99 results about "Curse of dimensionality" patented technology

The invention provides a text emotion classification method based on the joint deep learning model which relates to the text emotion classification method. The method is designed with the object of solving the problems with the dimension disaster and sparse data incurred from the existing support vector machine and other shallow layer classification methods. The method comprises: 1) processing each word in the text data; using the word2vec tool to train each processed word in the text data so as to obtain a word vector dictionary; 2) obtaining the matrix M of each sentence; training the matrix M by the LSTM layer and converting it into vector with fixed dimensions; improving the input layer; generating d-dimensional h word vectors with context semantic relations; 3) using a CNN as a trainable characteristic detector to extract characteristics from the d-dimensional h word vectors with context semantic relations; and 4) connecting the extracted characteristics in order; outputting to obtain the probability of each classification wherein the classification with the maximal probability value is the predicated classification. The invention is applied to the natural language processing field.

The invention discloses a nasopharyngeal-carcinoma (NPC) lesion automatic-segmentation method and nasopharyngeal-carcinoma lesion automatic-segmentation systems based on deep learning. The method comprises: carrying out registration on a PET (Positron Emission Tomography) image and a CT (Computed Tomography) image of nasopharyngeal carcinoma to obtain a PET image and a CT image after registration;and inputting the PET image and the CT image after registration into a convolutional neural network to carry out feature representation and scores map reconstruction to obtain a nasopharyngeal-carcinoma lesion segmentation result graph. The method carries out registration on the PET image and the CT image of the nasopharyngeal carcinoma, obtains a nasopharyngeal-carcinoma lesion by automatic segmentation through the convolutional neural network, and is more objective and accurate as compared with manual segmentation manners of doctors; and the convolutional neural network in deep learning isadopted, consistency is better, feature learning ability is higher, the problems of dimension disasters, easy falling into a local optimum and the like are solved, lesion segmentation can be carried out on multi-modal images of the PET-CT images, and an application range is wider. The method can be widely applied to the field of medical image processing.

The disclosed multi-classifier combination face recognition method based on feature sorting comprises: extracting face area from initial image for pre-process and feature extraction; feature sorting to obtain different face feature groups; designing component classifier for every group to recognize face and combine results for optimal effect. This invention overcomes dimension disaster, reduces algorithm complexity, and improves recognition performance.

The invention discloses a comprehensive evaluation method for wireless sensor network performance, relates to a least squares support vector regression (LSSVR)-based comprehensive evaluation method for wireless sensor network performance to comprehensively and reasonably evaluate running performance of small and medium-scale wireless sensor networks, and belongs to the technical field of wirelesssensors and network communication. The method is characterized by comprising the following four steps of: determining a network running performance index, determining an independent network running factor value by adopting a factor analysis method, establishing an LSSVR-based network performance comprehensive evaluation model, and performing LSSVR-based network performance comprehensive evaluation with low power consumption. The method solves the traditional problems of 'dimensionality curse' and 'over-learning' by skillfully using a kernel function, is quite suitable for the comprehensive evaluation of the small and medium-scale wireless sensor network performance with a plurality of indexes, assists an administrator in timely and accurately grasping the running condition and tendency ofthe WSN, and provides basis for network running evaluation and optimization.

The invention relates to a lithium ion battery cycle life prediction technology, in particular, a deep belief network and relevance vector machine fusion-based lithium battery residual life prediction method. An existing lithium battery residual life prediction method relies on accurate physical models or complex signal processing technologies, as a result, the existing lithium battery residual life prediction method needs heavy investment, or the existing method is based on a shallow structures, as a result, the performance of fault prediction will be limited, and the existing method is vulnerable to curse of dimensionality, while with the method of the invention adopted, the problems of the existing method can be solved. A charging and discharging period-based lithium battery capacity degradation data set is obtained; the data are pre-processed; the fusion models of a DBN (deep belief network) and an RVM (relevance vector machine) are built; a DBN model and a RVM model are trained; and the trained fusion models of the DBN and the RVM is adopted to predict the residual life of a lithium battery. The method of the invention is suitable for predicting the residual life of the lithium battery.

The invention provides an intelligent fault diagnosis method based on a rough Bayesian network classifier, which comprises the following steps: using standard fault feature data as a fault diagnosis condition attribute set, using a standard fault mode as a fault diagnosis decision attribute set, and adopting a rough set principle to construct an original fault diagnosis information table T1; adopting the minimum entropy method to carry out discrete processing on various continuous fault diagnosis condition attribute values in the T1, so as to form a discretization fault diagnosis information table T2; using a rough set discernable matrix and a nuclear theory to carry out attribute reduction and optimal feature selection on the T2, so as to form a reduction fault diagnosis information table T3; and using the T3 to establish the Bayesian network classifier, so as to realize efficient and quick intelligent fault diagnosis. The intelligent fault diagnosis method avoids the 'curse of dimensionality' problem existed in a Bayesian network diagnostic method, overcomes weaknesses of rigid reasoning and critical misjudgment in a rough set diagnostic method, and greatly improves the efficiency and accuracy of fault diagnosis.

The invention discloses a web service combination method based on depth reinforcement learning for overcoming the problems of long time consumption, poor flexibility and non-ideal combination effect of the traditional service combination method in large-scale service scenes. The depth reinforcement learning technology and the heuristic thought are applied to the service combination problem. In addition, by considering the partial observability of the real environment, the service combination process is converted into a partially-observable Markov decision process POMDP, the solution problem of the POMDP is solved by using a recurrent neural network, and the method still expresses high efficiency encountering the challenge of curse of dimensionality. By adoption of the method provided by the invention, the solution speed can be effectively improved, the dynamic service combination environment is automatically adapted on the basis of ensuring the quality of the service combination scheme, and the adaptability and the flexibility of the service combination efficiency is effectively improved in a large-scale dynamic service combination scene.

The invention relates to a pilotless automobile combination navigation method based on vision screening. The method comprises the following steps: carrying out the conversion for a coordinate system; recognizing a building occlusion angle; judging availability of a satellite signal in a non-line-of-sight environment; implementing an improved self-adaptive square-root volume kalman filtering algorithm. A combination navigation algorithm adopting vision information as a screening condition is provided, a concept of removing the non-line-of-sight transmission of satellite data is introduced, a judgment method is also provided, so that the GPS satellite data information with lower precision caused by the blocking of a building can be eliminated; different from the traditional combination method adopting the vision navigation to process the data in parallel in real time, the vision information is used for screening the GPS data on purpose, so that the dimensional disaster caused by adopting the vision navigation in the traditional method is avoided; the improved self-adaptive square-root volume kalman filter algorithm is provided, and the strong nonlinear problem of the navigation data when a pilotless intelligent car runs on urban roads is considered.

The invention discloses a network traffic abnormality detection method based on an SVM (Support Vector Machine), which comprises the steps of reading historical network traffic data; extracting network traffic features of the historical network traffic data; carrying out data standardization on the network traffic features; carrying out reduction on the network traffic features to obtain simplified and optimized feature subsets; and training the optimal feature subset by utilizing the SVM to obtain an SVM classifier; adding processed online test network traffic data into the SVM classifier, carrying out calculation by the SVM classifier to obtain a final classification result, and determining whether the processed online test network traffic data is abnormal network traffic data. Compared with the prior art, according to the network traffic abnormality detection method disclosed by the invention, network traffic feature data is subjected to feature reduction and dimensionality reduction by a PCA-TS (Principal Component Analysis-Tabu Search) method, and the optimal feature subset is selected. The problems of long classification detection time, low efficiency and occupation for a larger storage space, which are brought by the curse of dimensionality, are avoided; and moreover, processing time is reduced for subsequent processing, and classification accuracy of the classifier is improved.

The invention discloses a sorting method based on non-supervision feature selection. By means of the method, high dimensional data are expressed in similar diagrams, distances between sample points are obtained through the ITML, and a similar matrix of the original high dimensional data is set up; then the SM algorithm is executed on the similar matrix and a diagonal matrix corresponding to the similar matrix to achieve mapping of original sample sets to feather vector space; then through learning of sparse coefficient vectors and MCFS scores, weight coefficients of all attributes in the original sample set are obtained, and the attribute which can best express the original sample information is selected out; finally a support vector machine is used for setting up a sorting model of the selected data to predict fatigue states of a driver. The method selects features of the high dimensional data under the condition of maintaining data aggregate structures before the sorting model is set up, and the negative effect of curse of dimensionality on data sorting is avoided.

The invention discloses an image classification method based on a category correlated codebook and a classifier voting strategy. The method comprises the following steps of: expressing an image as a set of local salient region image blocks by an image data set pre-processing module; generating category correlated codebooks by a category correlated codebook generating module; expressing the image as an image vector by an image vectoring module, and training a classifier between two random categories by selecting the trained image vector and a category tag corresponding to the trained image through a category correlated classifier training module; and finally, determining the category tag of the tested image according to voting results by a classifier-voting-strategy-based tested image classifying module. The category correlated codebook generating module effectively solves the contradiction of dimension disaster caused by over large codebooks and judgment insufficiency caused by over small codebooks; and meanwhile, the category correlated classifier training module also gets rid of the problems caused by sample unbalance in multi-category classification, and the classification performance is improved.

The invention discloses a multi-feature locality sensitive hashing (LSH) indexing combination-based remote sensing image retrieval method and belongs to the technical field of remote sensing image retrieval. According to the multi-feature LSH indexing combination-based remote sensing image retrieval method disclosed by the invention, LSH indexing of one of the best indexing technologies in high-dimensional feature spaces is introduced into the field of the remote sensing image retrieval, so that the problems of curse of dimensionality and retrieval time consuming can be effectively solved on a large scale, and the rapid retrieval of remote sensing images is realized. Meanwhile, the invention provides a new indexing validation index-a feature discriminative-ness-based indexing validation index (FDIVI) by aiming at the LSH indexing, and features best capable of distinguishing targets and backgrounds are evaluated and selected by the LSH indexing in all feature spaces, and therefore, the accuracy of a retrieval result is effectively improved. Compared with the prior art, the multi-feature LSH indexing combination-based remote sensing image retrieval method disclosed by the invention is capable of more rapidly and accurately realizing the retrieval of a great amount of remote sensing image data.

The invention discloses an intelligent operation and maintenance statement similarity matching method based on a natural language processing technology. The method mainly comprises two parts of data processing in knowledge base construction and sentence similarity matching based on deep learning. Compared with the prior art, the method has the advantages that (1) the operation and maintenance management knowledge is subjected to word segmentation by utilizing the specific word library and the HMM to find the new word model, so that the text word segmentation accuracy is improved, and the moreperfect text word library is established; (2) word vectors are trained through a deep learning method, so that the phenomenon of'dimensionality disaster 'represented by the word vectors can be avoided, information of vocabulary contexts can be fully mined, and relations between words can be obtained; And (3) on the basis of the sentence vectors configured with the weights, not only can the importance measure of each word be obtained, but also the information of the sentence vectors can be richer through the combination of the word vectors, and the accuracy of matching on the basis of forming the sentence vectors can be guaranteed through a cosine similarity matching algorithm.

Provided is a gait recognition method based on a similar rule Gaussian kernel function classifier. The method comprises the steps that a camera collects a current background image and an original gait image sequence of a detection target in real time, image preprocessing is carried out by the adoption of an Euclidean distance method and the like, and a standard gait image sequence is obtained; one gait sequence is divided into three gait subsequences by application of an interval frame grabbing technology, feature extraction is carried out, and gait feature vectors are obtained; similar rule construction is carried out by utilization of the feature vectors in a gait feature vector database; the gait feature vectors of the detection target are classified through the Gaussian kernel function classifier corresponding to the similar rule construction, and a recognition result is counted and output. The method can rapidly remove the background, and adaptability under different situations is improved by application of the image normalization and the interval frame grabbing technology. In addition, the similar rule Gaussian kernel function classifier can effectively avoid the problems of overfitting, dimension disasters and the like, and improve the integral recognition precision.

The invention discloses an identifying method of brain cognitive states based on tensor locality preserving projection (Tensor Locality Preserving Projection, TLPP). The method comprises the following steps: 1) pretreating and grouping of fMRI (functional Magnetic Resonance Imaging) data of the brain cognitive states; 2) constructing a neighbor graph G and a corresponding incidence matrix S; 3) calculating characteristic decomposition of a training sample set, solving corresponding characteristic transformation matrix and calculating low dimensional imbedding of training samples; 4) classifying and identifying: calculating low dimensional imbedding of the training sample sets, and distinguishing and classifying the training sample sets by a tensor distance-based neighbor classifier. According to the method, dimensionality reduction and characteristic extraction are directly carried out on multidimensional tensors by TLPP algorithm, and characteristic dimensionality reduction is carried out on collected brain cognitive fMRI data, so that the brain cognitive states are effectively identified and classified. By combining the tensor distance-based neighbor classifier, the classifying accuracy is improved. The method not only inherits advantages of conventional methods, but also greatly reduces complex of time and space and overcomes curse of dimensionality. The method is less in calculated amount, less in memory consumption and shorter in time consumed.

The invention provides a semantic comprehension system and method oriented to a Chinese text. Based on deep learning, a deep learning text classification model is provided; the model is divided into an input layer, a convolutional layer, a pooling layer, a GRU (Gated Recurrent Unit) layer, a fully connected layer and an output layer; a pinyin characteristic sequence of text segmentation is used asinput; characteristics are obtained through multi-layer characteristic extraction; and an intention category is predicted to obtain a text classification result. According to the semantic comprehension system and method oriented to the Chinese text, the part of speech of a statement does not need to be judged, a complex preprocessing process such as a syntax analysis tree and the like does not need to be generated, the text only needs to be segmented and the segmented text is converted into pinyin, and the problems that the relation between words and words cannot be measured, a lot of external prior knowledge is needed and the curse of dimensionality is easily generated when large-scale corpuses are processed in a conventional characteristic extraction method are solved.

The invention, which belongs to the technical field of artificial intelligence and control, relates to a neural network and enhanced learning algorithm, particularly to an inverted pendulum control method based on a neural network and reinforced learning, thereby carrying out self studying to complete control on an inverted pendulum. The method is characterized in that: step one, obtaining inverted pendulum system model information; step two, obtaining state information of an inverted pendulum and initializing a neural network; step three, carrying out and completing ELM training by using a straining sample SAM; step four, controlling the inverted pendulum by using an enhanced learning controller; step five, updating the training sample and a BP neural network; and step six, checking whether a control result meets a learning termination condition; if not, returning to the step two to carry out circulation continuously; and if so, finishing the algorithm. According to the invention, a problem of easy occurrence of a curse of dimensionality in continuous state space as well as a control problem of a non-linear system having a continuous state can be solved effectively; and the updating speed becomes fast.

The invention provides a large-scale interconnected power grid spinning reserve risk assessment method based on a state space division method Along with the appearance of a large-scale interconnected power grid, the quantity of elements of a power grid system is sharply increased, and the network scale is further complicated, thus the traditional reserve risk assessment method cannot meet the development requirements of the large-scale interconnected power grid, has the problems of curse of dimensionality in the aspect of state selection and inability of assessment in the aspect of state analysis, and can be influenced by transmission capacity restraint of area connecting lines. In the invention, the system state is selected by using a state space division method and analyzed by using a load reduction technology based on linear programming, thus the method better solves the problems of huge system scale and transmission capacity restraint of the area connecting lines in the spinning reserve risk assessment, realizes the quick and precise calculation of a spinning reserve risk index of the large-scale interconnected power grid, can provide real-time system spinning reserve information for personnel of a dispatching system, is beneficial for the dispatching personnel to taking effective measures in time, improves the reliability level of the system, and reduces the power-off risk of the system.

The invention belongs to the field of target tracking and mainly relates to a target tracking nonlinear system target tracking method based on distributed volume information filtering. The existing volume Kalman nonlinear system target tracking method is achieved on the premise that premise noise and measurement noise are not relevant and each measurement noise is not relevant, so that using scope of the volume Kalman nonlinear system target tracking method is greatly limited. The target tracking nonlinear system target tracking method deduces noise related expanding Kalman information filtering, volume Kalman information filtering is embedded in a time updating process and a measurement updating process, a noise relevant problem is solved, and practical applicability of the method is greatly strengthened. In addition, the method is based on decentralization, a theory of matrix diagonalization is used, dimensionality of a matrix is reduced to great extent, and dimensionality curses caused by high dimensions are avoided.

The invention provides a large-scale data abnormity recognition method based on bidirectional sampling combination. The method includes the following steps of carrying out crosswise sampling on a sample data set to obtain a sub-sample data set, carrying out attribute sampling on the sub-sample data set to obtain a stripe data set, carrying out abnormity degree grading on the stripe data set, repeating the above steps, combining abnormity degree scores and calculating values of expectation of the abnormity degree scores. Through the bidirectional sampling method, the large-scale data abnormity recognition method solves the problems that the number of the samples is large and the complexity is high and also solves the problem in curse of dimensionality; the data set is cut based on the sampling method, and therefore the expansibility of the large-scale data abnormity recognition method is improved.

The invention provides a method for predicting defects of object-oriented software. The method comprises the steps of processing a training data set, acquiring valid feature attributes and building a new training data set according to the valid feature attributes; training a support vector machine (SVM) according to the new training data set and performing parameter optimization via particle swarm optimization (PSO), wherein parameters include a penalty factor and a Gaussian kernel bandwidth; using an SVM model to perform defect prediction on prediction data according to the optimized parameters and obtaining a prediction result. A training data set is processed to acquire valid feature attributes and a new training data set is built according to the valid feature attributes, so that curse of dimensionality is avoided effectively, processing cost is reduced and data processing speed is increased; PSO is employed for parameter optimization, so that optimal parameters can be selected and defect prediction accuracy is improved.

The invention discloses a dynamic reactive power optimization method of a large-scale alternating current and direct current power system. The method comprises the following steps of: firstly, solving a sensitivity coefficient matrix of an alternating current and direct current system state variable on a control variable according to an equality constraint; secondly, solving a sensitivity coefficient matrix of section power on the control variable in consideration of a section power inequality constraint, and thus establishing an alternating current and direct current reactive power optimization successive linear planning model of a section power constraint at a single time point; and finally, solving and considering coupling constraints of adjacent time section direct current line power fluctuation smoothness one by one according to each time point, and introducing a computation result of the previous time point at a solving time point, so a dynamic optimization idea is truly embodied. By adoption of the method, the coupling constraints of the adjacent time section direct current power fluctuation smoothness can be considered, the problems of high computation amount and curse of dimensionality caused by the conventional solving dynamic reactive power optimization complete model are solved.

The invention provides a video recognition method based on a space-time pyramid network. The method comprises steps of extracting characteristics of each video clip sample in a video sample set through the convolution neural network, carrying out time-space linear operator processing so as to acquire a first vector and through a second convolution neural network, acquiring image information of image samples and acquiring a second vector; carrying out time-space linear operator processing on the vector obtained by splicing the first vector and the second vector; carrying out weighting pooling on an output result and the second vector so as to acquire a third vector; through average pooling, acquiring a fourth vector and a fifth vector and then carrying time-space linear operator processingso as to acquire a sixth vector; and according to a loss value, recognizing a to-be-detected video. According to the invention, through the dimension reduction operation and inverse transformation operation, problems of curse of dimensionality of bilinearity fusion and high operation complexity are solved; and by improving the bilinearity fusion operators, under the condition that two videos havethe similar background or the similar short films, better recognition effects are acquired.

The invention discloses an LSB replacement steganalysis method based on grey co-occurrence matrix statistic features. The method includes the steps of image bit plane decomposition, grey co-occurrence matrix calculation, feature selection and extraction and classification. Firstly, a grey image is decomposed into eight bit planes, differential matrixes between the lowest bit plane and other seven bit planes are respectively calculated, then a sum matrix of the differential matrixes is calculated, a grey co-occurrence matrix of the sum matrix is generated, the obvious features are extracted and calculated by researching and analyzing the characteristics of the co-occurrence matrix, and a support vector machine is used as a classifier to distinguish carrier images and hidden images. According to the method, the number of feature dimensions is small, curse of dimensionality is effectively avoided, detection precision is high, an algorithm is stable, robustness of image processing of retaining operations of JPEG compression, median filtering, noise addition and the like is achieved, satisfying generalization ability is achieved, and calculation complexity is low.

The invention discloses a high-dimensional multimedia data classifying method based on maximum margin tensor study. The method includes the following steps that (1) a training data set of multimedia data is built; (2) the training data set is modeled and analyzed to obtain a classifying model; (3) according to a user inquiry data set and the classifying model, the inquiry data set is classified. According to high-dimensional performance and structure performance of the multimedia, the multimedia data is expressed through tensor, and high-dimensional multimedia data is classified through a maximum margin classifier method. Classifying is finished while the multimedia data is subjected to decomposition analysis, structural information in the multimedia data is reserved, dimensionality curse caused by high-dimensional data generated through a traditional splicing method is avoided, and the method is more accurate than a traditional multimedia data classifying method and facilitates calculation.

The invention opens a retrieval method of human motion data based on data-driven and the decision tree analysis. This method extracts a method of three-dimensional space-time characteristics based on the transformation rule of three-dimensional space, from various key points of the human body among data of movement capture, and introduces a key space-time concept based on the continuity of movement in time and space. Because the three-dimensional space-time characteristics to avoid dealing directly with data of high-dimensional primitive movement of human body, thereby reducing dimension of the characteristic level, avoiding a dimension disaster, to achieve lower cost, and aiming on that characteristics of the key points in time and space relative to maintain an independent identity, through the study method of data-driven decision tree to analyse the different effects On learning key points of similar campaigns, making retrieval process complete the matching calculation from an important key points to the minor key points in turn, thereby excluding a large number of unnecessary similarity computing of minor key points, and ultimately achieve an efficient campaign retrieval system.

The invention discloses a fuzzy two-dimensional uncorrelated discriminant transformation based face recognition method. The method includes: firstly, adopting the fuzzy k-nearest neighbor to realize initial fuzzy processing of face images; secondly, calculating a first discriminant vector of fuzzy two-dimensional uncorrelated discriminant transformation; thirdly, calculating an optimal discriminant vector set of the method; finally, subjecting the two-dimensional face images to fuzzy two-dimensional uncorrelated discriminant transformation so as to realize accurate recognition of faces. By the method, the problem that internal data structures of the images are destroyed due to the fact that the images must be pulled into vectors by line or column during fuzzy two-dimensional uncorrelated discriminant transformation of the two-dimensional face images is solved, the problem of 'curse of dimensionality' caused when the two-dimensional images are pulled to the vectors can be avoided, face discrimination information of the two-dimensional face images can be effectively extracted, and recognition accuracy is high.

The invention provides a partition optimization method and a partition optimization system for an UHV AC / DC feed-in receiving-end power grid. The partition optimization method comprises the steps of:determining a power grid key channel set of a power grid according to a power grid partitioning method based on a complex community structure theory; and performing multi-objective optimization iterative calculation on randomly generated partitioning schemes based on the power grid key channel set according to a non-dominated sorting genetic algorithm NSGA-II based on fast classification, so as toobtain the optimal partitioning scheme set. The partition optimization method and the partition optimization system combine the complex community structure theory with the non-dominated sorting genetic algorithm NSGA-II based on fast classification and apply them to grid planning, thus the problem of curse of dimensionality of a large power grid is avoided, and the method is simple, practical andhigh in operability.

The invention discloses a distributed power supply cluster dynamic partitioning method and system. The method comprises the following steps: analyzing different control requirements of a power distribution network according to voltage distribution of a dominant node and dividing an economic control cluster and an emergency control cluster, wherein different control clusters adopt different cluster partition indexes; calculating a similarity matrix between distributed power supplies by utilizing the cluster partition indexes, and enabling the distributed power supplies to be equivalent to a network structure based on the similarity matrix; and based on an analysis method of a community theory in a complex network, and by utilizing a Q-value function for evaluating a community structure, dividing distributed power supply clusters. The method and system solve the problems of application limitation of cluster control and "curse of dimensionality" of distribution network scheduling and operation control, enable the cluster control to be expanded to a power distribution network having various distributed power supplies, realizes dynamic division of the distributed power supply clusters, have the advantages of high flexibility and adaptability and the like, can effectively improve control efficiency of the distributed power supplies, improve voltage control level of the power distribution network and realize safe and economical operation of the power distribution network.