62results about How to "Achieve parallelization" patented technology

The invention relates to a biomedicine missing data compensation method based on an EM cluster-BP neural network and belongs to the technical field of biomedicine. According to the method, an iterative algorithm EM algorithm which is important in processing incomplete data problems is adopted, different missing data are partitioned in different clusters according to an EM cluster and primary compensation is completed; and models are established for complete data in each cluster through a BP neural network method and precise compensation of the missing data in each cluster is finished. According to the biomedicine missing data compensation method, adaptability of a compensation algorithm to any missing mechanism is enhanced to a certain degree, accuracy of compensation is improved and the method is suitable for the biomedicine missing data compensation field.

The invention discloses a method for obtaining electromagnetic scattering properties based on subregion adaptive integration. First, geometric modeling is performed on a radar conductor object by means of CAD software for cutting the whole object into multiple subregions and virtual surfaces among subregions are kept; secondly, all subregion surfaces are scattered with triangles and subregion grid information after dissection is output; Descartes grids are established in subregions respectively again, induced current of subregion surfaces is obtained by means of adaptive integration method and coupling among subregions; finally, far region scattered field is calculated by means of the induced current of subregion surfaces and further radar scattering cross section of the conductor object is obtained. Auxiliary point current amount is reduced on the premise that accurate calculation results are guaranteed, and memory requirement is reduced and convergence rate and simulation efficiency are increased.

The invention provides a polymer microcapsule characterized by being simultaneously provided with both a hollow lumen and a porous shell, and also provides a preparation method for emulsifying the polymer microcapsule. The preparation method includes the steps of multiple emulsification loading, the morphology control of emulsion dripping and macromolecule suspension polymerization, and the scale of the preparation can be enlarged easily. The microcapsule is suitable for loading wide core materials which include biological activity materials and materials with catalytic function. With the microencapsulation method provided by the invention, the problem of the hard recovery of a nanometer catalyst can be solved, the pollution to the environment by nanometer powder can be avoided, and the collocation and loading of multiple catalysts can also be realized to construct a composite catalyst.

The invention discloses a parallelization word alignment method based on a bilingual word embedding technology. The method comprises the steps of obtaining a bilingual word vector table through an MPS-Neg bilingual word embedding technology on a Spark platform, obtaining a word alignment model through a bilingual word vector table, then conducting a distributed word alignment task, updating the bilingual word vector table through the word alignment result and MPS-Neg, and repeating the operations of word alignment and bilingual word vector table updating till the specific iteration frequency is reached. The parallelization word alignment method based on the bilingual word embedding technology solves the problem that an existing word alignment method cannot well adapt to a large-scale corpus word alignment task.

The invention discloses a DCR-based dynamic vulnerability detection method and a system. The method comprises the following steps: 1) reconstructing a vulnerability detection system on the basis of a DCR parallel computed structure feature; 2) generating a test task according to a seed test case input to an initial task and putting into a task queue of a dispatching node; 3) distributing the task in the task queue to an idle calculation node by the dispatching node; 4) executing the corresponding test task by the calculation node, after ending the execution, taking the newly generated test cases and the test case resulted in target program crash as calculation results and sending to the dispatching node; 5) stipulating all the received calculation results by the dispatching node, organizing all the newly generated test cases into test tasks and transferring to the task queue, and storing the test case resulted in target program crash; 6) repeating the steps 3)-5) till the task queue is empty, and then confirming the vulnerability of the target program according to the stored test case resulted in target program crash.

The invention relates to an NCS algorithm parallelization method based on multiple FPGA platforms. The method includes the steps of subaperture processing and synthetic aperture processing, and the subaperture processing and the synthetic aperture processing are carried out in a paralleled mode through two-stage assembly lines. A pipelining paralleled processing mode is adopted in the subaperture processing. A plurality of data processing tasks independent from one another in the subaperture processing and the synthetic aperture processing are allocated in a plurality of chips for paralleled execution. The NCS algorithm parallelization method based on the multiple FPGA platforms has the advantages that a multi-core hardware architecture is utilized to divide operating tasks into a plurality of sub-tasks and the sub-tasks are allocated to processing units of different chips for being processed respectively, so parallelization of task processing is achieved, and the efficiency of on-chip software is effectively improved.

The invention relates to a parallel coupling marine acoustics forecasting system and an operation method, and belongs to the field of physical ocean and underwater sound field modeling fusion. The system comprises an ocean environment forecasting module, an ocean-acoustics coupling module and an ocean acoustics propagation solving module; the marine environment forecasting module provides initial environment input for the system; the ocean-acoustic coupling module has the functions of parameter initialization, terrain extraction and ocean environment parameter profile extraction, and an ocean model and an acoustic model are combined together by converting parameters, including temperature, salinity and sea surface roughness output, of an ocean module into input of the acoustic model; the marine acoustic propagation solving module is used for calculating a marine sound field and outputting results including the marine sound field and the sonar operating distance. The system realizes real-time forecast of an underwater sound field, can be transplanted to any high-performance computer system, and quickly realizes accurate calculation of a three-dimensional sound field and estimation of a global sea sonar operating distance.

The invention relates to a distributed real-time rule reasoning method and a distributed real-time rule reasoning device oriented to massive streaming data. The method comprises the following steps: establishing a rule graph corresponding to a rule set; converting the rule graph into a data operation graph; generating reasoning operations according to the data operation graph; setting the priorities of the reasoning operations according to a deadline, and putting the reasoning operations into a reasoning operation ready queue according to the descending order of the priorities; selecting a reasoning operation with a highest priority, creating a reasoning task, distributing and executing the reasoning task; when the reasoning operation is completed, deleting the reasoning operation from thereasoning operation ready queue and judging whether the reasoning operation ready queue is an empty queue or not; if the empty queue is not an empty queue, continuing to execute the above operation;if the empty queue is the empty queue, representing that an input fact is matched with all rules. According to the method, mass real-time data can be matched with rules in a distributed environment and the rules can be completed as much as possible before the deadline. The method and the device can be applied to the application of the internet of things with the time constraint, so that the intelligentization of the real-time application of the internet of things is improved.

The invention discloses a method for parallelly processing data in an extensible manner for network I/O (input/output) virtualization. The method has the advantages that data receiving and transmitting, protocol analyzing, data stream managing and upper-level application are regarded as network I/O links on the basis of hierarchical structures of networks, and parallel cooperative work is carried out in the various links in an asymmetric assembly line manner; sufficient computational resources are injected into computational resource bottleneck points in a symmetric parallel mode, and accordingly the data processing capacity is parallelized and is extensible; multi-core resources are flexibly managed according to system load, and accordingly performance bottleneck of I/O virtual resources due to the computational capacity of the traditional I/O driving methods can be eliminated.

The invention provides a multi-layer and multi-pass composite wire arc additive manufacturing method and system for a metal structure. The multi-layer and multi-pass composite wire arc additive manufacturing method for the metal structure comprises the following steps of: 1, selecting a welding wire and a substrate required for forming a specific metal structural part, and determining technological parameters; 2, generating a multi-layer and multi-pass wire arc additive path of outer wall, short straight line filling and layered turning filling; and 3, enabling a welding gun to move according to the generated multi-layer and multi-pass wire arc additive path under the driving of a robot. The welding gun is driven by the robot to conduct 3D printing according to the generated outer wall, short straight line filling and layered turning filling composite path; the outer wall path improves the size precision of a formed workpiece; the short straight line filling path solves the problem that the heights of complex structural parts are inconsistent in the printing process, and obvious macro buckling deformation is not prone to being formed; and through layered turning filling, interlayer defects are greatly reduced through, and the mechanical property of the workpiece is improved. Digitization, intellectualization and parallelization of part manufacturing are achieved.

The invention discloses a Chinese text classification method based on an MPI (Message Passing Interface) and adaboost.MH, and is used for solving the problem of long total Chinese text classification time caused by long adaboost.MH training time when a data size is large. The method comprises the following steps that: storing a Chinese text subjected to word segmentation processing into a training data set; then, combining a mutual information method with the MPI to realize feature word selection; then, carrying out reduction summation on all processes through an MPI_Reduce function in the MPI so as to obtain a similarity, and selecting a feature word according to the similarity; according to whether the feature word selected from the Chinese text contained in each process is in the presence or not, endowing the feature word with a weight by each process; and according to the communication function of the MPI, integrating process calculation results to obtain a text classification model, and utilizing the classification model to classify Chinese texts to be classified. By use of the Chinese text classification method, Chinese text classification time is greatly shortened.

The invention relates to a multithreading method of dynamic code conversion of loongson multi-core central processing unit (CPU) simulation. The multithreading method includes: initializing a loongson multi-core CPU needing simulation; preparing the corresponding dynamic code conversion environments for all cores of the CPU to be simulated; completing initialization of all dynamic code conversion through cross-thread communication; leading all the dynamic code conversion threads to start to work parallely. The multithreading method is characterized in that the cores of the loongson multi-core CPU needing simulation are detected, the corresponding multi-thread dynamic code conversion environments are prepared, parallel execution of dynamic code conversion is ensured by aid of the communication mechanism between multiple threads, unnecessary waiting of simulation cores when needing to obtain the execution chance and in the inter-core communication process is avoided, and the simulation efficiency of the loongson multi-core CPU is improved. The multithreading method has wide practical value and application value in the technical field of computer virtualization.

The invention relates to a cloud-computing-based smart recommendation method that is implemented on a Hadoop cloud computing platform based on collaborative filtering algorithm of items or articles. The method comprises the steps of S1, establishing a concurrent matrix of articles; S2, establishing a user's score matrix for the articles; S3, calculating recommendation results through the matrixes; S4, storing the recommendation results; S5, recommending the results. The invention also discloses a cloud-computing-based smart recommendation system. The Cloud-computing-based smart recommendation method and system have the advantages that the high complexity problem of an algorithm can be solved by using a cloud computing concurrent programming model, and complex operation is decomposed into simple task executions; additionally, by cloud computing, it is possible to provide a large-scale data storage system, and limits on hardware of a single computer are surpassed; the recommendation algorithm and the Hadoop cloud computing platform are integrated effectively, algorithm concurrence is achieved through the MapReduce distributed programming model, and operating efficiency of the algorithm is improved.

The invention discloses a local geometric feature-based efficient discrete geodesic line paralleling method. The method comprises the following steps of: constructing a local geodesic information transfer topological structure; inputting a network model, initiatively updating own distance values by all the peaks in the network model, and deriving new pseudo-source point elements, namely, new second-type window elements; initiating deriving new window elements by edge window elements in first-type window elements, and storing the new window elements into buffer areas corresponding to the edges;establishing a plurality of edge threads to process a window element derivation event on each edge in parallel; creating a plurality of edge threads to process a window element collection event on each process in parallel; initiatively obtaining own sub-window elements by the edges according to own topological relationships; initiatively collecting window elements in a corresponding buffer area by the edge e and writing the collected window elements into a rolling array corresponding to the next state of rolling arrays; exchanging the rolling array; and if updatable second-type window elements exist, re-inputting the network mode l, and otherwise, ending the operation. According to the method, the algorithm efficiency is improved, the operation time is shortened and complete paralleling is realized.

The invention provides an automatic trajectory planning method and system for multi-layer and multi-channel wire arc additives of a metal structure. The automatic trajectory planning method comprises the following steps: Step 1, conducting layered slicing treatment on a model of a printed workpiece; Step 2, after obtaining a layered relation matrix, calculating a contour line according to triangles recorded in the matrix by means of a topological relation; Step 3, calculating filling line segments; Step 4, sorting all the intersection points obtained in the Step 3 according to the size of the X coordinate; Step 5, connecting the filling line segments; and Step 6, carrying out layered direction-changing scanning. Multi-layer and multi-channel 3D printing of a metal structural part is carried out according to the composite printing path of outer wall + short straight line filling + layered direction-changing filling, chemical components of a formed workpiece are uniform, the purity is high, and the structure is almost free of anisotropy. Multi-layer and multi-channel 3D printing of the metal structural part is carried out according to the composite printing path of outer wall + short straight line filling + layered direction-changing filling, the grain size of the formed workpiece is small and uniform, the mechanical performance is high, and the level of the formed workpiece can exceed that of a casting with the same component.

The invention discloses an indoor positioning method based on a WiFi hot spot, which comprises an offline stage and an online stage, wherein the offline stage comprises the steps that a sample reference point and an AP node for receiving the WiFi hot spot are set in a designated area range, and information of all the AP nodes of all the reference sample points is collected; the mapping relation between the information of each AP node collected by each sample reference point and the position information of the corresponding reference point is calculated, and a fingerprint database is established; and a characteristic fingerprint space of the fingerprint database is extracted by adopting a PCA analysis method based on MapReduce. The online stage comprises the following steps: an RSSI signalof each AP received at each sample reference point is collected on line through a positioning terminal, and therefore an RSSI signal vector is formed; the PCA analysis method is adopted to extract themain positioning characteristic of the RSSI signal vector, so as to obtain a characteristic vector of the RSSI signal of each sample reference point; and accurate positioning is carried out accordingto the characteristic vector of the RSSI signal and the data information of the characteristic fingerprint space by adopting an EWKNN algorithm based on MapReduce.

The invention relates to a concurrent land resource quality evaluation factor space quantifying method which comprises performing concurrent discretization to an evaluated space region; layering land resource quality evaluation factors; and performing concurrent space quantifying to land resource quality evaluation factors. The concurrent land resource quality evaluation factor space quantifying method generally has the advantages of simplicity and quickness, and is applicable to being carried out on personnel computers with multiple CPUs (central processing units) and multiple cores, small work stations and computer clusters. The concurrent land resource quality evaluation factor space quantifying method plays an important practical significance in sufficiently playing hardware computing proficiency of a current computer, shortening land evaluation work time and increasing work efficiency.

According to an arc additive manufacturing method for partition printing and filling of a reinforcing rib metal structural part, a welding machine serves as a heat source, a metal wire serves as a forming material, and partition filling paths are sequentially planned for cladding printing. The method comprises the following steps: 1) selecting a welding wire and a substrate required by a specific formed metal structural part; 2) layering and slicing a formed metal structural part model, and partitioning the two-dimensional contour according to the characteristics of the metal structural part; 3) performing path planning on each partition; and 4) driving a welding gun by a robot to move according to the generated partition path. The invention provides a partition filling electric arc additive manufacturing method for a reinforcing rib metal structural part. A partition filling path is developed autonomously; the welding gun is driven by the robot to conduct 3D printing according to the generated partition filling diameter, the heat input amount of all the sub-partitions is balanced in the whole printing process, part deformation is reduced to a great extent, and digitization, intellectualization and parallelization of part manufacturing are achieved.

The invention provides a deep market quotation obtaining system and method based on an FPGA, and relates to the field of stock quotation systems.The method comprises the steps that a TCP protocol stack is built based on the FPGA, communication with an exchange quotation gateway is conducted through a 10-gigabit optical port, and original data of the exchange quotation gateway is received; the method comprises the following steps: firstly, transcoding an order book, then realizing deep market condition snapshot, commission and transaction transcoding based on a Binary protocol through FPGA logic, reconstructing the order book by using information of commission and transaction transcoding, and finally, multicasting transcoding data and reconstructed order book data to a downstream client through a 10-gigabit optical port in a UDP broadcast form. And the characteristics of hardware programmability, low power consumption and low time delay of the FPGA are fused. Based on RTL-level logic programming customization TCP protocol stacks and Binary protocol decoding of the deep public, parallelization and pipeline processing of data are achieved, and extremely low deep market quotation time delay and extremely high processing performance are achieved.

The invention discloses a method for parallelizing SHIFT module serial codes in GTC software. SHIFT module original serial codes are parallelized effectively through exclusive scanning, namely, the position, in integral data, of data of each thread is marked through exclusive scanning, follow-up calculation is performed directly according to the marked positions of all the threads to avoid the process that data dispersed at different positions in a memory in the original serial codes are converged into a continuous memory space, parallelization is achieved in the process of void filling, GTC program operation time is obviously shortened, and performance of a GTC program is effectively improved.