110results about How to "Explanatory" patented technology

The invention discloses a fuzzy rough set and decision tree-based track circuit red light strip fault positioning method. The method mainly comprises the following steps of: 1) establishing an initial decision table; 2) carrying out fuzzy discretization on continuous fault feature attributes to establish a fuzzy decision table; 3) inputting fault sample training data to obtain a reduced decision table; 4) establishing a diagnosis decision tree model; 5) inputting measured data into the diagnosis decision tree model, carrying out calculation to obtain a fault diagnosis result, inputting the measured data into a diagnosis positioning decision tree model, carrying out preliminary judgement to obtain a fault positioning result, judging faults of specific equipment by combining expert experiences, and giving corresponding fault maintenance suggestions. The method can rapidly and correctly position fault points of uninsulated frequency shift track circuit red light strip faults, greatly reduce the blindness and complexity of fault diagnosis, have relatively good rule explanation and relatively good robustness, improve the fault positioning speed and correctness and provide a new fault positioning technological means for intelligent fault diagnosis of track circuits.

The invention discloses a panoramic segmentation method based on a graph neural network. The panoramic segmentation method comprises the following steps: extracting a plurality of target features froma picture; segmenting the head network through an example to obtain a foreground category probability, a background category probability and a mask result of the picture, and semantically segmentingthe head network to obtain a preliminary semantic segmentation result of the picture; processing the new foreground image through the foreground category probability to generate an instance classification result, and extracting a target instance segmentation mask from the instance classification result according to a mask result; processing the new background image through the background categoryprobability and the preliminary semantic segmentation result to generate a target semantic segmentation result; and fusing the target instance segmentation mask and the target semantic segmentation result by adopting a heuristic algorithm to generate a panoramic segmentation result. The invention further discloses a panoramic segmentation system based on the graph neural network, computer equipment and a computer readable storage medium. By adopting the method and the device, the panoramic segmentation effect of the picture can be optimized by utilizing the mutual relation between the objects.

The invention relates to a data modeling based wearing monitoring and early-warning method for a variable pitch bearing of a wind turbine. The method includes the following steps: (1) data collection;(2) data pre-processing; (3) sample marking; (4) feature construction; (5) model construction; (6) algorithm verification; (7) model deployment. The method of the invention is high in effectiveness and accuracy, realizes a wearing monitoring and early-warning function for a variable pitch bearing of a wind turbine through modeling and predication by means of common sensor data, such as wind speed, power, rotational speed, pitch angle and pitch motor current, recorded by an SCADA system of the wind turbine, and has the characteristics of low cost, high efficiency, strong interpretation and thelike.

The invention belongs to the technical field of sleep monitoring, and relates to a single-lead electroencephalography (EEG) automatic sleep staging method. The method comprises the following steps: (1) performing signal pre-processing; (2) extracting classification features; (3) carrying out sleep staging. The method provided by the invention has the advantages that 1, a pre-processing algorithm is designed to obtain single-lead EEG signals with better quality; 2, the method extracts multiple features from the time domain, the frequency domain and the nonlinear domain, and screens out the representative features; 3, a random forest model is adopted, so that over-fitting does not need to be worried about; the method has a good anti-noise ability, and the staging results of all decision trees can be obtained, so that the confidence probabilities of the random forest for all sleep stages are obtained; 4, a D-S evidence theory is combined, so that the accuracy rate of the sleep staging isfurther improved.

The invention belongs to the technical field of education data mining, and discloses a knowledge measurement-oriented test question, knowledge and ability tensor construction and labeling method. In combination of a Q matrix and Bloom cognitive field education target classification, knowledge point mastering is divided into six cognitive ability levels: knowing, understanding, application, analysis, integration, evaluation and test question, knowledge and ability tensor construction. An interpretable test question label prediction model is constructed by adopting an active learning strategy, interpretable label prediction information entropies are obtained, unlabeled samples are input into the prediction model by utilizing the trained interpretable test question label prediction model, andthe label prediction information entropies with relatively high interpretability are fed back, so that man-machine coordination is carried out. According to the method, the influence of subjectivityof manual labeling on the TKA tensor is reduced, the labeling accuracy and efficiency are high, and the expert labor cost is greatly reduced. The method is high in mobility, can be applied to examination and annotation of test question knowledge points of all subjects, and is better in applicability.

The invention discloses a microsatellite instability detecting system and method based on genome sequencing. The method comprises the steps that microsatellite detecting site selection is conducted, wherein an effective detecting site is selected according to sequencing data of a certain tumor sample, the threshold value of the instability of a single microsatellite corresponding to the effectivedetecting site and the evaluation standard of the instability of the microsatellite of a certain tumor sample are computed; microsatellite instability detection is conducted on a detected sample according to the threshold value of the instability of the single microsatellite corresponding to the effective detecting site and the evaluation standard of the instability of the microsatellite of the certain tumor sample. The system and the method have the advantages that the system and the method do not rely on a control sample, the pain on a detected person caused by sampling can be reduced, all genetic information of the detected person is included in the control sample, the probability of privacy leakage of the detected person can be reduced by not using the control sample, and the detectioncost can be reduced by not detecting the control sample; the operation is convenient, the cost is low, and the confidence level is high.

The invention discloses a cigarette loose end rate prediction method based on an improved gradient improvement decision tree, which comprises the following steps: acquiring process parameters of tobacco shreds in the same batch in each process of tobacco shred making and loose end rate data in a final package process, and forming an original data set by the process parameters and the loose end rate data; dividing the original data set based on the original data set in combination with a correlation coefficient analysis method to obtain a key process parameter set; performing normalization processing on the data in the key process parameter set, and performing random division on the normalized data to obtain a training data set and a test data set; on the basis of training samples in the training data set, constructing an improved gradient improvement decision tree model; and inputting the test data set sample into the improved gradient improvement decision tree model, and predicting the cigarette empty head rate. A connection is established between each process parameter of the tobacco shred making process and the index of the cigarette packet vacancy rate, so that more accurate prediction of the cigarette vacancy rate is realized.

The invention relates to a credit assessment technology, and aims to provide a personal credit assessment method and system based on fusion neural network feature mining. The method comprises the steps that behavior data of an individual user are preprocessed and checked and then subjected to matrix processing, and the obtained data serve as input of an LSTM model and a CNN model at the same time; in the LSTM model, sequentially processing by an embedding layer, a bidirectional long short-term memory neural network and an attention mechanism layer, and outputting a time sequence behavior feature vector extracted from the data; in the convolutional neural network model, processing is carried out through a convolutional layer and a pooling layer in sequence, and local behavior feature vectors extracted from the data are output; and carrying out vector splicing on the two types of feature vectors, taking the spliced feature vectors as input of an XGBoost classifier, and carrying out training to finally obtain a personal credit evaluation result. Compared with the prior art, the method has the characteristics of comprehensive index coverage, wide processing index source, advanced modeling mode, flexible model expansion, complete and effective feature extraction and accurate result.

The invention discloses a conversational music recommendation method based on Meta-graph knowledge map representation. The method comprises the following steps: generating a feature vector offline based on the Meta-graph knowledge map representation, and performing online conversational recommendation based on the Bandit algorithm. By adoption of the method, the music recommendation based on the Meta-graph knowledge map representation in a conversation scene is realized, that is, in a scene of human-machine dialogue, the preference for music of a user is obtained in real time, the long-term and short-term preference of the user is modeled in combination with a knowledge map, a context-aware recommendation result is provided in time to achieve real-time recommendation, and contextual information, user needs and feedback can be well handled.

The invention discloses a remote sensing sub-pixel map-making method based on integrated pixel level and sub-pixel level spatial correlation characteristics. The method includes the steps that firstly, a frequently-used pixel level spatial correlation characteristic description method (a spatial attraction model) is utilized for extracting pixel level spatial correlation characteristics from soft classification information (category proportions) of a neighborhood pixel easily and rapidly without iteration; secondly, a widely-used sub-pixel level spatial correlation characteristic description method (an exponential decay model) is utilized for extracting sub-pixel level spatial correlation characteristics from a sub-pixel map-making iteration result; then, the spatial correlation characteristics of the pixel level and the sub-pixel level are normalized and fused to obtain the integrated spatial correlation characteristic value used for determining the sub-pixel categorical attribute; finally, the optimal spatial layout of the sub-pixel with the largest sum of all the category spatial correlation characteristic values in mixed pixels is obtained according to the integrated spatial correlation characteristic value through a classical binary branching-bounding integer programming algorithm. The remote sensing sub-pixel map-making method is high in calculation speed and simulation precision.

The invention relates to a Raman spectrum analysis method based on a convolution neural network. Firstly, the classification model is established. The Raman spectrum of the material is pretreated, andthen the pretreated Raman spectrum is input to the neural network for training to determine the weights of each layer of the network, and the classification model is named RS-CNN. Secondly, the Ramanspectra of the substances to be predicted are pretreated, and then the pretreated Raman spectra with predicted substances are inputted into the classification model, and the output of the classification model is the classification result. Convolution neural network denoising and baseline correction are integrated into the convolution neural network in a convolution manner, so that the preprocessing and identification problems are solved in a unified model framework, and the adaptive data processing is realized, which makes up for the shortcomings of the traditional methods.

The invention discloses a method and system for deriving risk control intelligent rules based on a decision tree. According to the importance of the features, the invention sorts its huge number of features, screens out important features, and builds decision trees of different depths based on these features. Then use the set threshold to filter the decision tree, and finally derive the rules according to the filtered decision tree. The method of the invention can ensure that under the normal operation of the business system, rules can be layered and derived according to different feature numbers, and fraudulent behavior can be detected to the greatest extent. Compared with the risk control system with artificial rules, the system of the present invention is more stable and intelligent, and the efficiency of intelligent rules is higher, so that the loss of enterprises can be minimized. Especially in systems with complex business and huge data volume, this advantage becomes more and more obvious.

The invention discloses an eye image segmentation method and device based on prior information, equipment and a medium, and the method comprises the steps: obtaining a fundus image data set, and calculating prior information according to the fundus image data set; constructing a machine learning model, and training the machine learning model based on the prior information to obtain an image segmentation model; obtaining a to-be-segmented target image, wherein the target image comprises an eye part; and inputting the target image into the image segmentation model to obtain a target image segmentation result output by the image segmentation model. Prior information is introduced to train an image segmentation model. The segmentation result of the to-be-segmented target image has high interpretability and accuracy. In addition, a classification model can be trained based on prior information, and the classification model is used to classify the segmentation result of the to-be-segmented target image so as to obtain the classification result with high interpretability and high accuracy, so that the scheme of the invention has a wider application prospect.

The invention discloses a remote sensing sub-pixel mapping method based on the spatial distribution characteristics of features, comprising the following steps: first, classifying remote sensing images into an area pattern, a line pattern and a point pattern according to the spatial geometric characteristics of features; then, under the hypothesis of spatial dependence, using a sub-pixel mapping method based on vector boundary to process area features, using a line feature template sub-pixel mapping method to process line features, and using a sub-pixel mapping method based on space pattern consistency matching to process point features; and finally, embedding the sub-pixel mapping results of the three spatial patterns to get a sub-pixel map of the images. According to the invention, a sub-pixel mapping theoretical model capable of processing point, line and area features at the same time and based on the spatial distribution characteristics of features is built theoretically, and the simulation precision is high.

The invention provides a content recommendation method based on a domain knowledge graph, and the method comprises the steps: inputting historical click contents of a user into a candidate content generation model based on entity representation, and generating first candidate contents which the user may be interested in; generating a content representation vector based on a content representationlearning model of knowledge graph interest sampling; according to the content representation vector, obtaining click probability distribution of the user to the content, and generating second candidate content which the user may be interested in; and sorting each content in the first candidate content and the second candidate content to obtain a content recommendation list. The knowledge graph-based content recommendation method has the advantages that the relationship between the contents can be established through the knowledge graph, so that the contents recommended to the user have an association relationship with the historical click contents of the user, and the recommendation result is more interpretable. According to the method, the content cold start problem can be solved, and meanwhile, the recommendation performance is improved under the condition of lack of user historical behavior data.

The invention relates to a random forest visualized data analysis method based on LargeVis. The random forest visualized data analysis method comprises the steps of preprocessing a training dataset; extracting important characteristics of the training dataset through a random forest; adopting LargeVis to reduce the dimension; based on the random forest of LargeVis, conducting visualized processing. By means of the random forest visualized data analysis method based on LargeVis, aiming at high-dimension data, through the characteristic importance trained by the random forest, new sub-high-dimension data is formed, and then through the data subjected to LargeVis dimension reduction, the sub-high-dimension data is sent into the random forest to be predicted and analyzed to form visualization,the classifying precision can be improved, the visualization time can also be prolonged, and meanwhile, the random forest visualized data analysis method adapts to different pieces of data.

The invention belongs to the technical field of image super-resolution reconstruction, and discloses an image super-resolution reconstruction method based on deep convolution sparse coding, and the method comprises: embedding a multilayer learning iteration soft threshold algorithm ML-LISTA related to a multilayer convolution sparse coding model ML-CSC into a deep convolution neural network DCNN; adaptively updating all parameters in the ML-LISTA by using the learning ability of the DCNN, and constructing an interpretable end-to-end supervision neural network SRMCSC for image super-resolution reconstruction; and introducing residual learning, extracting residual features by using an ML-LISTA algorithm, combining the residual and an input image to reconstruct a high-resolution image, and then accelerating the training speed and the convergence speed. The SRMCSC network provided by the invention is compact in structure, has good interpretability, can provide a result with visual attraction, and provides a practical solution for super-resolution reconstruction.

The invention discloses a cross-social media user identity recognition method and method based on an attention mechanism. The method comprises the following steps: acquiring data of different modalities of a plurality of users on different social media to serve as training data; for data of different modalities, respectively adopting different models to learn potential representations of the data,and training a user identity recognition model; calculating the similarity of the data between the users on different social media by combining a time sequence relation and confidence coefficients ofthe data of different modalities; mapping the similarity between the users to a probability space by using a multi-layer perceptron to obtain the probability that the users on different social mediapoint to the same user entity; constructing an objective function by adopting cross entropy, and carrying out iterative optimization solution on model parameters, wherein the model is used for determining whether to point to the same user or not according to different modal data on different social media. According to the method, the difference of data transmission of different modes of data is considered, and the user identity recognition accuracy is higher.

The invention discloses a traffic flow prediction method based on mixed feature mining. The method comprises: introducing mixed characteristic data on the basis of traffic flow data; wherein the dataspecifically comprises time characteristic data and traffic situation characteristic data; according to a traffic flow prediction target, mining corresponding features which are high in importance, large in feature difference and independent of one another from mixed features, eliminating features which are low in correlation and redundant and repeated, combining the mined features with traffic flow data to serve as model input, constructing a traffic flow prediction model, and achieving traffic flow prediction through the model. The prediction model with lower complexity and higher interpretability is constructed while rich features are introduced, and the prediction accuracy of the model is remarkably improved.

The invention relates to a stock index prediction method based on adaptive feature extraction, and the method comprises the steps: S1, obtaining stock index data, and obtaining the daily opening price, lowest price, highest price, closing price and transaction volume; s2, calculating an artificial index value proposed by a financial and economic expert; s3, constructing sample features and samplelabels, and dividing all samples into a training set, a verification set and a test set; s4, performing adaptive feature extraction on the sample; and S5, inputting self-adaptive extraction features and the artificial indexes calculated in the S2 into a neural network prediction model based on a factor machine, and outputting a prediction result. According to the method, the characteristics of thestock indexes are extracted in a self-adaptive manner, and the extraction method is simple and high in interpretability; the neural network based on the factor machine is used as a prediction model,interaction between features can be learned, the nonlinear expression capability is achieved, and the linear complexity is achieved; the accuracy of the stock index prediction technology can be effectively improved.

The invention relates to a hydrological prediction technology, in particular to a waterlogging prediction method and system and a storage medium, and the method comprises the steps: obtaining basic data, carrying out the generalization of a drainage basin and hydraulic facilities, and calibrating model parameters; constructing an SVM model, and predicting an urban waterlogging space-time range and a waterlogging submerging depth; constructing a pipe network hydrodynamic model of the research area as an AGswmm model, and simulating the waterlogging condition of the research area in combination with topographic data; training an SVM model by using simulation result data generated by the AGswmm model to realize model coupling so as to obtain an urban inland inundation rainfall runoff prediction model; performing data assimilation, calculating the prediction model by taking actually measured and predicted rainfall as input conditions, performing error analysis on actually measured data and prediction data of the prediction model to obtain a residual error, and correcting the residual error to obtain a prediction result closest to a real value. According to the method, model errors are reduced, the waterlogging simulation precision is ensured, and meanwhile, the processing and calculating time of mass hydrological data is shortened.

The embodiment of the invention discloses a message generation method, device and equipment. The method comprises the steps: obtaining user feature information, carrying out the coding of the user feature information, and forming a first feature vector corresponding to the user feature information; matching from a retrieval library according to the user feature information to obtain a plurality ofrisk description statements corresponding to the user feature information; performing characterization processing on the plurality of risk description statements to obtain a second feature vector corresponding to the user feature information; fusing the first feature vector and the second feature vector to obtain a fused third feature vector; inputting the third feature vector into a topic matching model to obtain a plurality of topics corresponding to the third feature vector; based on the plurality of topics and the third feature vector, generating a message containing the topics and the risk description statement.

The invention discloses a translucent automatic driving artificial intelligent system and a vehicle. The translucent automatic driving artificial intelligent system comprises a sensor module and a decision-making module. The sensor module is used for collecting various kinds of environmental information and comprises a plurality of direct sensor modules and an indirect sensor module, the multipledirect sensor modules collect current vehicle information in the various kinds of environmental information, and the indirect sensor module obtains current road information and other vehicle information in the various kinds of environmental information. The decision-making module is used for making decision of the vehicle be translucent by adding a deep learning network of people professional knowledge according to the various kinds of environmental information. The system can make the decision of the vehicle be translucent by adding the deep learning network of the people professional knowledge, not only is the hardware condition required by automatic driving lowered, but also the training efficiency is improved, accordingly the safety and the reliability of the vehicle are effectively ensured, and the robustness of the vehicle is improved.

The embodiment of the invention provides a recommendation method and device and a device for recommendation, and the method specifically comprises the steps: determining a demand word corresponding toexpression behavior data of a user by using a preset word bank; wherein the preset lexicon is used for storing preset words, and the commercial values of the preset words accord with a first condition; determining a recommendation object according to the demand word; and providing the recommendation object to the user, or sending access entry information of the user to a supplier terminal corresponding to the recommendation object, so that the supplier terminal transmits the recommendation object to the user terminal of the user through the access entry information. According to the embodiment of the invention, the interpretability of the recommended object can be improved, and the test cost can be reduced.

The invention discloses a building energy-saving expert design system and method based on sensitivity analysis, and the system comprises a building scheme input module which collects building design parameters inputted by a user, to-be-optimized parameters, and an optimization interval; a building energy consumption calculation module which can calculate a building energy consumption value and a human body comfort value through building design parameters; a data sampling module which is used for stratified sampling in an optimization interval of a to-be-optimized parameter to form a random data set; a sensitivity analysis module for performing sensitivity analysis on to-be-optimized parameters and building energy consumption values according to a random data set; an energy-saving suggestion module for providing energy-saving optimization suggestions according to a sensitivity analysis result; a parameter modification module used for collecting modification of a user on a to-be-optimized parameter; a judgment module used for judging whether current modification of a user is reasonable or not according to a building energy consumption value and a human body comfort value; and a result display module used for comparing building schemes before and after modification.

The invention discloses a subject embedding and document representation method based on network link and document content. By introducing topological link information in document network, the problemof fuzzy semantic terms (such as synonyms) affecting topic in topic embedding model can be solved effectively. Using the link information, combining the document content and links in the data with probability graph model, optimizing the parameters of the model by variational inference method, the negative influence caused by fuzzy semantics is basically solved, and more accurate topic embedding and document representation are obtained. The method of the invention combines the link relation and the content information in the document network, The problem of inefficient topic embedding caused byfuzzy semantics (such as polysemy) in the existing topic embedding model is effectively improved. The probability graph model is established to make the method more interpretable, and the variationalexpectation maximization algorithm is used to make parameter updating efficient, convergence time short, and can be applied to large-scale networks.

The invention discloses a method for automatically generating abbreviations of English paper titles in a computer. The method includes the steps of analyzing description content to obtain weight of each word, namely the degree of importance; according to the weight of each word, adopting a beamsearch method to query candidate abbreviations in the whole abbreviation space, and obtaining primary scores of the abbreviations; adjusting the scores of the abbreviations to obtain the final scores, and then sorting the final scores in a descending mode. According to the method, by improving existing abbreviation generation methods, the equal processing problem of each part of the description text can be avoided. Meanwhile, through syntactic parsing, semantic parsing, language models and other techniques related to natural language processing, related linguistics knowledge taken into account when people create the abbreviations can be learned to a certain degree, so that the explanatory degreeis higher during generation of the abbreviations.

The invention discloses a scene simulation method and system for a comprehensive energy system. The method comprises the steps: processing obtained meteorological data affecting a distributed power supply to obtain a simulation sample set; calculating based on the simulation sample set, a meteorological sensitive load and an electric vehicle charging parameter to obtain the power generation powerof a power system, the power generation power of a thermodynamic system and the gas supply power of a natural gas system of a comprehensive energy system; carrying out load flow calculation based on the power generation power of the power system, the power generation power of the thermodynamic system and the gas supply power of the natural gas system to obtain the typical scene set of the comprehensive energy system. According to the invention, the distributed power supply system, the meteorological sensitive load system and the comprehensive energy system are regarded as a whole for analysis,so that the importance of multi-energy-flow association coupling is manifested, energy operation management is promoted to be efficient, and the comprehensive utilization efficiency of energy is greatly improved.

The invention discloses a keyword extraction method and device, equipment and a medium, and relates to the field of data processing. The method comprises the steps of obtaining a first comment text from a plurality of comment texts, wherein an emotion tag of the first comment text is a first emotion tag; performing word segmentation on the first comment text to obtain a feature word set of the first comment text; calculating an information entropy set of the first comment text according to the feature word set, wherein information entropies in the information entropy set are obtained through calculation according to feature words in the feature word set; and determining a keyword of the first comment text according to the information entropy set. According to the method and device, the information entropy set of the comment texts can be obtained, the keywords in the comment texts are determined according to the information entropy set, and the information entropy represents the difference between the comment texts, so that the keywords obtained through the information entropy have higher interpretability for the sentiment classification result, and the modeling effect and interpretability are improved.