides a prediction model generation method and device for power grid dispatching future state early warning, and the method comprises the steps: collecting historical data which may face scenes in the dispatching process of a power grid enterprise and cost data of investment future state early warning; processing the historical data and the cost data through a z-score standardization method, and scaling the data of different orders of magnitude to the same order of magnitude; performing feature selection on different prediction targets by adopting Elastic Net to obtain respective high-correlation feature data sets, splitting the high-correlation feature data sets into training sets and test sets, and training a BP neural network model by using the training sets to obtain prediction models of different prediction targets, and performing performance test on the prediction model through the test set to obtain a final prediction model. The calculation speed and the decision-making efficiency can be improved.

The invention discloses a multifunctional PLC expansion device based on an FPGA and a configuration method thereof.The device comprises the FPGA, the FPGA comprises a communication module, a function configuration module, a pulse output module, a serial port module, a pulse counting module and an IO control module, the communication module is used for receiving a configuration instruction, and the function configuration module is used for receiving the configuration instruction; the function configuration module is respectively connected with the communication module, the pulse output module, the serial port module and the pulse counting module, and is used for selectively activating any one of the pulse output module, the serial port module or the pulse counting module according to the configuration instruction; and the IO control module is respectively connected with the pulse output module, the serial port module and the pulse counting module for function output.

The present application discloses a method for deducting inventory of goods. The method comprises: receiving a request for deducting inventory of goods, wherein the request for deducting inventory of goods comprises inventory deduction quantities and inventory identification information of multiple goods to be deducted; determining a database storing inventory information of each commodity to be deducted according to the inventory identification information of each commodity to be deducted; when there are multiple determined databases, creating multiple threads; executing corresponding inventory deduction operations in the corresponding database in parallel according to the inventory deduction quantities corresponding to each commodity to be deducted through multiple threads; when there are threads that fail to successfully execute the inventory deduction operation, generating corresponding compensation records according to all successfully executed inventory deduction operations, and writing the compensation records into the inventory compensation table; compensating and updating the inventory information stored in the corresponding database according to the inventory identification information and inventory deduction quantities of the commodity to be compensated in the compensation record. The present application can reduce the probability of deadlock and improve user experience.

The utility model discloses a four-axis numerical control pass-type nut tapping machine which comprises a machine box, a control system is installed in the machine box, and a working table used for machining nuts is arranged on the upper end face of the machine box. The more than one nut tapping mechanism is mounted on the working table side by side, and a plurality of nuts are automatically tapped at the same time through the more than one nut tapping mechanism; the feeding system is mounted at the top position of the working table of the case and is used for continuously feeding nuts to each nut tapping mechanism; and the material collecting system is mounted at the discharging ends of the more than one nut tapping mechanism and is used for collecting the tapped finished nuts. According to the automatic tapping machine, automatic tapping of a plurality of nuts can be achieved, the whole process is automatic, manual participation is reduced, the working efficiency is greatly improved, and the product percent of pass is improved.

The invention provides a multi-round accurate translation method and system based on a large model in the field of self-training financial supervision. The method comprises the following steps: S1, constructing a training data set of the large model in the field of financial supervision; s2, constructing a large model in the financial supervision field, and performing fine adjustment on the general large model; s3, performing multi-round translation on a plurality of to-be-translated texts by utilizing the fine-tuned financial supervision field large model, and constructing a context information base; s4, translating a target text by using the fine-tuned large model in the financial supervision field, and adjusting and optimizing a translation result by using a context information base; the large model in the financial supervision field is a fine-tuning translation model which is built based on a Gamma 227b large language model and aims at the financial supervision field. The translation process is highly automatic, a user only needs to input a text or a document needing to be translated, a high-quality translation result can be obtained, and complex operation is not needed.

The invention discloses a block cipher-based data detection method for a quantum neural discriminator. The method comprises the following steps: a sender encrypts a specific plaintext by using a block encryption algorithm to obtain ciphertext data and transmits the ciphertext data to a receiver; receiving and generating a specific data set, wherein the data set is composed of normal data conforming to characteristics and random abnormal data; thirdly, a receiver constructs a quantum neural network, input data are coded into a quantum state, convolution is achieved through a revolving door and a controlled door, and residual connection is achieved through control bit addition and unit matrix addition; the method comprises the steps that firstly, a network is obtained, then the network is trained and passes a verification test, so that an efficient quantum nerve discriminator is obtained, and finally, a receiver inputs received data into the quantum nerve discriminator for detection. According to the method, the powerful parallel processing capability of quantum computing and the advancement of the deep learning technology are fused, and normal data and abnormal data can be effectively distinguished, so that whether the data are leaked or tampered or not is judged.

The invention relates to a radar radiation source individual identification method based on large-scale streaming data processing, and the method comprises the steps: training a known radiation source intra-pulse characteristic waveform template library, calculating the similarity between waveforms in the template library, and determining the similarity threshold value of each radiation source individual; the Spark Streaming distributed computing platform analyzes and calculates the received binary system intermediate frequency data stream containing the radiation source intra-pulse information to obtain the instantaneous amplitude waveform and the instantaneous frequency waveform in the signal pulse; and taking two characteristics of instantaneous amplitude and instantaneous frequency as individual identification characteristics, performing similarity calculation on a waveform obtained by analysis and a waveform in a template library by using a dynamic time warping algorithm, and comparing a similarity calculation result with a similarity threshold to obtain an individual identification result. According to the method, the individual identification efficiency of the large-scale streaming radar radiation source data can be improved, and the parallel processing capability is enhanced.

The application provides an LDPC code decoding method and device, electronic equipment and storage medium. The method comprises the following steps: obtaining an input code word, a grouped check matrix and an initial confidence of an LDPC code, wherein the grouped check matrix is obtained by grouping an original check matrix according to a preset condition; presetting two initial quantization values with a given magnitude bit width, calculating an initial companion based on the input code word and the grouped check matrix; starting from the initial companion and the two initial quantization values, performing decoding iteration column by column, updating a current column variable node in the iteration, obtaining a soft decision message and transmitting the soft decision message to a next column, updating an associated grouped check node, and performing post-processing compensation on an oscillation correctable code word; updating the companion after each column is updated and judging a decoding success condition; and outputting a result when the condition is met or a maximum iteration number is reached. The scheme can realize the collaborative optimization of decoding performance and hardware implementation efficiency.

The present invention relates to the field of data processing technology, specifically a GPU cluster data sharing method for AI model training, comprising the following steps: optimizing the real-time monitoring of gradient information according to the gradient information processed by GPU nodes, adjusting the distribution of data in GPU cluster data sharing between GPU nodes, adjusting the data flow direction, optimizing resource allocation, and generating data flow optimization strategies. In the present invention, by real-time monitoring and adjustment of the gradient information processed by GPU nodes, the efficiency of GPU cluster data sharing and the speed of AI model training are significantly improved, real-time monitoring allows dynamic adjustment of data distribution, effective use of computing resources, reduction of redundancy and processing time, data flow and resource allocation optimization make data transmission more uniform in a multi-GPU environment, reduce the transmission bottleneck between nodes, and enhance the parallel processing capability of the training process. This fine data management supports larger-scale data set processing and improves model accuracy and efficiency.

The invention provides a lightweight deep learning target detection and recognition model and method, and the method comprises the steps: carrying out the lightweight processing of a network frame based on a YOLOv5s algorithm, obtaining an overall frame module, carrying out the processing and feature extraction of an input image through a convolution layer, an RELU activation function, dimension splicing, and summation operation through a CnTR module, and obtaining a feature extraction result, the loss function module adopts a GIOU loss function, measures the difference between a prediction value and a true value of the model according to a feature extraction result, and guides parameter adjustment of the model; the activation function module adopts an RELU activation function to carry out nonlinear transformation on a feature extraction result; the model quantification module quantifies parameters and data in the overall framework module, the feature extraction module and the activation function module, and reduces the storage demand and calculation amount of the model. The method has the advantages of being light in network weight, high in calculation efficiency, high in detection precision, suitable for FPGA platform application and the like, and an efficient target detection task can be achieved in a resource-limited environment.

The utility model relates to the technical field of machining and manufacturing, in particular to a grabbing robot rail type moving structure which comprises a rail system, a material conveying system, a moving platform, a grabbing robot and a control system. Cooperative operation of the multiple grabbing robots is achieved through the rail type moving structure, a single-line single-device mode is upgraded to a distributed cluster operation mode, operation efficiency is improved, meanwhile, single-point bottleneck is eliminated, material conveying and grabbing rhythm separation is achieved, restriction of starting and stopping of a traditional single device on a whole production line is avoided, production continuity is improved, and production cost is reduced. The method has wide applicability in the fields of industrial automation, logistics storage and the like, and economic benefits are remarkable.

The invention discloses an SM9-based verifiable fair broadcast proxy re-encryption method and a hardware architecture thereof, and relates to the technical field of information security. The re-encryption method comprises the following steps: encrypting an original message based on SM9 to generate an initial ciphertext, generating a re-encryption key based on a private key of a data owner and an authorized user identifier set, and sending the initial ciphertext and the re-encryption key to a proxy server; the proxy server performs re-encryption operation on the initial ciphertext based on the re-encryption key to generate a re-encrypted ciphertext, and broadcasts the re-encrypted ciphertext to all authorized users, and the authorized users perform decryption verification on the received re-encrypted ciphertext by using own private keys; and a public verifier can perform verification arbitration on the re-encrypted ciphertext based on a non-interactive zero-knowledge proof protocol. According to the re-encryption method provided by the invention, a multi-user application scene is expanded, direct verification of the re-encrypted ciphertext is realized, and a fair and neutral dispute judgment rule is established.

The invention discloses a field programmable gate array-based Xintf bus interface controller, which comprises a dual-port RAM module, a DSP chip assembly, an FPGA chip, an Xintf interface control line management module, an Xintf interface data writing management module and an Xintf interface data reading management module, and is characterized in that the FPGA chip is connected with the Xintf interface control line management module, the FPGA chip is connected with a daughter card through an Xintf bus, and the daughter card is connected with the DSP chip assembly. The FPGA chip is used for releasing the daughter card bus to the authorized DSP chip according to the authority state of the DSP chip assembly to realize daughter card conduction connection; the system has the advantages of being high in flexibility, expandability, reconfigurability and parallel processing capacity, function expansion can be carried out on the Xintf asynchronous bus of the DSP chip through the FPGA chip, external expansion chip access control of various function types is achieved, external expansion interface access diversification is achieved, function expansion of external expansion chip access control logic is achieved through the FPGA, and the system is convenient to use and popularize. And the method has higher flexibility and expandability.

The invention relates to a stacker rail transfer system and a stacker transportation management method, the stacker rail transfer system comprises a main rail, branch rails and a rail transfer mechanism, and the main rail comprises a plurality of segmented rails arranged at intervals; the plurality of branch rails are arranged on one side of the main rail, the plurality of branch rails are correspondingly arranged between two adjacent segmented rails of the main rail, and each branch rail and the segmented rails on the two sides of the branch rail are arranged at intervals; the multiple track transfer mechanisms are correspondingly arranged between the two adjacent segmented tracks and the branch tracks of the main track, each track transfer mechanism comprises a track transfer driving source, a main path splicing track and a branch path splicing track, and the track transfer driving sources drive the main path splicing tracks and the branch path splicing tracks to move synchronously. The communication between the main road splicing track and the two segmented tracks in the main track can be realized, or the communication between the branch road splicing track and one segmented track and the branch track in the main track can be realized. The invention aims to improve the operation efficiency and reliability of the stacking machine in a storage environment.

The invention discloses an emotion analysis method based on a convolutional neural network and a YOLO algorithm, and the method comprises the steps: collecting data, and carrying out the preprocessing of the data; constructing a convolutional neural network model; training and adjusting the model; training and optimizing the model; identifying and reasoning the emotion; and analyzing and fusing the emotion. Through the convolutional neural network, different data modes and problems can be automatically learned and adapted, features can be automatically extracted from input data, complex feature engineering does not need to be manually carried out, the YOLO algorithm is relatively simple in structure and easy to understand and implement, meanwhile, the training process is simple, and emotion analysis can be comprehensively and accurately carried out.

The invention discloses an FPGA (Field Programmable Gate Array) calling method based on a window sliding method, which comprises the following steps: a CPU (Central Processing Unit) monitors TCP / UDP (Transmission Control Protocol / User Datagram Protocol) data of a port, a plurality of new threads are created according to an intercommunication object, and each thread independently records interaction information with different equipment; when the multiple threads receive the service data at the same time, the CPU dynamically adjusts the sequence of the service data sent to the FPGA according to the current operation condition of the FPGA and the performance of processing different services; the method comprises the following steps of: continuously sending service data to be processed to an FPGA (Field Programmable Gate Array) by using a window sliding method according to the operational capability of the FPGA, and dynamically adjusting the size of each service sub-window in a window according to the operation condition of different services which are currently carried out in the FPGA and the processing condition of each different service. The invention further discloses an FPGA calling device based on the window sliding method, corresponding equipment and a storage medium. According to the invention, the high-concurrency processing capability of the equipment when a large number of different services are processed at the same time is improved, and the equipment is ensured to meet the use requirements of high reliability, high concurrency and low time delay.

The invention provides a multi-agent system DFS pseudo-tree communication method, structure and application thereof. The multi-agent system DFS pseudo-tree communication method comprises the following steps: constructing a multi-agent system DFS pseudo-tree; determining an optimal cut point and a root node of the DFS pseudo-tree, and determining an optimal neighbor node; the multiple agents initialize local data, a root node triggers a message-driven search process, and when the number of nodes to be searched is smaller than 2, the nodes are selected; and when the number of the to-be-searched nodes is greater than or equal to 2, preferentially selecting the optimal cut point as the next search node, and if the cut point does not exist, preferentially selecting the optimal neighbor node as the next search node until the search is finished, and terminating the operation of the root node. The method can effectively solve the problem of multi-agent collaboration in a complex environment, has high efficiency and good expansibility, and particularly has remarkable advantages in application to collaborative route planning of a large-scale missile system.

The invention relates to the field of medical data analysis, in particular to a privacy protection medical data analysis method based on a block chain and federal learning. Comprising the steps of collecting and preprocessing original medical data, and obtaining comprehensive features based on the preprocessed original medical data; training the medical model to obtain a prediction result; evaluating and optimizing the medical model, generating a medical model parameter update value, performing encryption and signature processing on the medical model parameter update value, and generating a signature; performing signature verification on the encryption parameter update values and the signature, integrating the encryption parameter update values to obtain an aggregation parameter update value, and encrypting and decrypting the aggregation parameter update value to obtain a final parameter update value; and optimizing the medical model based on the final parameter update value. The problem that federal learning may expose individual privacy during model training is solved; the block chain lacks security and expandability; and when the number of participants is huge and the medical data volume is huge, the calculation and communication overhead of model aggregation is high, and the efficiency and speed of medical data analysis are influenced.

A neural network system is provided that implements a generative model for autoregressively generating a distribution of a plurality of current filterbank samples of an audio signal, where the current samples correspond to a current time bin, and each current sample corresponds to a channel of the filterbank. The system includes a hierarchy of a plurality of neural network processing layers ordered from a top layer to a bottom layer, each layer trained to generate conditioning information based on previous filterbank samples, and for at least each layer other than the top layer, also based on conditioning information from a higher layer in the hierarchy; and an output stage trained to generate a probability distribution based on previous samples of one or more previous time bins and conditioning information from the lowest processing layer.