36 results about "Restricted Boltzmann machine" patented technology

A method in an illustrative embodiment comprises configuring a machine learning system with a multi-layer network architecture comprising at least one neural network and one or more additional network layers, training the neural network at least in part utilizing quantum sampling performed by a quantum computing device, obtaining data characterizing a monitored system, processing at least a portion of the obtained data through at least a portion of the multi-layer network architecture of the machine learning system to generate a prediction of at least one characteristic of the monitored system from the obtained data, and executing at least one automated action relating to the monitored system based at least in part on the generated prediction. The neural network may comprise, for example, a deep belief network (DBN) that includes at least first and second restricted Boltzmann machines (RBMs) of respective first and second different types, or at least one conditional restricted Boltzmann machine (CRBM).

This invention discloses a bearing remaining life prediction method based on BN-RBM and DA-BiLSTM. First, raw vibration signals from the bearing during operation are collected and preprocessed. Then, a bearing remaining life prediction model is constructed. This model includes a BN-RBM feature extraction network, a DA-BiLSTM feature fusion network, and a linear regression layer. The BN-RBM feature extraction network dynamically updates the prior weight distribution of the Restricted Boltzmann Machine (RBM) using posterior probabilities constructed from Bayesian BN, and extracts features using the trained RBM. The DA-BiLSTM feature fusion network performs bidirectional feature fusion, and the linear regression layer performs linear activation, outputting the predicted bearing remaining life. This invention effectively improves the accuracy of bearing remaining life prediction by providing timely and effective life prediction before failure, reducing the risk of downtime caused by failures.

The invention provides a dynamic modeling simulation method and system for the energy efficiency ratio of a solar seawater desalination system, and belongs to the technical field of seawater desalination and system modelling. Membrane surface resistance, selective permeability and direct-current bus voltage ripple data in the photovoltaic electrodialysis process are firstly obtained; secondly, Fourier transform is carried out on ripple data, then the ripple data and membrane parameters are spliced to generate an input matrix, the matrix is imported into a deep belief network, and a restricted Boltzmann machine is used for extracting an unsteady-state ion impedance vector reflecting the influence of voltage fluctuation; a nonlinear regression model of the vector and unit water production energy consumption is established through a least square support vector machine; finally, the water production rate and the energy efficiency ratio are calculated according to the predicted energy consumption and the photovoltaic power, and a dynamic simulation curve is generated. According to the method, the nonlinear influence of the photovoltaic voltage ripples on the membrane impedance can be quantified through deep learning, and the precision of predicting the energy efficiency ratio of the seawater desalination system under the fluctuating power supply working condition is remarkably improved.

The invention discloses a method for optically realizing a restricted Boltzmann machine. According to the invention, analog calculation of the restricted Boltzmann machine and optical Gibbs sampling can be realized, and the system has the advantages of wide application range, fast information transmission, simple structure, low cost, fast calculation and the like. According to the invention, a coherent wide-spectrum light source is used as signal input, a light field is subjected to light splitting by using a light splitting device such as a grating and then enters a modulator, then spinning, interaction and magnetic field parameters are coded on light wavefront at different positions through a time or space light modulator, and optical Fourier transform is carried out by using a time or space lens system, so that the optical field is obtained. The light intensity after Fourier transform is measured through the detector, and finally the difference between the light intensity measured twice is calculated, so that optical Gibbs sampling is realized, the calculation complexity is reduced, and the calculation efficiency is improved. The method provided by the invention has important application prospects in the fields of optical neural networks and the like, can realize applications of content generation, classification and identification and the like, and is convenient to integrate in optical chips and the like.

The invention provides an audio-visual multi-mode speech recognition method, a model training method and electronic equipment. The voice recognition method comprises the following steps: acquiring first video data and first audio data; a target model for audio-visual multi-modal speech recognition is obtained, the target model comprising a first target network, in which the first target network introduces a multi-scale packet sparse constraint comprising a plurality of norm constraints and a plurality of packets on the basis of a deep belief network (DBN), and the multi-scale packet sparse constraint comprises a plurality of norm constraints and a plurality of packets; the first target network divides hidden layer units of a restricted Boltzmann machine (RBM) into non-overlapping groups and performs feature extraction of the first voice by using a non-overlapping group lasso method; and inputting the first video data and the first audio data into a target model to obtain a voice recognition result of the first voice output by the target model. By introducing a multi-scale grouping sparse constraint and multi-head attention fusion mechanism, the accuracy and robustness of speech recognition in a complex scene are improved.

The invention discloses a coal mine catastrophe scene deduction method, which belongs to mine disaster emergency rescue, and comprises the following steps: collecting environmental data of a coal mine stope; performing data screening according to the environment data and the key elements, and extracting time sequence features; deducing time sequence characteristics by using an energy-based dynamic time sequence model to obtain probability energy distribution of disaster development; according to the probability energy distribution, constructing a Boltzmann machine model with a limited time sequence for predicting a mine catastrophe development trend; wherein the explicit layer unit corresponds to a sensor observation value at a current time point, the hidden layer unit captures potential time sequence characteristics and modes, correlation between a hidden layer at the current moment and hidden layers at the previous N moments is established through time sequence connection, and model training is conducted through a time-dependent conditional contrast divergence algorithm; aiming at the problem that a traditional Boltzmann machine model is mainly used for processing static distribution, so that the processing precision of a coal mine catastrophe scene with obvious time sequence evolution characteristics is low, the coal mine catastrophe scene deduction precision is improved.

The invention provides a flower quality grading quality inspection method and system based on AI and image processing, and belongs to the technical field of computer vision and pattern recognizing.The method comprises the steps that an original image of the surface of a flower is collected through high-resolution imaging equipment, and high-frequency sub-band data representing tiny physical characteristics are extracted through multistage discrete wavelet transform; meanwhile, a sliding window is adopted to traverse the image, the color information entropy of a local area is calculated, and a color entropy graph is constructed. And after vectoring and splicing the two types of features, inputting the two types of features into a deep belief network model based on a multilayer restricted Boltzmann machine, and extracting deep abnormal feature codes. And finally, performing linear discriminant analysis on the code by utilizing a classification projection vector based on inter-class and intra-class distance optimization to generate an insect attack infection index, and realizing automatic grading quality inspection of the flower quality according to the insect attack infection index. According to the method, precise recognition and quantitative grading of tiny insect pests and recessive lesions on the surfaces of the fresh flowers are realized.

Embodiments of the present disclosure relate to a question answering method, an electronic device, and a computer program product. The method includes: determining an answer library associated with a question; determining a restricted Boltzmann machine associated with the answer library, wherein the restricted Boltzmann machine is configured to determine a set of questions that the answer library can answer and association relationships between questions in the set of questions and the answer library; and determining, using the restricted Boltzmann machine, annotation information associated with the question and targeted to the answer library. With the technical solution of the present disclosure, it is possible to determine, while determining an answer library associated with the question, annotation information that is targeted to the answer library, and to enable customer service personnel to have a more thorough understanding of the determined answer library and obtain targeted recommendation information.

The invention discloses a bearing residual life prediction method based on BN-RBM and DA-BiLSTM, and the method comprises the steps: firstly collecting an original vibration signal in a bearing operation process, carrying out the data preprocessing, and then constructing a bearing residual life prediction model which comprises a BN-RBM feature extraction network, a DA-BiLSTM feature fusion network and a linear regression layer. The BN-RBM feature extraction network dynamically updates the weight prior distribution of a restricted Boltzmann machine RBM by using the posterior probability constructed by Bayesian BN, and performs feature extraction by using the trained RBM, the DA-BiLSTM feature fusion network performs bidirectional feature fusion, the linear regression layer performs linear activation, and the predicted value of the residual life of the bearing is output. The method effectively improves the prediction accuracy of the residual life of the bearing, and reduces the shutdown risk caused by the fault through timely and effective life prediction before the fault.

This invention discloses a method and system for evaluating the intelligence level of an agent based on a Restricted Boltzmann Machine (RBM), relating to the field of artificial intelligence evaluation technology. First, it acquires replay data from pairwise battles between multiple tested agents in an experimental task. Then, for each tested agent, multiple evaluation indicators are calculated based on its battle data from each replay. Using these indicators as input, a trained RBM network predicts the ELO score of the tested agent. The final ELO score is then obtained by averaging the ELO scores from multiple replays. Finally, for each tested agent, its intelligence level is determined based on its final ELO score. This eliminates the need for subjective level classification standards, achieving an objective and universal intelligence level evaluation.

The invention discloses a power system oscillation mechanism identification method and a model construction method, and relates to the field of power system low-frequency oscillation, and the method comprises the steps: collecting the active power data of a generator and the active power data of a tie line; performing linear normalization and principal component analysis dimension reduction processing on the data, and constructing a multi-dimensional time sequence to-be-tested sample; a sample is input into a pre-trained deep belief network (DBN) model, the model is formed by stacking a plurality of restricted Boltzmann machines (RBM), data deep features are extracted through unsupervised layer-by-layer greedy pre-training, and parameter optimization is performed in combination with supervised fine tuning. According to the method, negative damping oscillation and forced power oscillation can be automatically identified, the uncertainty of manual feature selection is avoided, the influence of noise and amplitude difference on the identification precision is effectively suppressed, and the adaptability of the model to an actual power grid complex disturbance condition is improved.

The application discloses a gear vibration noise prediction method based on a deep belief network, which comprises the following steps: firstly, denoising the gear vibration noise data collected through experiments; secondly, constructing a database based on a frequency analysis algorithm; thirdly, constructing a deep belief network model based on a restricted Boltzmann machine and training the model; fourthly, obtaining optimal parameters through a particle swarm optimization algorithm based on adaptive inertia weight and a learning factor; and finally, using the denoised and amplified gear vibration data to predict the noise by using the APSO-DBN method. The method amplifies the data through frequency analysis, overcomes the training difficulty caused by insufficient data, and has better iteration effect than the PSO-DBN algorithm. In the total sound pressure level prediction, the average relative error of the APSO-DBN algorithm is obviously reduced compared with the DBN network with manually selected network nodes.

The invention discloses a method and a system for intelligently identifying damage of a hydrogen conveying pipeline based on multi-sensor feature fusion. The method comprises the following steps: firstly, acquiring multi-source sensor detection signals including electromagnetic and ultrasonic; performing time domain, frequency domain and time-frequency domain feature extraction on each sensor signal, and constructing an original feature set; then, a restricted Boltzmann machine is adopted to carry out unsupervised deep feature learning and dimension reduction processing on the high-dimensional features, feature-level fusion of multi-sensor data is achieved, and a fusion feature vector with higher characterization capacity is obtained; and finally, inputting the fusion feature vector into a pre-trained BP neural network classifier to realize intelligent identification and classification of different damage types and damage degrees of the hydrogen delivery pipeline. The method has the advantages of being high in detection efficiency and accurate in recognition result, and the accuracy and reliability of hydrogen conveying pipeline damage recognition are effectively improved.

The invention discloses an environmental air quality monitoring method based on artificial intelligence. The method comprises the following steps: S1, collecting and processing air pollutant concentration data, meteorological parameter data and spatial geographic information data; s2, constructing an improved deep belief network model, wherein the improved deep belief network model comprises an input layer, a multi-layer restricted Boltzmann machine hidden layer, a multi-layer physical prior embedding layer, a space-time dependent modeling module and an output layer; s3, performing layer-by-layer unsupervised pre-training operation on the improved deep belief network model; s4, performing global fine tuning optimization on the improved deep belief network model; s5, performing adaptive dynamic adjustment on the weight of the physical consistency error term; s6, inputting monitoring data collected in real time into the improved deep belief network model after global fine tuning optimization, and executing forward reasoning operation; and S7, executing abnormal pollution event detection. According to the invention, the improved deep belief network model is adopted, and accurate prediction and intelligent early warning of the ambient air quality are realized.

Embodiments of the present disclosure relate to a question and answer method, an electronic device and a computer program product. The method comprises: determining an answer base associated with a question; determining a restricted Boltzmann machine associated with the answer base, the restricted Boltzmann machine being used to determine a question set that the answer base can answer and a relationship between a question in the question set and the answer base; and determining, using the restricted Boltzmann machine, description information associated with the question for the answer base. Using the technical solution of the present disclosure, the description information for the answer base can be determined at the same time as the answer base associated with the question is determined, and the customer service personnel can be made to have a more thorough understanding of the determined answer base and obtain targeted recommendation information by providing the determined description information to the customer service personnel, thereby improving the user experience of the customer service personnel using the question and answer system.

A first computing node of a system can receive sensor data about a physical environment. The first computing node can analyze the sensor data with a restricted Boltzmann machine (RBM) neural network to determine whether there is a fault condition in the physical environment, an identification of the fault condition being omitted from data used to train the RBM neural network. The first computing node can update the RBM neural network based on the sensor data to produce a first updated RBM neural network. The first computing node can send a first patch indicative of the first updated RBM neural network to a central server. The first computing node can receive, from the central server, information indicative of a second updated RBM neural network, the second updated RBM neural network being based on an aggregation of the first patch and of a second patch generated by a second computing node.

The invention discloses a high-precision position keeping method and device for cooperative control of a multi-dimensional motor and a storage medium. The high-precision position keeping method and device are used for keeping the high-precision position of the multi-dimensional motor. The method comprises the following steps: constructing a multi-dimensional data set; performing standardization processing on the multi-dimensional data set to obtain standard data; calculating a covariance matrix based on the standard data and performing eigenvalue decomposition; determining a plurality of principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; projecting the standard data into the principal component space to form a dimensionality reduction data set; inputting the dimension reduction data set into a pre-trained deep belief network, and outputting a comprehensive feature vector through hierarchical features of a restricted Boltzmann machine; inputting the comprehensive feature vector into a depth deterministic strategy gradient algorithm model, and constructing a reward function; outputting a real-time adjustment strategy based on the reward function, and obtaining an optimization control parameter; and realizing high-precision position keeping of the plurality of motors according to the optimized control parameters.

The invention discloses a distributed resource access-oriented power distribution network operation risk level assessment method, which comprises the following steps of: acquiring multi-source operation data of a power distribution network, and performing dynamic aggregation to generate an aggregation feature matrix; inputting the aggregation feature matrix into a double-target sparse auto-encoder to extract risk sensitive features, and outputting a low-dimensional dense feature vector after feature selection; clustering analysis is carried out on the feature vector data to construct an initial weight matrix of a first-layer restricted Boltzmann machine of the deep belief network, and then the feature vectors are input into a pre-trained deep belief network to obtain probabilities belonging to all risk levels; and comparing the probability of each risk level with a corresponding adaptive threshold value, excluding the risk level with the probability less than the adaptive threshold value, and if the risk level with the probability greater than or equal to the corresponding adaptive threshold value exists, determining that the risk level with the highest probability is an evaluation result. The method can effectively and accurately evaluate the operation risk of the power distribution network.

The application discloses a high-precision position keeping method and device for multi-dimensional motor cooperative control and a storage medium, and is used for keeping high-precision position of multi-dimensional motor. The application comprises the following steps: constructing a multi-dimensional data set; performing standardization processing on the multi-dimensional data set to obtain standard data; calculating a covariance matrix based on the standard data and performing eigenvalue decomposition; determining a plurality of principal component feature vectors according to cumulative variance contribution rates to form a principal component space; projecting the standard data into the principal component space to form a reduced dimension data set; inputting the reduced dimension data set into a pre-trained deep belief network, outputting a comprehensive feature vector through hierarchical features of a restricted Boltzmann machine; inputting the comprehensive feature vector into a deep deterministic policy gradient algorithm model to construct a reward function; outputting a real-time adjustment strategy based on the reward function to obtain optimized control parameters; and realizing high-precision position keeping of the plurality of motors according to the optimized control parameters.

This application provides a dynamic modeling and simulation method and system for the energy efficiency ratio (EER) of a solar-powered seawater desalination system, belonging to the technical field of seawater desalination and system modeling. First, this application acquires data on membrane surface resistance, selective permeability, and DC bus voltage ripple during the photovoltaic electrodialysis process. Second, the ripple data is Fourier transformed and concatenated with membrane parameters to generate an input matrix. This matrix is ​​then imported into a deep belief network, and a restricted Boltzmann machine is used to extract the unsteady-state ion impedance vector reflecting the influence of voltage fluctuations. Subsequently, a nonlinear regression model of this vector and unit water production energy consumption is established using a least-squares support vector machine. Finally, the water production rate and EER are calculated based on the predicted energy consumption and photovoltaic power, generating a dynamic simulation curve. This application can quantify the nonlinear influence of photovoltaic voltage ripple on membrane impedance through deep learning, significantly improving the accuracy of EER prediction for seawater desalination systems under fluctuating power supply conditions.

The application discloses a self-supervised deep representation method based on contrast learning, comprising the following steps: S1, obtaining a data set V to be trained; S2, training N pieces of data in one batch, and performing data enhancement on each piece of data v; S3, feeding the data v and the enhanced data vs into a restricted Boltzmann machine for forward propagation, and obtaining corresponding data h and data hs in the hidden layer; S4, performing contrast learning to improve the similarity between positive samples; S5, performing back propagation to obtain reconstructed data v' of the data v in the visual layer; S6, minimizing the distance between the data v' and the data v; S7, repeating steps S2-S6 until the reconstruction loss function converges, and realizing the training of the restricted Boltzmann machine; and S8, using the trained restricted Boltzmann machine in S7 to perform deep representation learning on the data set V, and obtaining deep feature representation thereof. The application can overcome the purposeless learning defect of the traditional restricted Boltzmann machine, thereby improving the representation capability of the model.

The application discloses a kind of multi-modal characteristic fusion's intelligent target identification method, on the basis of constructing typical aerial target high-resolution range image, infrared image sample library, respectively using attention bidirectional gate recurrent unit model from high-resolution range image time series sample, using light convolutional neural network model from infrared image sample learning extraction aerial target multi-modal deep feature representation;Using Gaussian distribution limited Boltzmann machine realizes aerial target multi-modal feature abstract fusion, removes redundancy, forms more discernibility, representative multi-modal feature joint representation;Using kernel ultra-limit learning machine with strong generalization ability under small sample condition, fast training speed and high classification accuracy as classifier, realize aerial target type identification.The application uses heterogeneous deep learning model to abstract extraction and fusion aerial target feature information in high-resolution range image, infrared image for target identification, improve recognition precision.

The application discloses a signal processing method and system of an S-band narrowband digital processor of a phased array, and the method comprises the following steps: receiving and pre-processing service signals from a comprehensive processor, and converting the service signals into zero intermediate frequency signals; channelizing and decomposing the signals into a plurality of sub-channels with different center frequencies and bandwidths through a multi-phase filter group; performing deep-level feature extraction on the signals of the sub-channels by using a deep belief network which is stacked by a plurality of restricted Boltzmann machines; searching for optimal beam forming weights in a feature space by using a particle swarm optimization algorithm; finally, performing beam synthesis by weighting and summing the signals of the sub-channels according to the obtained optimal weights, and converting the result back into an analog intermediate frequency signal for subsequent transmission processing. The application can improve the processing performance and the accuracy of beam forming of a phased array system in S-band narrowband digital processing.

The specification of the present disclosure relates to a device, method, and computer program for searching for genes in a synthetic lethality relationship by using a Gaussian restricted Boltzmann machine. According to any one of the above-described means for solving the problem, a gene in a synthetic lethality relationship with a target gene may be output through an artificial intelligence model trained by receiving a training data set including an mRNA expression value in RNA Seq data, the presence or absence of a mutation in WES data, and a dependency score in CRISPR KO data. In addition, it is possible to increase the efficiency of anticancer treatment by using the genes in a synthetic lethality relationship calculated by using the trained artificial intelligence model. In addition, it is possible to present a treatment route with low drug resistance by using the genes in a synthetic lethality relationship calculated by using the trained artificial intelligence model.

The application discloses a sedimentary microfacies recognition method based on a deep belief network, and relates to the technical field of intelligent information processing, and comprises the following steps: firstly, combining and optimizing logging curves; then, using original data of the selected logging curves to construct binary images reflecting curve shape features, taking the binary images as sample feature data, and using a deep belief network based on a restricted Boltzmann machine to capture the mapping relationship between the sample features and the categories thereof. The trained network can be used for single-well profile sedimentary microfacies recognition. With the powerful learning ability and generalization ability of the deep belief network, the recognition precision of the sedimentary microfacies can be effectively improved.

The present disclosure provides a bearing fault feature extraction method, which introduces a Gauss-Bernoulli restricted Boltzmann machine model, solves the problem that the input vector of the traditional restricted Boltzmann machine is restricted by the Bernoulli binary distribution and has poor reconstruction and fitting effect for non-binomial distributed data; uses the cosine loss function as the loss function, retains the advantage of the Softmax loss function in expanding the inter-class difference, and reduces the sensitivity to different signal strengths; combines the attention mechanism adaptively to more effectively extract features that are effective in describing the bearing status.

The invention relates to the technical field of computer data processing and artificial intelligence, and discloses a multi-type variable adaptive CRBM digital twinning modeling method, device and medium, the method comprises the following steps: obtaining modeling data and defining the modeling data as visible, conditional and hidden variable sets, the visible variables comprising non-standard distribution types; constructing a condition-restricted Boltzmann machine model, and directly constructing corresponding conditional probability distribution and interaction energy items according to original probability distribution characteristics for visible variables of non-standard distribution types; performing parameter updating on the model by calculating a weighted combination of a likelihood gradient and an adversarial gradient by adopting an adversarial training mechanism in which adversarial items are introduced; and using the trained model to generate digital twin data through Gibbs sampling based on a given condition input variable. According to the method, heterogeneous data can be directly processed, the original statistical characteristics of the data are reserved, and the precision of model parameter estimation and the fidelity of generated data are improved.

The application discloses a multi-modal critical edge biomarker identification method, takes an individual cancer patient as a dynamic network system, combines a dynamic network theory biomarker theory and a multi-modal evolutionary algorithm, performs hidden space search by using a restricted Boltzmann machine on the basis of an MMPDNB model, and is a new multi-modal PDENB identification model. Firstly, a PEN of the cancer individual patient is constructed. Then, an optimization objective function is designed. Finally, a multi-modal optimization algorithm is used to search for a PDENB set. The application can not only promote the researches on a mathematical model and an algorithm design of the PDENB identification problem, but also help to understand the individual heterogeneity of the cancer, and realize the early diagnosis and treatment of the cancer individual patient.

The invention provides a prediction control method and device for a system, electronic equipment and a computer storage medium, and the method comprises the steps: training a restricted Boltzmann machine in a deep belief network structure offline layer by layer through employing a training sample set in an unsupervised pre-training manner, and automatically extracting low-order to high-order abstract features from data, dependence on domain knowledge and calculation complexity are reduced, a trained offline model is evaluated by using a test sample set, and finally an intelligent prediction control model considering both accuracy and training time is determined. According to the method, the deep learning controller adopting the deep belief network structure is combined with the model prediction control, so that the calculation complexity is effectively reduced while the prediction accuracy is ensured in the actual application process, and the online processing time is shortened. Therefore, effective combination of energy control and optimal scheduling is realized, good energy-saving and emission-reducing effects are achieved, and the energy utilization efficiency and sustainability of the building are improved.

The invention provides a transformer capacitance compensation through-current method and system based on deep learning, and relates to the technical field of reactive power compensation of a power system. The method comprises the following steps: acquiring a transformer signal to obtain standardized input data; modeling of the double-layer restricted Boltzmann machine is completed; generating a deep hidden feature vector; generating a capacitor switching parameter; the control hardware module receives a control instruction and triggers a relay to act; and comparing the reconstruction error with a preset threshold value, maintaining the current switching state when the reconstruction error does not exceed the threshold value, and regenerating and adjusting the capacitor switching parameter when the reconstruction error exceeds the threshold value. According to the method, a complete process from standardized input, feature modeling and capacitance parameter mapping to control execution and error closed-loop updating is constructed, adaptive adjustment of capacitance switching action and whole-process data retention are supported, and decision consistency and execution stability under complex working conditions are improved.