34 results about "Support vector method" patented technology

The invention discloses a SVM (Support Vector Machine)-based multi-sensor target tracking data fusion algorithm. According to target information acquired by sensors, a compact combination mode is adopted, the SVM serves as a middle layer, an environmental variable and a measurement variance normalized vector serve as input of the SVM, output of the SVM serves as trust of the sensor, a known training sample is used for offline training, real-time filter information is used for online estimation, and a fusion knowledge base performs track fusion through real-time weighting according to the obtained trust of the sensor. According to the SVM-based multi-sensor target tracking data fusion algorithm and the system thereof, the SVM principle is adopted, the algorithm complexity is low, the environmental variable and the measurement variance normalized vector are introduced, the biological robustness and the fault tolerance are strong, extension is easy, and the algorithm and the system thereof are applicable to the field of multi-sensor tracking.

Disclosed is an improved technique for training a support vector machine using a distributed architecture. A training data set is divided into subsets, and the subsets are optimized in a first level of optimizations, with each optimization generating a support vector set. The support vector sets output from the first level optimizations are then combined and used as input to a second level of optimizations. This hierarchical processing continues for multiple levels, with the output of each prior level being fed into the next level of optimizations. In order to guarantee a global optimal solution, a final set of support vectors from a final level of optimization processing may be fed back into the first level of the optimization cascade so that the results may be processed along with each of the training data subsets. This feedback may continue in multiple iterations until the same final support vector set is generated during two sequential iterations through the cascade, thereby guaranteeing that the solution has converged to the global optimal solution. In various embodiments, various combinations of inputs may be used by the various optimizations. The individual optimizations may be processed in parallel.

The invention relates to an aero-engine reliability monitoring method based on mixed weibull distribution, and the method comprises the following steps of extracting state monitoring information and performance degradation information of an aero-engine which is already replaced and repaired; extracting a relation between each monitoring parameter and the performance degradation of the aero-engine by utilizing a support vector method, and realizing the monitoring of the performance degradation value of a wing engine; utilizing a degradation model to describe the accumulated degradation volume of the aero-engine on the basis of the monitoring result of the real-time degradation value of the wing aero-engine; estimating a random parameter in a linear degradation model of the aero-engine, and determining the variation of the random parameter and the upper limit and the lower limit of the accumulated performance degradation volume of the aero-engine; establishing an aero-engine reliability monitoring model based on the dual-parameter mixed weibull distribution; utilizing a maximal likelihood method to give an expression of each parameter in the aero-engine reliability monitoring module; estimating hyper-parameters of the aero-engine reliability monitoring module based on the dual-parameter mixed weibull distribution; and calculating the reliability monitoring value of the aero-engine, and realizing the real-time and precise reliability monitoring of the aero-engine.

The invention discloses a magnetic resonance parallel imaging method of a multi-support vector model, which belongs to the field of magnetic resonance parallel imaging. The method comprises completely sampling an intermediate region in a K space, dividing into a training set and a testing set, performing accelerated sampling on the other regions to obtain samples as a prediction set, and performing normalization processing on the data in the sets; dividing the training set into a plurality of training subsets, and selecting different parameters to train each training subset by using a support vector to obtain different combined weighting function models; testing the combined weighting functions on the testing set, and selecting a plurality of optimal submodels; and predicting the prediction set by using the optimal submodels, taking an average value as a value of uncollected points, performing reverse normalization processing, and converting K space data into an image. The parallel imaging method has good generalization and small overall reconstruction error by using weighting functions fitted by the support vector.

The invention discloses an intelligent online quality detection method for an automobile transmission, belonging to the field of automobile noise and vibration control. Specific to the important and complexity of online delivery testing, synchronous acquisition of vibrating voltage signals and rotating speed pulse signals is realized on the aspect of signal acquisition. The vibrating voltage signals acquired in the delivery testing process of the automobile transmission are unstable signals, so that the phenomenon of 'frequency alias' occurs when the conventional signal processing method is adopted. Processing of unstable signals of the transmission is performed by combining an order method with a cepstrum method for extracting characteristic signals, so that the problem of 'frequency alias' is solved, sidebands of power spectra are eliminated, and characteristic signals are convenient to observe and extract. A support vector method is used on the aspect of quality detection, and a support vector model is continuously enriched and updated by using a standard extremal curve method, so that the model comprises more information, and the aim of realizing high detection speed and high accuracy is fulfilled.

To provide a method capable of reducing support vectors without decreasing the performance of an SVM. The method includes: a step of learning an SVM by using a set of training samples for initial learning which have known labels; a step of evaluating a training sample for initial learning corresponding to an outlier (value greater than 0 and equal to or less than C) based on a parameter a value obtained by learning the SVM; and a step of removing the training sample for initial learning corresponding to the outlier from a set of the original training samples for initial learning.

The invention relates to a reciprocating compressor intelligent diagnosis method based on EMD-PCA. Intelligent diagnosis of faults of a reciprocating compressor is realized by comprehensively using an empirical mode decomposition (EMD) method, a principal component analysis (PCA) method and a multi-class classification support vector method; the method is characterized in that common faults of the reciprocating compressor are subjected to feature extraction, the fault alarm and diagnosis are performed according to different sensitive features of different faults, acceleration signals of a reciprocating compressor cylinder are subjected to feature extraction by using empirical mode decomposition, and each group of acceleration signals have 30 features; then dimension reduction is performed by using principal component analysis to obtain a three-dimensional feature space; finally, five working conditions obtained by experiments are diagnosed by using multi-class classification support vector, and the intelligent diagnosis process framework of the reciprocating compressor is summarized and proposed. Through the adoption of the method, fault diagnosis accuracy under the five working conditions is high.

The invention discloses a degradation data missing interpolation method based on a support vector machine and an RBF neural network. The method includes the following steps that firstly, a degeneration data trend model is established by means of the support vector machine; secondly, residual error sequences of observed degradation data are calculated; thirdly, the RBF neural network is set up, and the network is trained by means of the residual error sequences of the observed degradation data; fourthly, residual error sequences of missing data are estimated through the trained RBF neural network; fifthly, trend terms of the missing data and estimation results of the residual error sequences are merged, so that a degradation data interpolation result is obtained. A support vector machine method and an RBF neural network method are combined to obtain the degradation data missing interpolation method, and the problem of interpolation of performance missing degradation data in an accelerated degradation test is solved.

The invention relates to a data stream prediction method and a data stream device. The data stream prediction method comprises the steps of: updating an integration model index according to sample data, wherein the integration model index is used for storing the mapping relationship between a keyword and a support vector set, the support vector in the support vector set is one of an SVM (support vector machine) classifier in the integration model; performing word segmentation to an enter text to obtain the key word of the enter text, wherein the enter text is the data stream for predicting; according to the updated integration model index, searching the support vector set containing the key word and the information of the SVM classifier in which the support vector in the support vector set is arranged; and predicting the enter text by all support vectors in the searched support vector set. The data stream prediction method and the data stream device provided by the invention perform sublinear on-line prediction based on the integration model index; and because of aggregating the support vectors according to the key word by an inverted list, the prediction is obviously accelerated.

The present invention relates to a modeling method for a support vector machine based on data compression. The modeling method has the technical characteristics that the method comprises the following steps: sampling modeling data through an equidistant sampling method; compressing the modeling data; calculating the boundary of each cluster of data under leaf nodes of a clustering feature tree, and choosing a boundary point most possibly becoming a support vector as the modeling data of the support vector machine; and establishing a model of the support vector machine: establishing a model of the support vector machine according to the modeling data through a support vector machine method. In the modeling method of the present invention, the modeling sample quantity of the support vector machine is greatly reduced under the condition of ensuring the accuracy rate of the algorithm to the greatest extent through a pre-sampling strategy, a data compression technology, an increment sampling strategy and the like, so as to greatly improve the modeling speed of the support vector machine and lower the memory consumption, so that the support vector machine technology can be applied to a big data analysis scene, thereby remedying the defect that a neural network method, a Bayes method and the like in the big data analysis have low prediction accuracy.

The invention provides an island detection method based on binary tree complex wavelet transformation, and belongs to the field of electric power systems. The island detection method comprises the following steps of acquiring control variables of a three-phase voltage type inverter, and establishing a PQ decoupling control mathematical model of a grid-connected inverter based on the acquired control variables; acquiring electrical change characteristics of a public connection point of a microgrid island under a power matching condition according to the established PQ decoupling control mathematical model; extracting voltage and current harmonic signals at the common connection point by means of a wave trap, decomposing the voltage and current harmonic signals based on binary tree complex wavelet transformation to construct a characteristic vector corresponding to the micro-grid; and improving a support vector machine on the basis of a support vector selection and genetic algorithm, andadopting the improved support vector machine to identify the island state of the microgrid. By virtue of improvement of the support vector selection and genetic algorithm, damage of the active detection method to the electric energy quality can be eliminated, and the problem of detection blind areas existing in the traditional passive detection method also can be solved.

The invention discloses a text classification method based on sample scaling. The method comprises the following steps: calculating the distance from a data sample to a classification hyperplane, finding a sample far away from the classification surface of the support vector machine, deleting the sample, endowing corresponding weights to the remaining samples according to the distance, and training the support vector machine by using the weighted data samples. According to the classification method provided by the invention, the sample data is firstly reduced, and then the data is correspondingly weighted so as to be used for text classification in a support vector machine. The influence of noise data on support vector machine classification can be reduced, the noise immunity of the modelis improved, the number of support vectors is reduced, and better text classification accuracy is obtained.

The invention relates to a point cloud compression encoder key parameter optimization method based on a support vector machine, and the method comprises the steps: carrying out the preprocessing of the geometric information and color information of a point cloud; extracting a feature vector of the point cloud; for a given target code rate, finding an optimal parameter pair which minimizes distortion by using a full search method; extracting an optimal tag under a given target code rate from all point clouds in the training set; and writing the optimal label information and the feature vector information into a training set, performing training by using a support vector machine and the training set information to obtain a model, testing the feature vector information in the test set by using the model, and predicting an optimal test label on a continuous domain to obtain an optimal parameter pair of the test set. According to the method, the distribution characteristics of the point cloud are utilized, a support vector machine method is used for training to obtain the optimal coding parameter pair of the test point cloud, and the time cost is greatly reduced while the coding performance of the encoder under the condition of a given coding bit rate is ensured.

The invention discloses a distributed support vector clustering method which comprises the following steps: processing the input data set according to a preset processing rule, and initializing global parameters and tasks; distributing a preset data set or a specific calculation result to each computational node; initializing a weight vector of the preset data set when the preset data set is distributed to the computational node; performing iterative operation according to a preset formula, and calculating a weight coefficient value of each sample in the preset data set; finding a sample of which the weight coefficient value is higher than a preset minimal value to serve as a support vector, and numbering the support vector; and building a support function by utilizing the support vector and the weight coefficient of each support vector, performing cluster division to obtain a cluster label of the support vector, and calibrating a non-support vector sample as a clustering analysis result. According to the method, the support vector clustering efficiency can be effectively improved.

The invention relates to a hearth outlet NOx prediction system based on numerical simulation. The hearth outlet NOx prediction system comprises a prediction model building part and a real-time prediction part. The invention also relates to a prediction method which comprises the following steps: processing DCS data, obtaining boundary conditions required by numerical simulation calculation, establishing a three-dimensional geometric model of a hearth, calculating NOx concentration data of a hearth entrance under different working conditions, and determining a database; according to the database, establishing a hearth outlet NOx prediction model by adopting a support vector method; and finally, predicting the NOx concentration at the outlet of the hearth in real time according to the real-time inlet parameters of the hearth under the actual working condition. The method can predict the change of NOx at the outlet of the hearth and the distribution of NOx in the furnace when the combustion condition changes, namely the operation parameters of air/powder and the like are adjusted, can guide SCR ammonia injection in advance, and has important significance for improving the denitration efficiency of a coal-fired power plant, the operation safety of an air pre-heater and the economical efficiency.

The invention relates to the technical field of aviation generator testing, in particular provides a generator life prediction algorithm based on support vector base. According to the algorithm, structural risk minimization principle is adopted to replace the empirical risk minimization principle for generator life prediction, the problem of sample learning under limited data can be solved better,the idea of kernel function is adopted, the problem of nonlinear space is transformed into that of high dimensional feature space, the original problem can be solved by linear classification method in high dimensional feature space, the algorithm complexity is reduced, the maximal spacing principle is adopted, so that the effect of processing of classification samples is better, the effect of life prediction is good, and it is achieved that the maintenance of the aviation generator changes from the ex post maintenance or regular maintenance to the on-condition maintenance, so that the maintenance cost is lowed.

The invention discloses an SVM-based ultra-supercritical boiler heating surface pollution monitoring method, and belongs to the field of monitoring systems. Traditional calculation results based on heat balance and heat transfer principles often have large access to actual working conditions. An SVM-based ultra-supercritical boiler heating surface pollution monitoring method applies a support vector method to a regression prediction process, minimizes a convex function, and obtains a sparse solution: defining a loss function which can ignore a real value error, i.e., an epsilon insensitive loss function; an epsilon-SVM regression machine is established on the basis of the epsilon insensitive loss function; based on the epsilon-SVM regression machine, an upsilon-SVM regression machine is developed. The mean square error of the detection result is very small, and the curve can well track the actual process.

The invention provides a method for detecting waveform points of the voltage sag based on a space vector method. The method specifically comprises the following steps of: S1, obtaining an instantaneous value waveform of a three-phase voltage of a voltage sag to be detected; S2, extracting basic features of known waveforms; S3, constructing a space vector through the three-phase voltage vector, andcalculating a radius change rate of the space vector; S4, finding the position where the waveform point is located of a sampling point, through the change rate of the space vector radius; and S5, calculating waveform points based on the position of the sampling point, which is obtained by combining the angle of the space vector x (t) with the real axis. According to the method for detecting waveform points of voltage sag based on space vector method, the method combines the space vector by the three-phase voltage, detects waveform points according to the different characteristics of the spacevector when the sag does not occurs and the sag occurs, and solves the errors existing in the principle of the existing algorithm.