62 results about "Bayesian information criterion" patented technology

The invention discloses a complex audio segmentation clustering method based on a bottleneck feature. The method comprises the steps that a deep neural network with a bottleneck layer is constructed; a complex audio stream is read, and endpoint detection is carried out on the complex audio stream; the audio feature of a non-silent segment is extracted and input into the deep neural network; the bottleneck feature is extracted from the bottleneck layer of the deep neural network; the bottleneck feature is used as input, and an audio segmentation method based on the Bayesian information criterion is used, so that each audio segment contains only one kind of audio type and adjacent audio segments have different audio types; a spectral clustering algorithm is used to cluster segmented audio segments to acquire the number of audio types of complex audios; and the audio segments of the same audio type are merged together. According to the invention, the used bottleneck feature is a deep transform feature, can more effectively describe the feature difference of the complex audio type than a traditional audio feature, and acquires an excellent effect in complex audio segmentation clustering.

The invention discloses an expansion target tracking method based on GLMB (Generalized labelled multi-bernoulli) filtering and Gibbs sampling. The expansion target tracking method based on GLMB filtering and Gibbs sampling estimates the target number and the shape of the expansion target, provides a multiple expansion target tracking method under a labelled random finite sets (L-RFS) framework, and mainly includes two aspects: dynamic modeling of multiple expansion targets and tracking estimation of multiple expansion targets. The expansion target tracking method based on GLMB filtering and Gibbs sampling includes the steps: combined with a generalized label multi-bernoulli filter, establishing a measurement limit hybrid model of the expansion targets, by means of Gibbs sampling and Bayesian information criterion, deriving the parameters of the limit hybrid model to learn tracking of the state of the multiple expansion targets, using an equivalent measurement method to replace measurement generated from the expansion targets, and performing ellipse approximating modeling on the shape of the expansion targets to realize estimation of the shape of the expansion targets. The simulation experiment shows that the expansion target tracking method based on GLMB filtering and Gibbs sampling can effectively track the multiple expansion targets, can accurately estimate the state and theshape of the expansion targets, and can obtain the track of the targets.

A method learns a structure of a video, in an unsupervised setting, to detect events in the video consistent with the structure. Sets of features are selected from the video. Based on the selected features, a hierarchical statistical model is updated, and an information gain of the hierarchical statistical model is evaluated. Redundant features are then filtered, and the hierarchical statistical model is updated, based on the filtered features. A Bayesian information criteria is applied to each model and feature set pair, which can then be rank ordered according to the criteria to detect the events in the video.

The invention discloses an output-only linear time-varying structure modal parameter identification method and belongs to the technical field of structural dynamics. Firstly, a cost function of a least squares support vector machine vector time-varying autoregressive model is deduced; secondly, a function space is built by means of a Wendland compactly supported radial basis function; a regular factor is determined through the non-parameter method based on Gamma testing, and a basis function width reduction coefficient is given on the basis of actual experiences; a time-varying autoregressive model order is determined according to the Bayesian information criterion and the Akaike information criterion; a function space order is determined according to the ratio of residual sum of squares to sequence sum of squares; finally, the matrix expression of the least squares support vector machine vector time-varying autoregressive model is solved according to the cost function, modal frequency of a system is solved according to a time freezing method, and linear time-varying structure modal parameter identification is finished. The method can improve calculation efficiency, improves system robustness, and is widely used in linear time-varying structure modal identification in structural dynamic engineering application.

A method (200) and apparatus (100) for segmenting a sequence of audio samples into homogeneous segments (550 and 555) are disclosed. The method (200) forms a sequence of frames (701 to 704) along the sequence of audio samples, and extracts, for each frame, a data feature. The data features form a sequence of data features. Transition points in the sequence of data features are thin detected by applying the Bayesian Information Criterion to the sequence of data features. The transition points define the homogeneous segments (550 and 555). Preferably the data feature is single-dimensional and a leptokurtic distribution is used as an event model in the Bayesian Information Criterion.

The invention relates to the online speaking people cluster analysis and particularly relates to the online speaking people cluster analysis based on bayesian information criterion. The method comprises steps of collecting original audio signals and dividing the original audio signals into audio segments with boundaries through the bayesian information criterion, performing audio characteristic extraction on the audio segments, clustering the segments with audio characteristics through the bayesian information criterion, and forming a plurality of clustering groups like the clustering group 1, clustering group 2....clustering group n. An online speaking people cluster analysis system comprises two modules which are an audio signal segmentation module and a segment cluster analysis module, which greatly improve the accuracy of segmentation, guarantee high efficiency of clustering, realize high efficient parallel transcription, segmentation, classification and clustering of signals of the online speaking people on a premise that the audio materials of the original speaking people are not required and realize high efficiency transcription, segmentation, classification and clustering of the signals of online speaker.

The invention discloses a data prediction method and system. The data prediction method comprises the steps of acquiring data information of a single service through a database, wherein the data information comprises a multi-dimensional characteristic variable; performing data screening on the multi-dimensional characteristic variable by combining an Akaike information criterion (AIC) value and a Bayesian information criterion (BIC) value according to a multivariate regression method, and filtering data noise; and importing the data screened characteristic variable into a machine learning model, and performing modeling analysis on the data. The accuracy of data prediction is effectively improved according to the data prediction method and system.

The invention discloses a voice signal processing method. A non-voice part in a voice signal is cut off through endpoint detection, and a plurality of first voice signal fragments are obtained; Bayesian information criterion BIC detection is performed on the plurality of first voice signal segments to obtain speaker transformation points; the speaker transformation points are used as segmentationpoints to divide the plurality of voice signal segments so as to obtain a plurality of second voice signal segments; therefore, the problem that the traditional BIC-based segmentation method is low incalculation efficiency can be solved, and the effect of accurate and rapid recognition of the speaker transformation points of the voice signals is achieved.

A method and a system of predicting an electric system load based on wavelet noise reduction and empirical mode decomposition-autoregressive integrated moving average (EMD-ARIMA) are provided. The method and the system belong to a field of electric system load prediction. The method includes the following steps. Raw load data of an electric system is obtained first. Next, noise reduction processing is performed on the load data through wavelet analysis. The noise-reduced load data is further processed through an EMD method to obtain different load components. Finally, ARIMA models corresponding to the different load components are built. Further, the ARIMA models are optimized through an Akaike information criterion (AIC) and a Bayesian information criterion (BIC). The load components obtained through predicting the different ARIMA models are reconstructed to obtain a final prediction result, and accuracy of load prediction is therefore effectively improved.

The invention relates to a shot clustering method based on the spectrum segmentation theory, which comprises the following steps: utilizing, the spectrum segmentation theory for shot clustering; extracting feature vectors of each unspecified shot; calculating similarity between each two categories according to the extracted feature vectors; then constituting each shot cluster as a weighted undirected graph; segmenting each shot category into two shot categories by a using spectrum according to the similarity between each two categories; using Bayesian information criteria to judge whether the segmentation is effective or not, the effectively segmented shot sub-categories are iteratively segmented, the ineffectively segmented shot categories are terminals; finally syncretizing the classification results after the segmentation to get the optimal shot classification number and the classification result. The invention solves the difficult problem that the optimized classification number is difficult to estimate in the clustering algorithm, and improves the recall ratio and the pertinency ratio of the clustering result by utilizing the precise classification spectrum segmentation; the proposed overall fusion operation has a function of correcting the classification errors, thereby effectively avoiding the problem of local optimum relation.

The invention provides a fan prediction management method and device, an electronic device and a storage medium. The method comprises the steps of obtaining an original time sequence of fan quantity parameters of each unit time period of history of a to-be-predicted platform; using the ARIMA model as a modeling sample to predict the fan quantity of the platform in the future to-be-predicted time period, so that data reference is provided for operation management of the to-be-predicted platform, and the purposes of fan suction and fan benefit mining and conversion are achieved; on model parameter selection, using the distribution condition of autocorrelation indexes and partial autocorrelation coefficients in a stationary sequence after difference; rapidly and accurately determining an initial value of a model autoregressive item parameter and an initial value of a moving average item parameter; using the minimum information amount criterion and the Bayesian information criterion to select an optimal model to predict the fan amount of the platform in the to-be-predicted time period, so that the fan amount prediction precision is greatly improved, the fan amount prediction precision is almost consistent with actual measurement data, and effective reference can be provided for fan operation of the public account in advance.

The invention belongs to the technical field of digital audio signal processing and discloses an automatic digital audio tampering point positioning method based on the BIC (Bayesian information criterion). The method comprises the steps as follows: performing active voice detection on a to-be-detected tampering signal to determine a silence fragment in the voice signal; sequentially extracting the Mel-scale frequency cepstral coefficient characteristic of each frame after framing of the silence fragment, and performing long window framing in time sequence; calculating the BIC value of each long-term characteristic frame; taking all crest points in a sequence constituted by BIC values of all long-term characteristic frames as suspicious tamper points, and cutting off the silence fragment front and back with the suspicious tamper points as midpoints; calculating a BIC value sequence of each cut-off window containing suspicious points. Automatic positioning of digital audio tampering points is realized; compared with a traditional tampering detection method, the method has the advantages that the calculated amount is reduced, the omission ratio of the tampering points is reduced, thethreshold selection problem is solved; the method has robustness for the condition of covering of noise with the tampering points.

The invention discloses a feature extraction method for performance degradation evaluation of a rolling bearing. The method comprises the following steps of S1, acquiring vibration signal informationof the rolling bearing; S2, conducting self-adaptive EEMD decomposition on a vibration signal of the rolling bearing; S3, adopting a Bayesian information criterion and a correlation kurtosis method for screening sensitive IMF components, wherein firstly, the Bayesian information criterion is adopted for calculating the number of the sensitive IMF components, secondly, the sensitive components arescreened out according to the values of the correlation kurt (CK), finally, composite spectral analysis is conducted on the sensitive IMF components, and a calculated composite spectral entropy servesas a feature parameter of the performance degradation of the rolling bearing. According to the method, a composite spectral analysis method is adopted for fusing the selected IMF components, the composite spectral entropy is extracted as the degradation feature of the rolling bearing, the sensitivity to the degradation process is high, and the capability of characterizing the degradation processof the rolling bearing by the feature is improved.

The invention discloses a runoff probability forecasting method, wherein, the method mainly comprises the following steps: adopting a method based on K. Medoids clustering method is used to cluster the training set, and the initialization parameters of HMM are obtained. Using Baum-Welch algorithm to study HMM, the state transition probability matrix of HMM and the probability distribution of observation model are obtained. According to Bayesian Information Criterion (BIC), the model is selected and the number of HMM states suitable for the training set is selected. Finally, the conditional probability distribution function is obtained by Gaussian mixture regression GMR reasoning according to the given forecasting factors as the runoff probability forecast. The probability forecasting method of the invention introduces the concept of runoff hidden state, and can obtain the hidden state transition probability matrix by using hydrology, topography, meteorology and other factors, and obtain effective and reliable future runoff probability forecasting distribution, thereby providing scientific basis for reservoir optimal operation and decision-making.

The invention provides a soil environment quality zoning method and system. The method comprises the following steps: extracting soil environment quality comprehensive characteristics of monitoring points in a target area based on a principal component analysis method; screening out main influence indexes of the soil environment quality by adopting a geographic detector; establishing a series initialization pre-classification scheme, and determining an optimal pre-classification scheme according to a Bayesian information criterion; constructing a Gaussian mixture model according to the optimal pre-classification scheme, and estimating hidden variable parameters representing sample point categories in the Gaussian mixture model through an EM algorithm to obtain initial classification of the monitoring points; and obtaining an initial partition based on the corresponding Thiessen polygon of the monitoring point, and performing final partition on the target area in combination with natural boundary information of the target area. According to the method, on the basis of the comprehensive characteristics of the soil environment quality of the monitoring points, the Gaussian mixture model based on the EM algorithm is constructed, and comprehensive partitioning of the soil environment quality based on the high-dimensional attribute characteristics is achieved.

A load prediction method for an electric vehicle charging station based on a copula algorithm comprises the following steps of (1) classifying users; (2) classifying and fitting data; (3) extending the data; and (4) outputting a load curve of the electric vehicle charging station. According to the load prediction method for the electric vehicle charging station based on the copula algorithm, the users are firstly classified; a kernel function and a copula function are selected by use of an AIC (Akaike Information Criterion) and a BIC (Bayesian Information Criterion), data are fitted by use ofa kernel density function, extended data including a coupling relation between the data are obtained in combination with the copula algorithm, the extended data can really reflect actual charging behaviors of the users, and an obtained prediction curve better accords with the actual situation.

The invention discloses a signal source number estimation method based on Gerschgorin circle transformation and modified Rao score inspection. The method comprises the steps: firstly, calculating a sample covariance matrix of an observation signal; then, carrying out Gerschgorin circle transformation on the sample covariance matrix, and by utilizing the estimated value of the characteristic valueof the sample covariance matrix obtained after transformation and on the basis of the modified Rao score inspection thought, detecting the structural characteristics of the large-dimensional covariance matrix; and then, by detecting whether the covariance matrix of the noise part in the observation signal is in direct proportion to a unit matrix, constructing an observation statistical magnitude used for establishing an information theory criterion likelihood function, wherein the statistical magnitude is also the statistical magnitude of a sample characteristic value; and on the basis, carrying out signal source number estimation through a generalized Bayesian information criterion. The method provided by the invention has relatively wide applicability, is suitable for signal source number estimation under a classic asymptotic system, and is also suitable for signal source number estimation under a common asymptotic system; and the method is suitable for signal source number estimation in a white Gaussian noise environment and is also suitable for signal source number estimation in a color noise environment.

The invention provides a source number estimation method under the frame of the Bayesian Information Criterion (BIC), and is suitable for large-scale self-adaptive antenna scenes, under generalized asymptotic conditions, namely m and n are infinite, m / n is equal to c belonging to zero to infinity, wherein m and n respectively represent the number of antennas and the number of snapshots, and the reliable detection of the source number is provided under the condition. According to the source number estimation method disclosed by the invention, the prior probability is obtained through the co-calculation of a log-likelihood function and a cost function, and the source number is effectively obtained through maximizing the prior probability. Simulation results prove the superiority and the effectiveness of the source number estimation method disclosed by the invention.

The invention relates to an industrial process fault diagnosis method based on the Bayesian information criterion. The method comprises: collecting normal industrial data and calculating several kindsof detection statistics amounts based on normal data; carrying out fault detection on a to-be-detected sample; expressing a fault isolating task into a combinatorial optimization problem; convertingthe problem into a mixed integer nonlinear programming problem by combining the Bayesian information criterion; on the basis of a forward selection algorithm, simplifying the problem into a mixed integer quadratic programming problem; on the basis of a branch-and-bound algorithm, solving a series of similar mixed integer quadratic programming problem to obtain a fault variable combination causingthe sample fault. The industrial process fault diagnosis method has high universality; and the fault variable can be identified without predetermining a fault direction or a known historical fault data set. When the amplitude of the fault is small, an accurate diagnosis result is obtained. Besides, the combination optimization problem is transformed into the quadratic programming problem with sparse constraints for calculation, so that the computational efficiency is improved substantially.

The invention discloses a living energy consumption prediction method and system based on ARMA and regression analysis. The method comprises the steps of obtaining a per capita living energy consumption item and measured value thereof; establishing a first sample corresponding to the measured value of the per capita living energy consumption, and constructing a time series; determining the order of an ARMA model according to the Bayes information criterion and constructing the ARMA model; establishing a sample set with an impact factor of a reality factor and a time sequence as a second sample; performing a regression analysis on the second sample and obtaining a combined prediction model; and using the combined prediction model to carry out combined prediction on the time sequence. The combined machine learning prediction model based on ARMA and regression analysis in the invention can better adapt to the characteristics of time series and accurately describe the actual influencing factors and has a beneficial effect with high test accuracy.

The invention provides a Granger causality discrimination method based on a quantitative minimum error entropy criterion. According to the method, the coefficient and the order of a regression model are determined by adopting the quantitative minimum error entropy criterion and a Bayesian information criterion, a causality discrimination index is obtained by calculating the error entropy and coefficient, and the causality between two time sequences is determined according to a causality judgment standard. Compared with a traditional Granger causality discrimination method based on a minimum mean square error criterion, the method is more accurate in estimating coefficients of the regression model, the obtained error entropy is smaller, and the causality discrimination index can be more accurately calculated. Due to the adoption of a quantization method, the calculation complexity of the method is remarkably reduced. The method integrates the error entropy and the coefficient when calculating the causality discrimination index, which makes the calculation of the causality discrimination index more accurate and robust. Therefore, the Granger causality discrimination method based on the quantitative minimum error entropy criterion provided by the invention is more easily promoted and used in practical applications.

The invention discloses a fine classification and prediction method and system for a resident electrical load mode, and the method comprises the steps of collecting the electrical load data and weather data of a resident; screening the meteorological features based on a Bayesian information criterion, wherein the meteorological features meeting the conditions form a meteorological feature library;performing clustering analysis on the resident power load data to obtain a resident power consumption mode; improving the LSTM network by using a fusion activation function; and based on the improvedLSTM network, predicting the residential electricity loads in different electricity consumption modes. The beneficial effects of the invention are that the classification and prediction method provided by the invention can achieve the more precise classification of the power consumption modes of residents, and obtains a more precise prediction result according to the classification result.

The invention relates to a method and apparatus for performing time series forecasting through machine learning. The method herein includes the following steps: pre-processing the acquired time series data, and obtaining a result of the pre-processing; on the basis of the result of pre-processing, detecting whether the time series data contains seasonal cycles; if the result of the pre-processing determines that the time series data contains the seasonable cycles, in accordance with the Akaike information criterion and bayesian information criterion, selecting a time series model; if the result of the pre-processing determines that the time series data does not contain seasonable cycles, adding the time series data to a time series data pool, and if the number of the added time series data is greater than a preset threshold value, returning to the step of selecting a time series model based on the Akaike information criterion and bayesian information criterion. According to the invention, the method herein has the beneficial advantages that the method uses automation procedures to complete time series prediction, continues the optimization of models by means of machine learning, and increases prediction accuracy.

The invention discloses a fatigue driving judgment method based on an unsupervised extreme learning machine multi-clustering algorithm, belongs to the technical field of driving safety, and determinesan optimal classification cluster number and a probability density distribution function under each class through a Gaussian mixture model and a Bayesian information criterion, and determines an optimal identification model in a fatigue identification data set. Through a feature extraction non-iterative algorithm of an unsupervised extreme learning machine, a minimum value converged to the wholeenvironment is obtained, and an output matrix is obtained; the advantages of four clustering algorithms under unsupervised extreme learning machine feature extraction under different feature divisionlearning are fully utilized through a PCA algorithm; and component score coefficient matrix calculation is performed on the fatigue identification point identification accuracy matrix, and a normalized score coefficient is converted into a weight coefficient for balancing four clustering algorithms in the field of fatigue identification, so that the precision of training set data clustering tendsto be balanced.

The invention relates to signal processing and artificial intelligence, and aims to provide a rotating machinery fault diagnosis method based on independent component analysis and correlation criteria. The method comprises the steps of: measuring multi-channel vibration signals of the rotating machine, and performing transformation through an independent component analysis algorithm; selecting a separation signal based on a correlation criterion and a Bayesian information criterion, and extracting a fault feature vector; and training and diagnosing the diagnosis model of a rotary machine support vector machine, inputting the feature vector R into the model to perform fault diagnosis on the rotary machine, classifying diagnosis results according to a preset judgment condition, and giving analarm. According to the method, the problem that previous independent component analysis is only applied to artificial diagnosis and is rarely applied to artificial intelligence diagnosis is solved,and the accuracy of fault diagnosis of the rotary machine is improved to a greater extent, so that the method has a very good use value.

The invention relates to a Gibbs parameter sampling method applied to a random point mode finite hybrid model. The method comprises the steps that firstly, a random point mode finite hybrid model anda random point mode likelihood function are constructed, then random point mode finite hybrid model parameter prior distribution is constructed, and posterior distribution of model parameters is obtained according to the model parameter prior distribution; and finally, estimating the number of distribution elements in mixed distribution and model parameter values by adopting a sampling algorithm combining a Gibbs sampling algorithm and a Bayesian information criterion. Compared with the traditional FMM which only describes the characteristic randomness of the data, the random point mode distribution function also describes the cardinal number randomness of the data; on the basis of RPP-FMM, a Gibbs sampling algorithm is adopted to sample sample data to obtain model parameters, and the situation that parameter estimation may fall into a local extreme point all the time, and a global extreme point cannot be obtained is avoided. According to the method, the modeling precision and the parameter estimation precision are effectively improved.

The invention discloses a wind power non-parametric probability interval ultra-short-term prediction method. The method is based on adaptive LASSO and an extreme learning machine. Firstly, nonlinear quantile regression is carried out on a wind power sequence to obtain self-adaptive adjustment parameters; an optimal quantile regression model output coefficient is calculated based on an extreme learning machine by utilizing quantile regression based on adaptive LASSO and an improved Bayesian information criterion; and finally, a wind power time sequence is input to obtain an ultra-short-term prediction value. According to the quantile regression prediction model constructed through the method, the interval score is obviously superior to that of a traditional prediction model based on quantile regression, the prediction precision and interval width comprehensive indexes are good, and the reliability of wind power prediction is greatly improved.

The invention discloses a multi-state system reliability evaluation method considering state transition correlation. The complex correlations between state transitions are described by copula function, and the unknown parameters in the model are estimated by the state observation data of the system. The best model is selected by using Bayesian information criterion. On this basis, the reliabilityand conditional reliability of the system are calculated. Since the method of the invention uses a copula function to model the correlation between system state transitions, the failure law of the system in each state is different and the correlation forms between the state transitions are various. Compared with the prior method, the present invention can realize the reliability modeling of the multi-state system with complex failure law, thereby improving the accuracy of the system reliability evaluation.