131 results about "Recurrent neural network" patented technology

A method of representing a space of handwriting stroke styles includes representing writer-, character- and writer-character-level style variations within a recurrent neural network (RNN) model using decoupled style descriptors (DSD) that model the style variations such that character style variations depend on writer style.

This invention proposes a method, system, and electronic equipment for modeling the full-wind-speed operation characteristics of wind turbines. The method includes: collecting operational monitoring data of wind turbines; processing the operational monitoring data to form a sample set for modeling full-wind-speed operation characteristics; using a random forest algorithm to evaluate the importance of candidate features and select a subset of target features; using a recurrent neural network (RNN) to extract temporal dependency features from the subset of target features; using a convolutional neural network (CNN) to extract local pattern features from the subset of target features; and fusing the temporal dependency features and local pattern features to establish a model of the full-wind-speed operation characteristics of the wind turbine and output the predicted operation characteristics. By using the random forest algorithm to select the importance of multi-dimensional candidate features, and combining the ability of RNN to extract long-term temporal dependencies with the advantage of CNN in perceiving local fluctuation patterns, high-precision fusion modeling of the operation characteristics of wind turbines across the entire wind speed range is achieved.

This invention provides a real-time monitoring and early warning system and method for surrounding rock stress during drilling, belonging to the field of coal mine safety monitoring technology. The system includes an information acquisition module, a drilling information processing module, a video processing module, a microseismic information processing module, and a comprehensive information processing and early warning module. Through pressure sensors, displacement sensors, cameras, and microseismic sensors, it collects feed pressure, rotation pressure, cylinder displacement, drill cuttings images, and microseismic signals in real time. It performs filtering, noise reduction, and feature extraction on various types of data. A weighted average algorithm is used to fuse multi-source data to construct a fused dataset. A surrounding rock stress monitoring model is built based on an LSTM deep recurrent neural network to achieve real-time analysis and hazard level classification and early warning of surrounding rock stress. This invention realizes multi-source information fusion monitoring of surrounding rock stress state during drilling, effectively identifying strong, medium, and weak stress hazards, and improving the safety and reliability of drilling operations and early warning.

This application provides a method and system for monitoring the operational status of reservoirs based on deep learning, relating to the field of water conservancy project operation monitoring. The method includes: collecting multi-source time-series data on reservoir operation, including hydrological time-series data, engineering time-series data, and environmental time-series data; constructing a multi-source feature sequence of the reservoir's operational status based on the multi-source time-series data; inferring the multi-source feature sequence using a deep learning model to obtain reservoir operational status assessment results and trend prediction results; the deep learning model consists of convolutional neural network branches, recurrent neural network branches, and attention mechanism branches connected in parallel; and determining the reservoir's operational status based on the reservoir operational status assessment results and trend prediction results. This application, used in reservoir operational status monitoring, solves the technical problem of low accuracy in existing reservoir operational status monitoring methods.

The application discloses a kind of LED early fault diagnosis and life prediction method based on spectral power distribution;The present application is based on similarity detection method to carry out fault diagnosis, extracts spectral characteristic value from statistical model, is reduced dimension by principal component analysis, and is clustered using K-means++ method, finally using distance and threshold comparison determines abnormal time.On this basis, long short-term recurrent neural network is established to predict spectral characteristic value, remodel spectrum, and predict remaining useful life.The present application combines spectral power distribution, and carries out fault diagnosis and life prediction to accelerated step aging white light LED based on long short memory recurrent LSTM neural network method, which greatly improves the accuracy of LED early abnormal detection and remaining life prediction.

This invention relates to the field of pharmaceutical automation equipment and discloses an intelligent sampling robot suitable for drug production. The robot includes a multi-sensor fusion positioning system, a force control coordination system, and a cleanroom structure design. The multi-sensor fusion positioning system comprises a 3D laser positioning module, a binocular vision module, a sub-pixel edge detection module, and a sensor fusion algorithm. The force control coordination system includes a hybrid RNN-LSTM (Recurrent Short-Term Memory) and CNN (Convolutional Neural Network) network, an airbag gripper, a six-dimensional force sensor, and an underactuated rotation mechanism. The cleanroom structure design includes a mechanical gripper, a nano-sterile coating, and a stainless steel IP69K sealing structure. By employing a fusion algorithm combining extended Kalman filtering and fuzzy evidence theory, efficient fusion of multi-sensor data is achieved. Dynamic threshold adjustment is realized through the hybrid RNN-LSTM and CNN network, reducing force control fluctuations and lowering production costs.

In a system and method for enabling a user to identify the emotions of speakers during a telephone or online conversation, spoken audio input is pre-processed using a one-dimensional Mel Spectrogram and / or a two-dimensional Mel-Frequency Cepstral Coefficient (MFCC) matrix, reducing the two-dimensional matrix to a single dimension output, and identifying at least one emotion in the audio input using a convolutional or recurrent neural network.

The application discloses a power load multi-model integrated prediction method and device, equipment and medium, and relates to the technical field of power load prediction, which comprises the following steps: cleaning and normalizing the historical load time series data, then adopting a strict causal sliding window mechanism to construct multi-scale features, determining the optimal parameters of a tree model, a gated recurrent single-rank regression meta-model and a bidirectional recurrent neural network through a two-stage hyperparameter joint optimization strategy and constructing corresponding models, inputting structured features, a first time series matrix and a second time series matrix for orthogonal division prediction, and finally outputting results through enhanced stacked fusion. The method realizes root-cause avoidance of data leakage, does not depend on exogenous variables, reduces the parameter tuning calculation complexity, realizes model differentiation and complementation, and improves the accuracy and robustness of power load prediction.

This invention relates to the technical field of predicting the remaining creep life of high-temperature metallic materials, and provides a method for predicting the remaining creep life based on physical information and recurrent neural networks. The method includes: dividing relevant data of the high-temperature metallic material into static and dynamic features; calculating the correlation between static features and creep fracture life; incorporating the positive or negative sign of the correlation coefficient as a physical loss term into the loss function to construct a physical information neural network; inputting the normalized static features into the constructed physical information neural network and training and adjusting the parameters, using the trained physical information neural network to obtain the predicted creep fracture life of the high-temperature metallic material; inputting the normalized dynamic features into the recurrent neural network and training and adjusting the parameters, using the trained recurrent neural network to obtain the predicted proportion of the remaining creep life to the creep fracture life; and multiplying the predicted creep fracture life by the predicted proportion to obtain the predicted remaining creep life of the high-temperature metallic material.

The application discloses a crop growth function prediction method and system based on a deep operator network, and comprises the following steps: collecting environment parameter vectors aligned according to timestamps; inputting a trained prediction model to obtain a crop growth function and output a prediction result. The prediction model comprises a recurrent neural network module and an improved deep operator network module. In the recurrent neural network module, the environment parameter vectors are taken as inputs to obtain time sequence feature vectors. The improved deep operator network module comprises a branch network and a trunk network; the branch network takes the time sequence feature vectors as inputs; the trunk network takes evaluation points as inputs to generate trunk vectors. A growth state vector is obtained from the outputs of the crop growth operators at the evaluation points. Thus, the deep operator network and the recurrent neural network are fused to form a continuous dynamic crop growth function, the problems of low prediction accuracy and weak generalization ability of existing models are overcome, continuous spatiotemporal data can be processed with less required computing power, and the method can be deployed in most planting scenes.

This invention discloses an adaptive tree neural network structure and processing method for construction project management. The structure includes: an engineering data input module for collecting structured and time-series data of construction projects; a tree structure encoding module for constructing a multi-level decision tree structure based on project hierarchy and encoding node features; a recurrent neural network module for performing time-series modeling of node feature sequences to predict project status; an adaptive feedback module for feeding back the prediction results to the tree structure encoding module to update node features or weights, forming an adaptive closed loop; and an engineering management output module for generating construction management decision information based on the prediction results. This invention solves the problems of missing engineering structure representation, discontinuous time-series prediction, and lack of dynamic adjustment capability in existing technologies by integrating hierarchical engineering structure modeling and time-series prediction and introducing an adaptive feedback mechanism, thereby improving the intelligence level of construction progress prediction, resource scheduling, and risk warning.

This invention relates to the field of hydrological forecasting technology and discloses a medium- and long-term intelligent runoff forecasting method, comprising: collecting watershed data, preprocessing and extracting features, generating a feature vector sequence and dividing it into training sets; constructing a physical constraint recurrent neural network module and training it based on the total loss function of physical constraint loss; constructing a sliding window online adaptive correction module, initializing the sliding window to store the measured runoff values ​​and preliminary runoff forecast values ​​of the most recent O times; inputting the feature vector of the current time into the trained physical constraint recurrent neural network module to generate preliminary runoff forecast values, calculating the historical average deviation according to the window state and correcting it to obtain the final forecast result; finally, forming a new sample pair of measured runoff values ​​and preliminary runoff forecast values ​​and adding it to the window while removing the oldest sample to achieve dynamic window updates; this invention achieves high-precision, high-physical-consistency, and online adaptive intelligent forecasting of medium- and long-term runoff.

The application discloses a text classification method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining a first text; preprocessing the first text to obtain a word vector included in the first text and at least one word vector corresponding to the word vector; performing weighted fusion processing on each word vector in the word vector and the at least one word vector based on a lattice-LSTM model to obtain a word lattice vector of each position corresponding to each word vector; performing semantic aggregation on the word lattice vector based on a capsule network model to obtain a first semantic vector of each position; determining a category label corresponding to the first semantic vector of each position, and taking the category label as the classification of the first text.

The application discloses a rolling bearing degradation trend prediction method based on linear regression and TCN, characterized in that: firstly, the full life cycle vibration data of the rolling bearing is collected through an acceleration sensor, the sample entropy of the data is calculated and obtained as the performance degradation index of the bearing, the sample entropy is preprocessed smoothly, then the SPT point of the bearing is determined according to the linear regression model and mu+3delta, the influence of different window lengths on the determination of the SPT is analyzed, finally the preprocessed sample entropy value is input into the trained TCN network for degradation trend prediction. The application can effectively and timely determine the SPT point of the bearing, the TCN can well fit the performance degradation trend of the bearing, and the problem that the equipment cannot be safely operated due to the too late discovery of the bearing fault is avoided; compared with the method of adopting the recurrent neural network prediction, the accuracy of the method for the equipment health management is obviously improved, and a new idea is provided for the degradation trend prediction method of the bearing.

This invention relates to a closed-loop control method and system for improving the compaction density of LFP electrodes. The method involves real-time acquisition of images of the micro-particle packing of the coating slurry, synchronized with the rolling mill operation data; extraction of micro-pore distribution features from the images, establishing a mapping relationship between these features and the material's fractional-order constitutive parameters; combining real-time rolling pressure and roll gap data, obtaining the equivalent fractional-order derivative order and rheological viscosity of the electrode material through rapid inversion calculation; inputting the viscoelastic properties into a physically constrained recurrent neural network to predict the full-cycle thickness evolution curve of the electrode, including elastic rebound, creep, and relaxation processes, and calculating the predicted compaction density based on the steady-state value of the curve; constructing a planning model with the goal of minimizing the residual between the predicted density and the set target, and jointly optimizing the roll gap position compensation and hot roller heating power online to achieve closed-loop control of the compaction density.

Artificial intelligence for time-series data analytics is provided. A first time-series data set is provided to a pre-trained recurrent neural network trained based on a second time-series data set. A prediction of a ground truth state of the first time-series data set is received therefrom. The first time-series data set is provided to a dynamical recurrent neural network trained based on the second time-series data set and the pre-trained recurrent neural network. A noise-reduced prediction of a ground truth system state of the first time-series data set is received therefrom. An estimate of sensor noise is read. The estimate of sensor noise is generated based on the second time-series data set and the pre-trained recurrent neural network. A prediction of a state of the system is generated based on the pre-trained recurrent neural network, the dynamical recurrent neural network, and the estimate of sensor noise.

This application discloses a singing scoring method and related apparatus, equipment, and storage medium. The singing scoring method includes: acquiring the singing audio to be evaluated and the target score of the target song to which the singing audio belongs; extracting the acoustic features of the singing audio and extracting the singing sequence of the target score; wherein the singing sequence includes the pitch and beat length of the target song at each singing moment; performing feature extraction on the acoustic features based on a first extraction model to obtain a first feature representation, and performing feature extraction on the singing sequence based on a second extraction model to obtain a second feature representation; wherein both the first and second extraction models adopt a target network framework, which includes a sequentially connected convolutional neural network and a recurrent neural network; and making a prediction based on the first and second feature representations to obtain the singing score of the singing audio. The above scheme can improve the applicability, accuracy, robustness, and generalization ability of singing scoring.

The application provides a multi-directional stress prediction method, device and equipment in a milling cutter milling process and a storage medium. It relates to the field of numerical control machine tool processing digital twin technology. The method comprises: obtaining small sample experimental data based on an orthogonal test method, time-frequency decomposition of the milling force test signal, and extraction of multi-dimensional features of the milling force dynamic characteristics; analyzing the correlation between the processing parameters and the features, screening the key features, establishing a physical mapping model of the process parameters to the key features and solving the cutting coefficients; constructing a time-varying signal prediction model based on a recurrent neural network, predicting the multi-directional dynamic milling force of the milling cutter with the key features in the small sample test data; and based on the cutting coefficient, designing an adaptive filter to post-process and optimize the predicted signal and inverse normalize it, and output the final prediction value. Based on small sample data, the application can accurately predict the multi-directional dynamic stress of the milling cutter only with the process parameters, and effectively improve the virtual-real mapping and dynamic optimization capability of the processing process.

The present application provides a kind of flue gas CO2 capture system load fluctuation fast response control method, to solve the problem of poor adaptability of existing capture system to power load fluctuation, prone to "disturbance amplification effect".The present application first establishes CO2 capture dynamic model based on mass balance, energy balance equation, then constructs MTRNN-MPC predictive control system, integrates real-time optimization, multi-time domain recurrent neural network and model predictive control.MTRNN captures short-term disturbance and long-term trend through multi-time scale sub-module, RTO adopts "regular cycle + disturbance trigger" mechanism to update optimal set value, and MPC realizes multivariable collaborative optimization with "control time domain + 4 times prediction time domain" strategy.The method can quickly respond to flue gas flow, concentration and other fluctuations, improve system robustness and adaptability, and is suitable for CO2 capture device matched with flexible operation power system.

The application provides a RAW domain low-illumination image enhancement method and system, relates to the technical field of image processing, and can effectively avoid amplification of noise in the subsequent enhancement process, reduce noise and color distortion of the RAW domain enhanced image by separating the content signal graph and the noise residual graph of the RAW domain low-illumination image. Through the content signal graph, the light rays of the camera light path are simulated and light integration is carried out, so that the physical level detail reconstruction can be carried out on the region with serious structure loss in the RAW domain low-illumination image, and the global light consistency and edge sharpness better than the traditional two-dimensional convolution network are realized. By introducing the target double-flow recurrent neural network, the optimization conflict between denoising and color restoration in the low-illumination enhancement can be solved, the progressive collaborative improvement of the image quality can be realized, and the high harmony of the final enhancement result in the three dimensions of brightness, detail and color is ensured.

This invention proposes a method and system for reducing the order of FPSO models by integrating time-recurrent neural networks and fuzzy logic, belonging to the field of marine engineering structural analysis and intelligent modeling technology. Addressing the problem of large degrees of freedom and low computational efficiency in FPSO finite element models, this invention extracts key degrees of freedom from the high-dimensional finite element model to construct low-dimensional feature vectors. It then uses a fuzzy logic system to perform regularized modeling of the dominant mechanical properties of the structure, achieving initial order reduction. Furthermore, a time-recurrent neural network is introduced to compensate for high-order residual effects that are difficult to accurately characterize using the fuzzy logic model, ultimately resulting in a low-dimensional, high-fidelity reduced-order model. The proposed method significantly reduces the degree of freedom and parameter dimension of the finite element model while maintaining the main mechanical properties of the structure, effectively reducing computational complexity and improving the model's applicability and practical engineering value in applications such as rapid finite element analysis, online structural response prediction, and engineering decision support.

This invention relates to the field of intelligent floor temperature control technology, specifically disclosing an environmentally adaptive intelligent floor temperature control system and control method, including the following steps: Step S1: Collect historical user temperature setting data and floor temperature control operating data; Step S2: Analyze historical data to establish a correlation model between indoor perceived temperature and air temperature; extract user temperature setting curves, generate behavioral habit feature curves, and classify them through a recurrent neural network to obtain several categories and their average feature curves; Step S3: Match the current user behavior curve with the historical average curve to predict the target temperature curve for the next day; combine the temperature correlation model to dynamically adjust the floor temperature; This method achieves personalized and adaptive temperature control optimization through data-driven and machine learning approaches.

This invention relates to the field of blood glucose concentration prediction technology, specifically disclosing a method, system, and device for blood glucose concentration prediction based on a multi-branch attention mechanism. The method includes: acquiring multivariate time-series data related to blood glucose regulation; dividing the features into multiple feature subsets; inputting each feature subset into a corresponding independent recurrent neural network branch for time-dependent learning to generate a hidden state sequence; inputting each hidden state sequence into a corresponding attention submodule for processing, outputting a weighted feature representation; fusing the weighted feature representations to obtain a unified feature representation; and generating a blood glucose concentration prediction result through an output layer. This invention separates the temporal patterns of heterogeneous features through a multi-branch structure and focuses on key time steps using an attention mechanism, solving the problems of feature interaction interference and insufficient capture of key information in existing technologies, significantly improving the accuracy and clinical applicability of blood glucose prediction.

This invention discloses a knowledge-guided method for separating P-waves and S-waves in complex geological structures, applied to the field of seismic data processing. It addresses the problem that existing technologies struggle to effectively separate P-waves and S-waves in complex geological structures. First, this invention utilizes a recurrent neural network for elastic wave full-waveform inversion to obtain knowledge representations of P-waves and S-waves in complex structures. Second, using these knowledge representations, a dual-branch autoencoder network is constructed. One branch separates P-waves under the guidance of the P-wave knowledge representation, while the other branch separates S-waves under the guidance of the S-wave knowledge representation. Finally, an autoencoder network architecture is constructed that comprehensively constrains wavefield reconstruction error, P-wave knowledge-guided error, and S-wave knowledge-guided error. Theoretical analysis and numerical calculations demonstrate that the proposed method can effectively separate P-waves and S-waves in complex geological structures, reducing dependence on data samples and exhibiting a certain degree of robustness.

This application discloses a method for early warning of disc and shaft cracks in aero-engines, belonging to the field of aero-engine fault diagnosis technology. It includes: acquiring historical flight parameters and generating time series by component clustering; extracting initial node features of each component using a parallel recurrent neural network; constructing a topological graph with components as nodes and physical connections as edges, and outputting a set of node features with fused coupling relationships via graph attention network message passing; decoding vibration prediction values ​​using a fully connected network after global pooling; constructing a composite loss function containing data fitting, graph smoothing, attention sparsity, and physical conservation terms, and training the model by adjusting weights in stages; inputting the parameters of the flight to be tested to obtain the predicted values, calculating the relative residuals, and outputting a graded early warning based on dynamic thresholds and trend conditions. This application integrates engine physical topology and data to achieve reliable early warning of disc and shaft cracks.

This application discloses a resource scheduling method, apparatus, electronic device, and medium, belonging to the field of network resource scheduling technology. The method includes: acquiring energy consumption data, latency data, and load data from multiple servers, wherein the first server includes a cloud server and an edge server; inputting the energy consumption data, latency data, and load data into a preset resource scheduling model, and outputting multiple system data of the resource scheduling system composed of the multiple servers; inputting the multiple system data into a preset deep learning model, and outputting response data corresponding to the resource scheduling system to represent the network environment's response to the state; the deep learning model is a model combining a time-recurrent neural network and an attention mechanism; and scheduling the network resources of the multiple servers based on the response data. The embodiments of this application employ a resource scheduling model and a deep learning model to improve resource scheduling efficiency.

The application discloses an Internet risk prediction method and device based on domain name analysis, equipment and medium, relates to the technical field of artificial intelligence, can be applied to the field of finance and medical health, and mainly aims to solve the problem of poor prediction accuracy of the existing Internet domain name risk. Including: in response to an Internet data access request, calling a monitoring thread; determining the risk level of the user access domain name based on at least one of the blacklist rules, the whitelist rules and the risk classification rules in the monitoring thread; if the risk level is low risk or medium risk, a risk prediction model corresponding to the risk level is called, and the domain name reputation, historical behavior and IP address associated with the user access domain name are predicted based on the risk prediction model to obtain a risk prediction result, and the risk prediction model is constructed based on a convolutional neural network and a recurrent neural network.

Systems, methods, and computer program products train a residual neural network including a first fully connected layer, a first recurrent neural network layer, and at least one skip connection for anomaly detection. The at least one skip connection directly connects at least one of (i) an output of the first fully connected layer to a first other layer downstream of the first recurrent neural network layer in the residual neural network and (ii) an output of the first recurrent neural network layer to a second other layer downstream of a second recurrent neural network layer in the residual neural network.

The application relates to the technical field of lithium ion batteries, and provides a lithium ion battery early life prediction method and system based on a liquid state neural network, which comprises the following steps: constructing a multi-working-condition data set, and only adopting electric quantity as input features; based on the LNN of the CfC framework, adopting Embedding and LNN fusion or CNN and LNN fusion, mapping the features by adopting a full connection layer as a hidden layer, constructing a lithium ion battery early life prediction model, and performing model training; adopting the trained lithium ion battery early life prediction model to predict the early life of the lithium ion battery, and obtaining a prediction result. The Embedding and LNN fusion model has obvious performance improvement compared with the Embedding and LSTM fusion model of a discrete recurrent neural network, the CNN and LNN fusion realizes fusion of local feature extraction and continuous time dynamic modeling, and higher-precision lithium ion battery early life prediction is obtained.