707 results about "Deep belief network" patented technology

The invention discloses a power grid security situation predicting method based on an improved deep learning model and belongs to the technical field of power system safety. The power grid security situation predicting method includes: performing power grid security situation evaluation through power grid data collection and preprocessing; aiming at the characteristic that indicator data of power grid security situation evaluation are high in relevance and dimension, providing an improved self-coding network method to lower dimension of the indicator data, and utilizing a data sample after dimension reduction and a power grid security situation value corresponding to a next time monitoring point; adopting an improved deep belief network to build a deep learning situation predicting model with multi-input and multi-output for power grid security situation prediction. By the power grid security situation predicting method, speed and accuracy of power grid security situation prediction can be improved effectively.

The invention provides a fault diagnosis method for a rolling bearing based on a deep learning and SVM (Support Vector Machine). The method comprises using a manure learning algorithm in a deep belief network theory to complete a characteristic extraction task needed by fault diagnosis; automatically extracting the substantive characteristics of data input independent of manual selection from simple to complicate, from low to high, and automatically digging abundant information concealed in known data; in addition, classifying and identifying a test sample by adopting an SVM classification method, seeking and finding a global minimum of a target function through an effective method previously designed, so as to solve the problem that a deep belief network may be trapped into a locally optimal solution. According to the fault diagnosis method for the rolling bearing based on the deep learning and SVM provided by the invention, the accuracy and effectiveness of the fault diagnosis method for a rolling bearing can be improved, and a new effective way can be provided to solve the accuracy and effectiveness of the fault diagnosis method, therefore the fault diagnosis method can be extensively applied complex systems in chemistry, metallurgy, electric power, aviation fields and the like.

The invention discloses a short-term load prediction method of a variational mode decomposition and deep belief network. The method comprises the following steps of 1) using a variational mode decomposition method to decompose original historical load data into a series of mode functions with different characteristics; 2) using an approximate entropy to calculate each modal function complexity, merging the modal functions whose approximate entropy values are similar into a new component, and carrying out characteristic analysis on each component; 3) in order to calculating correlation of an influence factor and an output variable, carrying out normalization processing on data; 4) combining a period characteristic of a load, and using a mutual information theory to select an input variable set from aspects of a historical load, a meteorology factor, a date type and the like; and 5) constructing the short-term load prediction method based on the deep belief network (DBN), and verifying method validity through a load prediction scene before 24h. By using the method, short-term load prediction precision is effectively increased and an electric power system load prediction problem can be well solved.

The invention discloses a speaker recognition method based on depth learning. The method comprises the following steps: S1) carrying out pre-emphasis and overlapping-type framing windowing on collected voice signals; S2) carrying out endpoint detection on the collected voice signals by utilizing a dual-threshold endpoint detection method based on short-time energy and short-time zero-crossing rate, and judging and indentifying the staring moment, transition stage , noise section and ending moment of the voice; S3) carrying out feature extraction on the voice signals; S4) forming a depth belief network model based on restricted boltzmann machine hierarchy, training the established depth belief network model by utilizing layer-by-layer greedy algorithm and with speaker voice feature parameters being combined, and adding a Softmax classifier to the top layer of the depth belief network model; and S5) inputting the voice features of a speaker to the depth belief network model obtained after being subjected to training, calculating the probability that the model outputs voice features similar to the voice features of the other speakers, and selecting the speaker corresponding to the maximum probability as recognition result.

Deep belief networks are usually associated with a large number of parameters and high computational complexity. The large number of parameters results in a long and computationally consuming training phase. According to at least one example embodiment, low-rank matrix factorization is used to approximate at least a first set of parameters, associated with an output layer, with a second and a third set of parameters. The total number of parameters in the second and third sets of parameters is smaller than the number of sets of parameters in the first set. An architecture of a resulting artificial neural network, when employing low-rank matrix factorization, may be characterized with a low-rank layer, not employing activation function(s), and defined by a relatively small number of nodes and the second set of parameters. By using low rank matrix factorization, training is faster, leading to rapid deployment of the respective system.

Systems and methods are provided for fusing new evidence and displaying node parameters of a decision network. The decision network can be a belief network, such as a Dempster-Shafer belief network. The Dempster-Shafer belief network includes node parameters that conform to the Dempster-Shafer combination rule, which is based on an evidential interval: the sum of a belief value, a disbelief value and an unknown value is equal to one. A user override is provided to allow a user to override node parameters associated with a hypothesis or outcome and the decision network self-adjusts the appropriate link values or learned to instantiate the override. The back-propagation algorithm is used to adjust the links.

The invention discloses a behavior recognition method based on sparse spatial-temporal characteristics. The method comprises the steps as follows: step 1, convolving an input video with an original input video by using space-time Gabor to establish a scale space; step 2, using the expressions of different scales as values of different channels of a space-time depth belief network and associatively learning characteristics of multi-scale; step 3, recognizing and classifying the behavior characteristics. The behavior recognition method of the invention inputs depth network to associatively learn the characteristics of multi-scale by establishing the scale space to improve performance of behavior recognition. The behavior recognition method of the invention introduces the thought of spatial pyramid in terms of information loss problem of a pooling operation and performs multilevel expansion to pooling output, and combines with sparse coding to fuse pyramid multilevel characteristic, thereby improving characteristic dimension output by a pooling layer, further improving the performance of the original network and improving behavior recognition rate.

The invention discloses a gesture recognition method based on deep learning. The gesture recognition method comprises the following steps: carrying out noise reduction processing to a collected gesture image, and eliminating a color offset phenomenon in the gesture image; locking an area, where a gesture is positioned, in the image by adopting a frame difference method and a color characteristic detection method, and tracking a gesture by adopting a CamShift algorithm to obtain a gesture target; carrying out the deep learning on a gesture target image; and inputting the obtained gesture image to be recognized into a trained deep belief network model to finish the recognition and the classification of the gesture.

The invention discloses a deep learning-based vehicle path optimization method and system. The method includes the following steps that: real-time road data and historical road data are acquired, the acquired data are preprocessed, so that a tagged data set can be formed; a deep belief network model is constructed, and the deep belief network model is trained; the trained deep belief network model is utilized to predict all the paths from a vehicle to a destination, and the congestion coefficients of each path are output; and the paths are evaluated comprehensively based on two indexes, namely the congestion coefficients and distances, and an optimal path is outputted, wherein the optimal path is a path corresponding to the minimum linear accumulation result of the two indexes, namely a corresponding congestion coefficient and a distance. According to the method and system, through the powerful feature extraction function of the deep belief network model, required information can be obtained from multi-dimensional road traffic data, interference can be reduced, a congestion situation is predicted accurately and reasonably, path search efficiency can be improved, human-made mistakes can be reduced, and valuable time can be saved for disaster relief work.

The invention discloses a short-term wind power prediction method based on integrated empirical mode decomposition and a deep belief network. The short-term wind power prediction method comprises the steps of: decomposing an original wind power sequence into a series of intrinsic mode functions with different features by adopting integrated empirical mode decomposition, calculating sample entropy of the original wind power sequence and the intrinsic mode functions, combining the intrinsic mode functions with similar sample entropy values into a new sequence, and forming a random component, a detail component and a trend component; selecting an input variable set by adopting a partial autocorrelation function; constructing a training sample set according to the input variable set of each component; and establishing a deep belief network short-term wind power prediction model for each component, and superposing prediction results of the components, so as to obtain a final short-term wind power predicted value. The short-term wind power prediction method provided by the invention effectively improves the short-term wind power prediction precision, and can effectively solve the wind power prediction problem of the electric power system, so as to provide more reliable guarantee for large-scale wind power integration.

The invention discloses a human face verification method based on bilinear united convolutional nerve network. The human face verification method comprises steps of 1) using a human face image which is prepared in advance to perform convolutional nerve network (CNN) training, 2) using the human face image which is in a training set to perform bilinear CNN fine tuning, 3) inputting a human face image to be verified, segmenting the two images, extracting united characteristics outputted by the bilinear CNN, and 4) making an obtained vector go through self-encoding network training to obtain a final verification result. The human face verification method is based on the bilinear CNN, replaces two repeated inputs of an original bilinear nerve network with different human face verification input images, and brings forward a human face verification description factor. The human face description factor has robustness to illumination, shielding and posture change. Furthermore, the characteristic extracted by the bilinear CNN has a smaller dimensionality than the characteristic dimensionality of a common CNN fully connected layer, which reduces number of parameters, makes follow-up deep belief network training simple and improves accuracy of human face verification.

A fault diagnosis method for power electronic circuits based on optimizing a deep belief network, including steps. (1) Use RT-LAB hardware-in-the-loop simulator to set up fault experiments and collect DC-link output voltage signals in different fault types. (2) Use empirical mode decomposition to extract the intrinsic function components of the output voltage signal and its envelope spectrum and calculate various statistical features to construct the original fault feature data set. (3) Based on the feature selection method of extreme learning machine, remove the redundancy and interference features, as fault sensitive feature data set. (4) Divide the fault sensitive feature set into training samples and test samples, and primitively determine the structure of the deep belief network. (5) Use the crow search algorithm to optimize the deep belief network. (6) Obtain the fault diagnosis result.

The invention discloses a fault diagnosis method of a high-voltage circuit breaker based on a depth belief network, which comprises the following steps: step 1, selecting a data sample required by anexperiment, and dividing the unified standardized sample data into a test sample and a training sample according to a specific proportion; Step 2: building and initializing the DBN deep belief networkfault diagnosis model; Step 3, inputting a large number of unlabeled samples or unlabeled samples in the pre-training set from the bottom of the model, and pre-training the RBM in the model by usinglayer-by-layer unsupervised greedy learning; Step 4: the whole model being fine-tuned by genetic algorithm; Step 5, the fault diagnosis model of the high-voltage circuit breaker obtained by training being classified to the fault samples of the test set in step 1, so as to obtain the fault classification result, and the diagnosis accuracy rate of the model being counted. The invention discloses a fault diagnosis method of a high-voltage circuit breaker based on a depth belief network, which can train a large amount of data samples to realize the fault diagnosis function of the high-voltage circuit breaker.

The invention discloses a real-time monitoring method of the state of cutting tools for numerical control machining of complicated structural components based on deep learning. The method is characterized by comprising the following steps: constructing a two-level deep learning model including deep belief network and convolutional neural network, training a deep learning network based on a large number of numerical control machining monitoring signals, realizing real-time monitoring of the state of the cutting tools; firstly, adopting a large number of monitoring signal data to train the deepbelief network so as to realize automatic extraction of characteristics of the monitoring signals, constructing a signal characteristic input matrix, then establishing a relationship between the monitoring signals and process information and geometric information so as to construct the convolutional neural network, training the convolution neural network by a large amount of sample data, establishing the mapping relationship between monitoring information and the state of the cutting tools, finally, according to the real-time monitoring information during numerical control machining, determining the state of the cutting tools through the trained deep learning model. The method is suitable for monitoring the state of cutting tools for numerical control machining of complicated structural components in mass production of parts as well as in small batches or even single-piece production.

The invention discloses a human face age estimation method based on the fusion of deep characteristics and shallow characteristics. The method comprises the following steps that: preprocessing each human face sample image in a human face sample dataset; training a constructed initial convolutional neural network, and selecting a convolutional neural network used for human face recognition; utilizing a human face dataset with an age tag value to carry out fine tuning processing on the selected convolutional neural network, and obtaining a plurality of convolutional neural networks used for age estimation; carrying out extraction to obtain multi-level age characteristics corresponding to the human face, and outputting the multi-level age characteristics as the deep characteristics; extracting the HOG (Histogram of Oriented Gradient) characteristic and the LBP (Local Binary Pattern) characteristic of the shallow characteristics of each human face image; constructing a deep belief network to carry out fusion on the deep characteristics and the shallow characteristics; and according to the fused characteristics in the deep belief network, carrying out the age regression estimation of the human face image to obtain an output an age estimation result. By sue of the method, age estimation accuracy is improved, and the method owns a human face image age estimation capability with high accuracy.

The invention discloses a deep belief network-based short text feature optimization and sentiment analysis method. The method comprises the following steps of: 1, obtaining a microblog short text corpus set, a thesaurus, semantic progressive associated words, microblog expression dictionary and a word segmentation model; 2, expanding and reconstructing short texts; 3, carrying out word segmentation and preprocessing on the short texts; 4, constructing a word similarity calculation model; 5, expanding feature vectors of the short texts; 6, self-adaptively extracting an expanded candidate feature set on the basis of a feature depth of a deep belief network; 7, carrying out classified training on the feature set obtained by the deep belief network by utilizing a machine learning classification algorithm so as to obtain a classification and prediction model; and 8, carrying out sentiment annotation on a test data set by utilizing the classification and prediction model. The method is capable of finding potential feature semantic information more effectively and improving the quality of sentiment feature extraction so as to improve the correctness of sentiment classification.

The invention discloses a multi-feature fusion-based deep learning face recognition method. The multi-feature fusion-based deep learning face recognition method comprises performing feature extractionon images in ORL (Olivetti Research Laboratory) through a local binary pattern and an oriented gradient histogram algorithm; fusing acquired textual features and gradient features, connecting the twofeature vectors into one feature vector; recognizing the feature vector through a deep belief network of deep learning, and taking fused features as input of the deep belief network to layer by layertrain the deep belief network and to complete face recognition. By fusing multiple features, the multi-feature fusion-based deep learning face recognition method can improve accuracy, algorithm stability and applicability to complex scenes.

The embodiment of the invention provides an electrical power system load prediction method and device based on a depth belief network (DBN) and relates to the field of electric power systems. Through the method and device, the convergence rate can be increased, and the prediction error can be lowered. According to the specific scheme, the method comprises the steps that a training sample and a test sample are acquired; an energy function of an RBM model is constructed; the training sample is utilized to perform layer-by-layer training on at least one hidden layer and visible layer to obtain weights of the training sample among nodes of the hidden layers and the visible layers; and a prediction value of an electrical power system load is obtained according to output data obtained through the training sample and the DBN obtained after test sample input is trained. The method and device are used for electrical power system load prediction.

The invention discloses a deep belief network image recognition method based on Bayesian regularization and belongs to the field of artificial intelligence and machine learning. The deep belief network plays a more and more important role in the field of digital detection and image recognition. The invention provides a deep belief network based on Bayesian regularization on the basis of the network sparsity characteristic and changes of connection weights to solve the problem of overfitting in the training process of the deep belief network. By applying Bayesian regularization to the network training process, balance between error decreasing and weight increasing is effectively adjusted. The classification experiment of a digital script database proves effectiveness of the improved algorithm. An experimental result shows that in the deep belief network, the deep belief network image recognition method can effectively overcome the overfitting phenomenon and improve accuracy of digital recognition.

The invention provides a speech recognition model establishing method based on bottleneck characteristics and a multi-scale and multi-headed attention mechanism, and belongs to the field of model establishing methods. A traditional attention model has the problems of poor recognition performance and simplex attention scale. According to the speech recognition model establishing method based on thebottleneck characteristics and the multi-scale and multi-headed attention mechanism, the bottleneck characteristics are extracted through a deep belief network to serve as a front end, the robustnessof a model can be improved, a multi-scale and multi-headed attention model constituted by convolution kernels of different scales is adopted as a rear end, model establishing is conducted on speech elements at the levels of phoneme, syllable, word and the like, and recurrent neural network hidden layer state sequences and output sequences are calculated one by one; and elements of the positions where the output sequences are located are calculated through decoding networks corresponding to attention networks of all heads, and finally all the output sequences are integrated into a new output sequence. The recognition effect of a speech recognition system can be improved.

The invention discloses a natural language semantic analysis system and method based on the depth neural network. The method comprises the steps that a knowledge map is built, a training set is inputted, and an N-Gram probability model is obtained, a matrix is obtained as an input by representing words as vectors using the word2vec, a deep belief network model is used for the entity identification and the input validation set, the classifier parameters and the input test set are adjusted, the group abilities of the models are tested, the knowledge graph method is adopted to apply reasoning to the entities in the descriptions of the language, and corresponding conclusions are obtained. Compared with the prior art, the natural language semantic analysis system and method based on the depth neural network uses the knowledge graph method to apply reasoning to the entities in the descriptions of the language and to obtain the corresponding conclusions, so that our natural language understanding abilities are provided not only with the capacity to understand the literal meaning, but also with logical reasoning and the understand of the meaning on a deep level, and the method has promotable and practical value.

The invention discloses a bearing fault mode diagnosis method and system based on deep learning. The method comprises: a learning rate is adjusted automatically, noises are introduced, and a linear correction unit is introduced as an activation function, and thus a deep belief network is improved based on the activation function; a data set and long- and short-term memory networks are migrated toobtain a synthetic data set, a trained data set is extended based on the synthetic data set, and the improved deep belief network is trained based on training data set training to obtain a bearing fault diagnosis model; and a vibration signal of the bearing is collected and a bearing failure mode is diagnosed based on the bearing vibration signal and the bearing fault diagnosis model. On the basisof combination of a semi-supervised learning and a migration learning algorithm, the diagnostic accuracy is improved while no insufficient data are provided.

The invention provides a recognition method, based on deep belief network, of three-dimensional SAR images. The method comprises the following steps: firstly establishing a simulation sample bank of the three-dimensional SAR images, performing projection to different azimuthal angles and pitch angles through one or a small quantity of objective three-dimensional SAR images, so as to obtain a plurality of two-dimensional SAR images, ensuring that the small quantity of obtained three-dimensional SAR images are converted into two-dimensional images, and performing recognition through a two-dimensional image recognition method, and the method can greatly reduce the cost, and reduce the time for acquiring SAR imaging. According to the method, a splicing crossover verification method is proposed, and the deep belief network is improved, so that the deep belief network can automatically adjust parameters, self optimization of parameters is realized, the occurrence of over-fitting learning state and under-fitting learning state is effectively avoided, advance features of sample data can be accurately learnt, a better recognition result is obtained for the deep belief network, the complexity of manual setting of parameters is eliminated, and the recognition efficiency is improved.

The invention, which belongs to the field of voice signal processing and machine learning, relates to a deep-belief-network-characteristic-vector-based channel-robust voiceprint recognition system comprising a voice acquisition and preprocessing module, an original spectral characteristic extraction module, a deep belief network training module, a speaker voiceprint characteristic vector extraction module, a speaker acoustic model generation module and a speaker identification module. On the basis of voice data from different channels and corresponding speaker identity numbers, a deep belief network is trained in a manner of supervision; and a discrimination ratio is provided to select a deep belief network hidden layer output having an optimal class discrimination property, thereby constructing a speaker voiceprint characteristic vector having channel robustness. Compared with the traditional i-vector-based speaker confirmation system, the provided system has higher voiceprint recognition accuracy on the condition of channel mismatching.

The invention relates to a deep belief network model based cement clinker fCaO prediction method. The method comprises the steps that major variables capable of reflecting the firing situation of a cement clinker are preliminarily selected to form an auxiliary variable set, and a prediction variable is the cement clinker fCaO content; a field instrument and an operator recorder respectively acquires auxiliary variables and field data of the cement clinker fCaO content, a grey relational analysis method is adopted conduct dimensionality reduction on the auxiliary variable set; parameters in a deep belief network structure, namely parameters training the deep belief network are determined according to a deep belief network algorithm and sample data volume, and further optimization of weighting and bias of the whole network is achieved; a counter-propagation algorithm is adopted to conduct error correction on the determined parameters in a deep belief network structure, and further a prediction model of the cement clinker fCaO is determined; real-time data of the auxiliary variable set is acquired, and errors of the obtained real-time data of the auxiliary variable set are eliminated according to 3delta criterions; further, the cement clinker fCaO content is predicted.

The invention relates to a virtual network function dynamic migration method based on deep belief network resource demand forecasting, and belongs to the field of mobile communication. The method comprises the following steps: (S1) in view of the dynamic features of SFC business resource demand in a slicing network, establishing a system overhead model of comprehensive migration overhead and bandwidth overhead; (S2) in order to realize spontaneous VNF migration, monitoring the resource utilization condition of virtual network function or link in real time, and discovering the deployed bottom nodes or resource hot spots in the link in time by using an online learning based adaptive DBN forecasting method; (S3) designing a topology awareness based dynamic migration method according to the forecasting result, so as to reduce system overhead; (S4) proposing a tabu search based optimization method to further optimize the migration strategy. The forecasting method provided by the invention not only increase the convergence rate of a training network, but also realizes a perfect forecasting effect; by combining the forecasting method with a migration method, the system overhead and the violation frequency of the service level agreement are effectively reduced, and the performance of network service is improved.

The invention provides a water quality prediction method, which comprises the following steps of building a water quality prediction model on the basis of a deep belief network and a long short time memory network; training a water quality prediction model on the basis of the historical water quality time sequence data and weather data; predicting the water quality on the basis of the water quality prediction model. The invention also provides a water quality prediction device, which comprises a building module, a training module and a prediction module. The water quality prediction method anddevice provided by the invention have the advantages that the deep belief network and the long short time memory network are introduced into the water quality prediction model; a great amount of water quality time sequence data and weather data are processed by using the deep belief network; the relationship between all factors is mined; the groundwork is laid for high-precision prediction difficult to realize the multi-variable nonlinear parameter data; on the basis of the deep belief network, the mass water quality time sequence data and weather data processing results are predicted on thebasis by studying the long-time relationship between the water body quality and the time sequence by the long short time memory network; the optimum prediction effect is obtained.

The invention discloses a tobacco leaf grading method based on a hyperspectral image and a deep learning algorithm. The tobacco leaf grading method comprises steps of 1, obtaining hyperspectral image data of a tobacco leaf sample to be measured, 2, performing high level characteristic extraction on the image data to perform dimension reduction, and 3, performing classification on obtained image information and spectral information. A hardware platform of a hyperspectral imaging system comprises a light source, a light splitting module, an area array CCD detector and a computer provided with an image collection card; spectral information can be obtained while the imaging system is utilized to perform image information collection, separate collection is not needed and collection time is shortened; in the step 2, a convolutional neural network is utilized to perform pre-processing and then a deep belief network is utilized to perform characteristic extraction; in the step 3, a Sofmax layer is added on the top layer and obtained characteristics are inputted into a softmax regression classifier to realize classification. The tobacco leaf grading method based on the hyperspectral image and deep learning can maximally achieve lossless grading, accurately divides a tobacco leaf grade, and ensures benefits of a purchasing party.

The invention relates to the technical field of deep learning, and in particular to a method and a system for classifying vehicle-borne laser-point cloud targets. The method for classifying the vehicle-borne laser-point cloud targets comprises the following steps of preprocessing vehicle-borne laser-point cloud data to generate a to-be-classified target point cloud; constructing a basic training sample database; generating an input characteristic vector; constructing a deep belief network; training the deep belief network; generating the characteristic vector of the to-be-classified target point cloud, and completing classification of the vehicle-borne laser-point cloud targets by taking the characteristic vector as an input characteristic of the trained deep belief network. Automatic identification and classification of the vehicle-borne laser-point cloud data are realized by use of the deep belief network, the automation degree of identification and classification of the target point cloud is effectively improved, and the method and the system are strong in robustness and can be applied to vehicle-borne laser-point cloud data with complicated scenes.

The present invention discloses a face expression recognition method based on deep learning. The method comprises the steps of: decomposing obtained video signals into image sequences, and establishing a skin color model to perform skin detection and remove a background area; employing a face detecting algorithm of a space skin color model to train a face classifier and perform face detection of candidate face areas to obtain a face area; performing processing of the face area image such as affine transformation and correction to obtain a face expression image; and finally, extracting LBP features from the expression image, employing a deep belief network to perform learning and training of the LBP features to obtain higher-level extraction features, and taking the higher-level extractionfeatures as a classifier for face expression recognition. The face expression recognition method based on deep learning can be widely applied at occasions such as public safety, psychological lie-detection and smart home, and a computer can provide more humanized service for the mankind through expression recognition. The face expression recognition method has very wide application scenes and verygood application prospects.