861 results about "Recurrent neural network" patented technology

A system for detecting service interruptions in a cable broadband network using telemetry-based anomaly detection, wherein the system comprises the following: a telemetry acquisition unit configured to acquire multi-parameter telemetry data from heterogeneous broadband infrastructure elements, including cable modems, amplifiers, optical nodes and cable modem termination systems (CMTS), wherein the telemetry data includes the signal-to-noise ratio, modulation error ratio, forward error correction counter, power levels and latency statistics; a preprocessing and harmonization module that is operationally coupled with the telemetry acquisition unit, wherein the module is configured to normalize heterogeneous telemetry streams by adjusting sampling rates, synchronizing timestamps, interpolating missing data, and filtering out false outliers; an anomaly detection unit that is communicatively connected to the preprocessing and harmonization module, wherein the unit comprises a hybrid detection framework with statistical prediction models and machine learning models, wherein the statistical prediction models include ARIMA or Holt-Winters models to predict the expected telemetry behavior and the machine learning models include recurrent neural networks and autoencoders trained on historical telemetry; an ensemble evaluation subsystem within the anomaly detection unit, configured to combine the outputs of the statistical prediction models and the machine learning models to generate anomaly probability evaluations with adaptive confidence intervals; an interruption classification module configured to receive anomaly probability values ​​and correlate anomalies across multiple devices, geographic clusters, and time windows, wherein the interruption classification module differentiates between transient anomalies and service-impairing interruptions based on a multidimensional correlation; and an alerting interface configured to transmit outage alerts with severity, root cause metadata, and geolocation to a network management system so that the operator can intervene.

A system for real-time analysis of emotional feedback during motivational speeches, consisting of: a series of multimodal sensors, including at least one visual sensor configured to capture facial expressions of spectators, at least one directional microphone configured to capture the audio responses of the audience, and optionally one or more physiological sensors configured to capture biometric signals from spectators; an edge-based processing unit that is communicatively coupled to the arrangement of multimodal sensors, wherein the edge-based processing unit comprises the following: (a) a feature extraction module configured to extract visual features from captured facial images, acoustic features from voice responses, and physiological features from biometric signals; (b) an emotion inference machine configured to process the features using a deep learning-based emotion recognition model comprising a convolutional neural network (CNN) for classifying facial expressions, a recurrent neural network (RNN) for classifying voice emotions, and a multimodal late fusion layer configured to compute a composite emotion state vector representing the aggregated emotions of the audience; (c) a timestamp and speech alignment module configured to correlate the calculated composite emotion state vector with segmented portions of a live motivational speech based on real-time speech-to-text transcription and semantic analysis; and (d) a session-based storage unit configured to log time-indexed emotional state vectors and corresponding speech segments for post-event analysis; A speaker feedback interface comprising a portable display or a podium-mounted visualization panel, wherein the interface is configured to display visual indicators of emotional feedback in real time, the indicators being derived from the emotional state vector and including at least emotional trend graphs, threshold alerts, or engagement indices.

A system, a method and a processor are for AI-powered generation and delivery of video clips. The processor is configured to: load a first video file of a first video content item, the first video file comprising video frames associated with timestamps; load a first subtitle file of the first video content item, the first subtitle file comprising subtitle text associated with the timestamps; execute a natural language processing (NLP) model with the subtitle text as input, the NLP model including language pre-processing steps for classifying words, names or phrases in the subtitle text and associating initial classifiers with the subtitle text, the NLP model including one or more of a recurrent neural network (RNN), a Bidirectional Encoder Representations from Transformers (BERT) model, or a generative pre-trained transformer (GPT) model for a dialogue analysis comprising processing sequences of dialogue in the subtitle text in view of the initial classifiers to associate one or more portions of the dialogue with one or more first classifiers of first narrative elements; execute an image recognition model with at least some of the video frames as input, the image recognition model including a convolutional neural network (CNN) for an object detection analysis and a facial recognition analysis comprising processing video sequences to associate one or more of the video frames with one or more second classifiers of second narrative elements; generate a narrative map of the first video content item by temporally aligning the first narrative elements with the second narrative elements based on the timestamps associated with the video frames and the first subtitle file; and generate a video clip including at least one segment of the first video content item, the at least one segment including selected video frames associated with at least one of the first or second narrative elements identified from the narrative map and selected for inclusion in the video clip.

The invention discloses a cable optical fiber temperature measurement monitoring system based on big data AI fusion, particularly relates to the technical field of cable optical fiber temperature measurement monitoring, and aims to realize thermal-electric dynamic coupling monitoring through high-density distributed optical fiber temperature measurement and big data AI fusion modeling and introduction of current load synchronous acquisition, and realize high-density distributed optical fiber temperature measurement and big data AI fusion by means of time synchronization and space relocation processing. The method comprises the following steps: identifying a weak response signal of a thermal inertia hysteresis characteristic and an initial temperature rise stage, constructing a cross-scale time sequence recurrent neural network through multi-dimensional enhancement and vector fusion, outputting a continuous scoring sequence reflecting abnormal credibility, and constructing a risk factor group reflecting a fault evolution path in combination with a scoring fluctuation amplitude, a continuous period and a response delay characteristic, and carrying out positioning offset correction through correlation mapping of risk factors and along-cable temperature distribution and by utilizing a topological relation between optical fiber path coordinates and a cable space, and realizing space reverse projection of an abnormal peak value, thereby recovering an accurate position of a fault heat source and a latent hidden danger section.

The invention relates to a cross-attention mechanism optimization method and a cross-attention mechanism optimization system for multi-modal feature fine-grained alignment. The method comprises the following steps: constructing a space-time topological graph of a dynamic scene, and representing a space-time relationship among intelligent agents through nodes and various connecting edges; coding the heterogeneous spatio-temporal information by using a plurality of groups of parallel recurrent neural networks, and converting the heterogeneous spatio-temporal information into uniform dimension feature representation; establishing a multi-modal cross attention alignment component, and quantifying feature quality from multiple dimensions through a multi-criterion evaluation unit; an adaptive weight fusion system is adopted to dynamically integrate evaluation results, and a unified quality score is generated; constructing a progressive optimization architecture based on the quality score, and performing multi-task cooperative training on a double-layer attention mechanism by combining real-time sampling and a directional optimization strategy; and finally generating a movement decision. According to the method, fine-grained alignment and optimization of multi-modal features are realized, and the accuracy and adaptability of intelligent agent navigation and trajectory prediction in a complex dynamic environment are remarkably improved.

The invention belongs to the crossing field of artificial intelligence and process control, particularly relates to an intelligent coupling control strategy for chlorine production through seawater electrolysis based on digital twinning, and aims to solve the problems of response lag, weak disturbance suppression and lack of electrode aging early warning in a traditional control method. According to the strategy, a high-fidelity digital twin fusing electrochemistry, fluid and thermodynamics is constructed, characteristics of current, temperature, impedance spectroscopy and the like are collected and fused at high frequency through a multi-source sensor, a state space model driven by a deep recurrent neural network is established, and current efficiency and electrode aging coupling evolution recognition is achieved; when salinization or flow mutation is detected, the twinborn body predicts and dynamically calibrates the current density and the voltage set value in advance, and the safety process window is synchronously scaled. According to the strategy, the chlorine gas yield fluctuation is reduced by 60%, the energy consumption standard deviation is reduced by 45%, the fault early warning is advanced by 2-5 hours, the fault-free operation time of the system is prolonged by 50%, and high-precision, self-adaptive and preventive intelligent control of the electrolysis process is realized.

A system for real-time creditworthiness assessment of small and medium-sized enterprises (SMEs) based on agent-based artificial intelligence (AI), the system includes: a large number of autonomous agents connected to each other via secure communication protocols; a data ingestion processing unit configured to capture and preprocess heterogeneous data streams from structured and unstructured sources, including APIs for financial transactions, tax databases, e-invoicing registers, inventory management systems, social sentiment feeds, and IoT-enabled devices, with preprocessing including schema harmonization, anomaly detection, and encryption-based integrity preservation; an ensemble of scoring control units configured to generate dynamic credit scores, wherein the ensemble includes machine learning models, including gradient-enhanced decision trees, recurrent neural networks, and graph neural networks, the graph neural networks being configured to assess relational dependencies between SMEs, suppliers, and customer nodes in a business ecosystem; an audit controller configured to generate regulatory-compliant justification paths for each credit score; a feedback processing unit configured to update the evaluation control unit ensemble using reinforcement learning signals derived from observed repayment behavior, payment default trends, and fraud detection markers; and a user interface configured to display timestamped credit histories, confidence intervals and explanatory justifications, with the entire system running on a containerized cloud-native infrastructure with blockchain-based ledgers for version control of each point update.

The invention discloses a settlement amount time sequence prediction method and system for deep foundation pit excavation adjacent buildings. The method comprises the steps that original monitoring data of on-site building settlement are acquired; performing data preprocessing on the obtained original monitoring data; constructing a recurrent neural network; model input and output parameters are determined through principal component analysis; optimizing the recurrent neural network based on an optimizer; performing hyper-parameter optimization based on an optimization result; performing settlement time sequence prediction by using the optimized recurrent neural network; the system comprises a data acquisition module, a preprocessing module, a model construction module, an analysis module, an optimization module, a parameter optimization module and a prediction module. By constructing the settlement prediction model, dynamic modeling and accurate prediction of the settlement trend of the building in the excavation process of each stage of the foundation pit are realized; historical settlement monitoring data and multi-layer soil body excavation information are combined, and multi-source input parameters are introduced, so that the adaptability of the model to complex working conditions is enhanced.

The invention discloses a microseismic source positioning method, device and system, and a storage medium. The method comprises the following steps: dividing a microseismic data sample data set into a training set and a test set; according to the training set, constructing a microseism inversion subnet containing a Swin Transform encoder, and according to the training set, constructing a microseism inversion subnet containing a Swin Transform encoder; the micro-seismic forward modeling subnet realizes seismic wave field continuation by inputting a micro-seismic source position and a speed model and utilizing a recurrent neural network structure and a convolution operator, and establishes a forward modeling subnet based on a wave equation; constructing an inversion-forward modeling closed-loop neural network according to the micro-seismic inversion subnet and the forward modeling subnet based on the wave equation; and inputting the test set into the inversion-forward closed-loop neural network to carry out micro-seismic source positioning. By adopting the technical scheme of the invention, the limitations on physical constraint, feature modeling and anti-noise capability in the prior art are overcome.

The invention provides a coal mine roof dynamic monitoring and early warning method and system based on artificial intelligence, and relates to the field of coal mine safety, and the method comprises the steps: collecting a stress waveform and a displacement waveform of a coal mine roof, extracting stress loosening characteristics, and constructing a three-dimensional stress conduction strength field; identifying a stress sudden change transition point by using a recurrent neural network to form a stress conduction chain; analyzing a regional topology migration rule and extracting a stress field instability path; matching with historical data to determine a prediction window; and continuously verifying the instability path and generating a support reinforcement scheme. The coal mine roof instability risk can be predicted in advance, and the early warning accuracy and timeliness are improved.

The invention discloses a model-based adaptive optical dynamic modeling and control method and device, and belongs to the cross field of adaptive optics and model-based reinforcement learning, and the method comprises the steps: defining the state of an AO system as a combination of multi-mode observation, and carrying out the corresponding coding; fusing the multi-modal observation data through a cross attention mechanism to form a unified observation potential state rich in dynamic information; a deep dynamic world model based on the recurrent neural network is constructed, the model maintains a hidden state and deduces a discrete potential state, and the hidden state and the discrete potential state jointly form complete potential state representation of the world model; rebuilding original observation and predictive rewards according to the complete potential state representation of the world model; and training a reinforcement learning agent in the learned world model, learning an optimal strategy through an imaginary trajectory, and outputting key parameters of the AO system in real time. According to the method, the adaptability, the control precision and the robustness of the AO system in a complex dynamic environment are improved.

The invention relates to the technical field of intelligent gateways, and discloses a multi-mode industrial Internet of Things intelligent gateway based on edge computing and an implementation method thereof, and the method comprises the steps: obtaining sensor data streams of a door magnetic sensor, a human body sensor, a temperature and humidity sensor and a smoke detector in an intelligent region; constructing a space-time fusion data matrix based on the sensor data stream; inputting the space-time fusion data matrix into an edge layer quantization neural network, a fog layer recurrent neural network and a cloud layer large model for distributed reasoning to obtain a reasoning result set; and performing priority queue scheduling and zero-copy transmission on emergency events, important events and conventional events in combination with the reasoning result set to generate a control instruction sequence, so that parallel operation of data acquisition and processing is realized, and the intelligent level, the response performance and the operation reliability of an intelligent regional Internet of Things system are improved.

A VEM-Token emotion synchronization function hierarchical fusion method is different from a traditional NLP-Token method, an emotion synchronization function VEM-sync is innovated for the first time, the emotion synchronization function VEM-sync is synchronized with a VEM-Token sequence of music beats, multiple high-dimensional emotions are directly described by adopting mathematical languages, and therefore deviation of discretized natural language characters on description of a high-dimensional emotion analog quantity function is avoided, and the emotion synchronization effect is improved. The method comprises the steps of defining VEM-sync and synchronous content, defining rhythm attributes, emotion attributes and emotion functions, adopting one or combination of multi-layer weighted scanning, a recurrent neural network, a long and short-term memory network, a self-attention mechanism and an RAG network generated by retrieval enhancement, and performing hierarchical fusion calculation to output a time emotion function of a vocal music file. According to the phonetic function or the emotional function, the effects of emotional texts, emotional expressions, emotional languages and emotional multi-dimensional animations are directly driven by crossing discrete text tokens, a model context protocol (MCP) and a function calling function are supported, and copyright management and encryption and decryption or interfaces are provided.

The invention discloses an abnormal sleep signal identification method for an intelligent sleep monitoring pillow, and relates to the technical field of abnormal sleep signal identification. Comprising the following steps of collecting multi-modal sleep physiological data, preprocessing the multi-modal sleep physiological data, extracting multi-dimensional features from the preprocessed data, constructing a fusion analysis model to judge abnormal sleep signals, triggering a hierarchical intervention mechanism and outputting associated intervention suggestions. According to the method, the time dimension alignment technology and the sliding window matching technology are adopted, the problem of data synchronization caused by multi-sensor sampling frequency difference in a traditional abnormal sleep signal recognition technology is solved, multi-source features are fused through deep learning, the accuracy of abnormal sleep signal recognition is remarkably improved, and the recognition accuracy of abnormal sleep signals is improved. An intelligent analysis model based on deep learning multi-network fusion is constructed, deep extraction and fusion judgment of complex sleep features are realized by utilizing cooperative processing of a convolutional neural network, a recurrent neural network and an attention mechanism, and the reliability of abnormal sleep signal recognition is improved.

The invention discloses a key parameter long-time-sequence prediction method for a complex process industry, and the method comprises the steps: collecting multivariable sensor data in the process industry, and constructing a high-dimensional long-time-sequence prediction data set; constructing a PatchConvRNN prediction model by combining time slice embedding, dimension decoupling convolution, depth separable convolution and a recurrent neural network based on a sequence-to-sequence normal form; a point value-statistical mixed loss function is adopted, the point prediction precision, the sequence mean value and the standard deviation consistency are optimized at the same time, a prediction model is trained in combination with an optimization algorithm, and network model parameters are adjusted; and comprehensively evaluating the prediction model through a root-mean-square error, an average absolute percentage error and a standard deviation average absolute error. According to the method, high-precision prediction and fluctuation maintenance of the key time sequence variables under the complex working condition of the industrial process are achieved, and powerful support is provided for quality control and predictive maintenance of the production process.

The invention relates to the technical field of metallurgical industry simulation and intelligent control, and discloses an intelligent metallurgical process virtual simulation method and system based on digital twinning. The method comprises the following steps: acquiring multi-modal high-dimensional data in a metallurgical process and reducing dimensions to obtain a feature vector set of a hidden layer space; performing disturbance injection simulation by using the feature vector set of the hidden layer space to obtain a multi-path state sequence with time dependence; predicting the abnormal path by using a recurrent neural network to obtain a predicted state evolution trajectory; utilizing a preset inverse mapping function to obtain multi-path state representation in the physical space; screening to obtain a risk path set; key evolution nodes are extracted from the set to be processed, and a virtual simulation scene is obtained; and performing optimization simulation on preset process adjustment parameters according to the virtual simulation scene, and determining optimized parameter configuration. The method can solve the problem that it is difficult to construct a comprehensive virtual simulation scene which truly restores the physical production rule.

The invention relates to a meteorological-based dynamic graph network photovoltaic power station group ultra-short-term prediction method, which is characterized in that a dynamic space-time graph network is constructed, photovoltaic power stations in a region are regarded as a complex network system connected by meteorological fluctuations, each power station is taken as a node, and the propagation relationship of the meteorological fluctuations among the power stations is represented by edges of a graph. The weight and time delay of the edge are dynamically adjusted according to real-time meteorological data, and dynamic interaction between power stations, the overall trend under the stable meteorological condition and rapid fluctuation caused by sudden weather events are accurately captured in combination with a mixed model of a graph neural network and a recurrent neural network. Compared with a traditional method, the method has the advantages that the prediction precision, the real-time response capability and the physical interpretability are remarkably improved, scattered photovoltaic power stations are integrated into a mutually associated dynamic system, an efficient and reliable solution is provided for power prediction of a power station group, and the operation stability and the power grid dispatching efficiency of a large-scale photovoltaic system are enhanced.

The invention relates to the field of motor fault diagnosis, and particularly discloses a permanent magnet synchronous motor fault diagnosis method and system, which dynamically update motor physical model parameters through an online identification algorithm so as to reflect the characteristic change of a motor in real time. Thirdly, predicting the theoretical voltage of the healthy motor under the working condition by using the self-adaptive model and the real-time operation data, and calculating a residual sequence subjected to working condition normalization between the theoretical predicted voltage and the actual voltage; the residual signal is essentially free from the influence of working condition change, and only the abnormal characteristics caused by the fault are highlighted. And finally, inputting the high-robustness residual error sequence into a convolutional recurrent neural network, deeply fusing space-time fault features, and realizing accurate judgment on a motor fault type and confidence thereof, thereby effectively improving variable working condition adaptability and weak fault detection capability of diagnosis.

According to the intelligent recommendation method and device based on driving behavior analysis provided by the embodiment of the invention, accurate analysis of user interests is realized by innovatively constructing a multi-dimensional feature fusion mechanism and integrating registration information, video data, driving data and scene perception data. And designing a trajectory prediction model based on a recurrent neural network, and establishing a feature weighted fusion strategy for intelligent matching in combination with a user interest modeling network of an attention mechanism. And an online learning mechanism is introduced, and model parameters are continuously optimized through clicking, staying duration and visit record data, so that dynamic adjustment of personalized recommendation contents is realized. According to the method, the defects of the traditional technology in the aspects of feature extraction, interest modeling, recommendation strategies and the like are effectively overcome, and the recommendation service level and the user experience in the driving scene are remarkably improved.

The invention discloses an electroencephalogram signal recognition method based on a graph recurrent neural network, and the method specifically comprises the steps: carrying out the preprocessing of electroencephalogram signal data, and obtaining a data segment electroencephalogram signal; performing feature extraction and graph embedding processing on the data segment electroencephalogram signals, and dividing the data segment electroencephalogram signals into a training set and a test set; inputting the training set data into a diffusion convolution recurrent neural network for training; and inputting the test set data into the trained diffusion convolution recurrent neural network, if epilepsy detection is carried out, outputting a binary label and probability that whether the fragment contains the attack or not, and if epilepsy classification is carried out, outputting a label and probability distribution that the fragment belongs to four attack types. According to the method, the influence of the multi-channel space relation on prediction and classification of the electroencephalogram signal attack is deeply explored from three dimensions of time domain, frequency domain and space domain by taking the multi-channel space relation as an entry point, the problem of cross-domain feature deficiency in the prior art is effectively solved, the electroencephalogram signal attack mode can be more accurately recognized, and the recognition rate of the electroencephalogram signal is remarkably improved.

The invention discloses a battery state-of-charge estimation method and system based on an ultrasonic transmission signal, and relates to the field of industrial detection.The method comprises the steps that the ultrasonic transmission signal is obtained, segmentation and normalization preprocessing are carried out, and preprocessed input sample data are obtained; constructing an initial battery state-of-charge estimation model used for extracting local features, extracting global features, extracting a time dependency relationship and performing classified output; inputting input sample data into the initial battery charge state estimation model for training, and calculating a loss function; obtaining a battery state-of-charge estimation model in combination with the training iteration number and / or the loss error threshold value; and estimating the state of charge of the battery through the battery state of charge estimation model. The method comprises the following steps: providing a battery charge state estimation model, and capturing a time dependency relationship of a signal by using a recurrent neural network and a time attention mechanism; an ultrasonic transmission signal is used as input data, and the estimation precision is improved by using information contained in the signal.

The invention discloses a transformer substation hidden danger comprehensive monitoring method, system and equipment integrating multiple sensors and a medium, and belongs to the technical field of transformer substation hidden danger monitoring. Cleaning and fusing the collected data, and constructing a unified data set; inputting the unified data set into an identification structure constructed by combining a convolutional neural network and a recurrent neural network, respectively processing image type and time sequence type data, and completing hidden danger type identification according to a preset training set proportion and training parameters; carrying out risk value calculation and grade judgment in combination with an identification result and hidden danger related characteristics; and early warning information is generated based on a judgment result, classified pushing and display are performed, early warning processing feedback information is received, and state updating and recording and archiving are completed. Fusion sensing and recognition of multi-source monitoring data are achieved, hidden danger early warning capacity and good environmental adaptability are achieved, and the operation safety and management efficiency of the transformer substation are improved.

The invention discloses a heat exchange station control method and device based on neural network decoupling and a medium, and relates to the field of neural networks, and the method comprises the steps: carrying out the feedforward decoupling through a trained time-delay recurrent neural network, and outputting a decoupling control instruction; performing prediction through the trained prediction model, and outputting a prediction parameter output value at a future moment; outputting a frequency conversion increment through a rolling optimizer according to the predicted parameter output value; determining an output difference value between the second actual parameter output value and the predicted parameter output value; and adjusting model parameters and / or variable frequency increment of the prediction model. Feedforward decoupling is carried out by using a time-delay recurrent neural network, and dynamic characteristics of the system are learned through a data driving mode without depending on an accurate mathematical model. Compared with the traditional feedforward decoupling and frequency conversion decoupling strategies, the coupling relationship between the quality regulation channel and the quantity regulation channel can be processed more thoroughly, and the problem that multi-variable dynamic response is difficult to coordinate by single pressure difference control is solved.

The invention discloses a nickel-metal hydride battery performance prediction model construction method based on machine learning, and the method comprises the steps: aligning electrochemical test data, environment monitoring data, operation log data and material characterization data through a unified time reference, and forming a synchronous original data flow; a structured data matrix is generated through missing value filling, anomaly correction and dynamic range compression, and the data quality and comparability are improved; voltage platform duration, temperature rise rate and principal component scores are extracted and given semantic tags, and physical interpretable feature construction is achieved; fusing the multi-source features by using a potential space projection network to generate a unified potential representation; a recurrent neural network, an attention mechanism and a multi-scale fusion strategy are combined, historical dependence is mined, and comprehensive state representation is generated; and a performance prediction value is synchronously output, a degradation track is deduced, independent verification and dynamic calibration are matched, and the prediction stability is ensured. Multi-source information deep fusion is achieved, and the accuracy and adaptability of nickel-metal hydride battery performance degradation trend prediction are remarkably improved.

The invention belongs to the technical field of electricity price prediction, and particularly discloses an electricity price prediction method and system based on dynamic subgraph learning, a terminal and a medium. Comprising the steps of collecting multi-source electricity market data such as load, weather and market transaction, performing normalization and missing value filling, and constructing a dynamic electricity price information graph; dynamic sub-graph division is executed based on the edge weight calculated in real time among the nodes, and a density peak value clustering method is adopted to determine the center of the sub-graph and periodically update the center of the sub-graph; extracting spatial features in the sub-graph through a graph convolutional network, generating a sub-graph embedded vector, and inputting the sub-graph embedded vector into a bidirectional recurrent neural network to obtain time sequence features; utilizing a multi-head attention mechanism to realize interactive fusion among different sub-graphs to obtain global feature representation; and predicting the future electricity price in combination with the global features and the historical electricity price sequence. And in the face of new energy output fluctuation, load sudden change or market mechanism adjustment and the like, the prediction flexibility and accuracy are improved.

The invention provides a power short-term load prediction system based on multi-source information adaptive fusion and a working method thereof, and the system comprises a multi-view data construction module which is used for organizing heterogeneous input data into a plurality of independent semantic views; the spatial dependence coding module is used for learning spatial dependence among nodes through a graph attention network and generating spatial embedding representation; the time dynamic modeling module is used for processing the time sequence view through a recurrent neural network and generating time feature embedded representation; the auxiliary information modeling module is used for processing other auxiliary views through independent feedforward neural network processing and generating auxiliary feature embedded representation; the cross-view attention fusion module is used for carrying out adaptive weighted fusion on all the embedded representations so as to inhibit representation degradation caused by low-quality views; and the load prediction module is used for generating a load prediction result based on the fusion representation and carrying out end-to-end training by taking a mean square error as a target.

The invention discloses a respiratory intensive care method and system, and the method comprises the steps: deploying an acoustic sensor on a respirator loop, and synchronously obtaining the ventilation parameters and optional temperature and humidity data of a respirator; preprocessing the acquired multi-modal data, and converting an acoustic signal into a time-frequency spectrogram; the preprocessed data are input into a multi-modal fusion fault recognition model, the model comprises a convolutional neural network branch used for extracting acoustic features and a recurrent neural network branch used for capturing time sequence dynamics, and the probability of a specific fault precursor is output by combining the features of all the branches through a fusion layer; and finally, performing early warning decision according to the probability. The invention also comprises a system for implementing the method. According to the invention, through deep fusion of acoustic, pneumatic and thermodynamic data, accurate prediction of fault precursor is realized, and an earlier time window is provided for clinical intervention, so that the monitoring safety is improved.

A system, a method and a processor are for AI-powered generation and delivery of video clips. The processor is configured to: load a first video file of a first video content item, the first video file comprising video frames associated with timestamps; load a first subtitle file of the first video content item, the first subtitle file comprising subtitle text associated with the timestamps; execute a natural language processing (NLP) model with the subtitle text as input, the NLP model including language pre-processing steps for classifying words, names or phrases in the subtitle text and associating initial classifiers with the subtitle text, the NLP model including one or more of a recurrent neural network (RNN), a Bidirectional Encoder Representations from Transformers (BERT) model, or a generative pre-trained transformer (GPT) model for a dialogue analysis comprising processing sequences of dialogue in the subtitle text in view of the initial classifiers to associate one or more portions of the dialogue with one or more first classifiers of first narrative elements; execute an image recognition model with at least some of the video frames as input, the image recognition model including a convolutional neural network (CNN) for an object detection analysis and a facial recognition analysis comprising processing video sequences to associate one or more of the video frames with one or more second classifiers of second narrative elements; generate a narrative map of the first video content item by temporally aligning the first narrative elements with the second narrative elements based on the timestamps associated with the video frames and the first subtitle file; and generate a video clip including at least one segment of the first video content item, the at least one segment including selected video frames associated with at least one of the first or second narrative elements identified from the narrative map and selected for inclusion in the video clip.

The invention relates to the technical field of power business intelligent diagnosis, in particular to a combined switch detection method and system based on a neural network, equipment and a medium, and the method comprises the steps: collecting the multi-mode operation data of a combined switch in real time through various types of sensors, and carrying out the data preprocessing and feature enhancement; a one-dimensional convolutional neural network is adopted to extract deep local features from the time series data; a time sequence recurrent neural network integrating a plurality of specialized sub-networks is used for conducting self-supervised learning and time sequence dependency modeling on the feature sequence, each sub-network is specially responsible for specific type fault mode recognition, and self-adaptive fusion of multi-professional domain knowledge is achieved through a dynamic weight distribution mechanism; and finally, outputting a health state evaluation result and a specific fault type of the combination switch through a full connection layer and a Softmax classifier. According to the method, hardware anti-interference design and a multi-professional-domain self-supervised learning algorithm are closely combined, accurate, real-time and robust online monitoring of the health state of the combination switch is achieved, and the detection accuracy and the anti-interference capacity of the system are remarkably improved.

The invention discloses a multi-task NILM low-voltage transformer area energy management system, and aims to solve the problems that a traditional low-voltage transformer area energy management system (EMS) is low in load identification precision and does not give consideration to operation cost and user satisfaction. According to the system, firstly, through an NILM module based on a multi-task recurrent neural network (taking a GRU as a core), power utilization states and power consumption of various electric appliances are decomposed from total load data of an intelligent electric meter, and a user power utilization behavior portrait is constructed; and taking the portrait as an input, and carrying out cooperative scheduling on photovoltaic, energy storage and controllable loads in a transformer area through a multi-objective optimization model considering system operation cost and user satisfaction. Experimental verification shows that compared with a traditional EMS, the operation cost of the system is reduced by 32.59%, the user satisfaction degree is improved by 65.89%, the load identification precision of an NILM module is remarkably superior to that of CNN, LSTM and a single-task GRU model (MAE is as low as 1.574 W, and the F1 score is as high as 0.973), and intelligent and economical operation of a low-voltage transformer area can be effectively promoted.