52 results about "Additive model" patented technology

A method and an apparatus for improved duration modeling of phonemes in a speech synthesis system are provided. According to one aspect, text is received into a processor of a speech synthesis system. The received text is processed using a sum-of-products phoneme duration model that is used in either the formant method or the concatenative method of speech generation. The phoneme duration model, which is used along with a phoneme pitch model, is produced by developing a non-exponential functional transformation form for use with a generalized additive model. The non-exponential functional transformation form comprises a root sinusoidal transformation that is controlled in response to a minimum phoneme duration and a maximum phoneme duration. The minimum and maximum phoneme durations are observed in training data. The received text is processed by specifying at least one of a number of contextual factors for the generalized additive model. An inverse of the non-exponential functional transformation is applied to duration observations, or training data. Coefficients are generated for use with the generalized additive model. The generalized additive model comprising the coefficients is applied to at least one phoneme of the received text resulting in the generation of at least one phoneme having a duration. An acoustic sequence is generated comprising speech signals that are representative of the received text.

The method of the present invention provides a fast, robust, and automated multivariate statistical analysis of gas chromatography / mass spectroscopy (GC / MS) data sets. The method can involve systematic elimination of undesired, saturated peak masses to yield data that follow a linear, additive model. The cleaned data can then be subjected to a combination of PCA and orthogonal factor rotation followed by refinement with MCR-ALS to yield highly interpretable results.

The invention provides a joint fractal-based method for detecting a small target under a sea clutter background. The joint fractal-based method is higher in detection probability. The detection problem of a non-additive model is transformed into a classification problem, i.e. whether a target exists or not is equivalent to belong to a class in which a pure sea clutter exists, and a characteristic joint detection algorithm is provided. A bilogarithmic graph is established by using a trend fluctuation method through sea clutter data, a slope, namely a Hurst index, is fitted by using a least square method within a scale-free interval, and is used as a characteristic scalar, a nodal increment of a keypoint in the bilogarithmic graph is used as another characteristic scalar, therefore, a double-scalar obtained by each group of sea clutter data corresponds to one point in the bilogarithmic graph, n groups of corresponding points (i=1,...n) of the pure sea clutter data are obtained by using the steps, a space optimal classification line omega is obtained by using a convex hull function, sea clutters of regions in which the target possibly exits are obtained by using the same steps, and finally, by using whether the points exist in the space optimal classification line omega or not as a criterion, when the points exists in the space optimal classification line omega, no target exists, and when the points are outside the space optimal classification line omega, the target exists.

The present invention provides a microalloyed steel mechanical property prediction method based on globally additive model, including the following steps: determining some influencing factors of the microalloyed steel mechanical property prediction model; calculating the components and contents of carbonitride precipitation in a microalloyed steel rolling process; expressing the microalloyed steel mechanical property prediction model as an additive form of several submodels according to generalized additive model; estimating the microalloyed steel mechanical property prediction model; and verifying reliability of the submodels. The microalloyed steel property prediction models obtained in the foregoing solution have advantages such as high prediction precision and a wide adaptation range, and may be used for design of new products and steel grade component optimization, so as to reduce the quantity of physical tests, shorten the product research and development cycle, and reduce costs.

The invention relates to a big data-based time-space confusion exposure degree assessment system and method. The system comprises a time-space data mining block, a multi-source heterogeneous data fusion module, a final variable selection module, a time-space generalized additive model building module, a re-sampling model module, a variation function time-space modeling module and a concentration estimation module; massive time-space data is mined; a relationship between multiple influence factors and pollutant concentration is established by adopting an accumulative nonlinear method; and through residual variation function fitting, spatial autocorrelation is considered, so that the prediction precision and effect are greatly improved.

Image representation is performed by dividing a source image into foreground, background and selector planes. The foreground plane is selected to contain mainly line type art or textual type information, the background plane mainly contains image data, and the selector plane identifies whether the image data is maintained in either a specific plane or a combination of planes. A color is selected, by averaging or selecting an appropriate value based on overflow or other criteria, to replace each color in the foreground plane. Error in portions of the foreground plane resulting from replacing foreground colors is fed into corresponding portions of the background plane. Each plane is then compressed using compression schemes appropriate for the type of data maintained in each plane (LZW for the foreground, and JPEG for the background and lossless fax LLITT, for example). Image reconstruction is performed by decompressing each of the foreground and background planes, and selecting pixels from each of the foreground plane and an additive image produced by combining image data from both the foreground, background, and selector planes. The selection is made based on the selector plane (selection mask), which identifies where image data is maintained for the reconstructed image (i.e., the upper plane or a combination of planes).

The invention provides a completion method of pavement temperature response missing values and aims to solve the problems that traditional missing value completion methods are low in reliability, inconvenient to apply and the like. Under the conditions that pavement temperature monitoring is interrupted due to failures and temperature response has missing values, the missing values are completed. According to the method, the natural law that the atmosphere temperature fluctuates periodically and seasonally is used, an addition model adopting sine and cosine curves is adopted to simulate change features of pavement temperature response at the same moment every day, and corresponding mathematical models are established; on the basis of the randomness characteristic of pavement structure temperature response, a random number sequence related to the burying depth of a temperature sensor is put forward to correct the models, and the real situation of temperature response monitoring data can be better simulated. With the adoption of the method, the pavement temperature response missing values can be scientifically and effectively completed, and the continuity, stability and reliability of pavement temperature monitoring data can be well guaranteed.

The invention belongs to the field of video processing technology, and specifically discloses a method for sensing stereoscopic video coding based on a parallax just-noticeable difference model. The method comprises the following steps: (1) estimating parallax; (2) estimating a JND model based on the parallax; (3) calculating a brightness, texture and time weighted JND model; (4) combining the JND model based on the parallax with a spatial domain-time domain JND model by using a nonlinear additive model to obtain a binocular stereoscopic JND model based on the parallax; and (5) using the binocular stereoscopic JND model based on the parallax on a stereoscopic residual preprocessor for resetting a residual. The method disclosed by the invention can be used for effectively eliminating inter-view redundancy of time, space and binocular stereoscopic videos, and the information of the brightness, a texture region or an object edge keeps a very natural visual effect. Therefore, the method disclosed by the invention can be used for greatly reducing the stereoscopic video code rate on the premise of generating no influence on the stereoscopic visual perceptual quality.

The embodiment of the invention provides a method and device for conducting self-adaption spectrum reduction and wavelet packet noise elimination processing on voice signals. The method mainly includes the steps of calculating power spectra of an additive model of the voice signals; obtaining a power estimation value of the voice signals according to the power spectra of the additive model of the voice signals and power estimation values of signals with noise and noisy signals; conducting inverse Fourier transformation on the power estimation value of the voice signals to obtain strengthened voice signals; conducting wavelet transformation processing on the strengthened voice signals through the adoption of a wavelet transformation matrix; and conducting threshold processing on the wavelet transformation matrix, conducting inverse transformation on the wavelet transformation matrix after threshold processing, and obtaining the voice signals without noise. By means of the method and device, the problem of the background noise of the voice signals in a teleconference is effectively solved; in addition, other kinds of noise like music noise cannot be caused.

A method for modeling a system configuration, including isotonizing an unconstrained additive model of a cache architecture to obtain an isotonic additive model for the cache architecture, wherein the isotonic additive model is of the same functional form as the unconstrained additive model, smoothing the isotonic additive model using a flat spot technique to obtain a characterization of the cache architecture, and modeling a system configuration using the characterization of the cache architecture.

The invention provides a WLAN indoor positioning algorithm based on the linear discriminant analysis and the gradient lifting tree. In order to reduce the influence of the time-varying characteristicof the received signal strength value (RSSI) in the indoor WLAN environment on the positioning accuracy, a positioning feature for extracting the original RSSI signals by using the LDA is provided, and the prediction of the position coordinates is realized by constructing a GBDT model. The method is mainly divided into four processes. (1) an RSSI signal value of the AP is collected at a referencepoint, and fingerprint data is formed at the position coordinate of the reference point, and is saved in a fingerprint database; (2) an intra-class divergence matrix and an inter-class divergence matrix of the fingerprint data are solved so as to obtain a projection matrix, and the extracting of the RSSI signal positioning features can be realized; (3) a forward distribution algorithm and an addition model are utilized, and a GBDT positioning model is generated through iteration; (4) in the online stage, the RSSI signal values of the AP at the periphery of the test points are collected, and anLDA is used to perform feature extraction and the GBDT positioning model is input to calculate position coordinate.

An embodiment of the invention provides a self-adaption spectral subtraction and noise elimination processing method and device for voice signals. The method mainly comprises establishing an additive model of the voice signals according to signals with noise, noise signals and voice signals; calculating a power spectrum of the additive model of the voice signals and obtaining a power estimation value of the voice signals according to the power spectrum of the additive model of the voice signals and power estimation values of the signals with noise and the noise signals; performing inverse Fourier transformation on the power estimation value of the voice signals and obtaining the voice signals with noise being removed. The self-adaption spectral subtraction and noise elimination processing method for the voice signals effectively solves the background noise problem of the voice signals in a teleconference and not causes other noise such as musical noise.

The invention provides a microalloyed steel mechanical property prediction method based on a global additive model. The method comprises the steps that influence factors of a mechanical property prediction model are determined; precipitated components and content of carbonitride in the rolling process of microalloyed steel are calculated; the mechanical property prediction model of the microalloyed steel is expressed into an additive form of a plurality of submodels according to the generalized additive model theory; the mechanical property prediction model of the microalloyed steel is estimated; and the reliability of all the submodels is verified. Through the scheme, the obtained product property prediction model of the microalloyed steel has the advantages of high calculation precision, wide adaptive range, etc., can be used for new product design and steel grade component optimization and achieves the purposes of reducing the number of physical tests, shortening the product development cycle and lowering cost.

The invention relates to a brand new method for predicting the acid-base dissociation constant (pKa) of organic small molecules based on a hierarchical atom additive model. Based on the linear relationship rule of free energy changes of acid-base dissociation balance at a given temperature, the method firstly establishes the hierarchical atom additive model based on a Hammett-Taft equation and a processed substituent effect of 'dissociation center-rest part' brought forward by Cherkasov and the like, and then calculates the acid-base dissociation constant (pKa) value of a corresponding compound according to the concrete structure of the compound through the model. The method does not have the difficulties of large number of substituent electronic effect constants and correction factors relating existing in the prior method, ensures the speediness and accuracy of the prediction, shows good data fitting and predictive ability on a plurality of sample sets, breaks through the prior method which studies the drug pKa at an early research and development stage of a new drug, reduces the drug research and development cost on a large scale, and improves the discovery efficiency of the new drug.

The invention discloses a medium-and-long-term power load prediction method and device, computer equipment and a storage medium. The method comprises the steps of performing wavelet transform on historical daily power consumption data to obtain corresponding approximate components and detail components; performing feature extraction on the approximate component and the detail component to obtain feature data; inputting the feature data of the approximate component into a time series prediction model based on an additive model for learning to obtain a prediction result; inputting the feature data of the detail components into a time-based convolutional neural network model for learning to obtain an output result; taking the output result as a target feature, and learning the target featureby using a lightweight gradient boosting decision tree model to obtain a prediction result; and summarizing the prediction result of the approximate component and the prediction result of the detail component to obtain a final power consumption prediction result, thereby constructing a prediction model about the medium and long term power load. The invention can effectively improve the medium andlong term power load prediction precision.

The invention discloses a monocular video-based human skeleton tracking method. The method comprises the following steps: in a training stage, a, feature descriptor definition: quantifying the difference between a sketch and an initial skeleton by utilizing a distance-based method, b, feature extraction and classification: extracting features having max-relativity with a regression target through a relativity-based random fern method and classifying the features to obtain a regressand which decides the skeleton adjustment range of the current stage, and c, skeleton adjustment: adjusting the skeleton for proper times according to an addition model and outputting a cascading regressand; and in a test state: a, inputting a sketch and an initial skeleton of the first frame of a monocular video, b, gradually adjusting the initial skeleton to a final skeleton according to the cascading regressand obtained through training, and c, searching 5 skeletons mostly similar to the current frame of predicted skeleton, respectively regressing the 5 skeletons, and taking the regressed average skeleton as the initial skeleton of the next frame to predict the skeleton of the next frame. According to the method, the tracking of skeletons can be effectively realized, and the error accumulation phenomenon in the tracking process can be avoided.

The invention provides a modeling method of an evaluation model and an apparatus thereof. The method comprises the following steps of collecting modeling samples from a plurality of modeling scenes respectively, wherein each modeling sample includes a scene variable and several basic variables, and the scene variable indicates the modeling scene which the modeling sample belongs to; based on the modeling samples collected from the plurality of modeling scenes, creating a modeling sample set; and based on the modeling samples in the modeling sample set, training the evaluation model, wherein the evaluation model is an additive model, and the evaluation model is formed through adding a model portion formed by the several basic variables and a model portion formed by the scene variable. In the invention, a modeling complexity can be reduced and a grade acquired based on the model can be suitable for different service scenes.

The invention discloses an individualized living yield prediction model and a construction method. The construction method comprises the following steps: a, collecting living yield outcome, individualized ovarian function information, individualized biological information and individualized ovarian excretion promoting information of each sample patient; b, extracting parameter variables accordingto the accumulated live birth outcome and the individualized information; c, constructing a semi-parameter generalized additive model gm2 of the parameter variable; d, performing verification and sensitivity analysis on the model gm2, and training and evaluating the stability of the model gm2; and e, drawing a living yield prediction model curve graph. The method has the advantages that guidance can be provided for doctors and patients receiving ART treatment, the possibility of obtaining at least one live delivery infant can be predicted according to the age of the patients and the egg takingnumber, and the patients can be informed whether the live delivery possibility of the patients is remarkably increased or not after the patients receive extra stimulation periods.

The embodiment of the invention relates to a short-term load prediction method and system based on principal component analysis, and terminal equipment. The method comprises the steps of: dividing data obtained from an electric power system and an electric power meteorological system into training sample data and prediction sample data; performing correlation analysis on the training sample data to obtain a factor characteristic value influencing a load; performing dimension reduction processing on each factor characteristic value influencing the load through adoption of a principal componentanalysis method to obtain a principal component characteristic value influencing the load; and using a semi-parameter additive model for superposing the non-linear influences of all principal component characteristic values on the load to establish a load prediction model, so that the influence of interaction between factor characteristic values on load prediction is effectively reduced, the prediction precision of the load prediction model is improved, and the problems of limited usage scenarios and low prediction precision of an existing power load prediction model are solved. The method scientifically and comprehensively extracts factor characteristic value variables influencing the load to provide a more practical reference basis for a power market load predictor to make a scheme.

The invention discloses a hand-foot-and-mouth disease epidemic situation prediction method based on case, weather and aetiology monitoring data. The prediction method includes: analyzing the relationship of the hand-foot-and-mouth disease cases with weather and aetiology factors and the hysteresis effect, and screening the indexes included in a model; on the basis of multi-source data of the hand-foot-and-mouth disease case, weather and aetiology, constructing a hand-foot-and-mouth disease model through a time series generalized additive model; dividing the multi-source data into a training data set and a verifying data set, thus evaluating the fitting situation and prediction result of the hand-foot-and-mouth disease epidemic situation prediction model. By combining the case, weather, aetiology and population data, a prediction model is constructed by using the time series generalized additive model. The data then is separated into different data sets to train and verify the fitting situation and prediction result of the model. Therefore, foresight prediction and risk warning on the trend of the hand-foot-and-mouth disease are conducted in real time. The method is more reliable inprediction result and is higher in timeliness and practicability.

The invention discloses an image correlation detection method based on time sequence sparse regression and an additive model. The method comprises four steps. According to the method, a plurality of weights related to the time pattern smoothing function and ROI are obtained through the four steps, so that correlations between the brain region time progress track and the regional difference of the phenotype are obtained by describing the specific region through the contribution of a plurality of SNPs to the phenotype. The minimum root-mean-square error RMSE is used as an evaluation index for judging whether a specific brain area degeneration progress track is matched with the model or not, the RMSE of the method is 0.15, the RMSE of a model based on a time sequence sparse additive model in the prior art is 1.14, the RMSE of a sparse additive model and the RMSE of a group sparse additive model are 1.33, and therefore the effect is better.

The invention discloses a process industry fault diagnosis method based on a bidirectional long-short-term neural network, and the method comprises the following steps: S1, carrying out data set preparation, wherein a TE model introduces a fault from 160 groups of data, and a data set is used for building a monitoring model; S2, carrying out feature extraction, wherein the feature extraction adopts data feature extraction based on a gradient elevator, and an additive model composed of a plurality of classifiers is found out in the gradient descent direction; S3, establishing an experiment platform; S4, carrying out an experiment: building a bidirectional long-short-term neural network model by adopting a Keras framework, wherein a research object is a Tennician Eastman model; and S5, giving out experimental results. According to the method, the strong generalization ability of the bidirectional long-short-term neural network is utilized, the defects of gradient disappearance and gradient explosion of a long sequence can be avoided, and the problems of low accuracy, frequent missing report and false report phenomena and low generalization ability in process industry fault diagnosisare solved.

A local wind speed condition based discrimination method of vertical distribution pattern of Microcystis flos-aquae in the large shallow lake is as below: arranging monitoring points in lake with Microcystis flos-aquae, sampling stratified water under different wind conditions, and determining the proportion of algal biomass in the total biomass in different water layers by chemical analysis and microscopic examination; selecting a regression function using the AICc criteria for screening, and using the proportion as a dependent variable to build a unified function for algae vertical distribution pattern under different wind conditions; extracting the unified function coefficients, using a general additive model to determine contribution and a threshold of wind speed as a main factor, and using an regression analysis to determine the relationship between the unified function coefficients and the wind speed; and after integration, constructing a prediction model of wind speed based vertical distribution pattern of Microcystis flos-aquae. The present invention can obtain the vertical distribution pattern of Microcystis flos-aquae, and provides scientific and technological support for the estimation of the total amount of cyanobacterial bloom and control of cyanobacterial bloom.

The invention provides an in-pipeline corrosion growth prediction method and device based on a generalized additive model. The method comprises the steps: screening independent variables, obtained inadvance, of an in-pipeline corrosion growth prediction model through a Lasso compression estimation algorithm; based on a generalized linear additive model, establishing an in-pipeline corrosion growth prediction model according to the screened independent variables and the in-pipeline corrosion growth rate; and predicting the in-pipeline corrosion growth rate according to the in-pipeline corrosion growth prediction model. According to the method, modeling can be carried out on the basis of existing corrosion data in the pipeline, and the future corrosion depth is accurately estimated. Therefore, the internal detection period can be determined, a maintenance plan can be made, and safe pipeline operation and cost saving are facilitated.

The invention discloses a component software reliability analysis method based on an improved additive model, relating to software reliability analysis methods. When a unified framework based on a system structure model is built, the problem of the additive model is that the additive model has no regard for the system structure and execution features of component software application. The additive model solves an application system structure model without using a white-box analysis method, and realizes modeling for the service condition of the component in an application execution process. For the component processed by unit testing, the following steps are executed: defining the service condition of the component in a pi i modeling integration test, wherein the pi i refers to the execution time proportion of a component ci in the steady state of a component software system; when the execution time of the component software system is t, expressing the accumulated execution time of the component ci as ti=pi i t; so, expressing the number of failures of the component software system as Formula until the time t; expressing the failure speed of the component software system at the time t as Formula; and improving the component software system aiming at the currently raised problems of the additive model.

The invention discloses an SAR image speckle reduction method based on a directional wave domain HMT model, which is mainly used for solving the problem that detail information such as the edge structure of the image cannot be effectively kept and the radiation characteristic of an SAR image cannot be kept well in the existing method. The SAR image speckle reduction method comprises the following realization steps of: (1) inputting an image; (2) establishing a non-logarithm additive model; (3) constructing a directional wave domain HMT model parameter set; (4) establishing the directional wave domain HMT model parameter set; (5) estimating the directional wave high-frequency sub-band coefficient of a spotless image; (6) carrying out directional wave inverse transformation; (7) outputting the estimated spotless image. The SAR image speckle reduction method disclosed by the invention has the advantages of being capable of effectively keeping the edge and detail information of the image, and more obvious in speckle reduction effect, and can be used for speckle reduction processing for the SAR image.

The invention provides a multi-source heterogeneous data fusion architecture system based on an additive model. The multi-source heterogeneous data fusion architecture system comprises a multi-sourceheterogeneous data preprocessing module, a multi-source heterogeneous data fusion module and a model summing module. When the architecture system works, by calling a matched meta-model from a pre-established meta-model base, data of different formats can be processed, and data fusion is achieved through a model summing part. By means of the multi-source heterogeneous data fusion architecture system based on the additive model, the problem that during a multi-source heterogeneous data processing process, data easily gets lost, and needs to be matched with a target data source format according to the format of a data source can be effectively solved; meanwhile, the contribution of each newly added external data source to a target domain problem can also be analyzed, and high-speed processingand analyzing of mass heterogeneous data can be achieved.

The invention discloses a model interpretability method for detecting interaction characteristics in the financial risk control field, and relates to the field of graph neural network technology and model interpretability, and the method comprises the following steps: 1, training an additive model according to the existing training characteristics and training labels; step 2, using an interaction feature detection module to detect an existing interaction feature pair, using the obtained interaction feature pair, the training label and the residual error in the step 1 to construct a GAM model, and adding the additive model and the GAM model to obtain a GAM2 model containing the interaction feature pair; 3, according to the principle that when high-order interaction exists and only when all low-order interaction exists, multi-round iteration is conducted on the GAM2 model till the interaction feature order of a certain round of iteration is not increased any more, and a GAMn model is obtained; and step 4, realizing visualization and report export.

The invention discloses a knowledge and capability binary tracking method based on continuous matrix decomposition. The method comprises the steps of constructing a training set based on historical learning behaviors, and determining a first likelihood function and a first log-likelihood function of the training set; determining knowledge model parameters according to the first log-likelihood function, and constructing a knowledge model based on the knowledge model parameters; determining a second likelihood function of a to-be-constructed joint model based on the output data, and determininga target function of a capability model according to the second likelihood function; determining capability model parameters based on the target function, and constructing a capability model based onthe capability model parameters; and the knowledge model and the capability model are combined to obtain the joint model, and the joint model is an additive model or a multiplicative model. Accordingto the method, the implicit capability model is constructed on the basis of the continuous matrix decomposition model, the two models are fused and trained through the lifting algorithm, and comparedwith a traditional model, the method has higher interpretability and model accuracy.

A method for spatially authoring data in a data processing system, may include constructing one or more input spatial geometry regions and iterating through each input spatial geometry region to create current cumulative result data and rejecting geometry groups from the current cumulative result data. The method may also include for each particular input spatial geometry region of the one or more input spatial geometry regions, constructing minimal-split BSP trees from the particular input spatial geometry region and current cumulative result data, performing geometry processing by applying an additive modeling comparison rule to keep geometry outside of the particular input spatial geometry region with the current cumulative result data, and keep geometry outside of the current cumulative result data with the particular input spatial geometry region, and performing geometry processing by applying a subtractive modeling comparison rule to keep geometry outside of the input spatial geometry region with the current cumulative result data, and keep geometry inside of the current cumulative result data with the particular input spatial geometry region. The method may further include generating final result geometry after iterating over all of the one or more input spatial geometry regions.