47results about How to "Save computing space" patented technology

The invention relates to an intelligent association analysis method and an intelligent association analysis device for security events. The intelligent association analysis method comprises the steps of decomposing the attribute characteristic of a security event which is acquired in real time and standardizing the attribute characteristic values; and traversing an offline generated unified inference structure by means of the standardized attribute characteristic value, thereby determining an attack type. The intelligent association analysis method and the intelligent association analysis device improve association analysis efficiency.

The invention provides a method and device for simulating a diffusion process of a contrast agent. The method comprises steps as follows: a three-dimensional blood vessel model is built according to a blood vessel CT scanning image; the three-dimensional blood vessel model is pre-processed; the pre-processed model is rendered to a screen, and particles are initialized; the three-dimensional blood vessel model rendered to the screen and the initialized particles are subjected to collision detection, the action force of the three-dimensional blood vessel model on the particles is calculated according to a collision detection result to obtain external force; internal force among the initialized particles is calculated according to smoothed particle hydrodynamics; the state of the particles in the three-dimensional blood vessel model rendered to the screen is updated and rendered to the screen. The method and device solve the technical problems of large calculation amount, difficulty in acquisition of parameters and low authenticity of a simulation effect with a conventional method for simulating the diffusion process of the contrast agent in the prior art, and realize the technical effects of good instantaneity, high simulation effect accuracy and simple simulation process.

The invention discloses an optimization method of entity alignment based on graph partition . Candidate entity pairs are mined from all entities by composite index, and whether the candidate entity pairs are aligned or not is judged by entity similarity measurement method, and then an optimization algorithm based on graph partition is proposed to improve the accuracy of equivalent entity alignmentby utilizing the similarity relationship between entities. The method of the invention solves the entity alignment problem of the large-scale Internet data, and can accurately and completely mine theentity set equivalent to each other in the original data.

The invention belongs to the technical field of air-ground integrated Internet of Vehicles communication, and discloses an air-ground integrated Internet of Vehicles relay selection method based on state transition probability. The method comprises the steps of constructing an air-ground integrated Internet of Vehicles model composed of V vehicle nodes and U unmanned aerial vehicle nodes; constructing an information transmission mechanism between the mobile nodes based on the air-ground integrated Internet of Vehicles model; constructing a state transition probability between the mobile nodesbased on the information transmission mechanism between the mobile nodes; based on the state transition probability between the mobile nodes, constructing an information identification method based ona multiple-input-multiple-output technology; and constructing a relay selection method based on the state transition probability on the basis of the information transmission mechanism between the mobile nodes, the state transition probability and the information identification method. The method can be applied to an urban intelligent traffic system, the relation among vehicles, unmanned aerial vehicles and users is enhanced, and a real-time, accurate and efficient comprehensive transportation and information transmission system is built.

The invention provides a multi-satellite task planning method based on K-means clustering, the method comprises the steps of: S1, collecting task requirements of users T= {t1, t2, t3... Tn}, and obtaining an orbital working time set O= {o1, o2, o3,... Om} corresponding to each circle of sun illumination region of all currently available satellites;S 2, calculating the distance Disij between the task ti and each element oj in the set O, forming a distance set D= {di1, di2, di3... Din} between the task ti and the orbit set O, and clustering the task ti to the orbit k, Disik=Min (D) which is theshortest distance from the orbit set O; S3, judging whether the current clustering scheme sk belongs to the set S= {s1, s2, s3,... Sz}, outputting the clustering scheme sk if sk belongs to S, otherwise adding the scheme sk to the scheme set S, and returning to step S2. By analyzing the factors affecting the assignment of multi-satellite tasks and quantifying these factors and by using the K-meansclustering algorithm, a multi-satellite cooperative task assignment scheme is designed, which has fewer iterations and faster computational speed, and can meet the constraints of time complexity for large-scale optimization problems, and greatly improve the quality of imaging, and improve the completion rate of the task.

The invention belongs to the technical field of cognitive radio, and particularly relates to a frequency spectrum prediction sensing method based on an RCS-GRU model. The method comprises the following steps: simulating a channel by adopting an M/M/N queuing theory model; combining and splicing the channel state sequences simulated in the step 1 into a matrix; establishing a convolutional neural network model based on residual CBAM, inputting the spliced channel state matrix set into the model, and extracting features of spectrum occupancy states among channels; inputting the extracted feature data set into a GRU model, mining the features of the channel state in the time sequence, performing spectrum prediction, and outputting the channel state of the next time slot; adopting an Adam optimization algorithm to set a variable learning rate optimization cross loss function to train an RCS-GRU network, and adding a dropout method in the training process; and evaluating the prediction performance of the RCS-GRU by adopting a relation curve of the false alarm prediction probability and the detection prediction probability and a root mean square error (RMSE).

The invention discloses an open-circuit fault diagnosis method for a switching tube in a half-bridge sub-module of an MMC converter. The diagnosis method is composed of an event triggering program and an IGBT open-circuit fault diagnosis program. The IGBT open-circuit fault diagnosis program is started when the capacitive voltage of a certain half-bridge sub-module exceeds a threshold value interval set by the event triggering program; and when the capacitive current state observation value, the capacitive current theoretical value and the sub-module driving function Si of the half-bridge sub-module meet the conditions, the fault IGBT of the half-bridge sub-module can be accurately positioned. A sub-module capacitive current sensor does not need to be additionally added, and the method can be widely applied to existing MMC commercial products; capacitive current is observed in real time through a capacitive current state observer; the observation value of the capacitive current is symbolized through a symbolic function. By means of the method, the high anti-interference capacity is achieved for system parameter changes and sampling disturbance.

The invention discloses a steam turbine component health state assessment method based on ResNet, which comprises the following steps: carrying out a simulation experiment of a steam turbine component, and carrying out vibration data acquisition through a plurality of measuring points, wherein the vibration data comprises fault signal data and normal working condition data; carrying out label assignment on the fault signal data, wherein the label information comprises a fault type and residual available life; segmenting the fault signal data assigned by each label and carrying out normalization processing to obtain a sample set; dividing the sample set according to a preset proportion to obtain a training set and a verification set; training a pre-constructed ResNet-based multi-task learning neural network model through the training set by adopting a strategy of adaptively updating the network learning rate to reach a preset convergence condition, and obtaining a trained ResNet-based evaluation model; steam turbine part health state assessment is achieved through the assessment model. A multi-task learning mechanism is adopted, the fault type and the health degree of the steam turbine can be judged at the same time, and the accuracy is high.

The invention provides an alternating-current fault detection method for high-voltage direct-current power transmission. The alternating-current fault detection method for high-voltage direct-current power transmission comprises the following steps of obtaining the three-phase voltage of a current conversion bus and three-phase currents of outgoing lines of the current conversion bus; obtaining a voltage zero-sequence component value and a rotating vector amplitude of the current conversion bus according to the three-phase voltage of the current conversion bus, and obtaining current zero-sequence component values and rotating vector amplitudes of the outgoing lines of the current conversion bus according to the three-phase currents of outgoing lines of the current conversion bus; obtaining power zero-sequence component values and power rotating vector amplitudes of the outgoing lines of the current conversion bus; comparing the values and the amplitudes with the set values and the set amplitudes respectively to obtain the fault detection result. The alternating-current fault detection method for high-voltage direct-current power transmission is high in speed and accuracy, and meets the requirement of high-voltage direct-current power transmission for fault detection, and valuable information is provided for the defense commutation failure of a control system for high-voltage direct-current power transmission.

The invention discloses a device clustering method, a terminal device and a storage medium, and the device clustering method comprises the steps: obtaining a connection relationship between devices, the devices forming network nodes of a GAT network; obtaining an adjacent matrix of each device according to the connection relationship among the devices; performing feature extraction on each device to obtain a feature matrix of each device; inputting the adjacent matrix and the feature matrix into a pre-trained graph attention auto-encoder GATE model to obtain a reconstructed node feature matrix; and inputting the reconstructed node feature matrix into a clustering algorithm to cluster each network node to obtain the category of each device. The invention provides an equipment clustering scheme which does not need to update the node features of the whole graph and re-calculate the whole graph under the condition that the equipment nodes are increased, so that a large amount of calculation space is saved.

The invention discloses a Depthwise fast convolution system based on image processing. The Depthwise fast convolution system comprises an image pixel reading unit and a DW convolution calculation unit. The image pixel reading unit can store an image to be processed into the storage unit in a multi-dimensional form; and the DW convolution calculation unit reads the image data in rows and then performs multiply-accumulate calculation with the convolution kernels of the three rows to obtain calculation results of all pixel points. According to the system, a rapid algorithm of image convolution isrealized, the process of repeatedly reading the image storage unit can be reduced, the calculation space used by convolution is reduced, and the result of image convolution can be rapidly and accurately obtained.

The invention discloses a factory processing flow recommendation method based on singular value decomposition. The method comprises the steps of converting a factory processing flow 'reference flow 'which is being modeled and all factory processing flows in a flow library into corresponding process structure trees based on task nodes; extracting an m, n-gram set of the TPST corresponding to each processing flow in the flow library, wherein one m, n-gram represents a partial structure and behavior characteristics of one processing flow; taking a union set of the m, n-gram set of each factory processing flow in the flow library, and establishing an incidence matrix based on all the processing flows in the flow library; performing singular value decomposition on the incidence matrix, and dividing flows in the flow library into a plurality of classes according to a decomposition result; selecting any processing flow from each class of flow to be subjected to similarity calculation with the'reference flow', and using all the flows in the class of flow with the highest similarity as recommendation flows of the 'reference flow'. According to the invention, the most recommended flow of the 'reference process can be obtained only through one-time calculation, so that the calculation space is greatly reduced.

The invention discloses a splitting surface pre-screening method and a splitting surface pre-screening system applied to an alternating-current/direct-current system. The splitting surface pre-screening method includes simulating asymmetrical faults in three-phase voltage of a bus bar i; acquiring three-phase asymmetric indexes of all bus bars; constructing a correlation matrix according to the three-phase asymmetric indexes of all the bus bars; subjecting the correlation matrix to K-core decomposition so as to obtain a K-core value of each bus bar; screening out the bus bar with the maximum degree of influence on the alternating-current/direct-current system according to the K-core value of each bus bar; deleting the bus bar with the maximum degree of influence on the alternating-current/direct-current system from a splitting surface composed of multiple lines, and splitting the alternating-current/direct-current system according to the splitting surface. The splitting surface pre-screening method and the splitting surface pre-screening system applied to the alternating-current/direct-current system have the advantages that on one hand, stability of the alternating-current/direct-current system subjected to splitting control is guaranteed, and on the other hand, the size of a splitting surface search space is decreased, so that operation time is saved and operation efficiency is improved.

The invention discloses a novel abnormal mutation point detection method, and particularly relates to a data abnormal mutation point detection algorithm based on a limited accompanying group mechanism. A finite accompanying group mechanism is utilized to help to efficiently and accurately identify and detect the mutation points in the industrial data stream. According to the method, a limited accompanying group thought is provided, and effective identification and detection of abnormal mutation points of the data stream are realized by selecting a limited high-quality accompanying group and constructing a to-be-detected point group. The algorithm has low delay, and abnormal mutation points can be timely identified in real-time data streams.

The invention discloses a motor target prediction torque obtaining method. The method comprises calibrating an A/B corresponding relation table and a motor efficiency table; calculating according to the motor predicted voltage uppred and the motor predicted current ipred to obtain motor predicted power Ppred, and checking a motor efficiency table according to the motor predicted power Ppred and the motor rotating speed n to obtain motor predicted efficiency eta pred; obtaining a motor table prediction torque Tqpred, map according to the motor prediction power Ppred, the motor rotation speed n and the motor prediction efficiency eta pred; calculating to obtain A according to the motor predicted current ipred and the motor actual current iact; checking an A/B corresponding relation table according to A to obtain B; and according to the predicted torque Tqpred, map of the motor table and the actual torque Tqact, A and B of the motor, obtaining the target predicted torque Tqpred of the motor, nammely Tqpred = Tqpred, map * B + A * Tqact * (1-B). The invention further discloses a motor control system. The precision of the target prediction torque Tqpred of the motor is improved, meanwhile, the calculation amount is reduced, the calculation space of the controller is saved, and it is avoided that the service life of the battery is shortened due to over-current of the battery.

The embodiment of the invention provides a micro-service architecture calling method and device, electronic equipment and a storage medium, and the method comprises the steps: receiving protocol request information of an external micro-service; obtaining interface data according to the protocol request information; integrating the interface data according to the protocol request information to obtain an initial response result; filtering the initial response result to obtain a response result; and sending a response result to the external micro-service. By implementing the embodiment of the invention, a plurality of interfaces with single functions can be freely combined together for calling, so that the calling of multiple interfaces is simplified, and the maintenance cost is reduced.

A method, a device, an apparatus, and a storage medium for de-duplication of feature word are disclosed. The method includes: acquiring a set of phrases associated with a current feature word in the set of feature words; calculating a sum of a specified part of a phrase based on an ASCII code corresponding to the phrase one by one to obtain a first set of sum values; the deduplicated feature wordsbeing determined by judging the number of minimum values in the first set of sum values. According to the technical proposal of the embodiment of the present application, by calculating the sum of ASCII codes, the feature words with the same meaning are de-duplicated, thereby reducing the computational complexity of the current feature word de-duplication method, saving computational space, and significantly improving the generalization ability of the current feature words to the text.