49 results about "Denoising autoencoder" patented technology

The invention discloses a single-cell data deep clustering method and device based on double auto-encoders and a medium, and relates to the technical field of single-cell RNA sequencing. The method comprises the following steps: acquiring original gene data of a single cell; an scDAEC model is constructed; the scDAEC model comprises a dynamic combination multi-head attention mechanism, a denoising auto-encoder and a graph auto-encoder; the de-noising auto-encoder is constructed based on zero-expansion negative binomial distribution loss adjusted by a zero-expansion proportion adaptive weight; the graph self-encoder is constructed based on a graph convolutional network; and inputting the original gene data into the scDAEC model for clustering processing to obtain a single cell data clustering result. The framework constructed by the method disclosed by the invention has very strong generalization ability and high fault tolerance.

The application discloses an online monitoring method of a voltage transformer based on independent component analysis, which samples information of historical data, steady-state data and real-time data of signals output by the voltage transformer, constructs a data set from the historical data, and imports the data set into an initial SDAE network model based on a sparse denoising autoencoder to perform dynamic training, uses the steady-state data to fine-tune parameters based on the SDAE network model obtained through offline training, and thus obtains encoding data of the SDAE network model; uses an independent component analysis method to perform independent component decomposition by taking the encoding data as input; calculates sample statistics and overall statistical threshold, compares real-time statistics with the overall statistical threshold, and if the real-time statistics are lower than the overall statistical threshold, it is determined that the state of the voltage transformer is normal in this round of judgment; otherwise, it is considered that there is an abnormal voltage transformer in the voltage transformer group.

The application discloses an intelligent online laboratory process data anomaly detection method, in particular to an intelligent online laboratory process anomaly detection method based on a particle swarm optimization algorithm and a deep sparse denoising autoencoder, which comprises the following steps: S101, collecting running process data based on an intelligent online laboratory equipment and building a sample database; S102, performing Min-Max normalization preprocessing operation on the data; S103, building an anomaly detection model based on a deep sparse denoising autoencoder (DSDAE), so that the model can learn the data characteristics of normal data after training; S104, optimizing the number of hidden layer neurons of the DSDAE model by using a particle swarm (PSO) algorithm, and establishing an anomaly detection model based on PSO-DSDAE; S105, determining a model reconstruction residual tolerance threshold by using a cumulative sum algorithm, and establishing an anomaly judgment mechanism; S106, training the anomaly detection model based on PSO-DSDAE by using dataset data, and realizing anomaly data detection by comparing whether the reconstruction residual between the model output and the input exceeds the threshold.

The application provides a natural gas pipeline anomaly detection method and system based on a hybrid deep neural network, wherein the method comprises normalizing feature data of a natural gas pipeline to obtain input feature values; constructing a stacked sparse denoising autoencoder deep neural network model as a first hybrid deep neural network according to the input feature values; constructing a cost function according to the input feature values, performing unsupervised feature learning of the first hybrid deep neural network by using the cost function, and obtaining a second hybrid deep neural network; adding a supervised classifier to the second hybrid deep neural network to obtain a third hybrid deep neural network; inputting the input feature values into the third hybrid deep neural network to obtain output feature values; calculating a maximum probability value of the output feature values by using the supervised classifier; and reducing a difference between the maximum probability value of the output feature values and a label. The application can improve anomaly detection accuracy and reduce the false positive rate of anomaly detection when an intrusion attack or an anomaly occurs.

The application discloses a CNN denoising method based on distance-Doppler information, comprising the following steps: acquiring continuous multiple frames of radar distance-Doppler images, and constructing a convolutional neural network model; introducing an effective loss solving function to the convolutional neural network model for training; inputting a current frame image, a previous frame image of the current frame and a previous two frame image of the current frame into the trained convolutional neural network model to obtain a denoised image of the current frame image, wherein the convolutional neural network model comprises a self-adaptive moving encoder module combined with a convolution block attention module, effectively avoiding the problems of redundant parameters and serious information loss caused by forward convolution and maximum pooling of a convolutional autoencoder and other convolution-based denoising autoencoders. Meanwhile, the feature map is adaptively encoded, and those feature maps containing target information are selectively emphasized, and those feature maps containing interference information are as much as possible ignored.

The embodiment of the present application provides a kind of training method for trajectory observation data denoising model of motor target, the model includes preprocessing unit and denoising autoencoder, the method comprises: obtaining training set, each sample in training set includes input data and label, input data is the multiple frames of observation data obtained by sensor in predetermined time interval observing target movement track, label is the real movement track of target in corresponding time;Model is trained using training set, and the parameters of model are updated based on the loss calculated, wherein, preprocessing unit carries out denoising preprocessing to input data, and denoising autoencoder carries out coding and decoding to input data after denoising preprocessing, and the movement track after denoising is obtained based on the output of coding and decoding of denoising autoencoder Model, the model capable of effectively denoising observation data is obtained by the training method of the present application, to improve the accuracy of model to motor target movement track estimation.

The application provides a power system inertia analysis method, device, equipment and storage medium, wherein the method comprises: estimating the equivalent inertia of the power system using a physical model; correcting the equivalent inertia of the power system using a pre-trained data model to obtain updated equivalent inertia, wherein the data model adopts a denoising autoencoder, training labels are generated by including fine models of various types of inertia sources in the training process, and the update speed is adjusted by a momentum factor when updating the gradient; input the updated equivalent inertia into a simulation platform, add perturbations to each inertia node in the simulation platform to obtain the frequency of each inertia node, and calculate the distribution of each inertia node based on the updated equivalent inertia, the frequency of each inertia node and the center frequency of the inertia of the power system. The application proposes a hybrid intelligent driving scheme guided by a physical model and corrected by a data model, which can balance the calculation efficiency and prediction accuracy as a whole.

The invention relates to a training method of an unsupervised language translation model and a language translation method and device. The implementation scheme is as follows: performing grammatical analysis and coding processing on unsupervised training corpora to obtain a grammatical feature vector; respectively inputting the training corpora into the corresponding monolingual word embedding layer and the shared word embedding layer to obtain monolingual word embedding and shared word embedding, and generating semantic vectors according to the monolingual word embedding and the shared word embedding; the shared word embedding layer is constructed based on a mixed corpus containing a source language and a target language; performing fusion processing on the grammar feature vector and the semantic vector to obtain a fusion vector of the training corpus; training an unsupervised pre-training model by using the fusion vector; and initializing an encoder of the language translation model by adopting a weight parameter of the trained unsupervised pre-training model, and performing unsupervised training on the language translation model by alternately executing a de-noising automatic encoder and a reverse translation task. According to the invention, the accuracy, fluency and robustness of translation can be improved.

The invention discloses a turbine fault diagnosis method and related device, and the method comprises the steps: collecting an original vibration signal of a turbine, carrying out the coding processing of the original vibration signal of the turbine through a denoising self-encoder, and achieving the signal dimension reduction compression through the hidden layer dimension constraint, and obtaining a coding signal; inputting the coded signals into a convolutional neural network, extracting local features through a convolutional layer, and combining the local features into a feature map with a higher depth dimension after the local features are screened by a pooling layer; and inputting the feature map with a higher depth dimension into a full connection layer, finely adjusting network parameters through a back propagation algorithm in a supervised learning mode, and outputting a classification result of turbine fault types and positions. According to the method, huge operation parameters of the turboset are trained through deep learning, and automatic output of fault feature type positions and intelligent diagnosis of operation health conditions are completed without depending on an existing expert diagnosis system.

A computing system may include a processor and a memory having a set of instructions, which when executed by the processor, cause the computing system to execute actions. The actions include identifying an estimate of a distribution of missing block patterns, generating a noisy dataset by removing first data from an original dataset based on the estimate and training a denoising autoencoders (DAE) based on the noisy dataset.

The application discloses a multi-source data processing method and system of a cigarette detection instrument, relates to the technical field of tobacco, and realizes non-discriminatory access to heterogeneous data sources through a unified interface protocol and automatic connection, avoids the huge workload of customized development and the problem of system rigidity, provides stable real-time data flow for subsequent processing, uniformly converts unstructured original data into standard structured records by using a formatted description template and automatic analysis technology, automatically and efficiently associates multi-instrument data of the same detection event by using a similarity matching algorithm based on local sensitive hashing, applies a weighting aggregation algorithm based on an attention mechanism to give differentiated weights to time series data and fusion, generates a fusion feature matrix that more comprehensively represents the quality state of a sample, performs automatic feature extraction and dimension reduction by using a stacked denoising autoencoder, and finally outputs results in the form of a standardized data packet with complete check information, thereby providing a ready-to-use unified high-quality data basis for upper-layer quality analysis and the like applications.

The application provides a state evaluation method and device of an intelligent high-voltage switch, a terminal and a storage medium. The method comprises the following steps: obtaining a sensing data set of the intelligent high-voltage switch; training a stacked denoising autoencoder (SDAE) and a deep belief network (DBN) respectively according to the sensing data set to obtain the weight of the SDAE and the weight of the DBN; fusing the weight of the SDAE and the weight of the DBN by using a weighted fusion method to obtain a fusion weight, initializing the weight of a target evaluation model according to the fusion weight, and testing the feasibility of the target evaluation model; when the feasibility of the target evaluation model meets a preset condition, obtaining operation data of the intelligent high-voltage switch, inputting the operation data into the target evaluation model, and obtaining the operation state of the operation data. The application can accurately and efficiently evaluate the state of the high-voltage switch.

The application provides a missing genotype filling method based on an automatic encoder sample matching, realizes low-cost and accurate filling of missing genotypes, and can provide more accurate genetic data support for various genetic analysis work. In the application, the genotype information value of each sample at each position in the target data file is converted and finally encoded into a one-hot encoding, a training set and a test set are divided, and then a convolutional denoising autoencoder model is constructed. The application uses an automatic preprocessing strategy to segment the sample data set participating in filling, reduces the device memory occupation, so that the user can successfully perform high-precision genotype filling using a low-cost device. The application has high filling precision, simple and reliable model structure, high training efficiency, and has a wide application prospect in the field of genetic sequence analysis, and can be used for subsequent biological whole genome association analysis and whole genome selection work.

The embodiment of the application relates to a kind of processing method and device for optimizing electron micrograph quality, the method comprises: one complete model pre-training DDPM model is recorded as first pre-training model;And by the way of fusing multiple LoRA adapters on first pre-training model, obtain first fusion model;Collect multiple spherical aberration electron micrograph to form corresponding first training atlas;The parameter set L of first fusion model is fine-tuned based on first training atlas;After fine-tuning, corresponding electron micrograph quality optimization processing is carried out to any ordinary transmission electron micrograph based on the second denoising autoencoder of first fusion model, and corresponding optimized electron micrograph is obtained.Through the application, the image quality of ordinary transmission electron micrograph can be optimized to reach the quality level of spherical aberration electron micrograph, and the model fine-tuning efficiency can be improved, and the model use cost is reduced.

The invention provides a three-dimensional metallogenic prediction method, medium, equipment and product, and relates to the technical field of metallogenic prediction.The method comprises the steps that geophysical, geochemical and remote sensing data are collected and preprocessed in a metallogenic area, and a three-dimensional multi-source attribute data set is constructed and divided into a training set, a verification set and a test set; preliminarily training the denoising auto-encoder by using the training set added with the noise; a discriminator is added behind the preliminarily trained encoder, and the preliminarily trained de-noising auto-encoder and the discriminator are trained; verifying and testing the trained encoder, decoder and discriminator by using the verification set and the test set; and adding a full-connection classification layer behind the trained encoder, freezing parameters of the trained encoder, training the full-connection classification layer by using the marked metallogenic sample, and inputting noise-containing data of a region to be predicted into the finally trained encoder and the full-connection classification layer to obtain a metallogenic prediction result. The method is high in anti-noise capability.

This invention relates to a gyroscope error compensation method based on a stacked denoising autoencoder. The method is as follows: Multiple gyroscope output data are collected for each rotation axis of the gyroscope. The output data from the three rotation axes are then concatenated, normalized, and degraded to obtain the input vector of a trained network model. The network model has M encoding layers and a fully connected layer. The output vector of the Mth encoding layer serves as the input vector of the fully connected layer. The output vector of the fully connected layer is then inversely normalized to obtain the compensated output vector for the three rotation axes of the gyroscope. This invention can compensate for both random and deterministic errors simultaneously, and it reduces reliance on human experience, enabling end-to-end learning and saving time and effort.

This invention provides a method and system for comprehensive performance evaluation of hot-rolled strip steel based on spatiotemporal topological representation, belonging to the field of intelligent monitoring of industrial processes. The method first acquires historical production time-series data and corresponding strip steel performance grade labels, and preprocesses and segments them to obtain local time sub-series; it constructs a twin denoising autoencoder network to extract slow feature representations, and then constructs a working condition state topology graph structure; it constructs a comprehensive performance evaluation model for hot-rolled strip steel based on graph convolution; it inputs the working condition state topology graph structure into the graph convolutional layer, and under the supervision of the strip steel performance grade labels, learns discriminative spatiotemporal features that integrate state dependencies along the time sequence dimension, then inputs them into a fully connected layer and a classifier to output the hot-rolled strip steel performance grade; it then evaluates the output results based on a loss function and optimizes the model parameters; finally, it performs a comprehensive evaluation of the hot-rolled strip steel performance based on the trained model. This invention improves the real-time performance and accuracy of comprehensive performance evaluation of hot-rolled strip steel.

The invention discloses an offshore wind turbine blade pitch control method and device, computer equipment and a medium, and the method comprises the steps: receiving preprocessed multi-source heterogeneous monitoring data transmitted by an offshore wind turbine end, and obtaining the multi-source heterogeneous monitoring data through a high-frequency multi-dimensional monitoring network fusing a mechanical state and marine environment characteristics; a physical mechanism and data compensation are fused, and a digital twinborn model is constructed; continuously purifying multi-source heterogeneous monitoring data through a potential denoising auto-encoder model, and calibrating a digital twin model; a multi-dimensional health assessment index system is established in combination with a digital twin model, and health assessment index calculation of the health state of the key component of the offshore wind turbine is achieved; according to a real-time health assessment result, calculating a safe operation boundary of the offshore wind turbine in the current health state; and the dynamic safe operation boundary and the real-time health state are converted into a variable-pitch instruction for each blade, and variable-pitch regulation and control of the blades are achieved. According to the invention, the safety and power generation economy of the offshore wind turbine can be improved.

A system and method include 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.

This application discloses a hyperspectral anomaly detection method based on a stacked denoising autoencoder and collaborative representation using a spectral loss function. Addressing the issue of large datasets and abundant redundant information in hyperspectral images, this method extracts features from the hyperspectral images without compromising detection accuracy. The invention improves and optimizes the autoencoder's network structure and loss function to enhance its feature extraction capability. It also reduces redundant computation by selecting the results of the intermediate hidden layers of the autoencoder network as output. Finally, a collaborative representation algorithm is used to obtain the final anomaly detection result. Compared with several representative algorithms, this method exhibits better detection performance.

The invention provides a thermal coherence scattering real-time spectrum unfolding method based on a stacked noise reduction auto-encoder, and relates to the technical field of magnetic confinement fusion plasma microwave diagnosis, and the method comprises the steps: 1, randomly dividing a thermal coherence scattering signal into a training set, a verification set and a test set, and training a stacked noise reduction auto-encoder 1 according to the training set, the verification set and the test set; 2, calculating a Spearman correlation coefficient between each output prediction error of the stacked noise reduction auto-encoder 1 and a main ion temperature prediction error, removing an output target with relatively low correlation with the main ion temperature, and training a stacked noise reduction auto-encoder 2 on the basis of the Spearman correlation coefficient; and step 3, randomly disrupting the arrangement of an input feature, calculating the influence of the process on the performance of the stacked noise reduction auto-encoder 2 until all features are traversed, removing the input feature which has small influence on the performance of the stacked noise reduction auto-encoder 2, and further training the stacked noise reduction auto-encoder 3 on the basis. The invention provides a real-time spectrum unfolding method for a thermal coherence scattering system for measuring the main ion temperature in magnetic confinement fusion.

This invention discloses a method and system for foreign object impact warning of power lines based on a spatiotemporal denoising autoencoder. The method includes: acquiring raw time-series signals from multiple sources and inputting them into a spatiotemporal denoising autoencoder; processing the raw time-series signals and reconstructing a reconstructed waveform of the line health state using a spatiotemporal mask; calculating the reconstruction residual between the raw time-series signals and the reconstructed waveform of the line health state, extracting current environmental background noise features and historical residual features, and inputting them together into a spatiotemporal risk tracker; using a multi-head attention mechanism and a convolutional gated recurrent unit to perform spatiotemporal dependency modeling on the input, and outputting multidimensional spatiotemporal features representing the current line operating state; calculating an adaptive dynamic judgment threshold based on the multidimensional spatiotemporal features to obtain a dynamic judgment threshold under the current operating condition; and performing a foreign object impact warning by comparing the offset between the dynamic judgment threshold and the reconstruction residual. This invention can significantly reduce the false alarm rate of warnings.

The application provides a kind of concrete-filled steel tube void knocking acoustic detection method suitable for low signal-to-noise ratio environment, belongs to the field of bridge nondestructive testing, the method comprises: collecting actual engineering field noise signal, combined with simulated noise to produce various noise data sets, and collect clean knocking signal in a quiet environment, use clean signal to superimpose noise signal as training set, and train to obtain stacked denoising autoencoder model;Knocking and collecting concrete-filled steel tube surface audio signal in actual engineering environment, use the trained stacked denoising autoencoder model to denoise the signal to be identified;Use CNN-LSTM model to extract the time-frequency characteristics of the denoised signal, and complete the identification of void defects. This method overcomes the shortcoming that the precision of traditional knocking acoustic detection is greatly affected by environmental noise, and is suitable for concrete-filled steel tube void detection in complex and low signal-to-noise ratio environment, with high detection precision.

This invention discloses a method and system for foreign object impact warning of power lines based on a spatiotemporal denoising autoencoder. The method includes: acquiring raw time-series signals from multiple sources and inputting them into a spatiotemporal denoising autoencoder; processing the raw time-series signals and reconstructing a reconstructed waveform of the line health state using a spatiotemporal mask; calculating the reconstruction residual between the raw time-series signals and the reconstructed waveform of the line health state, extracting current environmental background noise features and historical residual features, and inputting them together into a spatiotemporal risk tracker; using a multi-head attention mechanism and a convolutional gated recurrent unit to perform spatiotemporal dependency modeling on the input, and outputting multidimensional spatiotemporal features representing the current line operating state; calculating an adaptive dynamic judgment threshold based on the multidimensional spatiotemporal features to obtain a dynamic judgment threshold under the current operating condition; and performing a foreign object impact warning by comparing the offset between the dynamic judgment threshold and the reconstruction residual. This invention can significantly reduce the false alarm rate of warnings.

The invention relates to a spectrum occupancy prediction method fusing an auto-encoder and bidirectional LSTM-Attention, and belongs to the field of intelligent processing of radio signals. The method comprises the following steps: preprocessing a time sequence data set based on a sparse auto-encoder; designing a spectrum occupancy rate prediction model based on BiLSTM-Attention (BiLSTM-Attention); and predicting the frequency spectrum occupancy rate based on the iLSTM-Attention. According to the method, for the problems of data redundancy and relatively high noise when the model processes high-dimensional data, the stacked noise reduction auto-encoder is used for preprocessing the input data, so that the generalization ability of the model is improved. And then, a bidirectional LSTM network is adopted, so that past and future information can be utilized, the model can learn more fully and effectively, and the accuracy of spectrum occupancy prediction is improved.

The application discloses a pear internal quality prediction method based on an autoencoder and multi-task learning, and belongs to the technical field of nondestructive detection of agricultural products. The method comprises the following steps: obtaining sample-level and pixel-level visible / near-infrared hyperspectral data sets of pears; constructing a denoising autoencoder, and migrating the encoder part of the pre-trained denoising autoencoder to sample-level average spectrum modeling; and constructing a DAE_multi multi-task network, which comprises the encoder part of the denoising autoencoder and a multi-task prediction module, the multi-task prediction module comprising a shared feature layer, a soluble solids content prediction head and a hardness prediction head, the shared feature layer receiving deep spectral features output by the encoder, and the soluble solids content prediction head and the hardness prediction head receiving outputs of the shared feature layer respectively to output a soluble solids content prediction value and a hardness prediction value. The application provides a feasible modeling idea for nondestructive detection of the soluble solids content and hardness of pears under small sample conditions.

The invention discloses an image clustering method and system based on a self-adaptive multi-manifold automatic encoder, and the method comprises the steps: training a denoising automatic encoder to reconstruct original image data, and learning the low-dimensional expression of the image data in a potential space; by minimizing the regularized multi-manifold divergence index, adaptively determining the manifold number of the image data in the high-dimensional space; respectively calculating reconstruction loss, clustering loss and multi-manifold topology consistency loss, and weighting to obtain overall loss; wherein the multi-manifold topological consistency loss is used for measuring the topological structure difference of each manifold in the original space and the potential space; the overall loss function of the network is optimized, pre-trained noise reduction auto-encoder parameters are adjusted, optimized network parameters and image data potential expressions are obtained, and finally the optimized image clustering effect is obtained. According to the method, the de-noising auto-encoder and multi-manifold learning are effectively combined, and a more excellent solution is provided for clustering of image data.

This invention primarily addresses the performance issues caused by the limited computing power of wireless terminal devices in industrial control systems, as well as the high latency problems of traditional authentication methods. It proposes a physical layer authentication scheme based on ShuffleNet and a self-attention mechanism. First, a method based on an improved variational autoencoder and generative adversarial networks is proposed to augment CSI data, and a denoising autoencoder based on particle swarm optimization is used to reduce the dimensionality of the data. Then, a lightweight SE module is introduced as the implementation of the self-attention mechanism, further enhancing the model's focus on key features and improving its discriminative power and feature learning ability in complex scenarios. Finally, the ShuffleNet-SE model is used to identify CSI data features, classifying legitimate and illegitimate devices in the industrial control system.

This invention discloses an intelligent operation and maintenance management method and system based on multimodal AI, belonging to the interdisciplinary field of data artificial intelligence and industrial intelligent operation and maintenance. The method includes collecting and preprocessing multimodal data, using an improved VMD method to perform modal component decomposition on vibration signals, calculating first-order and second-order energy moments to generate initial vibration feature vectors, and using a denoising autoencoder (SDAE) to obtain the final vibration feature vector. The multimodal data includes vibration signals, energy consumption, images, and text data. By introducing an improved VMD method combined with Lagrange multipliers and an adaptive step-size mechanism, a frequency domain iterative optimization solution process is constructed. Simultaneously, a denoising autoencoder is used to achieve nonlinear, deep-level feature enhancement, thereby effectively improving the system's ability and stability in extracting key modal components from complex vibration signals.

This application discloses a method for detecting magnetic anomalies in background noise based on a sparse denoising autoencoder, belonging to the field of intelligent signal learning and perception technology. The method includes: acquiring noisy magnetic field data; inputting the data into a sparse denoising autoencoder for sparse denoising processing to obtain denoised magnetic field data; establishing a loss function; training the sparse denoising autoencoder using the loss function; acquiring magnetic field data to be detected; inputting the magnetic field data to be detected into the trained sparse denoising autoencoder to obtain the corresponding real-time reconstruction error; comparing the real-time reconstruction error with a set reconstruction error threshold to determine whether the magnetic field data to be detected contains magnetic anomaly signals. Compared with existing machine learning methods, the method in this application does not require manual annotation of massive amounts of data; it only requires training using unlabeled ocean magnetic field noise data collected in the ocean.