51results about How to "To achieve the purpose of dimensionality reduction" patented technology

The invention belongs to the technical field of path or flight path planning of robots as well as low-altitude flight aircrafts, specifically relates to a path planning method of a passable area divided at unequal distance, and is used for solving the problem that existing planning algorithm has large time complexity in time and space complexity. The path planning method comprises the following steps of: calculating convex extreme points of each barrier curve; dividing the passable area by using each convex extreme point as a horizontal line; abstracting each small area obtained by dividing into a peak of a graph; forming an undirected graph by all peaks; finding out a peak serial number corresponding to the small area at which a starting point and a final point are located; finding out all paths for the undirected graph by breadth-first or depth-first scanning; finding out an actual to-be-travelled path of a moving object according to the situation on an actual map. The path planning method disclosed by the invention has the beneficial effect of overcoming the problems of algorithms of A* and the like on memory space and operation time, and overcoming a convergence problem of an ant colony algorithm at the same time. Besides, time complexity and space complexity are improved greatly in comparison with other algorithms.

The invention discloses a semi-supervised learning-based method for filtering junk users in a social network. A cooperative training algorithm is applied to the detection of the junk users in the social network. Massive information in the social network is classified mainly by utilizing a supervised learning algorithm at present; and the algorithm is based on a classification model built based on annotated data, but the social information scale is huge, the labor cost required for data annotation is high, and few methods for solving the problem of user data annotation of the social network exist. A method is proposed; and by referring to the cooperative training algorithm, multiple views and multiple classifiers are applied to a large amount of non-annotated social network data or a small amount of annotated social network data, so that the classifiers on different views learn mutually and the purpose of data annotation is achieved.

The invention relates to the technical field of sewage treatment quality monitoring, and provides a sewage treatment process monitoring method based on KPLS and RWFCM. The method comprises the steps that firstly, collecting sewage treatment process data samples containing normal working conditions and abnormal working conditions wherein data of sewage treatment operation variables and data of effluent quality variables serve as an input data matrix and an output data matrix respectively, and the two matrixes are standardized; constructing a KPLS model, and solving a score matrix; then, based on an RWFCM algorithm, clustering the score matrix to obtain a membership matrix, and according to the membership matrix, carrying out abnormal working condition monitoring on the sewage treatment process; and finally, establishing a linear regression model of the membership matrix and the sample variables, solving a variable contribution matrix, and performing abnormal condition identification onthe sewage treatment process according to the variable contribution matrix. According to the invention, dimensionality reduction can be carried out on high-dimensional data, nonlinear data can be processed, the method is insensitive to outliers, and timeliness and accuracy of monitoring and identification in a sewage treatment process can be improved.

The invention discloses a novel channel estimation method for a multi-user 3D MIMO system, and the method comprises the steps: carrying out the modeling through employing an included angle between a direction of arrival and an x axis and an included angle between the direction of arrival and a y axis of a planar array antenna, and projecting a 3D MIMO channel to the x axis and the y axis; in the uplink preamble stage, allocating an orthogonal pilot frequency sequence to each user, and obtaining paired spatial features and optimal rotation angles of channels projected to the x axis and the y axis by each user; grouping the obtained paired spatial features, distributing the same pilot frequency information in the groups, and distributing orthogonal pilot frequency sequences among the groups;carrying out channel estimation through the spatial feature and the optimal rotation angle of the (n-1)th coherence time in the group of the nth coherence time after the leader stage, and then dynamically updating the spatial feature and the optimal rotation angle of the user; and repeating the intra-group channel estimation method to obtain spatial features and optimal rotation angles of all users in the cell, reconstructing channels of an x axis and a y axis of all users in the cell, and generating a 3D MIMO channel through a Kronecker product.

The invention discloses a micro-blog short text classification method based on lexical chain feature extension and an LDA (latent Dirichlet allocation) model, and provides a lexical chain feature extension method according to short length, less content, sparse feature and the like of a micro-blog text. A basic lexical chain is generated based on Chinese thesaurus, the micro-blog text is extended by the aid of the basic lexical chain, the lexical chain can cover lexicons recorded by the Chinese thesaurus and can further cover other lexicons without being recorded by the Chinese thesaurus, and the lexical chain can be continually richened when the micro-blog text is extended. The micro-blog text is expressed by the aid of subject probability distribution of an LDA subject model according tohigh dimension and unobvious semantic feature of a vector space model in micro-blog text classification, a similarity calculation dimension is effectively reduced, and a certain semantic feature is fused. According to the method, advantages of lexical chain feature extension and the LDA model are combined, and a micro-blog classification method is provided. Experiment results show that the methodeffectively improves classification performance of the micro-blog text.

The invention discloses a cause-and-effect structure learning method based on flow characteristics. The method comprises following steps of 1: generating and distributing any new characteristic in a flow manner; 2: carrying out a correlation analysis on each of newly generated characteristics; 3: carrying out a redundancy verification analysis on a characteristic set; 4: carrying out searching orientation based on each of the characteristics; 5: repeating steps of 1-4 until the numbers of the generated characteristics exceed a limit value, finally obtaining a corresponding cause-and-effect structure. According to the invention, a cause-and-effect structure relation can be found in linearly randomly distributed data with flow characteristics and time complexity of learning is reduced, thereby satisfying timeliness requirements of online learning.

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 invention discloses an encrypted cloud data-oriented multi-keyword matching sorting search method for a cloud network. According to the method, an index is established in a multi-keyword form, sothat multiple keywords extracted from a document are mapped to one dimension of the index through a comprehensive scoring algorithm; meanwhile, a score matching algorithm is adopted in a search process; and the score matching algorithm is combined with the comprehensive scoring algorithm, so that search results can be sorted according to matching types and the quantity of matching keywords. Therefore, the method reduces the index storage overhead and is high in sorting precision and high in encryption and search efficiency.

The invention discloses a feature selection method and device for model training, electronic equipment and a computer readable storage medium. The method comprises the steps:calculating the mutual information amount of to-be-selected features and labels in an original feature set to obtain the correlation degree between each to-be-selected feature and the corresponding label; calculating the average mutual information amount of the to-be-selected features and the selected feature subsets of the original feature set to obtain the redundancy of each to-be-selected feature and the selected feature subset; and selecting a feature subset from the to-be-selected features as input data for model training according to at least one of the correlation degree and the redundancy. According to the invention, the model training time can be reduced, and the training precision is improved.

The invention relates to a power system operation state simulation method and a power system operation state simulation system. The method comprises the following steps of: performing statistical analysis on the basis of historical data of the operation state of a power system; obtaining probability distribution curves of three kinds of random variables including the power generator active power output, the power generator voltage and the active load of a load node of a power generator; and using a clustering analysis method for respectively clustering the probability distribution curves with similar distribution in the three kinds of random variables. The dimension reduction goal is achieved; the time period for generating mass power network operation state simulation samples can be greatly shortened; and the efficiency is improved. The clustering probability distribution curves subjected to clustering merging dimension reduction are sampled; the power flow distribution is subjected to simulation calculation; a power network operation state simulation sample is obtained; the obtained power network operation state simulation sample is similar to the power network actual operation state; the problem of low precision of the calculation result due to random sampling of each random variable in a designated range in the existing method is effectively solved; and the precision of the power flow distribution calculation result can be effectively improved.

The invention discloses a continuous information forecasting method based on a filter, comprising the following steps: on the basis of multipoint geostatistical method, a filter is utilized to realize the dimensionality reduction of training images, filter score spaces are generated by the filter, all training patterns having similar filter scores are classified into a category in the filter score spaces, training patterns belonging to the same category are extracted in a random manner in the forecasting process, and then the patterns are pasted in an area to be stimulated. In the invention, the method is applicable to the reappearance of the structure feature information of the training images and has better effect in predicting continuous variables.

The invention discloses a clustering thought-based multi-view dynamic brain network feature dimension reduction method, which relates to the technical field of image processing, and comprises the following steps of: clustering vertexes and edges of a built dynamic brain network by using a clustering method, and enabling correlation time sequence rules among the vertexes distributed in the same cluster to be similar; processing each cluster by using a central moment method to obtain a central moment correlation time sequence; then, based on a central moment thought of a central moment, constructing a low-order brain network by using the central moments of the correlation time sequences; furthermore, a high-order dynamic network is constructed on the basis of correlation, and a high-order brain network is constructed by adopting the principle of a low-order dynamic network. According to the method, the reduced low-order dynamic brain network and the high-order dynamic brain network are constructed by using a clustering thought, so that the network scale and the calculation complexity are reduced to a great extent; and establishing a plurality of brain networks by utilizing the central moment characteristics, and providing diagnosis information for disease diagnosis from a plurality of perspectives.

According to the CBL feature extraction and denoising precise webpage classification method, feature extraction is performed on a data set, and noise data in the data set is removed. Firstly, feature extraction is performed based on a feature extraction method of a CBL model, original high-dimensional spatial features are mapped or converted into new low-dimensional spatial features, useless noise data is mapped to a weak dimension, feature items in an original space are greatly reduced, representative feature items are selected according to relevance of the feature items, and the purpose of dimension reduction is achieved. Secondly, noise data is removed based on a noise processing method of a CBL model, a data set is divided into a plurality of subsets according to categories to which the data set belongs, a probability feature topic model corresponding to each subset is constructed, information entropy values of webpages in the data set and the probability feature topic models of the subsets are calculated, and if the information entropy values of the webpages are larger than a given critical value, the webpage belongs to noise data, the junk information is cleared, and the accuracy and precision of webpage classification are greatly improved.

The invention provides a modeling method for a power transmission line icing thickness prediction model based on PR-KELM, and the method comprises the steps: a first stage: converting image data intoLBP image data, carrying out the dimension reduction through employing a PCA algorithm, calculating the gray histogram cascade, and obtaining the extracted image data features; performing feature screening on the meteorological data and the mechanical data by adopting a ReliefF algorithm, and removing highly related redundant features to obtain extracted meteorological and mechanical feature data;and a second stage: forming sample data by using the feature data obtained in the first stage and the icing level in the original image data, training a PR-KELM model by using the training data, testing the trained PR-KELM model by using the test data, and finally obtaining a power transmission line icing thickness prediction model. The method has the advantages that the PR-KELM model is adoptedto predict the icing thickness, selection of the learning rate is not very sensitive, the method is not prone to falling into a local optimal solution, and therefore the accuracy of the prediction model is improved.

The invention discloses a speech recognition method and device, computer equipment, and a storage medium. The method comprises the steps: carrying out the conversion of the original voice information,and obtaining a digital speech signal; obtaining a Hankel matrix based on the digital speech signal; carrying out the SVD (singular value decomposition) of the Hankel matrix to obtain at least two singular values; carrying out the inverse calculation of the SVD of at least two singular values to obtain a target speech signal; and carrying out the reduction processing of the target speech signal,and obtaining the target speech information. The method can effectively inhibit the noise interference, so as to improve the recognition accuracy of the target speech information in the speech recognition process.

The invention discloses a block level cache prefetching optimization method based on deep learning. The method comprises the steps of obtaining IO data with bytes as a unit from a test data set, converting the IO data into IO data with a block as a unit; judging whether the converted IO data is hit in a cache or not; if not, performing sequential prediction on the converted IO data to obtain a plurality of storage blocks, storing the converted IO data in an IO queue of a memory, judging whether the IO queue is full or not, if yes, inputting all the IO data in the IO queue of the memory into the trained Seq2Seq model to obtain predicted IO data, and obtaining a plurality of corresponding storage blocks according to the predicted IO data. According to the method, correlation of IO is mined through a deep learning method, prediction of an IO sequence is completed through the Seq2Seq model based on LSTM, and finally IO sequence prediction and sequential prediction are combined, so that cache prefetching is completed, and the hit rate of a cache is increased.

The invention discloses an epilepsy detection integrated circuit based on a sparse limit learning machine algorithm. Signals in electroencephalogram data are divided by a window containing 256 points,and the data type is a 16-bit fixed-point number comprising an 8-bit integer part and an 8-bit decimal part; four-dimensional signals are generated from the electroencephalogram data of the known type through a wavelet transform circuit module, then the generated signals are input into a feature extraction circuit module to obtain eight-dimensional feature vectors, and then the eight-dimensionalfeature vectors are input into a classifier circuit module for training, classified and output; brain electric signals of the unknown type sequentially pass through the wavelet transform circuit module and the feature extraction circuit module to obtain eight-dimensional feature vectors, and the eight-dimensional feature vectors are input into a classifier circuit module, classified and output. Lifting type wavelet transform is used for processing the electroencephalogram signals, features with different frequency bands in a signal time domain and in a frequency domain can be obtained, and results are superior to those of a traditional filter and Fourier transform.

The invention provides an optical performance monitoring method and device, electronic equipment and a storage medium, and belongs to the technical field of optical performance monitoring in short-distance optical communication, and the method comprises the steps: carrying out the preprocessing of a to-be-processed SVDD signal, and carrying out the resampling; quantizing the trajectory information, and extracting hidden features in the trajectory information; and matching the hidden features. According to the method, accurate joint modulation format recognition, OSNR monitoring and RCD estimation can be realized on SVDD-BPSK, SVDD-QPSK, SVDD-8QAM, SVDD-16QAM, SVDD-32QAM and SVDD-64QAM. The requirements of short-distance communication on low calculation complexity, multi-parameter combined monitoring and high monitoring precision are met. Based on abundant difference characteristics contained in trajectory information, the method has the potential of being applied to backbone optical network and other linear or nonlinear damage monitoring.

The embodiment of the invention discloses a default electricity risk model characteristic selection method and device and equipment. The method comprises the steps that S101 the default electricity tag of a user and the characteristic factor of the user are acquired; S102 an LASSO penalty function is constructed according to the default electricity tag and the characteristic factor; S103 the LASSO penalty function is solved through modified LARS to acquire the valid set of arguments of the LASSO penalty function; and S104 the arguments are filtered according to a set filtering rule and the valid set, so as to acquire a selected characteristic factor. According to the technical scheme provided by the embodiment of the invention, the selection efficiency and the effectiveness of the characteristic factor are improved.