37 results about "Relief algorithm" patented technology

The invention discloses a psychological stress assessment method based on multi-physiological-parameter integration. The method includes: designing a reasonable stimulation program, acquiring four types of electrophysiological signals, namely electrocardiogram signals, electromyographic signals, pulse wave signals and electroencephalogram signals from people suffering psychological stress; extracting affective features of the four types of electrophysiological signals; subjecting the extracted features to feature selection by means of Relief algorithm, genetic algorithm optimization and the like; acquiring related integration functions on the basis of basic probability assignment mass. According to the method, multi-parameter signals are subjected to acquisition, preprocessed, feature selection and psychological stress affective recognition and are integrated; compared to single-parameter classified recognition or multi-parameter data-level or feature-level integration, the method allows data information to be more fully utilized and psychological stress emotions to be more accurately recognized.

The present invention discloses a cancer-related genes finding method by using miRNA expression data, based on a pan-cancer program PanCancer under The Cancer Genome Atlas (TCGA), and uses statistical analysis and a machine learning algorithm to carry out analysis and processing on the gene expression data, and to identify complex diseases related genes. The method comprises: sorting out sample data; carrying out statistical analysis on the miRNA expression data; sorting miRNA in order of an average change rate; selecting a target gene; extracting a corresponding disease sample and normal sample; and using a Relief algorithm to sort genes in the extracted miRNA sample. The method disclosed by the present invention can find a plurality of risk genes related to cancer and other complex diseases, and has important significance to a biological target therapy, biomedical research, pathogenesis explanation, risk prediction, and the like.

The invention discloses a feature selection method facing to SNP (Single Nucleotide Polymorphism) data, which specifically comprises the following steps: firstly, carrying out data pre-processing; secondly, removing unrelated SNP features by using a newly-designed Relief algorithm; thirdly, carrying out critical degree sorting on the SNP features by using an improved SVM-RFE algorithm; and finally, screening the critical SNP sorting by using cross validation. The feature selection method has the beneficial effects that the advantages of Filter feature selection and Wrapper feature selection are combined, and a secondary division method is used in the machine learning process, so that the problems of a high-dimensional small sample in the SNP feature selection and a SNP pathogenic combination mode are solved, and the analysis efficiency and the accuracy are improved.

The invention relates to a feature selection-based arrhythmia classification method. The method comprises the following steps: preprocessing an ECG (electrocardiograph) signal; extracting a morphological feature and a time-frequency feature according to a detected position R, and constructing an original feature vector; calculating feature weights, namely calculating the weight of each feature in the original feature vector by using a Relief algorithm; instructing a population to be initialized according to the feature weights, respectively performing selection, crossover and mutation operation according to a selection probability, a crossover probability and a mutation probability according to the adaptation degree of an individual to obtain a next generation, repeating the operation until the maximum number of iteration times termination condition is met, and then outputting the individual with the highest adaptation degree as an optimal feature; and enabling a plurality of di-classifiers to form an identification classifier through a multi-classification strategy to realize identification of various arrhythmias. According to the feature selection-based arrhythmia classification method, the dimensions of the features can be reduced and the accuracy of identification of the various arrhythmias can be improved.

The invention discloses a congestion control method based on equitable distribution of a communication channel. The method comprises the following steps: firstly, the queue increase rate of an upstream neighbor node and the average transmission delay of a data packet are estimated according to the acquired data of the upstream neighbor node; then, the queue length of the upstream neighbor node is calculated according to the estimated queue increase rate, the total queue length of the node and the upstream neighbor node is further calculated, and the congestion detection is carried out in combination with the average transmission delay of the data packet; if congestion occurs, the network congestion degree (namely the congestion degree) is classified, the length of the congestion relief period is determined, and the node is equitably distributed with the communication channel occupation time; finally, a congestion relief algorithm is carried out according to the congestion degree so as to relieve or ease the congestion; if the congestion does not occur, no treatment is carried out. The use equity of the channel is improved, the collision and the packet loss probability are effectively reduced, and the handling capacity is effectively increased.

The invention discloses a method for predicting the residual life of an aeroengine, The method firstly screens the measurable multi-dimensional performance characteristic parameters in the performancedegradation process of the aero-engine through the Relief algorithm, Principal component analysis (PCA) is used to reduce the correlation between features and extract more concise and effective low-dimensional orthogonal degenerative features, and Kernel smoothing is used to fit the actual degenerative trajectory. And then the similarity matching of the degradation trajectory is carried out, anda feature section set which is most similar to the track section of the to-be-predicted sample in one group of historical samples is found. In order to predict the residual life of aeroengines, the data distribution of the residual life of the similar characteristic segments is obtained by kernel density estimation, and the life prediction estimates of the samples are obtained by density weightedmethod.

The invention relates to a distributed generation island detection method based on meta learning. The distributed generation island detection method based on meta learning comprises the following steps that (1) key features for island detection are recognized by using an RELIEF algorithm, so that a key feature set is obtained; (2) classification prediction is carried out on the original key feature set through a plurality of basic learning devices; (3) the classification results of the basic learning devices serve as feature items to be added to a training set, and a new sample set T is generated; (4) a meta learning device uses the sample set T as a training set, and carries out relearning on the classification results of the basic learning devices to obtain a final detection result. Compared with the prior art, the distributed generation island detection method based on meta learning solves the problems that due to weak correlation features in samples, accuracy of a classification algorithm can be reduced, and overfitting is caused, and has the advantages of being good in applicability, high in detection accuracy and the like.

The invention discloses a novel feature gene selection method based on logistic and relevant information entropy. The method comprises the following steps that a dataset is subjected to logistic regression, a gene variable with great influence on the classification is obtained, a Relief algorithm is used for giving a value on the gene variable, sequencing is carried out, a maximum feature value gene is added to an initial feature gene set, and the relevant information entropy is calculated. The novel feature gene selection method has the advantages that a logistic regression model in the machine study is introduced into the feature gene selection method, and a high-quality gene expression profile is obtained; the correlation between gene variables is measured by the relevant information entropy, redundant genes are deleted, and a feature gene sub set with high classification capability and fewer genes is obtained through searching a feature gene space set.

The invention is applicable to the field of remote sensing technology, and provides an object-oriented change detection method based on multi-feature fusion, comprising the following steps of: S101, image preprocessing; S102, texture feature extraction; S103, image segmentation; S104, feature extraction of that object; S105, generating a difference image; S106, acquire an initial change detectionresult; S107, calculating feature weights; S108, the object change detection result is obtained, and each detected object is clustered into two classes of variable and invariant by fusing multi-dimensional features through weighted fuzzy C-means method. As that embodiment of the invention carry out the processing of the above step on the two-phase images, analyze and determine the weight of each dimension feature, and the weights of each dimension feature are extracted by Relief algorithm, The weighted fuzzy C-means method is formed by adding weights into the fuzzy C-means method, and the weighted fuzzy C-means method is used to fuse the multi-dimensional features, and the detection objects are clustered into two categories: variable and invariant, which effectively fuses the different features to carry out object-oriented change detection and improves the accuracy of the change detection results.

The invention discloses a selection method for a sensitive characteristic index of a persistent attention level. The selection method includes: acquiring an electroencephalogram signal through a plurality of electrodes, recording a reaction time parameter corresponding to the electroencephalogram signal, and acquiring subjective fatigue evaluation data and behavior performance evaluation data; calculating an electroencephalogram entropy parameter according to the acquired electroencephalogram signal; adopting the Kruskal-Wallis detection method and the Relief algorithm to select index sets having the sensitivity different from the persistent attention level respectively; and using an intersection G of the acquired two index sets as the sensitive characteristic index of the persistent attention level. The selection method for the sensitive characteristic index of the persistent attention level can achieve selection of a persistent attention level electroencephalogram e characteristic index sensitive to a bullet train driver.

The invention discloses an intelligent prediction method for cutter suction dredger construction productivity. The method comprises the steps of (1) collecting construction monitoring data of all parts of the dredger; (2) establishing a feature selection model influencing the productivity of the cutter suction dredger according to a Relief algorithm to obtain an optimal feature set; (3) performingnoise processing on the optimal feature set according to a continuous mean denoising method to obtain a denoised feature data set; (4) selecting an extreme boosting tree ensemble learning predictionmodel according to the denoised feature data set, and predicting the productivity of the cutter suction dredger; and (5) evaluating and measuring the productivity prediction result according to various evaluation indexes. According to the method, characteristic elements which are considered to influence the productivity in traditional experience are broken through, new effective factors which influence the productivity are selected, main characteristics are selected by using a data mining method, and the characteristics are subjected to machine learning, so that the productivity of the cuttersuction dredger is accurately predicted, and a new way is found for predicting the productivity of the cutter suction dredger.

The invention provides a post-fault electric power system transient stability prediction method, which comprises the following steps of: 1, on the basis of considering power generator terminal voltage fluctuation, acquiring a feature data set through a time domain simulation method so as to analyze the influence of the power generator terminal voltage fluctuation on electric power system transient stability evaluation accuracy; 2, performing feature selection on the feature data set, and performing dimension reduction processing on an original data set by applying a Pearson correlation coefficient and combining a Relief algorithm so as to quickly screen out a better feature subset; and 3, defining an uncertain sample based on a distance value between a sample and an SVM classification hyperplane so as to reduce the calculation amount. According to the invention, the uncertain samples are screened out and are analyzed by using the feature data at the subsequent sampling moment after the fault is removed, so that the number of the uncertain samples can be gradually reduced, and the prediction accuracy of the classifier is improved.

The invention provides an underground carry-scraper fault diagnosis method based on a PSO-BP neural network, and belongs to the technical field of underground carry-scraper fault diagnosis. According to the method, firstly, original data of mine equipment operation are extracted from a database and classified and transferred according to the model of the equipment, after the obtained data are preprocessed, feature parameters of faults are mined by adopting a Relief algorithm, and a BP neural network model optimized by a particle swarm optimization (PSO) is constructed for fault diagnosis. Compared with the traditional manual screening, the invention has the advantages that the fault features are screened and filtered through the Relief algorithm, a large amount of analysis time can be saved, a more accurate analysis result can be provided; in addition, the optimized PSO-BP neural network overcomes the defect that only the BP neural network is used to converge to a local optimal solution, and the fault diagnosis accuracy of the mine equipment is improved.

The embodiment of the invention discloses a method and device for recognizing a camouflage person through a multi-spectral camera carried by an unmanned aerial vehicle, and a medium. The method comprises the following steps: preprocessing a DN value original image collected by a multispectral camera to complete band registration and radiation correction, and obtaining a preprocessed reflectivity image corresponding to the DN value original image; generating an original feature set based on the preprocessed reflectivity image; screening from the original feature set through cross validation based on a Relief algorithm to obtain an optimal feature subset; based on a preset data set, training the candidate supervised classifier models to obtain classification evaluation results corresponding to the candidate supervised classifier models, and selecting an optimal model from all the candidate supervised classifier models; classifying the preprocessed reflectivity image through an optimal model to obtain a predicted binary image; and identifying a camouflage person target from the prediction binary image, and determining the position of the identified target according to the position information in the DN value original image.

The invention discloses a walk acceleration information segmentation-based personnel identification method. Walk acceleration data of personnel is collected by utilizing an acceleration sensor arranged in a mobile phone; for further mining acceleration information, an eigenvector is detailed, and an acceleration change rate is explored; an HMM-based information segmentation method is proposed; andthe acceleration data is divided into a dynamic part and a steady part according to a relationship between continuous acceleration data sampling points. Features are extracted from the two parts to form dynamic and static feature sets, and the features are selected from the two sets by applying a Relief algorithm to form a new feature set, so that the personnel identification rate is increased. The personnel identification is performed by utilizing a mobile phone, so that the identification timeliness is improved and the cost is low.

Various methods, apparatuses/systems, and media for detecting a target behavior are disclosed. A processor implements a machine learning cadence model that implements an algorithm to obtain, on a per session basis, cadence data that indicates average time between each call and a standard deviation of times across each call across all active sessions of a desired target. The processor compares the cadence data to predefined background cadence data to identify whether the desired target is a new threat target or a background traffic; generates an internet protocol (IP) address of the new threat target; inputs the IP address of the new threat target into a machine learning behavior model that implements an algorithm to generate a fingerprint of all known places that the new threat target is operating; and applies a mitigation algorithm to all active sessions of the new threat target.

The invention belongs to the technical field of power grid fault diagnosis, and particularly relates to a distributed power generation island detection improvement method based on data mining, which adopts a meta-learning mode in machine learning to determine a threshold value of island detection, and the method comprises the following steps: step 1, adopting a RELIEF algorithm to identify key features of island detection before classification, wherein the key features comprise steady state quantity features and transient state quantity features; and step 2, performing a classification algorithm based on the base learner and the meta learner, and implementing distributed power generation island detection. The step 2 comprises the following steps: step 21, data stream mining: adopting an increment mode, and considering time and space efficiency of the algorithm; step 22, meta learning: utilizing complementarity of different classifiers to improve adaptability of data mining and machine learning; and step 23, online self-learning: adopting a sliding data window and a multi-classifier integration method to realize online self-learning, and keeping the classification precision at a relatively high level all the time.