44 results about "Structural risk minimization" patented technology

The invention discloses a migration classification learning method for maintaining a sparse structure of image classification. The method includes the steps of finding two different source and targetdomains with similar distribution, the source domain containing label data, firstly, training a classification classifier on the source domain by using a supervised classification method, and predicting a pseudo label of target domain data by using the classifier; secondly, constructing edge distribution and conditional distribution terms of the source and target domain data respectively by usingthe maximum mean difference, and combining the both to form a joint distribution term; thirdly, constructing a sparse representation matrix S on all the data by using an effective projection sparse learning toolkit, to construct a sparse structure preserving term; fourthly, constructing a structural risk minimization term by using the structural risk minimization principle; and fifthly, combiningthe structural risk minimization term, the joint distribution term, and the sparse structure preserving term to construct a uniform migration classification learning framework, and substituting into the framework using a classification function representation theorem including a kernel function to obtain a classifier that can be finally used to predict the target domain category.

The invention relates to a multi-resolution network characteristic registration-based method for sorting face values and face directions of notes in a sorter, which comprises the following steps of: firstly, before performing image recognition and image measurement, balancing the luminance of different points on sampling images; secondly, performing Haar wavelet transform on note images, and extracting approximate picture contents of the note images; thirdly, extracting network characteristics; fourthly, analyzing the separability of the network characteristics on different note images; and finally, establishing a comprehensive sorter by using the network characteristics, wherein a structural risk minimization-based Gaussian mixture model is applied in the sorter. By the method, the real-time processing requirement of the note sorter can be met, and the requirement of processing 1,000 notes per minute can be met. Simultaneously, the defect of inconsistent note images in the same currency and face direction caused by abrasion and printing can be overcome effectively by adopting local characteristic registration so as to perform better recognition.

The invention discloses a knowledge learning and privacy protection based big-data user purchase intention predicating method which comprises following steps of: (1) performing normalization processing on a large number of historical data and a small number of current data; (2) grouping the data and establishing a training sample set; (3) counting user purchase intention probability of each group; (4) calculating group labels; (5) training the training set by using an improved support vector machine; (6) constructing a prediction function; (7) inputting to-be-predicted data into the predication function to obtain a prediction result. As the improved support vector machine is used in the method, the small number of current data set probability information and the large number of historical data set probability information are blended into a structural risk minimization learning framework, learning of knowledge in different periods is realized by virtue of constructing similar distance items among data, and accordingly, the knowledge learning and privacy protection based big-data user purchase intention predicating method which is applicable for learning problems of big samples is constructed.

The invention relates to a mechanical fault identification method for subspace embedding feature distribution alignment under different working conditions. The method comprises the following steps: firstly, aligning a source domain feature with a target domain feature in a target domain subspace by utilizing a related alignment method so as to prevent domain offset; then, directly predicting a pseudo label for a target domain in the spatial training base classifier, and quantitatively estimating respective weights of edge distribution and conditional distribution of two domains so as to adaptto the distribution difference between a source domain and the target domain; and finally, transmitting the learning rules of the two steps through a structural risk minimization framework, constructing a kernel function to establish a classifier, and performing iterative updating to obtain a coefficient matrix of a final framework to complete fault diagnosis. Quantitative estimation of respectiveweights of two-domain edge distribution and conditional distribution is of great significance in cross-domain mechanical fault diagnosis, and feasibility and effectiveness of the method are proved through multi-class composite fault diagnosis examples. The method is suitable for the fields of state monitoring, fault diagnosis and the like of mechanical equipment.

The invention discloses a sample label missing data classifier training method which is suitable for processing classified data of two groups of samples, wherein all label data of one group of the samples are lost. An optimized solution technology is provided, label reliability of unlabeled samples is taken as a decision variable to be solved, an optimizing model is established on the basis of a structural risk minimization principle, the model can directly call a nonlinear-programmed toolkit on a small and medium-sized data set for solution, two convex-programmed sub-problems can be solved respectively by an alternate search algorithm on a large-scale data set, and two part variables of the model are solved by means of iteration. The method is high in universality in the different data sets, and good promotion performance on individual test sets is achieved.

The invention discloses a lithium iron phosphate cell lifetime prediction method based on MIV and an SVM model. The method comprises steps of by applying the MIV algorithm, obtaining the influence importance degree of an input variable to an output variable; then, screening out the most important variable to serve as the input variable; preventing unimportant independent variables to be introducedinto training and test processes of the prediction model; after variable optimization, obtaining a new training set and a new test set only including the optimal variables; and by use of the optimaltraining set and the SVM, carrying out training to obtain the prediction model. According to the invention, the structural risk minimization is used as the optimal criterion in the SVM, so the globaloptimal solution can be acquired; and by combining the prediction model which is subjected to the optimization, only includes variables of circulation times, resistance and the like and is obtained through training, prediction efficiency and prediction precision can be effectively improved.

The invention discloses a lead acid battery service life prediction method. The lead acid battery service life prediction method comprises following steps of obtaining the discharge electric quantity of a battery to be predicted when the a lead acid battery works in different environments, and working out the attenuation ratio of the current battery capacity; working out the practical application invalidation coefficient of the lead acid battery to be predicted according to the attenuation ratio, furthermore establishing an invalidation average speed model of the battery with the invalidation coefficient; setting the self-learning battery state-of-charge to be equal to the original amount, and setting that when the battery invalidation coefficient is a constant, a least square support vector machine LS-SVM decision function is obtained according to the structural risk minimization principle and a least square support vector machine; wherein a kernel function of K(xi, xj) is an inverse function of an exponential function, employing internal operation, solving a regression prediction function, substituting experimental data into the above formula to obtain a plurality of a, b values of the battery to be detected; and substituting a, b values which are subjected to error optimization into a battery residual life prediction model to obtain the service life of the battery.

The invention discloses an oil-well pump running condition recognition method and an oil-well pump running condition recognition device. The method comprises the following steps: inputting target characteristic vector of a target grayscale image to RWELM, carrying out RWELM operation to output a running condition, and training the RWELM by taking the RWELM as a model for oil-well pump running condition recognition and using an indicator diagram and the running condition as training data, wherein structural risk minimization theory is introduced to RWELM in the training process, namely a hidden layer is regulated by virtue of regulating parameters; therefore, an over-fitting problem of a conventional extreme learning machine is solved; and by replacing general hidden layer excitation function through a wavelet function, a problem of local optimum of the extreme learning machine is solved, and the method has the advantages of high diagnosis speed and high accuracy. The device disclosed by the invention, which is embedded in oil-well pump running condition recognition equipment or system, can discover the running condition in real time, thus providing a basis for running condition maintenance of an oil-well pump.

The embodiment of the invention discloses a mineral product detection method, which is characterized by comprising the following steps of: analyzing the ore deposit type of an area to be detected to obtain an ore-forming model of the area; inputting geological information, remote sensing information and geochemical information into the mineralization model to obtain characteristic information; and optimizing the feature information by adopting an interpolation algorithm and a factor analysis algorithm to obtain an evidence layer; carrying out classification prediction on the feature information by adopting a pre-trained support vector machine model to obtain a mineralization detection map. According to the method, calculation of an evidence layer is adopted, so that few samples are needed for prediction, and the method conforms to the actual geological conditions with few known mine points during mineralization prediction; by combining a kernel function, data dimension conversion can be carried out, the solving difficulty of a high-dimensional space problem is reduced, and the calculation rate is increased; and based on a structural risk minimization principle, an over-learning problem is avoided, and the generalization capability is high.

The invention discloses an online fuzzy least square support vector machine sintering process kinetics modeling algorithm. The algorithm fuzzily divides online input k+L groups of data vector space by using a fuzzy c mean value objective function, provides an input space fuzzy membership computing method, constructs a Langrange function according to a structural risk minimization principle, designs a model optimization object of the k+L groups of data according to a KKT optimal solution condition, and solves a model parameter and a model solution, determines the rolling characteristic of the data and whether to roll the online data to k+L+1 to update the model. The modeling algorithm introduces an input data fuzzy clustering idea so as to enhance the generalization capability of the model, and introduces the concept of a rolling time window so as to enhance similar solid sintering process online modeling practical performance.

The invention discloses a power grid network topology structure changing and parallel compensation device selecting method. The power grid network topology structure changing and parallel compensation device selecting method includes S1, grading input vectors; S2, simulating according to network structure parameters, load and small hydropower data, finding out the corresponding network structures and compensation modes of different running modes of a power distribution network, and obtaining several groups of training sets (X, Y) and testing sets (X', Y') of an extreme learning machine (ELM); S3, selecting the hidden layer node number combination Ls of ELM and a structural risk minimization rule constant set gamma s, and selecting an RBF function as an excitation function g(x); S4, training ELM, and testing to obtain the optimal L and gamma to obtain an optical network model of the ELM; S5, outputting the switch combination state with the minimum network loss. The power grid network topology structure changing and parallel compensation device selecting method reflects the nonlinear mapping relationship between input variable and output variable and generalization ability by the aid of ELM to build the network topology structure of which the varied load levels, small hydropower generating capacity and voltage meet demands and build the correspondence between parallel compensation modes.

The invention discloses a method for quantitatively analyzing a maltose mixture based on terahertz spectrum and image information fusion. The method includes the following steps: 1) performing featureextraction on the spectrum and image sample data of the maltose mixture, and fusing the extracted data; and 2) modeling the fusion data modeling of spectra and images in order to quantitatively analyze the maltose mixture. The method has the following advantages: 1, the prediction precision in the invention is significantly better than the prediction precision achieved by using a single spectrumor a single image; 2, a PCA algorithm is used to extract individual features from the spectral data and image data in the invention, so noises are effectively removed, and modeling related feature vectors are effectively extracted, thereby the prediction precision of a multi-source information fusion model is effectively improved; and 3, a Boosting iteration termination judgment index is providedaccording to a structural risk minimization theory, so the automatic optimization of base model parameters of a least squares support vector machine is achieved.

A lead acid battery application life prediction system disclosed by the present invention comprises a parameter obtaining unit, an arithmetic unit and an output unit. The parameter obtaining unit obtains the discharge quantity value, the environment temperature and the residual capacity of a to-be-predicted lead acid battery under different environments and during a work engineering; the arithmetic unit comprises a failure average speed model calculation module for calculating an actual application failure coefficient of the to-be-predicted lead acid battery via an attenuation ratio and further establishing a failure average speed model of the battery with the failure coefficient, and a prediction decision function calculation module for setting a self-learning battery state of charge to be equal to an original bulk, and setting to obtain a least square support vector machine (LS-SVM) decision function when the battery failure coefficient is a constant and according to the structural risk minimization principle and a least square support vector machine; and the output unit outputs a life prediction result.

The invention provides a stem cell activity detection system based on a svm algorithm. The system comprises a subset generation unit, a subset evaluation unit, a stop condition unit and a result verification unit. A support vector machine is widely applied to each filed of machine learning. Correlation problems of classification and mode identification can be solved, and problems of regression, fitting, density estimation and the like can be effectively processed. Compared to other existing machine learning algorithms, the SVM has a solid theory foundation and research and application are wide. The SVM is a small sample learning algorithm based on a statistics learning theory, structure risk minimization is taken as a target and a generalization capability is increased, and under limited sample information, an optimal solution of the problem can be acquired.

The invention relates to a robust oil leakage sea area identification method. A classifier is constructed through a cost-sensitive structural risk minimization model, and a robust oil leakage sea areaidentification method is designed for the unbalanced classification problem that the difference between the number of positive samples and the number of negative samples is too large and the common label exception problems that a large number of labels are lost, and a large number of wrong labels exist and the like in an oil leakage sea area identification task. The robust oil leakage sea area identification method can deal with the situation that the types of the complete polarization synthetic aperture radar samples in the oil leakage sea area are unbalanced, can improve the classificationprecision of the classification problem of label abnormal data, and can meet the actual requirements of the oil leakage sea area identification problem. The robust oil leakage sea area identificationmethod overcomes the problem of class imbalance caused by too large difference between the number of normal sea surface samples and the number of oil leakage sea surface samples, also overcomes the problems that an existing classification algorithm is difficult to process a large number of label missing, has a large number of wrong labels and other label anomalies, and can be effectively used foridentification and classification of offshore oil leakage areas and other practical application problems.

The invention discloses a near-infrared spectroscopy characteristic wavelength selection method for a least squares support vector machine model. The wavelength selection process is guided by the distribution case of a spectral information original sample reflected to a kernel function space, and the key is wavelength selection for reflecting the original sample to the kernel function space so that a large support vector boundary, a small distribution radius R and a large sample space distance are obtained and thus after the original sample is reflected by the kernel function space, a high degree of recognition is obtained. The method has a simple principle, can be realized easily and can easily acquire a model having a high distinction degree. The method carries out wavelength selection based on a structural risk minimization principle, is only based on a training sample set and guarantees model generalization ability after wavelength selection.

The invention discloses an automobile engine failure detection method including the following steps: determining a failure characteristic parameter set and a failure type set of an engine, wherein the failure characteristic parameter set contains multiple failure characteristic parameters, and the failure type set contains multiple failure types; building a structural risk minimization machine learning model based on a kernel function according to the failure characteristic parameter set and the failure type set; using a sample set to train and optimize the structural risk minimization machine learning model based on a kernel function; detecting multiple failure characteristic parameters of the engine, and taking the multiple failure characteristic parameters as a data source of failure detection; and inputting the data source to the structural risk minimization machine learning model based on a kernel function to get a failure detection result. According to the method, by combining a structural risk minimization machine learning algorithm based on a kernel function and principal component feature extraction, rapid and accurate engine failure detection is realized. Moreover, the method has the advantages of strong failure detection capability, high rate of definite diagnosis, and low misjudgment rate.

The invention discloses a method for predicting the parameters of a lithium battery impedance model. The method comprises the steps of: establishing a least squares support vector machine model for predicting target parameters; obtaining a voltage response value of a lithium battery to be tested; obtaining the target parameters of the lithium battery to be tested according to the least squares support vector machine model and the voltage response value. Therefore, the target parameters can be obtained according to the pre-established least squares support vector machine model and the obtainedvoltage response value; and the use of an artificial neural network to predict the parameters of the lithium battery impedance model is avoided, and the least squares support vector machine uses the principle of structural risk minimization, when applied to the prediction problem, the upper limit of a generalization error of the model is reduced while minimizing the sample point error, and the generalization ability of the model is improved, so that the parameters of the lithium battery impedance model can be predicted more accuracy.

The invention relates to a pipeline leak diagnosis combined algorithm based on large data. The pipeline leak diagnosis combined algorithm comprises the following steps that, pipeline pressure data, acoustic wave data and flow data are obtained in real time through a pressure transmitter, an acoustic wave sensor and a flowmeter; by connecting the PPS interface of the GPS system with the communication interface, pressure data, acoustic data and flow data are synchronously stored; measurement data are obtained by using negative pressure wave method, acoustic wave method and mass balance method, neural network algorithm is used to carry out leakage diagnosis after pressure data, acoustic wave data and flow data are fused under multiple scales, and multiple algorithm diagnosis results are obtained at the same time; the final diagnosis results are obtained by mode discrimination with structural risk minimization through synthesizing the diagnosis results of the above algorithms; and the location result of the combined algorithm is obtained through statistical analysis by synthesizing the location distance of each route. According to the pipeline leak diagnosis combined algorithm, the technical means of combining multiple detection methods are adopted, the advantages of multiple algorithms are fused, mutual compensation is achieved, and the sensitivity of leakage detection and the positioning accuracy of leakage positions are improved.

The invention discloses a novel practical support vector machine (SVM) technique used in quality control on a traditional Chinese medicine. The support vector machine (SVM) is a supervised machine learning technique based on Vapnik chervonenks dimensions and the structural risk minimization inductive principle, and is characterized by expressing a sample as a point in a higher dimensional space through an optimal kernel function, generating a hyperplane boundary and realizing sample classification; and different from other machine learning techniques, the SVM can minimize sampling point errors and improve the generalization capability of a model based on the structural risk minimization inductive principle. The SVM technique provided by the invention can detect the quality of a traditional Chinese medicine more sensitively and accurately than fingerprint chromatography, so that similarity evaluation limitations of the fingerprint chromatography are effectively compensated for, thereby achieving more sensitive and reliable quality control on the traditional Chinese medicine.

The invention discloses a structural risk minimization based weighted least squares power system state estimation method. According to the characteristic that the measurement number of power system state estimation is limited, a structural risk minimization based weighted least squares estimation model is provided through a statistical learning theory, so that the norm of a residue can be minimized, and meanwhile the confidence interval of a state variable can be minimized. Besides, a detailed solving process of the method is provided. The method accords with the structural risk minimization thought in the statistical learning theory, can obtain an estimation result more approximate to the true value of the state variable under limited measurement conditions, and has good engineering application prospects.

The invention discloses a source domain selection method for multi-source electroencephalogram migration, which comprises the following steps of: firstly, extracting tangent space characteristics and Grassmann epidemic characteristics, and minimizing marginal probability distribution difference between a source domain and a target domain; after the popularity features are obtained, performing classification model training on each source domain by taking structure risk minimization and conditional probability distribution difference minimization of the source domain and the target domain as a target function, predicting the target domain by each classifier, integrating prediction results of different source domains in a voting manner, and after the first iteration, performing classification model training on the target domain; the method comprises the following steps of: respectively training a classifier for each source domain, and finally voting to generate a multi-source classifier, so that the condition of LSA is met, carrying out LSA once to obtain mobility estimation values of different source domains, removing k source domains, and in the subsequent iteration, only repeatedly training classifier iteration for the remaining source domains, thereby improving the operation efficiency.

The invention provides a multi-source manifold electroencephalogram feature transfer learning method. Firstly, distribution mean values of covariance matrixes of a source domain and a target domain are aligned in a symmetric positive definite (SPD) manifold, tangent space features are extracted, Grassmann manifold features are extracted through Grassmann manifold learning, and marginal probability distribution differences of the source domain and the target domain are minimized. After the popular features are obtained, performing classification model training on each source domain by taking structural risk minimization and conditional probability distribution difference minimization of the source domain and the target domain as a target function, predicting the target domain by each classifier, and integrating prediction results of different source domains in a voting manner. And finally, iteration is carried out to obtain a classification result of multi-source transfer learning.

The invention relates to the field of data correction, in particular to a data synchronization correction method based on material balance, which comprises the following steps of: S1, establishing a model by adopting a support vector regression method based on a significant error related to measurement and material movement information loss; S2, using VC dimension to represent a measurement-related significant error and a material movement information missing quantitative relation; And S3, detecting and correcting measurement-related significant errors and material movement information loss byadopting a mixed integer linear programming method through minimizing the structure risk. According to the invention, measurement-related significant errors and material movement information loss inthe material balance problem can be synchronously detected and corrected.

The invention relates to a wind power plant power curve modeling method based on support vector regression, and the method comprises the steps: firstly the wind speed-power data of a wind power plantis collected, and a regression curve for fitting the upper and lower boundaries of an interval model is obtained; then, when the upper and lower boundaries of the wind power plant interval model are fitted, a convex quadratic programming problem needs to be solved, conversion from a second norm to a first norm is carried out according to the equivalence principle of the norm, and the conversion into the first norm is completed to obtain a linear programming problem; secondly, an optimization problem of control model structure complexity control is applied to regression model identification, and an optimization problem of regression model identification is established by taking minimization of a maximum approximation error as an evaluation criterion; and finally, the optimization problems of the upper and lower edge regression models corresponding to the interval regression model is fused into an optimization problem based on structural risk minimization, and a polynomial kernel is introduced to obtain a new optimization problem of the upper and lower edge regression models. Compared with the traditional convex quadratic programming solution of support vector regression, the methodhas the advantages of high operation efficiency and high solution speed.

The invention relates to a dynamic load model modeling method adopting a support vector machine. The method comprises the following steps: collecting power grid failure data of a specific power branch of a certain power grid according to power grid data acquired in real time; summarizing data of a single-phase earth failure, an inter-phase short-circuit failure and a three-phase short-circuit failure, training load model data by the support vector machine to obtain a support vector machine model, and setting a structure of the model; acquiring an active power load difference equation model of the power branch according to a model equivalence principle; and acquiring a parameter theta and a parameter theta' of a system to be determined according to a linear kernel model and the structure of a system to be recognized. Through adoption of the dynamic load model modeling method, comprehensive consideration is given to experience risk minimization and structural risk minimization, so that an obtained power load model has higher generalization capability, can reflect the dynamic state of the system better, and is more approximate to a response feature of a real power system.

The invention discloses a multi-view classifier based on local features, comprising: an unlabeled multi-view large data set generation module, a global and local structural risk minimization classifier realization module, and a multi-view data local feature extraction module. Its advantage is that it effectively improves the classification performance of multi-view datasets through three aspects: effective data enhancement, construction of classifier design principles, and local feature extraction.