37 results about "Denoising autoencoder" patented technology

The invention provides to a method and a system for fault diagnosis of a gearbox of a wind turbine generator based on stacked denoising autoencoders and relates to fault diagnosis. Signals obtained by pre-processing original vibration signals collected when the gearbox of the wind turbine generator is in different working states are used as training data. The training data are input into stacked denoising autoencoders. Meanwhile, a quantum-behaved particle swarm optimization algorithm is introduced to optimize the structure and parameters. Then, pre-processed test signals are input into the stacked denoising autoencoders that are trained to extract high-dimensionality fault features contained in the original vibration signals. Then, the extracted fault features are input into a least squares support vector machine to complete the fault diagnosis of the gearbox.

A system and method for automatically interpreting EEG signals is described. In certain aspects, the system and method use a statistical model trained to automatically interpret EEGs using a three-level decision-making process in which event labels are converted into epoch labels. In the first level, the signal is converted to EEG events using a hidden Markov model based system that models the temporal evolution of the signal. In the second level, three stacked denoising autoencoders (SDAs) are implemented with different window sizes to map event labels onto a single composite epoch label vector. In the third level, a probabilistic grammar is applied that combines left and right context with the current label vector to produce a final decision for an epoch. A physician's report with diagnoses, event markers and confidence levels can be generated based on output from the statistical model. Systems and methods for dealing with channel variation or a missing EEG electrode valve are also disclosed. A feature-space boosted maximum mutual information training of discriminative features or an iVectors technique to determine invariant feature components can be implemented for generating a plurality of EEG event labels. An optional GUI allows scrolling by EEG events.

The invention discloses a power system probabilistic-optimal power flow calculation method based on a stacked denoising autoencoder. The calculation method comprises the following main steps that: 1)establishing a SDAE (stacked denoising autoencoder) optimal power flow model; 2) obtaining the input sample X of a SDAE optimal power flow model input layer; 3) initializing the SDAE optimal power flow model; 4) training the SDAE optimal power flow model so as to obtain a trained SDAE optimal power flow model; 5) adopting a MCS (Modulating Control System) method to carry out sampling on the randomvariable of a power system to be subjected to probabilistic power flow calculation so as to obtain a calculation sample; 6) inputting training sample data obtained in S5 into the SDAE optimal power flow model which finishes being trained in S4) in one time so as to calculate an optimal power flow online probability; and 7) analyzing the optimal power flow online probability, i.e., drawing the probability density curve of the output variable of the SDAE optimal power flow model. The method can be widely applied to the probabilistic-optimal power flow solving of the power system, and is especially suitable for an online analysis situation that system uncertainty is enhanced due to high new energy permeability.

The invention discloses an insulation fault diagnosis method for high-voltage equipment. The method comprises the steps of (1) acquiring a polluted training sample for representing partial discharge signals of a plurality of insulation fault types of the high-voltage equipment, extracting phase resolution pulse sequence data based on the polluted training sample, and normalizing the phase resolution pulse sequence data; (2) training a constructed deep sparse denoising autoencoder through the normalized phase resolution pulse sequence data; and (3) inputting the polluted partial discharge signals of the to-be-identified high-voltage equipment into the trained deep sparse denoising autoencoder, and obtaining the defect type of the high-voltage equipment based on output of an output classification layer of the deep sparse denoising autoencoder. In addition, the invention further discloses an insulation fault diagnosis system for the high-voltage equipment. The system comprises a data preprocessing module and a fault identification module.

The invention provides a hyperspectral image robustness classification method based on segmentation depth characteristics and low-rank representation. The method comprises the following steps that: firstly, in order to lower an influence of noise on feature extraction as far as possible, using a stack-based denoising autoencoder netework to carry out unsupervised feature extraction on a hyperspectral image; then, fully mining an intra-class similarity and an inter-class difference in the hyperspectral image to establish a robust classifier based on the low-rank representation; and finally, adopting an effective optimization method to carry out optimal solving on a target function. Under a situation that training data contains noise, a good classification effect can be obtained.

The invention discloses a planetary gear fault identification method based on an SDAE and a GRUNN. The method comprises the following steps of step 1, constructing a mixed model based on the SDAE andthe GRUNN, and eliminating noise components of input data, processing the time sequence data which are related before and after, and automatically extracting robust fault features from noisy samples;step 2, taking the training samples of the fault diagnosis of the planetary gears as input data of the mixed model constructed in the step 1, training the mixed model through an Adam optimization algorithm and a dropout technology, and preventing the occurrence of an over-fitting phenomenon; and step 3, identifying the state of the planetary gears in the to-be-diagnosed samples through a softmax classifier according to the trained mixed model. According to the method, a good diagnosis effect can be obtained under the condition that the training sample number is small, so that the method has relatively high anti-noise capability and time-varying rotating speed adaptability, thereby providing a novel solving idea for the fault identification of the planetary gears.

The invention discloses a blast furnace molten iron silicon content online prediction method and system based on a deep migration network. The method comprises the following steps: training de-noising autoencoder networks through molten iron temperature data in an unsupervised manner, and stacking a plurality of de-noising autoencoder networks, thereby obtaining a deep de-noising autoencoder network; embedding a dynamic attention mechanism module into the front end of a deep denoising autoencoder network, obtaining a deep network based on a dynamic attention mechanism, migrating a pre-trained deep network based on the dynamic attention mechanism, and obtaining a molten iron silicon content online prediction model. According to the method and the system, the technical problem of low online prediction precision of blast furnace molten iron silicon content in the prior art is solved, and a dynamic attention mechanism module is embedded into the front end of the deep denoising auto-encoder network, so that a dynamic attention score can be calculated for a process variable of each input sample in real time, the model can dynamically distribute more attention to effective and valuable process variables in each sample, and the molten iron silicon content can be further predicted online more efficiently and accurately.

The invention discloses a probabilistic load flow calculation method for an electric power system. The method comprises the following steps that: firstly, obtaining source load data and correspondingelectric power system topological structure data as load flow samples, and utilizing an SDAE (Stacked Denoising Autoencoder) load flow model to calculate the above load flow sample to carry out calculation to obtain a corresponding load flow result so as to carry out statistical analysis to obtain a probabilistic load flow calculation result. Since the SDAE load flow model is obtained in a way that SDAE model training is carried out in advance according to the target source load data and the corresponding electric power system topological structure data, the SDAE model can effectively extracthigh-dimensional nonlinear characteristics in the load flow sample by virtue of a deep stacking structure and a coding and decoding process so as to obtain a corresponding load flow result, and therefore, the calculation accuracy, speed and cost of the probabilistic load flow result can be comprehensively improved. The invention also provides a probabilistic load flow calculation system for the electric power system, and also can realize the above technical effects.

Embodiments of the invention are directed to systems, methods, and computer program products for cross-technology code analysis for redundancy identification and functionality recognition. In particular, the novel present invention provides a unique platform for analyzing software code across multiple coding language using a unique approach involving the use of denoising autoencoders. Embodiments of the inventions are configured to leverage a marginalized stacked denoising autoencoder approach to analyze software code, identify code redundancies, and improve efficiency for code storage and query ability by the use of a trained autoencoding module to autoencode software code attributes into vectorized data that can be compared to determine cross-platform functionality and redundancy within a software library.

The invention discloses a modulation signal denoising method based on a self-encoding neural network, and the method comprises the following steps: 1, simulating a general communication link structure through MATLAB simulation software, and generating a noisy sample data set and a pure sample data set of various communication modulation signals; step 2, performing [0, 1] normalization on each sample set; 3, constructing a modulation signal denoising automatic encoder based on the self-encoding neural network, and setting hyper-parameters; and 4, training the denoising model, and optimizing and updating the values of the parameters in the neural network by using a back propagation algorithm and a gradient descent method to obtain the denoising model. According to the method, the denoising network model based on the self-encoding neural network is used, the complex signal preprocessing process in a traditional modulation signal denoising algorithm is avoided, the overall structure process is simple, the network calculation amount is small, and the denoising speed is high.

The invention discloses a system and a method. The system includes a memory storing processor executable code for a denoised autoencoder, and one or more processors coupled to the memory to execute the processor executable code to receive raw signal data comprising signal noise, encode, by the denoised autoencoder, the raw signal data by performing a denoising autoencoder operation to produce a latent representation, and decode, by the denoised autoencoder, the latent representation to produce clean signal data reconstructed without the signal noise. A first filter is applied to a signal to emphasize activity within the signal and to produce a first modified signal, a rectifier and a second filter are applied to the first modified signal to smooth areas of the first modified signal with clinical importance and to produce a second modified signal, and high frequency energy zones of the second modified signal are automatically detected using an energy threshold to produce a weights vector.

The invention provides an unmanned aerial vehicle image target detection method and system based on a deep denoising autocoder. The method comprises: firstly, extracting radial gradient features of anunmanned aerial vehicle image; sending the extracted gradient features to a deep denoising autocoder; by the autocoder, adding random white Gaussian noise to the extracted radial gradient features; through multi-layer coding, generating high-level features of an unmanned aerial vehicle image with strong representation capability, using a back propagation algorithm to minimize a reconstruction error of the deep denoising autocoder solving model parameters, using a softmax classifier to classify the high-level features, and obtaining a high-precision and high-robustness target detection result.

The invention discloses a turbine engine performance degradation evaluation method based on a stacked-denoising auto-encoder, and belongs to the technical field of engine performance degradation evaluation. The turbine engine performance degradation evaluation method solves the problems that traditional multi-sensor data selection needs to rely on complex information evaluation criteria, extraction of degraded features in HI construction depends on a large number of signal processing techniques and expert experience, supervised training method label selection relies on manual participation, and a method is low in universality. Four denoising auto-encoders build stacked-denoising auto-encoders to extract the single node value of input data. Training set data carries out pre-training to thenetwork and uses a BP algorithm to fine-tune parameters. The extracted single node value is considered to be the health factor value at each cycle, and an HI curve of a training set is established. The test set is input to the trained stacked-denoising auto-encoder to obtain the health factor value at each cycle and construction an HI curve. The HI curves of training set and test set are subjectedto smoothing processing respectively, and the HI curves after smoothing processing are evaluated.

The invention discloses a piRNA-disease incidence relation prediction method based on a convolution denoising autoencoder. The method comprises the following steps: selecting and establishing a data set; generating the similarity characteristics of the Piwi protein interaction RNA sequences; generating semantic similarity features of the diseases; generating nuclear similar characteristics of Piwi protein interaction RNA and a disease Gaussian interaction spectrum; mining deep hidden features; constructing a training set and a test set; and constructing a classifier model. According to the method, excellent performance expression is achieved under a five-fold cross validation experiment, and the effectiveness of the convolutional denoising self-encoding neural network in prediction of Piwi protein interaction RNA and disease association is proved. Case research proves that the practical application capability of the method in discovering the incidence relation between potential Piwi protein interaction RNA and diseases is better proved.

The invention discloses a fault diagnosis method based on incremental compensation and dynamic self-adaptive enhancement, which comprises the following steps: 1. collecting vibration signals in various states from a vibration sensor at the motor drive end; 2. predicting fault data of bearing equipment Processing; 3. Add random noise to the training samples as the input of the denoising autoencoder for unsupervised greedy layer-by-layer pre-training; 4. When there is new equipment status data, use the existing trained DAE model to add Extract the failure mode, and use the mode similarity algorithm to compare similar modes, then use the incremental active fusion algorithm to incrementally merge the newly added failure mode, and use the weight dynamic compensation algorithm to calculate the dynamic weight; And the unlabeled fault data weighted by dynamic deep learning training is used as the input vector to train the SVM classifier; 6. Use the BP algorithm to globally fine-tune the relevant parameters in the entire model; 7. Classify and diagnose fault types.

The invention discloses a sound emission signal denoising method based on a denoising autoencoder (DAE), which is characterized in that the denoising autoencoder is trained through unsupervised learning to learn more stable invariance features, so that the error between a reconstructed signal and an original signal is converged to a minimum value, and the denoising purpose is achieved. A denoising experiment is carried out on the basis of processing 3011 corrosion sound emission signal samples; the experiment result shows that when the number of hidden layer neurons is 300, the denoising model has a good denoising effect, and the denoising model of the denoising autoencoder has better denoising performance and generalization compared with a wavelet threshold denoising method. The denoising autoencoder denoising model is applied to sound emission signal denoising, can effectively remove noise, and is of great significance to subsequent sound emission signal recognition processing.

The invention discloses a fault diagnosis method for a high-speed heavy-load input stage under an unbalanced sample. The specific steps are: use different simulated fault modes of a low-speed gearbox experimental platform to monitor vibration signals, obtain low-speed fault spectrum data, and obtain different fault modes through spectrum shifting. The fault spectrum data of the high-speed gearbox experimental platform; the fault spectrum data of the high-speed gearbox experimental platform is migrated to the fault mode of the helicopter input stage gearbox, and the fault spectrum data of the input stage of different fault modes are obtained; the fault spectrum data of the input stage obtained by two migrations and the normal In the mode, the actual running state monitors the vibration signal spectrum data, pre-trains the deep stack denoising autoencoder SDAE, and fine-tunes the SDAE with a small number of fault data samples at the input level during actual operation. The trained SDAE is used for automatic feature extraction and fault classification of high-speed heavy-duty input stage vibration signals. The invention solves the fault diagnosis problem of the helicopter's high-speed and heavy-load input stage under the condition of unbalanced samples.

The invention discloses an insulation fault diagnosis method for high-voltage equipment. The method comprises the steps of (1) acquiring a polluted training sample for representing partial discharge signals of a plurality of insulation fault types of the high-voltage equipment, extracting phase resolution pulse sequence data based on the polluted training sample, and normalizing the phase resolution pulse sequence data; (2) training a constructed deep sparse denoising autoencoder through the normalized phase resolution pulse sequence data; and (3) inputting the polluted partial discharge signals of the to-be-identified high-voltage equipment into the trained deep sparse denoising autoencoder, and obtaining the defect type of the high-voltage equipment based on output of an output classification layer of the deep sparse denoising autoencoder. In addition, the invention further discloses an insulation fault diagnosis system for the high-voltage equipment. The system comprises a data preprocessing module and a fault identification module.

A denoising autoencoder training method, an electrocardiography signal denoising method, and related apparatuses and devices, which use artificial intelligence to perform electrocardiography signal denoising, and may be applied to fields such as smart electrocardiogram detection. A reference electrocardiography signal to be denoised comprising significant features such as R peak positions and R-R distances of an electrocardiography signal to be denoised is extracted from said electrocardiography signal, and denoising is performed on a remaining electrocardiography signal to be denoised by means of a target denoising autoencoder after said reference electrocardiography signal has been removed from said electrocardiography signal, to prevent the target denoising autoencoder from performing denoising processing on the significant features in said electrocardiography signal, so that said reference electrocardiography signal and the denoised remaining electrocardiography signal are superimposed to obtain a denoised electrocardiography signal, wherein the R peak positions in said electrocardiography signal are better maintained, and distortion of the denoised electrocardiography signal is reduced.

A rolling early warning method and system of dynamic security risk situation for large scale hybrid alternating current / direct current grids, which includes: constructing the original feature set of the fast total transfer capability (TTC) estimation model; generating the training sample set of the fast TTC estimation model based on the network topology and the forecast information of a period of time in the future; constructing the fast TTC estimation model based on stacking denoising autoencoders and extreme learning machine; generating the future operating conditions (OCs), and determine the type of preventive control actions needed to ensure the system security based on the fast TTC estimation model and the heuristic search algorithm; and conducting the layered and hierarchical early warning for OCs according to the type of operating state and the type of preventive control actions that is needed.

The invention discloses a wireless transmission method for high-dimensional damaged data based on a noise-reduction self-encoder. The method of the invention includes model training and end-to-end transmission. Model training firstly performs data preprocessing on the historical perception data set, and divides it based on the K-fold cross-validation method; then builds a noise reduction autoencoder model, and trains based on the proposed new noise adding method of introducing random Gaussian noise in batches Denoising autoencoder models. In end-to-end transmission, firstly, the trained noise reduction autoencoder is split into two parts and deployed at the sending end and the receiving end, and then preprocessing and dimensionality reduction operations are performed on the perception data of unknown types of noise interference at the sending end, and the dimensionality reduction The data is transmitted to the receiving end, and finally the reconstruction operation is performed on the receiving end to obtain the reconstructed data of the undamaged sensing data. The method of the invention can effectively perform dimension reduction transmission, noise reduction processing and reconstruction of high-dimensional damaged perception data, and can filter out noise interference and dimension reduction transmission when collecting data in harsh environments.

The invention discloses an image recognition method based on human vision. This method constructs an image recognition structure capable of cross-problem recognition based on deep learning and human vision. Using this model structure, one model can recognize images in multiple problem domains, which is a further simulation of the human visual system. This method uses the nature of human visual perception, that is, the HMAX method, to directly extract the features of the original image, which reduces the complicated preprocessing steps and improves the calculation efficiency and feasibility of the method. The SDA method reduces the number of parameters in deep learning and improves the generality of the algorithm. To improve the training performance of general forward BP. From the actual experimental results, the classification accuracy of this method is also higher than other classification methods. Therefore, this method is an efficient and feasible image recognition method, and has universal applicability in the field of image recognition.

The invention relates to a method for establishing a SCMA codec model based on a noise reduction autoencoder, comprising: establishing an SCMA encoder based on a noise reduction autoencoder and a fully connected neural network, and mapping the user's original input data into a codeword ; Superimpose and transmit the codewords of all users on each resource block, and then superimpose the channel noise on the signal on each resource block; establish a SCMA decoder based on a fully connected neural network at the receiving end to decode the original input of all users Data; train the denoising autoencoder-based SCMA codec model; test the BER performance of the above denoising autoencoder-based SCMA codec model. Compared with the traditional SCMA system, the present invention reduces the complexity of encoding and decoding; compared with the existing SCMA system model based on deep learning, the present invention further reduces the bit error rate; compared with the existing deep learning-based SCMA system model, the invention has faster training convergence speed.

The feature that the existing voice lie detection algorithm often adopts is a combination feature, and the feature redundancy is relatively large. In view of this problem, the present invention discloses a stacked denoising self-encoder and deep neural network (SDAE-Deep Neural Network) for voice lie detection. DNN) structure. It consists of a two-layer encoding and decoding network followed by a DNN network. The structure first uses a two-layer denoising autoencoder structure to reduce feature redundancy. In order to prevent overfitting, dropout is added to each network layer of the stacked denoising autoencoder, and then a layer of DNN network is used for further learning. Features, and finally use the softmax classifier to fine-tune the network to obtain more representative features, thereby improving the network's ability to identify lies.

A gamma-ray spectrum classification apparatus, comprising circuitry configured: to provide a denoising autoencoder to receive gamma-ray spectrum data representing a gamma-ray spectrum of a material to be classified and to determine feature data indicative of one or more features representative of the gamma-ray spectrum data; and to provide a classification neural network to receive the feature data and to classify the material to be classified as one of a plurality of predetermined classifications using the feature data.

The invention discloses a chip surface defect detection method based on a convolutional de-noising auto-encoder. The method comprises the following steps: 1, constructing defect-free image reconstruction based on the convolutional de-noising auto-encoder; 2, constructing a residual image based on the overlapped area; and 3, performing defect detection based on the residual image. According to themethod, weak defects with the characteristics of low defect and background contrast ratio, small defect and the like can be effectively detected, the robustness is very high, and the chip surface detection precision is improved.

The invention discloses an LSTM fiber optic gyroscope temperature compensation modeling method based on deep embedded clustering, which includes: collecting temperature and fiber optic gyroscope zero bias data to construct a training data set, and training a denoising autoencoder layer by layer; Noisy autoencoder, constructing a deep autoencoder; based on a deep autoencoder, mapping the input x to obtain an embedded point z; calculating the soft allocation between the embedded point z and the cluster center, and constructing an auxiliary target allocation; using soft allocation and auxiliary target allocation The kl divergence is the objective function, iterates between calculating the auxiliary objective function and minimizing the kl divergence, updating the depth autoencoder parameters and cluster centers; segmenting according to the clustering results, using LSTM neural network on each segment The temperature compensation model of the fiber optic gyroscope is obtained through training. The invention can realize the temperature compensation of the gyroscope output zero bias error, obtain good fitting and prediction effects and high temperature environment adaptability, and improve the product precision of the optical fiber gyroscope.