47results about How to "Strong modeling ability" patented technology

The invention discloses a method for locating faults in a power communication network. The method comprises the steps that a bipartite graph model is built according to the many-to-many uncertainty of faults and symptoms; fault effect weight factors are led in on the basis of the bipartite graph model; the fault effect degree is calculated and corrected through credible parameters to obtain suspected fault sets. According to the method, the many-to-many uncertainty of faults and symptoms is modeled through a weighting bipartite graph, so that the causal relationship between the faults and the symptoms is expressed, the modeling capacity is good, noise immunity is high, and the method can adapt to real environment with fault burstiness and network complexity. The fault effect weight is led in, the total probability and the Bayes thought are utilized under the bipartite graph model, prior fault probability is transformed into condition probability, and the fault effect degree is calculated; finally, credible parameters are added to control effect of suspected faults, the cover degree and the contribution degree are combined, and the suspected fault sets of which the effect degree is within the controllable parameter range are picked out.

The invention discloses a mid-air gesture recognition method based on an inertial sensor. The method includes the steps that a mid-air gesture signal sequence is extracted aiming at sensing signals acquired by the inertial sensor, then data preprocessing is carried out, then a training sample set, a verification sample set and a test sample are acquired, an LSTM-RNN model is subjected to parameter initialization, the training sample set is used for training the LSTM-RNN model, in the training process, verification samples in the verification sample set are input in the LSTM-RNN model trained in the iteration process, the iteration frequency is controlled according to the recognition error rate of the verification sample set, and a final LSTM-RNN classifier is obtained; finally the test sample is input in the final LSTM-RNN classifier, and a gesture corresponding to the test sample is recognized through the final LSTM-RNN classifier. The method has the advantages of being higher in mid-air gesture recognition precision and accuracy.

The invention discloses an aerial handwriting character recognition method based on an inertia sensor. The method comprises steps of carrying out data preprocessing on aerial handwriting character action sensing signals acquired by the inertia sensor and then acquiring a training sample set, a verification sample set and a test sample; carrying out parameter initialization on an LSTM-RNN model; training the LSTM-RNN model subjected to the parameter initialization through each training sample; in the training process, inputting verification samples in the verification sample set into an iteration process to train the obtained LSTM-RNN model; according to error rate of the recognition of the verification sample set, controlling the iteration frequency so as to obtain a final LSTM-RNN classifier; and finally, inputting the test samples into the final LSTM-RNN classifier and recognizing corresponding characters of the test samples through the final LSTM-RNN classifier. The method is advantaged by quite high recognition precision and accuracy of aerial handwriting characters.

The invention discloses a personalized route recommendation method based on A star search and deep learning, which comprises the following steps: step 1, taking a historical track data set D, a starting point ls, an end point ld, a departure time b and a user u as input, and then inputting a recurrent neural network; step 2, modeling a cost function (n) from the starting point to the current n node and a cost function h (n) from the current n node to the terminal point; and step 3, in the process of finding the optimal path, extending one node each time, using the f (n) to evaluate the score of the node, and recommending the personalized optimal path track p *. The invention provides a personalized route recommendation method based on A star search and deep learning. In the personalized route recommendation method, a transfer rule between track points is learnt through a recurrent neural network, an attention mechanism based on historical data is utilized to help to learn the current cost in the A * algorithm, and finally a graph attention neural network is introduced to model the future cost in the A * algorithm.

The invention discloses a supervision-based industrial process fault detection method of a linear dynamic system model, and is applied to fault detection under the condition of obtainable quality variables in an industrial process. By use of the linear dynamic system model and the quality variables, the linear dynamic system model with supervision is established, and the dynamics of a process and the randomness of data are taken into consideration. Compared to other conventional methods, the method provided by the invention not only improves the fault detection effect of the industrial process, but also enhances the grasp of process operators for process states, and enables industrial production to be safer and product quality to be more stable; and the reliance of the conventional fault detection method on process knowledge is improved to a quite large degree, and automatic implementation of the industrial process is better facilitated.

The invention provides a small-scale corpus DNN-HMM acoustic model. For small-scale corpus speech recognition in a DNN-HMM acoustic model, feature extraction is carried out on an inputted small-scalecorpus speech; the DNN-HMM acoustic model is trained by using the extracted feature and the DNN-HMM acoustic model is obtained; a language model is trained by using text information corresponding to the small-scale corpus speech to obtain a small-scale corpus language model; and a decoder is obtained by using the acoustic model, the language model and dictionary construction, so that an overall small-scale corpus speech recognition frame is obtained.

The invention discloses a modeling simulation verification language (MSVL)-based asynchronous communication system and an MSVL-based asynchronous communication method, belongs to the technical field of system formalized modeling and verification, and mainly relates to a formalized method for modeling and verification of asynchronous concurrent systems. The MSVL synchronous communication system comprises process modules, channel modules and communication commands, wherein the process modules are used for modeling of behaviors of each component in the asynchronous concurrent systems; the channel modules are used for modeling of communication media among different components; and the process modules are used for executing the communication commands to realize asynchronous communication processes among the different components. The number of channels among the process modules can be randomly finite according to the requirement of the system, the channel capacity can be defined according to the communication requirement of the system components, and meanwhile, two sets of communication commands are defined to adapt different asynchronous communication processes. The system and the method can be used for modeling and verification of a distributed system.

The invention relates to a thin-wall metal support leg which comprises two punch-formed support members, wherein the two support members are oppositely arranged and welded into a whole, the joints of the two support members are identical. The invention also relates to a production method for the thin-wall metal support leg, which comprises the following steps of: a, forming a metal plate with proper material and dimension into the support members with openings by using a punching technology through one or more times of processing according to the demand on the shape of the support leg; b, cutting off the opening edges of the two support members; c, welding the two support members together by using a welding technology; d, polishing a welding surface to form a support leg black blank with an integrated shape and a structure function; and e, carrying out metal surface processing on the black blank. The thin-wall metal support leg manufactured by using the production method has the advantages of simple process, low production cost and high production efficiency, and is easy to manufacture a complicated surface pattern.

The invention provides a mobile robot positioning method based on onsite field line features. The method comprises steps S1-S10. In the mobile robot positioning method provided by the invention, the existing field line features in the working environment are used to ensure the consistency of a global environment map and the real working environment; the calculation of a matching degree is performed based on a manner of discretizing the field line features into sampling points, thereby effectively overcoming the problem of insufficient effective feature information, which may exist; a global probability map is constructed to facilitate the subsequent quantitative calculation of a robot in the positioning process and to shorten the calculation time; an incremental side boundary extraction algorithm is scarcely interfered by the external noise, few parameters are used, and the calculated amount is reduced; sampling is performed based on the Monte Carlo method, so that the filtering accuracy can be approximated to the optimal estimation, and the computational complexity is greatly reduced; the particle filter algorithm has strong modeling ability, can effectively overcome the restriction of Gaussian distribution under nonlinear conditions, can adapt to the requirements of realistic and complex environments, and can improve the self-positioning accuracy of the robot.

The invention discloses a node vulnerability estimation method based on a heterogeneous information network. The node vulnerability estimation method comprises the following steps: step 1, constructing the heterogeneous information network; 2, setting a virtual host with a known vulnerability value; 3, obtaining an adjacent matrix of the network host and the virtual host under each meta-path; 4, calculating a similarity value between each network host and the virtual host under each meta-path; step 5, performing weighted summation on similarity values between the corresponding network hosts and virtual hosts under each meta-path; step 6, extracting the mutual access relationship among the network hosts from the computer network, then constructing an access relationship matrix among the network hosts, and carrying out normalization processing; and step 7, performing node vulnerability iterative processing on each network host. The invention further discloses a storage medium, a system and calculation equipment. According to the invention, the accuracy of the node vulnerability estimation result is ensured.

The invention discloses a particle filtering-based tumor motion estimation and prediction system and method of a radiotherapy robot. The method comprises the following steps of (S1) collecting three-dimensional motion data of a skin-marker and an in vivo tumor by using a breathing tracking unit and an image positioning unit separately; (S2) building a motion relationship model between the tumor at a current moment and the tumor at a historic moment according to the three-dimensional motion data, and building a motion relationship model between the skin-marker and the in vivo tumor within a period of time as an observation equation of particle filtering by taking the model as a state transfer equation of particle filtering; and (S3) estimating the motion position of the in vivo tumor by using a particle filtering algorithm according to motion data of the skin-marker at the current moment on the basis of the state transfer equation and the observation equation. The particle filtering-based tumor motion estimation and prediction system and method can be applied to the state space model in any form, has stronger modeling capability on nonlinear characteristics of variable parameters and is relatively high in prediction accuracy.

The invention relates to a flight emergency prediction method and apparatus based on an LSTM neural network. The method comprises the steps of obtaining sample data of historical accident flight records; preprocessing the sample data, then performing dimension reduction on the sample data by utilizing an MDDM algorithm, and mapping the sample data into a multi-dimensional vector; constructing a prediction data set by taking the multi-dimensional vector as a sample; dividing the prediction data set into a training set, a verification set and a test set, and training an LSTM neural network until the error of the LSTM neural network is lower than a threshold value and tends to be stable; and inputting the current flight data into the trained LSTM neural network to obtain the flight emergency probability. According to the invention, related data of historical flight emergencies are subjected to preprocessing, dimensionality reduction and feature extraction and are used as samples to train the LSTM neural network, so that various data in the flight process are automatically and comprehensively monitored, early warning is given out for the flight emergencies in time, and safe execution of flight tasks is guaranteed.

The present invention discloses an industrial process fault diagnosis method based on a switching supervised LDSM (Linear Dynamic System Model), which is used for fault diagnosis on the condition that a key quality variable is obtainable in an industrial process. According to the fault diagnosis method, a supervised LDSM is expanded to a multi-modal form, and a switching supervised LDSM is established, thus dynamic characteristics and random characteristics of process data are considered, and important process operation information included in quality variables is also fully utilized. In comparison with the conventional method, the fault diagnosis method improves the capability of describing industrial process operation states by the model, improves a fault diagnosis effect, shortens delay time of diagnosis, enables fault processing to be more timely and effective, and is more beneficial to automatic enforcement of industrial process.

The invention discloses a real-time identification method and system for the dominant oscillation mode of an electric power system under fault disturbance, which specifically comprises the steps of carrying out fitting analysis on an electric power system oscillation signal to be measured by using a Prony model and combining a WLAV method; iteratively solving the fitting analysis model, stopping the calculation when the preset accuracy requirements are met, obtaining a calculation result, determining an optimal solution of the model order, and thus acquiring an optimal solution p-order Prony model; analyzing the electric power system oscillation signal to be measured by using the optimal solution p-order Prony model, calculating the energy of each oscillation mode, carrying out classification, sorting according to the energy values to determine the oscillation type of the system under the fault disturbance, and identifying the dominant oscillation mode of the system by combining the oscillation type. The identification method disclosed by the invention does not need to establish an accurate mathematical model of the electric power system, and can identify the dominant oscillation mode of the system by only analyzing response signals of each electrical quantity of the system after the occurrence of the fault disturbance.

The invention provides an intelligent unmanned chariot position loss finding method based on fog calculation. A main control unit globally controls the intelligent unmanned chariot according to the information and control algorithms of all intelligent unmanned chariots; meanwhile, each intelligent unmanned chariot can form a formation with the adjacent intelligent unmanned chariot; the position ofthe intelligent unmanned chariot is acquired by means of sensor information of the intelligent unmanned chariot. The intelligent unmanned chariot cannot obtain the position information due to weatheror geographical reasons; the main control unit adopts particle filtering and Bayesian prediction based on fog calculation; the intelligent unmanned chariot group is communicated with other intelligent unmanned chariot around the intelligent unmanned chariot at the lost position to obtain the position information of the intelligent unmanned chariot and the associated information of the intelligentunmanned chariot at the lost position, so that the position information of the intelligent unmanned chariot at the lost position is calculated, the intelligent unmanned chariot group is notified, andcoordinated fighting of the intelligent unmanned chariot group is guaranteed. Fog calculation is used for storing, analyzing, processing and mining data; data forwarding and processing are carried out; the characteristics of nonlinearity and non-Gaussian are mostly presented under the interference of various noises; particle filtering and Bayesian prediction are adopted, motion estimation and tracking are conducted through a particle filtering method, position information of the intelligent unmanned chariot group and intelligent unmanned chariot associated information of the lost position areobtained, and therefore the position information of the intelligent unmanned chariot at the lost position is calculated.

The invention discloses a vision-based human body state judgment algorithm, which belongs to the technical field of computer vision, and comprises the following specific steps of moving human body detection, feature extraction and description, behavior recognition and feature extraction and description: S1, obtaining an original image, and obtaining a foreground binary image of the original imagethrough a motion detection algorithm to serve as an optical flow calculation region; s2, performing black edge filling on the optical flow calculation area to obtain a square optical flow graph, dividing the square optical flow graph into 16 equal parts by using a 4 * 4 grid, taking an optical flow average value of each area, and finally representing an optical flow vector of the frame of pictureby using a 16-dimensional vector; assuming that the silhouette feature vector obtained from the frame is s and the optical flow feature vector is 0, the fusion feature V of the two is equal to [s; 0].The method is simple in structure and convenient to use, improves the accuracy of behavior recognition, reduces the complexity and feature dimension of feature extraction through a quick feature fusion method, is high in practicality, and is suitable for popularization and application.

The invention provides a hypergraph neural network classification method. The method comprises: obtaining to-be-predicted label data; constructing a hypergraph incidence matrix and an original feature matrix of the to-be-predicted label data; constructing a hypergraph neural network for different modes in the to-be-predicted label data, and generating a first hypergraph neural network model; matching and merging a preset second hypergraph neural network model and the first hypergraph neural network model, and replacing untrained parameters of each layer in the first hypergraph neural network model with trained feature conversion parameters; and inputting the hypergraph incidence matrix and the original feature matrix into the first hypergraph neural network model to obtain a final classification prediction result. Different types of node connections are quantified by defining the association rules, and complex association in a single mode-multi-mode is modeled hierarchically, so that knowledge learning of a complex network under multi-mode collaboration becomes faster and more accurate, and meanwhile, the classification prediction capability of complex association data is improved.

The invention relates to an infrared satellite radiance data cloud detection quality control method based on particle filtering. The method comprises the following specific steps: (1), constructing asatellite view field, and defining a mode layer and the cloud coverage proportion of each mode layer; (2) fitting an all-sky radiance value; (3) updating and normalizing all the particles; and (4) comparing the simulated brightness temperature with the clear sky brightness temperature, and judging whether the channel is not affected by the cloud or not, thereby completing the control of the infrared satellite radiance data cloud detection quality based on particle filtering. Due to the characteristic of non-parameterization, the condition restriction that the random quantity must meet Gaussiandistribution when the nonlinear filtering problem is solved is effectively avoided. The distribution wider than that of a Gaussian model can be expressed, and the higher modeling capacity is achievedfor nonlinear characteristics of variable parameters. The algorithm is quick and effective. The obtained cloud detection identifier can provide effective reference information for a business identification weather system and assimilation of numerical data.

The invention discloses a news recommendation method based on comparative learning. The method comprises a user interest extraction step based on comparative learning; the user interest extraction step comprises the following steps: providing a user interest encoder, wherein the user interest encoder is configured to encode a news sequence browsed by a user to obtain an interest vector; encoding the news sequence browsed by the user to obtain a first interest vector; performing data enhancement on the news sequence browsed by the user, and encoding the news sequence after data enhancement to obtain a second interest vector; training the user interest encoder, and in the training process, introducing interest comparison learning loss which enables the first interest vector to be close to the second interest vector and enables the first interest vector to be far away from interest vectors of other users.

The invention discloses a controllable general dialogue model for intention generalization, and belongs to the field of natural language processing. The system specifically comprises an encoding-decoding structure consisting of a dialogue encoder, an NLU decoder and an NLG decoder, an external database and a rewriter for controlling a text style, for an actual dialogue round of a user, firstly, a dialogue encoder reads a dialogue history, a previous round dialogue state and a current round user input, encoding and feature extraction are carried out to obtain a hidden state, and the hidden state is output to an NLU decoder and an NLG decoder after being preprocessed; the NLU decoder generates a sequence fragment reflecting the intention of the user, maps the sequence fragment into a database checking statement of a database according to the intention of the user, and returns a matching result DB Status by querying an external database; and the NLG decoder generates a reply statement in a natural language form according to the matching result DB Status, and finally feeds back the reply statement to the user. The algorithm complexity is low, the maintenance cost and the expansion cost are reduced, and the prediction efficiency is higher.