79 results about "Sequence modeling" patented technology

A text sequence recommendation method and system based on a large language model is disclosed, belonging to the technical field of recommendation algorithms. The method includes: a data preprocessing stage, a large language model pre-training stage, a sequence model fine-tuning stage and a matching stage. According to this disclosure, a large language model is introduced into a text sequence recommendation task, so that text can be better modeled by utilizing rich pre-training corpus of the large language model; meanwhile, sequence modeling is performed on the text, the capability of sequence recommendations modeling in a large model is activated, an ID-based recommendation paradigm in a traditional recommendation algorithm is eliminated, and recommendation task learning processing is better performed in a cold start scenario and a knowledge transfer scenario; and finally, a recommendation result is finally optimized by a sequence model.

This invention discloses an IoT-based energy management system for carbon black production equipment. The system includes: collecting energy consumption data and production condition data and unifying timestamps to form a multi-source synchronous data stream; performing standardization and resampling to form an online input dataset; constructing a time-delay structure to expand samples and form a causal discovery training sample set; initializing a directed sparse causal gating matrix and time-delay selection parameters and embedding a MambaTS selective state-space long sequence modeling model to form a parameterized association model; performing state recursive calculations to generate a state representation sequence and a model output dataset; and updating the causal gating parameters based on the model output dataset to form an updated state representation sequence. This invention achieves online consistent association modeling of energy consumption data and production condition data, improving the reliability and operational stability of online energy management associations.

A quality-aware dynamic ECG Mamba state update method is proposed. The method inputs the signal quality index, motion state information, and preprocessing confidence level into the quality control branch of a state-space sequence modeling network. Based on the output of the quality control branch, at least one of the state writing intensity, state forgetting intensity, and state propagation intensity of the state-space sequence modeling network at each time point is adaptively modulated. The method outputs a long-term dynamic ECG representation based on the modulated state update result, along with event detection results or classification results. By implementing quality constraints on the state-space writing, retention, and forgetting of long-term dynamic ECG sequences, with control variables derived from SQI, IMU, and preprocessing confidence level, the method avoids low-quality segments contaminating the long-term memory state, effectively improving the problem of false accumulation in long Holter / patch / motion monitoring sequences.

The present application relates to the technical field of remote sensing image processing and artificial intelligence, in particular to a state space model driven remote sensing image change caption generation method, the method comprises the following steps: obtaining image data and word mapping table of remote sensing image intelligent interpretation; extracting double time phase features through a double time sharing branch feature extractor; constructing a joint input sequence of a decoder according to the difference enhanced features, multi-scale visual features and text sequences; taking the constructed joint input sequence as input, calculating word probability distribution by autoregressive calculation based on a pre-trained language model decoder, completing prediction and cyclic iteration according to the word mapping table, and obtaining complete image change description sentences; the present application solves the problem of losing high frequency details in long sequence modeling of state space model through frequency domain modulation, and effectively suppresses background noise interference by using difference perception aggregation, thereby improving the accuracy of remote sensing image difference description generation.

This invention discloses an intelligent detection and location method for underground hard foreign objects based on dynamic feature fusion, belonging to the field of ground-penetrating radar technology. The technical solution includes: acquiring and processing ground-penetrating radar images; using a lightweight model based on an improved YOLO11 for real-time identification, which reconstructs the backbone network through a GhostNetV3 module and introduces a dynamic weighted attention mechanism to achieve feature fusion; utilizing an adaptive feature fusion algorithm to enhance the identification of small targets; employing a Transformer architecture for spatiotemporal sequence modeling and combining multi-source sensor fusion of GPS-RTK and IMU to achieve centimeter-level positioning; and finally, automatically marking the target on the ground using an integrated spraying execution unit. This invention achieves a fully automated closed-loop process for the detection, identification, location, and marking of underground hard foreign objects.

Various methods and processes, apparatuses or systems, and media for classifying human behaviors in various domains by using ensemble learning to perform sequence modeling with respect to sequence data are disclosed. The method includes: receiving a first set of data; partitioning the first set of data into a set of respective data streams, each respective data stream corresponding to a respective agent; extracting, from a first data stream, a first sequence of observations that relates to a first agent; inputting the first sequence of observations to each of several models that are trained by using historical data relating to the first agent; using the models to generate a composite score that relates to the first sequence of observations; and determining, based on the composite score, whether the first sequence of observations indicates at least one anomaly that relates to a behavior of the first agent.

The application discloses a critical patient sedation state evaluation method based on a multi-modal network model, relates to the technical field of medical monitoring and signal processing, and comprises the following steps: analyzing a plurality of channel physiological signals of a patient, constructing a dynamic time-varying function network, and performing multi-scale community detection to identify stable function modules in the network and dynamic connection modes thereof. The network topology mode is fused with instantaneous phase information of the signals to form a fusion feature space-time atlas. After the atlas is subjected to deep feature extraction and structured coding, a sequence modeling network is used to capture time sequence dependence, and finally a quantitative sedation evaluation signal is output by a decoder. Through dynamic modular structure analysis of the brain function network and cross-level feature fusion, the method realizes more continuous and accurate objective evaluation of the sedation state.

The application discloses a deep learning-based time-space fusion early warning method for deep rock burst, and constructs a dual-branch Transformer-CNN time-space feature fusion early warning model. The model captures local mutation features in a microseismic sequence through a parallel multi-scale CNN network branch, combines a Transformer network branch to extract global evolution features, realizes comprehensive mining of rock burst precursor information and identification of key precursor information, and further enhances the dynamic adaptability of the early warning model to complex microseismic sequences by using a self-adaptive gating fusion mechanism, thereby enhancing the adaptability of the early warning model in a complex monitoring environment. In addition, a learnable position coding is introduced to replace a traditional fixed coding mode, so that the early warning model can more flexibly capture time sequence dependence in mining stress evolution, and the problems of gradient vanishing and low calculation efficiency of a traditional cyclic network in long sequence modeling are overcome. Finally, the classification accuracy of the rock burst danger level and the early warning reliability are effectively improved.

This invention discloses a welding defect segmentation method based on frequency domain information guidance and Mamba sequence modeling, belonging to the field of image processing technology. It includes: constructing a welding defect image dataset; preprocessing the image dataset and dividing it into a training set, validation set, and test set in a 6:2:2 ratio; constructing a welding defect segmentation network based on frequency domain information guidance and Mamba sequence modeling; training the network using the training and validation sets, employing a cross-entropy loss function and the Adam optimizer, and using precision, recall, F1-score, and accuracy as evaluation metrics; finally, inputting the test set into the trained model to output high-precision welding defect segmentation results. This invention significantly improves the accuracy and robustness of welding defect segmentation by introducing wavelet transform to decompose frequency domain information, combining Mamba for efficient sequence modeling of low-frequency components, and utilizing dynamic serpentine convolution to enhance high-frequency details.

The application discloses a refrigerated cabinet multi-sensor heterogeneous data fusion method based on artificial intelligence, which comprises the following steps: collecting multi-source heterogeneous data in the operation process of the refrigerated cabinet, and constructing a unified event time axis; pre-processing multi-modal data, combining a heat-air flow-electricity model to generate a physically enhanced multi-modal tensor; constructing a disturbance activation vector and a modal mask graph and embedding them into the multi-modal tensor; inputting the disturbance enhanced input data into an improved Tide model to perform long sequence modeling, and outputting a state prediction sequence; performing prediction residual analysis, constructing a dynamic anomaly scoring function, and fusing the disturbance activation vector and the modal mask graph to generate an abnormal type label and a confidence fusion state representation; and using a federal training framework for sparse gradient upload, parameter aggregation and structure synchronization among multiple devices. The application can improve the multi-modal data fusion precision and abnormal recognition reliability of the refrigerated cabinet, and realize cross-device collaborative intelligent optimization.

This invention discloses an adaptive prediction method and system for shield tunneling parameters based on the Transformer architecture. The method includes data acquisition, data preprocessing, differentiation between active and passive variables, construction of time-series samples, Transformer model construction and training, joint prediction of multiple parameters, adaptive learning, and output and visualization of prediction results. The system includes a data acquisition unit, a data preprocessing unit, a unit for differentiating between active and passive variables, a unit for constructing time-series samples, a Transformer model training and prediction unit, an adaptive learning unit, and a prediction result output unit. By acquiring multi-source operating parameters in real time and performing systematic data preprocessing, decoupling control features and geological feedback features, constructing time-series samples using a sliding window, and inputting them into the Transformer architecture for training and achieving joint prediction of multiple variables, this method solves the problems of weak long-sequence modeling ability, insufficient characterization of multi-variable coupling, and poor real-time performance of traditional methods, significantly improving prediction accuracy and engineering applicability.

This invention discloses a knowledge graph completion method and system based on path-structure fusion, relating to the field of data storage technology. The method includes: at the data preprocessing level, a path filtering strategy based on PMI (Progressive Mindset Analysis) is used to accurately eliminate redundant and noisy paths by quantifying the co-occurrence correlation between entities, providing high-quality semantic context for subsequent inference. Secondly, during path modeling, the Mamba state-space model is used for path sequence modeling, and multi-level attention is used to aggregate path information. Thirdly, at the decoding and scoring level, a dual-tower architecture of path-structure is constructed, adaptively fusing and scoring global path semantic features with local graph topological features. Finally, through a mechanism of attribute completion—similar node determination—bidirectional path search, high-scoring interpretable paths are automatically mined from known entities and their similar nodes, and a hybrid scoring formula is used for comprehensive evaluation, ultimately outputting the inference link.

The application discloses a power load prediction method and system based on multi-stage time sequence preprocessing and double-branch wavelet Mamba, and relates to the technical field of intelligent power grids and time sequence analysis. In view of the problems that the existing power load prediction technology has insufficient long sequence modeling capability, low prediction robustness and precision, and cannot simultaneously consider global trend fitting and local mutation capturing when facing high-noise non-stationary data, a high-quality target sequence is first constructed through multi-stage time sequence preprocessing, and after channel independence and block embedding processing, the high and low frequency characteristic components are decoupled through discrete wavelet transform, the global long-range dependence is extracted through a bidirectional trend Mamba module, the local mutation characteristics are purified through a detail Mamba module, and noise is suppressed, and finally, the prediction result is output through inverse wavelet reconstruction. The application significantly improves the precision and robustness of long sequence power load prediction and reduces the computational complexity.

The application discloses a fall risk assessment method and system based on individual information, and relates to the technical field of medical assessment and artificial intelligence, and comprises the following steps: a data acquisition step, a support period detection step, an individual feature fusion step, a gait cycle segmentation step, a feature extraction step, a time sequence coding step, and a risk assessment step. First, multi-channel foot pressure time sequence data of a subject and individual information of the subject are acquired, then through five-layer progressive processing, problems of support period adaptive detection, gait cycle accurate segmentation, multi-dimensional feature extraction, time sequence dynamic modeling and adaptive model training are solved respectively, and finally precise individual fall risk assessment is realized. The application automatically adapts to the threshold of subjects with different physiological characteristics, captures the gait deterioration trend caused by fatigue through cycle-level double-flow time sequence modeling, and realizes adaptive personalized feature learning through layer-by-layer fusion of cascaded individual information.

The application discloses a liquid level height automatic calculation method and system based on spectral characteristics, a medium and equipment, and particularly relates to the technical field of container liquid level automatic calculation and dynamic monitoring. The method improves the algorithm through spectral curve change point detection, accurately captures the spectral feature difference between gas phase and liquid phase through spectral sequence modeling, introduces the running length index and dynamic updating mechanism, filters out the real gas-liquid interface change point through the multiple check mechanism to filter the false change point, infers the type according to the mean difference before and after the change point and the running length change, takes the clear mutation point or the midpoint of the gradual transition zone as the interface, combines the pixel resolution and the bottom starting position, and obtains the actual liquid level height through formula calculation, establishes the conversion relationship between the pixel quantity and the actual liquid level height through the proportional relationship between the image axis pixel number and the absolute height of the container, and finally takes the average value of the calculation results of three wave bands as the absolute height of the liquid level based on the pixel number of the spectral curve feature point.

The application provides a probiotic tolerance characteristic prediction method and device, electronic equipment and storage medium, relates to the technical field of probiotic screening, and comprises the following steps: obtaining genomic sequence data of a to-be-predicted probiotic; performing nucleotide subsequence fragment extraction on the genomic sequence data based on a plurality of preset length sliding windows to obtain a nucleotide subsequence fragment set corresponding to each preset length; mapping each nucleotide subsequence fragment to an integer token based on the frequency of each nucleotide subsequence fragment in the genomic sequence data to obtain a token sequence; and inputting the token sequence corresponding to each preset length into a tolerance characteristic prediction model to obtain a tolerance characteristic prediction result. The method and device provided by the application convert complex genomic sequences into digital sequences containing rich information, utilize the powerful sequence modeling capability of a deep learning model, and realize automatic, high-throughput and high-precision prediction of probiotic tolerance characteristics.

This application discloses a method, apparatus, equipment, and storage medium for predicting the moisture content of tobacco shreds at the outlet, relating to the field of tobacco technology. The method addresses the challenge of accurately predicting the moisture content of tobacco shreds at the outlet due to the strong coupling of multiple variables, nonlinearity, and large time delays in the tobacco drying process. It also addresses the pain points of existing mechanistic models (insufficient generalization), traditional recurrent neural networks (RNNs) prone to gradient anomalies in long-sequence modeling, and the high computational cost and difficulty in adapting to the real-time and deployment requirements of mainstream attention models for industrial online soft measurement. This method constructs a standardized prediction model system adapted to the time-series characteristics of tobacco drying. Through full-process time-series data normalization and core feature mining, feature biases caused by noise and time delays are eliminated. Lightweight long-sequence modeling optimization balances long-range dynamic capture capability and computational cost, avoids gradient anomalies, breaks through quadratic complexity limitations, reduces resource consumption, and provides stable soft measurement data for outlet moisture content.

The application relates to the technical field of deep learning, in particular to a memory state prediction method based on contrast learning and a stabilized long short-term memory network and application thereof. In the data preprocessing stage, three disturbances of label flipping, time disturbance and probability offset are performed on the sequence of review sequence data pairs, so that the sequence data has robustness when facing noise labels, generalization ability under different review time offsets, and stable convergence when observation errors and recording deviations occur; the stabilized long short-term memory network is used to extract features of positive and negative samples after review sequence and data enhancement processing, so that the numerical stability is ensured when a long sequence is modeled; the positive and negative sample features and the overall features of the sequence are combined for loss calculation and model optimization, so that the model can shorten the distance between positive sample pairs and lengthen the distance between negative sample pairs in the embedding space. The application aims to solve the problem of how to realize memory state prediction with noise robustness and stable time sequence modeling capability.

Embodiments of the present application provide a transaction behavior identification method, device, equipment, storage medium and program product, relating to the fields of artificial intelligence and financial technology. The method comprises: acquiring time series data corresponding to a transaction behavior, the time series data comprising transaction feature vectors of a plurality of time steps; performing feature extraction on the transaction feature vectors through a self-attention mechanism to generate global features, the global features containing global dependency relationships; inputting the global features into a bidirectional time series modeling module to extract local time dependencies of the global features through forward and reverse sequence modeling to generate bidirectional hidden states; and performing classification processing on the bidirectional hidden states to obtain an abnormal probability of the transaction behavior being abnormal. The method of the present application significantly improves the identification capability of hidden abnormal transaction behaviors, solves the performance deficiency problem caused by single models in traditional methods, and achieves high-precision, low-false-alarm-rate abnormal transaction behavior detection effects.

The application belongs to the field of traffic flow prediction, and provides a city traffic flow prediction method, system and product under complex road network conditions, and the technical scheme is as follows: time mixing features are extracted based on city road network traffic flow time series data, intersection information between different traffic features is captured based on the time mixing features, mixed feature representation is obtained by fusing the time mixing features and the intersection information between different traffic features; global dependence in a long sequence is captured by selective state space modeling based on the mixed feature representation, and dynamic time series features are output; after the dynamic time series features are enhanced, a trailing time dimension is projected to obtain prediction values of each feature variable of the traffic time series data. In city traffic flow prediction, data can be dynamically processed according to real-time traffic conditions, multivariate information can be effectively utilized, and efficient long sequence modeling can be performed, and the prediction accuracy of time series data is significantly improved.

This invention provides a lightweight adaptive attack chain extraction method, system, and storage medium for cross-text-log heterogeneous data sources. The method includes: Step 1: Attack chain extraction across text-log heterogeneous data sources; automatically extracting standardized attack technique chains and tactical chains from unstructured CTI reports and structured network alarm logs; Step 2: Adaptive prediction of tactical chains based on TCKC graph constraints and multi-reward PPO; dynamically predicting the attacker's next tactic using reinforcement learning to generate a high-confidence tactical evolution path; Step 3: Attack technique chain prediction based on sequence modeling and dynamic filtering using mapping rules; guided by the tactical chains predicted in Step 2, using a Transformer model to predict the specific technique sequence the attacker will use. The beneficial effects of this invention are: improved automation capabilities for heterogeneous data sources and realization of high-confidence dynamic tactical prediction.

Systems and methods of analyzing and predicting emotion changes for early detection, prediction, and proactive intervention in mental health care are provided. A method for analyzing and predicting emotional dynamics may include collecting, into a sequence-modeling computing system, a plurality of information channels comprising contemporaneous Instant Emotional State Information (IESI) captured over time from an individual and further data from the individual; training, by the sequence-modeling computing system, a set of model parameters using the plurality of information channels, wherein the set of model parameters are jointly learned through a multi-task learning process applied across the plurality of information channels; and performing machine-learning analysis configured to generate at least one detection, prediction, or intervention-related output and to facilitate at least one early detection, prediction, and proactive intervention in mental-health contexts based on the IESI and the further data.

The user preference analysis system based on shopping cart data of the present application, the system comprises a data acquisition module for collecting the interactive operation of the user and the shopping cart in real time; a behavior sequence modeling module for receiving the operation event signal, encoding the user operation sequence through a pre-trained time sequence analysis model, and generating a behavior feature vector signal; a preference analysis module for receiving the behavior feature vector signal, generating a dynamic preference signal reflecting the real-time interest tendency and decision state of the user through a multi-dimensional preference calculation model; a strategy generation module for receiving the dynamic preference signal, generating a strategy control signal containing personalized recommendation or marketing instructions according to pre-defined business rules; and a user portrait updating module connected to the data acquisition module and the strategy generation module respectively, receiving the operation event signal and the strategy control signal. The user preference analysis system based on shopping cart data of the present application can solve the problem of shallow, static and one-sided use of user shopping cart data in the prior art.

The application provides a panoramic image super-resolution reconstruction method and system based on improved Mamba, and the method comprises the following steps: shallow layer features are extracted by using convolution operation; the shallow layer features are input into a deep layer feature extraction module, and deep layer features are extracted by using a feature processing unit integrating three collaborative mechanisms; and after the shallow layer features and the deep layer features are fused, upsampling reconstruction is performed; the application realizes spatial domain local geometric perception enhancement through a geometric distortion perception window attention mechanism; the geometric distortion modulation Mamba mechanism uses a geometric distortion degree graph to perform affine transformation on the features, dynamically modulates key parameters of a state space model, and realizes spatial domain global distortion adaptive sequence modeling; the block DCT domain scanning Mamba mechanism constructs frequency-based ordered association between feature blocks through a specific scanning path in the frequency domain, and realizes frequency domain structured sequence modeling; through the collaboration of the three mechanisms, the application effectively improves the feature representation capability and reconstruction quality of the model for panoramic images.

This invention relates to a deep learning-based method and medium for recognizing VIN codes on vehicle body stamps. The constructed character recognition model includes: an input layer for receiving VIN code images; a feature extraction layer for performing multi-branch convolutional fusion and downsampling on the VIN code image to obtain a main feature map, and then performing edge enhancement and sequence modeling on the main feature map to obtain an enhanced feature map and character space features; a feature fusion layer for performing residual fusion on the main feature map, enhanced feature map, and character space features to obtain fused features; and an output layer for outputting VIN code characters based on the fused features. This provides a novel recognition technology solution that can adapt to complex working conditions, fully utilize the structural characteristics of VIN codes and the logical relationships between characters, and possesses high accuracy and high robustness.

This invention discloses a method for generating missing modalities in brain MRI based on hypergraphs and attention mechanisms. This method addresses the issue of missing modalities in multimodal brain MRI sequences such as T1, T1ce, T2, and FLAIR in clinical scenarios by constructing a unified multi-input multi-output translation framework. The method introduces hypergraph convolution and region-level self-attention into the generative network, aggregating group-level features from different tumors and healthy tissues and modeling structural relationships between sub-regions, preserving the spatial tissue structure of the tumor. Combined with bidirectional Mamba sequence modeling, it enables the 2D network to efficiently capture long-range dependencies between layers of 3D volumetric data, ensuring voxel-level spatial continuity. Through a teacher-student knowledge distillation mechanism, the student network learns structurally perceptual features without a tumor mask, achieving high-fidelity generation. This method can improve the overall quality of generated images and the detail fidelity of tumor lesions, thereby enhancing the performance of downstream tumor segmentation tasks and showing promising application prospects.

The application discloses a kind of two-dimensional space to three-dimensional space key point mapping conversion method and system, including sequence preprocessing module: the length of two-dimensional space key point sequence is obtained, carries out sparse preprocessing, obtains sparse two-dimensional space key point sequence;Mapping dimension-increasing module: sequence modeling is carried out to sparse two-dimensional space key point sequence, obtains three-dimensional space key point sequence after preliminary mapping;Sequence aggregation module: three-dimensional space key point of center target is obtained by compressing and aggregating three-dimensional space key point sequence after preliminary mapping;Training network module: end-to-end training is carried out to network using supervised learning, and mapping accuracy of three-dimensional space key point estimation network is improved using space-time constraint strategy;Visualization module is used to obtain visual result.The application has good robustness and universality, and can be widely applied in key point detection of a variety of objects.

The application discloses an intelligent grating microstructure parameter measurement method and system based on a MaLSTM hybrid model, and belongs to the fields of optical precision measurement, micro-nano manufacturing detection and artificial intelligence application. The method comprises the following steps: collecting a diffraction efficiency spectrum of a grating through a front-end data acquisition module; analyzing the spectrum data by using an intelligent processing center with a built-in MaLSTM hybrid model; the MaLSTM hybrid model extracts long-sequence global features of the spectrum through a linear complexity selective state space mechanism of a Mamba sequence modeling module, and finely captures local dynamic time sequence dependence of the spectrum features through a gating mechanism of an LSTM time sequence dynamic module, and then directly maps and outputs parameters such as duty ratio and groove depth of the grating by using a full-connection regression module; and finally, the real-time presentation of the measurement result is realized through a back-end visualization module. The application realizes pure data-driven end-to-end measurement, does not need physical model iterative inversion, and has the characteristics of non-destructiveness, high precision and strong robustness.

This invention relates to the field of material deodorization control technology, and more particularly to a continuous deodorization adaptive control method and system based on deep learning. It involves simultaneously collecting multi-source sensor data from the silo and heating chamber during continuous deodorization production, constructing a time-series feature sequence based on a sliding time window, and then predicting the current deodorization completion rate and future short-term production capacity range through time series modeling and regression calculation. Furthermore, it calculates a residence time correction coefficient based on the predicted deodorization completion rate and determines the maximum feasible feeding speed by combining it with the predicted production capacity range, forming dynamic control parameters. These parameters are then integrated with real-time weighing data to adjust the feeding frequency, transfer timing, and transfer weight. Additionally, it vectorizes the residence time distribution, filling rate, temperature, vacuum degree, and torque fluctuations within the heating chamber, performs regression assessment on the risk of agglomeration, and adjusts the control parameters according to the risk level to reduce the probability of agglomeration and improve the stability and automation level of the continuous deodorization process.