344 results about "Applicability domain" patented technology

The invention discloses a neural network training method and device so as to solve a technical problem that a student network trained by the existing knowledge transfer mode cannot give consideration to a large application range and accuracy. The neural network training method comprises the steps of selecting a teacher network which realizes the same functions as the student network; and iteratively training the student network based on distribution of first intermediate layer features and second intermediate layer features corresponding to matching with the same training sample data so as to acquire a target network, and thus transferring knowledge of intermediate layer features of the teacher network to the student network, wherein the first intermediate layer features are a feature map outputted from a first specific network layer of the teacher network after inputting the training sample into the teacher network, and the second intermediate layer features are a feature map outputted from a second specific network of the student network after inputting the training sample data into the student network. The neural network acquired by adopting the technical scheme disclosed by the invention is not only wide in application range but also excellent in performance.

The invention discloses an advanced geological prediction method for underground engineering. With geology as pivot, the advanced geological prediction method for underground engineering comprises four stages of long-term advanced geological prediction, middle-term advanced geological prediction, short-term advanced geological prediction and impending advanced geological prediction. The advanced geological prediction method comprises the following steps of: (1) making advanced geological prediction tasks for underground engineering for the underground engineering clear; (2) collecting data; (3) carrying out field geological review; (4) perfecting the modification of a longitudinal section drawing of the underground engineering; and (5) adopting combined geological method, geophysical prospecting method and horizontal drilling method to obtain advanced geological prediction. The invention provides the advanced geological prediction method for underground engineering; with geology as pivot, comprehensive geological analysis is carried out in the whole process of advanced geological prediction for underground engineering and the long-term, middle-term, short-term and impending advanced geological predictions are closely combined together to implement combination of geological method, geophysical prospecting method and horizontal drilling method; the comprehensive application of the geophysical prospecting method is optimized; and the advanced geological prediction method for underground engineering, provided by the invention, has the advantages of improving the predication accuracy, reducing predication cost and obtaining wide application range, etc.

The invention discloses a method for fitting and interpolating a G01 code based on a quadratic B spline curve, comprising the following steps of: by an adaptive approach selecting each characteristic point of each group of small line segment which is described by the G01 code; fitting a route which is to be processed with the quadratic B spline curve of all the characteristic points; according to the characteristic of the quadratic B spline curve and the limit of the acceleration of each driving shaft of the numerical control machine, simultaneously obtaining the maximum permissible machining velocity curve (VLC curve) of the quadratic B spline curve and the each speed key point on the VLC curve; according to the each speed key point, the control axis of the each key point, the maximum permissible machining velocity and the VLC curve, computing real machining velocity; according the real machining velocity curve and a interpolating error computing interpolating point and completing real-time interpolation. The invention has fast computing velocity, high machining precision, stable working performance and wide application range, can complete the interpolating computation of the spline curve in real time and meet digital control processing requirement of fast velocity and high precision under a premise that the preset precision of the system is met.

The invention discloses a training method and a training device of a convolutional neural network (CNN) model, and belongs to the field of image recognition. The training method comprises the steps of respectively carrying out a convolution operation, a maximum pooling operation and a horizontal pooling operation on a training image so as to acquire a second feature image; determining a feature vector according to the second feature image; carrying out processing on the feature vector so as to acquire a category probability vector; calculating a category error according to the category probability vector and the initial category; adjusting model parameters based on the category error; and continuing the model parameter adjusting process based on the adjusted model parameters, and using model parameters at the moment when the number of iterations reaches a preset number of times as model parameters of the well trained CNN model. According to the invention, the convolution operation and the maximum pooling operation are carried out on the training image on different levels of convolution layers, and then the horizontal pooling operation is carried out. The horizontal pooling operation can extract a feature image marking a horizontal direction feature of the image from the feature image, so that the well trained CNN model is ensured to recognize images of any size, and the application range of the well trained CNN model in image recognition is expanded.

Novel tools and techniques for generating and / or implementing an applied semantic knowledgebase. Some tools allow for data integration into coherent, semantically connected networks and for generation of sets of query-based models describing complex functional relationships as sub-networks. In an aspect, an applied semantic knowledgebase may comprise collections of SPARQL network queries describing a specific set of sub-network relationships and their applicable ranges for each element in the query.

The invention discloses a method for detecting changes of SAR (synthetic aperture radar) images on the basis of multi-scale product and principal component analysis ( PCA ), mainly solving the problems that the adaptability is poor, the application range is narrow and the change detection results are subject to image misregistration. The method comprises the following specific implementation procedures: firstly, conducting the logarithmic ratio operation on two inputted time phase SAR images to obtain a difference image; carrying out the wavelet transform on the difference image; carrying out the multi-scale product de-noising on the high-frequency information of each decomposition layer; then, combining the de-noised images of each layer and carrying out the PCA transform, wherein, a first PCA image is used as a new difference image; and finally classifying the new difference image by using the minimum error ratio threshold value of the generalized Gaussian model to obtain the final result image of changes. The experiment shows that the invention can enhance the change information, have strong antinoise performance and reduce the influence of image misregistration, thus having high applicability and can be applied to the disaster detection of SAR images.

The invention relates to a method for realizing optimization of a data source extensive makeup language (XML) query system based on sub-queries. The method comprises the following steps: receiving corresponding XML query input information; carrying out morphological analysis and syntactic analysis, and verifying the correctness and the validity; if the analysis is successful and the verification is passed, generating an XML analytical syntactic tree; translating the XML analytical syntactic tree, and converting the XML query input information into intermediate logical representation; carrying out rewriting treatment, and generating a target query expression; and calling supported querying and computing engines for querying and computing, and acquiring output query results. By adopting the method for realizing the optimization of the data source XML query system based on the sub-queries, the query rewriting can be used for converting a procedural query into a descriptive query, the sub-queries are optimized, and certain specific sub-queries are rewritten into equivalent attended operation of a plurality of tables by the combination of the sub-queries, thus the levels of query sentences are reduced as much as possible, the treatment of planning optimization can be carried out conveniently, and the method has stable and reliable working performance and wider application range.

The invention discloses a fault diagnosis method for a rolling bearing, and belongs to the technical field of fault diagnosis and signal processing analysis. High-frequency sampling and preprocessing of vibration signals are firstly performed on the rolling bearing with faults, the preprocessed signals are sequentially filtered by the aid of a Morlet wavelet filter, spectral kurtosis and unit spectral kurtosis of the filtered signals are calculated, and a filter parameter corresponding to the maximum value of the unit spectral kurtosis is selected by comparison, namely, a global optimization filter parameter is selected. According to the fault diagnosis and detection method, an operator can extract the optimal filter parameter in the diagnosis process without a lot of detection experience and numerous historical data, the method is wider in application range, errors caused by human errors are greatly decreased, the extracted optimal filter parameter can be more accurate, a diagnosis result is more correct, fault diagnosis and detection automation is more facilitated, more time is saved, and efficiency is higher.

The invention discloses a runoff prediction method based on an attention mechanism and LSTM by using a deep learning algorithm. The method comprises the following steps: firstly, collecting influence characteristics related to runoff in a drainage basin, then constructing a time sequence data set corresponding to the characteristics and runoff, obtaining a runoff prediction model based on an attention mechanism and LSTM through training, and predicting the subsequent runoff according to the obtained runoff prediction model. Meanwhile, some short-term important features are ignored when the LSTM memorizes a long-term sequence mode, so that an attention mechanism is added, key elements in a runoff sequence are selectively concerned, the capability of capturing effective features by the LSTM is improved, and the prediction precision is relatively high. In addition, a deep learning method driven by data is used, dependence on a hydrological and physical mechanism in a drainage basin is reduced, and the application range of the model is effectively expanded.

Using techniques of image processing, machine learning, and semantic processing natural language etc, the method combines semantic labeling for visual character of image with semantic labeling text character of image to carry out semantic labeling content of image. Moreover, based on labeling characteristic of specific user, the method also supports to correct mapping rule base of label at bottom layer so that the labeled result is more accorded with labeling requirement of specific user. The method is widely applicable to each applications of need to carry out image searches. The method raises labeling precision, and expands range of application.

The invention provides a side slope stable reliability sensitivity analysis method based on Monte Carlo simulation. The side slope stable reliability sensitivity analysis method based on the Monte Carlo simulation includes: step 1, constructing a joint probability density function of uncertain parameters; step 2, using a Monte Carlo simulation method to obtain a side slope failure probability, and obtaining a failure sample; step 3, designing various sensitivity analysis schemes, and respectively constructing joint probability density functions of uncertain parameters under all the sensitivity analysis schemes; step 4, obtaining side slope failure probabilities under all the sensitivity analysis schemes; step 5, obtaining a variation trend of the side slope failure probabilities along with statistical characteristics of the uncertain parameters according to the side slope failure probabilities under all the sensitivity analysis schemes. The side slope stable reliability sensitivity analysis method based on the Monte Carlo simulation is wide in application range, simple in computation process, high in computation efficiency, and capable of effectively revealing a response regularity between the reliability level of a side slope and the statistical characteristics of the uncertain parameters, and has certain guiding significance for side slope risk control, design optimization and the like.

A method and apparatus for training a model is disclosed in embodiment of the present application. One embodiment of the method includes obtaining a configuration file, wherein the configuration filecomprises a data set identification, a machine learning framework identification, a machine learning algorithm identification and parameter information; selecting data from the data set indicated by the data set identification as training data to generate a training data set; selecting the machine learning framework indicated by machine learning framework identification as the target machine learning framework from the preset set of machine learning frameworks; in the framework of target machine learning, using the machine learning algorithm to identify the indicated machine learning algorithmand the parameters indicated by the parameter information, and obtaining the classification model by training based on the training data set and the labels associated with the training data in the training data set. The embodiment enables users to select different machine learning frameworks, corresponding machine learning algorithms and parameters according to actual needs, and expands the application scope of model training.

The invention relates to a human body behavioral modeling identification method based on a priori knowledge cluster in a computer system. The method comprises the steps of human body behavioral modeling processing and human body behavioral identification processing. By the adoption of human body behavioral modeling identification method based on the priori knowledge cluster in the computer system, a constructed modeling system has a self-learning capacity and can continuously perfect a classifier along with the classification of observed samples. In cluster distribution obtained in the process of human body behavioral modeling, if outliers exist, the outliers serve as an abnormal behavioral model. When the system is designed, precaution is given as required, the method has great realistic meanings, important influence of priori knowledge on a human body behavioral identification result is considered, an improved cluster method is adopted to model and identify the human body, and foundation can be laid for identifying abnormal behaviors. The method is simple and efficient, can accurately identify different behavioral modes, and has good expandability. Working performance is reliable and stable, and the application range is wide.

The invention provides a soft measuring method based on an improved SVM (Support Vector Machine) for measuring boiler unburned carbon content in fly ash. The soft measuring method is based on particle swarm optimization and carries out parameter optimization on support vector regression, two parameters affecting the validity of a regression model are selected, firstly, values of related auxiliary variables are collected by sensors and are subjected to data preprocessing, two main parameters of the support vector regression model are identified according to the history data in the past 6 hours in order to determine a soft measurement model for the unburned carbon content in fly ash, the soft measurement model is updated every hour according to the updated history data, and the real-time measured values of the auxiliary variables are inputted to the built soft measurement model, so that the output value of the unburned carbon content in fly ash is obtained. The soft measuring method can be used for measuring the unburned carbon content in fly ash generated in the combustion process of a boiler of a fire power plant in real time, the real-time measurement on the unburned carbon content in fly ash is realized, and meanwhile, the soft measuring method has the advantages of high precision, low calculation time consumption, wide application range and the like.

The invention relates to a method for automatically extracting character relations from a Chinese text or a text set, and belongs to the technical field of computer science and information extraction. In the method, by means of sentence meaning model characteristics, the relation attribute affiliation is determined; the character relations scattered in the text or the text set are automatically extracted by combining methods such as relation attribute disambiguation, character relation strength calculation and the like; and the character relations are organized by a character relation network, and the character relations comprising character relation attributes and the relation strength are displayed in a character relation graph manner. According to the method, sentence meaning model characteristics are introduced, so that the accuracy of the method for extracting entity relations is improved, and the method for extracting the character relations is enriched. Besides, as the number of texts about a central character in the text set is increased, the method can extract character relations of the central character more accurately and comprehensively, and the application range is wider.

The invention provides a recording device clustering method based on Gaussian mean super vectors and spectral clustering. The method comprises the steps that the Melch frequency cepstrum coefficient MFCC characteristic which characterizes the recording device characteristic is extracted from a speech sample; the MFCC characteristics of all speech samples are used as input, and a common background model UBM is trained through an expectation maximization EM algorithm; the MFCC characteristic of each speech sample is used as input, and UBM parameters are updated through a maximum posteriori probability MAP algorithm to acquire the Gaussian mixture model GMM of each speech sample; the mean vector of all Gaussian components of each GMM is spliced in turn to form a Gaussian mean super vector; a spectral clustering algorithm is used to cluster the Gaussian mean super vectors of all speech samples; the number of recording devices is estimated; and the speech samples of the same recording device are merged. According to the invention, the speech samples collected by the same recording device can be found out without knowing the prior knowledge of the type, the number and the like of the recording devices, and the application scope of the method is wide.

The invention discloses a reinforcement learning based anaphora resolution method, which comprises the following steps: data preprocessing: carrying out word segmentation, sentence segmentation, part-of-speech tagging, part-of-speech reduction, named entity identification, syntactic analysis and word vector conversion on text data to obtain candidate preceding words and analogy word related characteristics; constructing a neural network model: combining the characteristics of the word vectors and the relevant characteristics which can learn the fingering pairs and the relevant semantic information, better sorting and scoring the candidate preceding words and the fingering words, and finally obtaining an fingering chain; and using the trained model to carry out anaphora resolution, inputting text data, and outputting a resolution chain. According to the method, deep learning training is carried out by adopting a reward measurement mechanism for overcoming the defects of a heuristic lossfunction, the model effect is improved, hyper-parameter setting is automatically carried out for different language data sets, the necessity of manual setting is avoided, the practicability of the model is improved, and the application range is expanded.

The invention belongs to the technical field experimental instruments, which relates to a multi-functional reaction device used for chemical engineering experiment. The multi-functional chemical engineering experiment device comprises a chemical reaction kettle, a control system and a feeding and discharging and metering system of reaction materials. The chemical reaction kettle adopts a double-layer high-transparent visible kettle body. An upper kettle cover and a lower kettle cover are made of stainless steel materials. The double-layer kettle body adopts conduction oil with high degree of transparency to heat, and the agitation is carried out by adopting strong magnetic force under static sealing. The control system comprises an in-kettle temperature measuring device, a kettle sleeve temperature measuring device, an in-kettle pressure measuring and transmitting device, a motor rotation speed measuring and transmitting device. The control system is connected with a control panel, andcan be connected with a computer to realize remote operation. The feeding and discharging and metering system of reaction materials comprises a feeding and metering system and a backflow separating and metering system of liquid and gaseous reaction materials. The multi-functional chemical engineering experiment device adopts a clasp or bolt connecting structure, wherein clasps and bolts are standardized and general and can be interchanged. The components can be added or cut voluntarily according to the change of the technique. The invention can be applied widely.

The embodiment of the invention provides a deep-learning artificial intelligence model establishment method and system. The method comprises: a deep-learning artificial intelligence model establishment request of a user is received and a model establishment job stream is generated based on the deep-learning artificial intelligence model establishment request; and according to the model establishment job stream, a deep-learning artificial intelligence model corresponding to the deep-learning artificial intelligence model establishment request is generated. According to the method and system provided by the invention, the deep-learning artificial intelligence model is generated automatically based on the deep-learning artificial intelligence model establishment request sent by the user and thus the user does not need to carry out software development, so that the user can carry out operation conveniently and the professional request of the user is reduced; and the application range is wide.

Provided is a pavement identification method for a reinforced road in an automobile proving ground. The method comprises steps of: acquiring the motion information of a vehicle body and axles by an acceleration sensor and an angular rate gyroscope, and acquiring CAN message data on a vehicle bus via an OBD interface of the vehicle; preprocessing acquired multi-source data to obtain stable time-domain data; computing the attitude information of the vehicle body and the axles and analyzing the protocol of the CAN messages; transforming the vibration acceleration data and attitude data from the time domain into the spatial domain by using vehicle speed; performing time-domain and frequency-domain feature extraction on the data in the spatial domain; designing a artificial neural network pavement classifier based on the extracted time-domain and frequency-domain features to identify the reinforced road in the automobile proving ground. The method solve a problem that an independent suspension is insensitive to landforms, compensates the defect of a conventional method that identifies the pavements just by using acceleration data, and greatly increases the identification accuracy and application range of the test road landform and ground.

The invention relates to a multi-robot system fault diagnosis method, comprising the following steps of: 1) obtaining real-time robot motion data; 2) adopting wavelet packet transform to perform feature extraction on the motion data; 3) inputting to-be-diagnosed data subjected to feature extraction into a trained fault diagnosis model, and calculating a corresponding real-time likelihood probability of current robot motion data; and 4) according to a relation between the real-time likelihood probability and a state threshold value, obtaining a current hidden state of the multi-robot system and obtaining a fault diagnosis result. Compared with the prior art, the method has the advantages of high robustness, accurate diagnosis result, wide application range and the like.

A structure finite element model correcting method based on multi-element uncertainty comprises the following steps: (1) building an initialized parameterization equivalent finite element model in finite element software; (2) screening out significance parameters; (3) obtaining sample points, and constructing an incomplete variable high-order response surface model; (4) judging validity of the response surface model, if the validity of the response surface model meets the requirement, executing the next step, and if the validity of the response surface model does not meet the requirement, executing the step (3) again; (5) building a rapid random sampling analysis model with the combination of a high-order response surface and the Monte Carlo method, and conducting statistics on a mean value and a covariance matrix of simulation output responses; (6) conducting statistics on a mean value and a covariance matrix of test output results; (7) constructing a weighting objective function of the mean values and covariances of tests and simulation; (8) reversely estimating a mean value and a covariance matrix of input parameters; (9) judging whether the mean value and the covariance matrix of the input parameters meet correction accuracy or not, if yes, stopping iteration, and if not, executing the step (8) again. According to the structure finite element model correcting method, the calculated amount of the iteration is reduced, the application range is wide, and optimization of a large-scale parameter range is achieved.

The present invention provides an information providing method and apparatus, which have advantages of being objective, high in efficiency, wide in application scope, and good in scalability. The method comprises: extracting a sample feature vector and a corresponding sample decision from historical consultancy session data, wherein an element of the sample feature vector is an attribute value extracted from the historical consultancy session data according to a preset attribute, and the sample decision is a user questioning sentence; performing training by using a plurality of sample feature vectors and corresponding sample decisions to obtain a probabilistic classification model; extracting a to-be-tested feature vector from a current client consultancy request, wherein the to-be-tested feature vector is the same as the sample feature vector in format; inputting the to-be-tested feature vector into the probabilistic classification model, then receiving one or more candidate decisions and corresponding probabilities output from the probabilistic classification model, wherein the candidate decisions are candidate user questioning sentences; choosing K candidate decisions of greatest probabilities as forecasted decisions, then providing a client with a standard reply corresponding to the forecasted decisions.

The invention discloses a pharmaceutical activity prediction and selection method based on genetic expressions and drug targets. The method comprises following steps: 1) obtaining target genes corresponding to drugs to be detected according to information in drug target database; obtaining genetic expression data of disease tissues of patients and corresponding compression data and obtaining important genetic lists of morbidity processes of patients by evaluating importance of genes through a systems biology method; 2) searching for whether the target genes obtained in the step 1) target the important gene lists during mobility of the patients or not through statistical analysis of in order to predict pharmaceutical activity of patients and select drugs suited to patients. The method is easily used with high efficiency and broad application scope. The prediction method can be used for selecting drugs suitable for individual patients such that a personalized treatment schme can be provided for patients.

The invention relates to the field of micro-grid photovoltaic power generation system research, in particular to a micro-grid photovoltaic power generation short-term prediction method based on deep learning, which comprises the following steps of: 1, data preprocessing: performing supplementary correction on bad data and missing data in historical data, and then performing normalization processing to form training sample data; 2, dividing the processed sample data into sample subsets of different seasons and different generalized weather types, and inputting the sample subsets into an LSTM prediction model as training samples; 3, establishing an LSTM short-term prediction model according to the weather type, and training is completed; And 4, selecting a corresponding LSTM prediction sub-model according to the season to which the prediction day belongs and the weather type prediction information, and outputting a photovoltaic power generation short-term prediction value at a future moment. According to the method, the deep learning algorithm is adopted to deeply mine the output characteristics of the photovoltaic power generation and establish the prediction model, expensive meteorological measurement equipment and instruments are not needed, the cost is reduced, and the method has a wide application range.

The invention discloses an offline handwritten and printed Chinese character identification method and system. The method comprises the steps of making a training set: loading the training set, loading a model for performing training, identifying the training set according to a weight of training in a model training stage, finding error parts in the training set, deleting the error parts, and adjusting parameters to obtain a final training set; and identifying an offline handwritten Chinese character: according to the final training set, loading the model and the parameters, reading a Chinesecharacter image to perform binarization and graying processing, extracting pixels of binarized and grayed images, and obtaining a new Chinese character image. The method and the system have the advantages that the realization method is simple; the application range is wide; the model adopts a convolutional neural network structure totally containing 44 convolutional layers, 10 pooling layers, 9 fusion layers, 9 batch standardization layers, 2 dropout layers, 1 full connection layer, 1 input layer and 1 output layer; and the offline Chinese character identification is accurate.

The invention relates to a method for synchronously realizing seismic lithofacies identification and quantitative assessment of uncertainty of the seismic lithofacies identification. The method comprises the steps of determining the type of the lithofacies and performing logging lithofacies definition, establishing a rock physical response relation between logging physical parameters and elasticity, establishing a probability statistical relation between the lithofacies and logging attributes, establishing the probability statistical relation of well-seismic scale elasticity parameters and constructing the statistical relation of the lithofacies and the seismic scale elasticity parameters, inverting the information of the probability distribution of the elasticity parameters of a target layer, obtaining the lithofacies probability information of the target layer by combining the inverted probability information of the elasticity parameters of the target layer and the statistical relation of the lithofacies and the seismic scale elasticity parameters, obtaining the maximum posterior probability solution of the lithofacies distribution according to the probability information of the lithofacies and outputting final model parameters. The method is capable of quantitatively characterizing the uncertainty of each link of lithofacies identification and the propagation and accumulation characteristics of the uncertainty in the lithofacies identification process, and also capable of performing uncertainty analysis on the seismic lithofacies identification; as a result, the reservoir evaluation risk is reduced; in short, the method is wide in application range.

The invention discloses an industrial process monitoring method based on a missing variable PCA model, so as to use a missing variable processing method to online estimate principal component information and realize the purpose of executing monitoring on estimation errors. Firstly, the principal component is estimated by assuming that each measurement variable data is missing one by one; and then, with the estimation error of the principal component and the error estimation value of the PCA model as monitored objects, online process monitoring is executed. Although sampling data under a normal working condition do not necessarily satisfy the Gauss distribution hypothesis, the estimation errors generally obey the Gauss distribution. From the point of view, although the method is based on the PCA algorithm, the method does not require hypothetical training data to obey or approximately obey the Gauss distribution, and the application range of the traditional PCA-based process monitoring method is extended to a certain extent. Besides, the method also has the advantage of a high multi-model generalization capability as multiple fault detection models are adopted.

The invention discloses a training method for classification models, and a device and a computer server thereof, so as to solve the technical problems that the prior art is low in calculation efficiency and narrow in application range by a semi-supervised learning technology training classification model. The method comprises steps of: constructing an initial classification model, wherein the initial classification model comprises at least one single-mode classification models comprising same classification tasks, and a mode data training set corresponding to each single-mode classification models comprising label training data and label-free training data; and training the initial classification models to obtain target classification models, based on a method of the feature coding distribution that aligning the label training data and the label-free training data in the mode data training set of each single-mode classification models. According to the training method for classification models, the training efficiency of the classification model can be improved, and the application range is wider.