121 results about "State sequence" patented technology

The invention relates to the technical field of exception prediction, in particular to a storage cabinet exception trend prediction system based on time series data analysis, which comprises a state monitoring module, an interval sensing module, a path reconstruction module, a symptom activation module and an evolution prediction module. According to the method, the state vectors including the temperature, the voltage, the current and the door lock state are constructed and combined with the timestamp information to form the time sequence data sequence, and the dynamic expression mode of state change is established; a jump characteristic is analyzed by using a ratio of a time interval to a state change amplitude, a short-time disturbance path and a trend evolution path are distinguished by combining a jump rate statistical index, and an evolution activation signal is identified based on trend maintenance and non-fallback characteristics. On the basis, a neural network structure with long-time dependent learning ability is introduced to capture an aperiodic thermal anomaly trend in a state sequence, and the accuracy and timeliness of anomaly recognition are improved through multi-dimensional parameter cooperative processing and path construction logic.

The invention provides a material detection operation behavior identification method based on multi-modal perception, and belongs to the field of artificial intelligence and electric power operation and maintenance detection, and the method comprises the steps: carrying out the feature extraction of the preprocessed visual, motion and voice modal data, and obtaining visual, motion and voice modal features; the visual, action and voice modal features are input into the trained fine-grained behavior segmentation network, the multi-modal input layer is used for carrying out weighted fusion on the visual modal features, the action modal features and the voice modal features to obtain multi-modal global feature representation, and the time sequence modeling layer is used for carrying out frame-by-frame analysis and labeling on the multi-modal global feature representation to obtain the multi-modal behavior segmentation network. The hidden state sequence is output, and the classification output layer is used for classifying the hidden state sequence and recognizing behavior categories; matching and aligning the behavior type with a preset standard operation instruction, and judging whether the operation behavior is abnormal or not; the invention further provides an identification system. And accurate identification and compliance evaluation of the operation process are realized.

The invention discloses a short video content intelligent generation method based on deep learning, and the method comprises the following steps: obtaining short video data and candidate material frame data, and organizing and generating a candidate material frame sequence; inputting the candidate material frame sequence into an improved Slot-VAE model to generate an object-level potential slot representation set; inputting the short video data into a shot-level semantic coding network, and constructing a double-path latent variable structure; establishing an object-level slot evolution module for time sequence modeling; executing joint training, and updating parameters of the model and the network; inputting the target short video data into a shot-level semantic coding network, and generating a target shot-level semantic slot set and a target object-level slot time sequence potential state sequence; and generating a target short video frame sequence in the object-level decoding network and executing post-processing to generate a short video content intelligent generation result. According to the invention, the lens semantic consistency and the object time sequence continuity are improved.

The invention relates to the technical field of encrypted communication, and discloses an intelligent analysis method and system based on multi-source data, and the method comprises the steps: obtaining multi-dimensional metadata of an encrypted communication flow, and generating a feature vector set; calculating a state transition probability of a communication session by using a Markov chain model, dividing the communication session according to a time window and mapping the communication session to a discrete state, and generating a session state evolution sequence; analyzing a session state evolution sequence based on a hierarchical topic model, mapping the state sequence to a behavior topic through potential Dirichlet allocation, mapping the behavior topic to an intention category through a hierarchical Dirichlet process, and outputting a communication intention category and a confidence score; according to the method, the limitation that static feature analysis cannot reflect the complete life cycle of the session is overcome, the problem of semantic gaps caused by limited metadata information dimensions is solved, and the accuracy and robustness of intention recognition are improved.

A system uses a machine learning based model such as a neural network for transcribing audio inputs. The system receives a set of audio inputs representing utterances of a conversation. For each conversation, the system determines a dialogue state for each utterance. The system uses a hierarchical language model for transcribing audio inputs of an online conversation using the received conversations. The hierarchical language model includes a top-level language model and a plurality of lower-level language model. The training is performed by (1) training the top-level language model using sequences of corresponding dialogue state, each sequence of dialogue states for a conversation, and (2) for each dialogue state, training a lower-level language model using utterances having that dialogue state. The system executes the hierarchical language model to transcribe audio input of new conversations.

The invention relates to the technical field of learning education, in particular to an AR-based personalized learning education auxiliary method and system. By collecting and synchronously processing multi-modal behavior data such as eye movement, gestures and head orientation, time sequence characteristics capable of accurately reflecting the learning state of a user are constructed, and then a hidden cognitive state sequence is decoded by using a hidden Markov model and inflection points of the hidden cognitive state sequence are recognized; finally, dynamic and automatic adjustment of learning contents is realized by means of a reinforcement learning model, deep cognitive state changes can be captured from continuous and dynamic user behaviors, accurate teaching intervention is timely performed at'inflection points' of key transition of cognitive states, personalized adaptive learning path planning is realized, and the learning efficiency is improved. The pertinence and effectiveness of learning are obviously improved; the technical problem that an existing AR learning system is difficult to intervene in time at an inflection point where a user cognition state is changed during path planning, so that real personalized learning path planning is realized is solved.

The invention relates to the field of man-machine interaction, and discloses a man-machine interaction method and system based on a large language model, and the method comprises the steps: extracting original semantic features from the current input of a user; obtaining a context representation of the current dialogue; extracting a historical emotional state sequence from the multi-round dialogue historical record; generating a current user emotion representation vector based on the context representation of the current dialogue and the historical emotion state sequence; constructing a hierarchical memory structure based on the current user emotion representation vector and the context representation of the current dialogue; adjusting the attention degree of the historical dialogue content stored in the semantic memory component according to the hierarchical memory structure; and generating a response sequence through a large language model based on the adjusted attention degree, the hierarchical memory structure and the context representation of the current conversation, and outputting the response sequence to the user. According to the technical scheme, the problems of stiff switching and dialogue breakage during emotion turning of a traditional dialogue system are solved.

The invention relates to an IEC61499 intelligent model segmentation training method based on DCU state enhancement, and belongs to the technical field of computer distribution. The method comprises the steps of constructing an intelligent model state enhancement module, transplanting an intelligent model to a DCU accelerator card, adjusting the number of attention heads through state vector complexity, carrying out parallel calculation, fusing and optimizing a multi-head attention matrix, sampling an optimal historical state based on Euclidean distance, and obtaining an enhanced state sequence in combination with a current state. Providing optimization state representation for segmentation learning; taking the enhanced state sequence as initial input, constructing and training a reinforcement learning agent, and enabling the reinforcement learning agent to learn and recognize an optimal segmentation strategy of the intelligent model; and deploying the trained intelligent agent, and performing intelligent splitting on the intelligent model according to the calculation load and the communication cost between the acceleration cards to realize distributed parallel training. The invention aims to solve the technical problems of unbalanced resource allocation, low model segmentation efficiency and redundant fixed attention mechanism calculation in a dynamic environment in the prior art.

The invention relates to the technical field of hoisting control, in particular to a hoisting control system and method for modularized fast-installed electrical equipment, which forcibly limits the sequence relation of the whole clamping and propelling process through a finite-state machine, so that the propelling direction, segmented propelling, syn-position recording and propelling and maintaining switching of a clamping component are all controlled by state constraint; through Markov decision, subsequent state transfer is associated only according to current state information in a holding stage, so that reverse displacement change in the holding stage is not judged by a single-point threshold value any more, but is continuously associated in a state sequence form, the conditions of syn-position jumping, advanced holding and repeated advancing in clamping action are avoided, and the clamping efficiency is improved. The development trend of reverse displacement is described through inter-state transfer records, so that formation of a keeping interval is established on the basis of continuous state consistency, continuous transfer association based on the current state is formed in the keeping stage, and the clamping process has controllability in time dimension and space dimension at the same time.

The invention relates to a rotating position coding design method suitable for large-model long context understanding, which comprises the following steps of: performing linear mapping on a hidden state sequence in a large-language model to obtain a query vector sequence, a key vector sequence and a value vector sequence, and copying one query vector sequence, wherein a self-attention layer comprising multiple attention heads is arranged in the large language model; performing traditional rotation position coding on the query vector sequence and the key vector sequence to obtain a query vector sequence and a key vector sequence with position codes; based on the copied query vector sequence, rotation is carried out, then traditional rotation position coding is carried out, and a query vector sequence of position coding after rotation is obtained; and based on the query vector sequence of the position code after rotation, the query vector sequence with the position code, the key vector sequence and the value vector sequence, carrying out self-attention operation to obtain an output result of a self-attention layer, and completing the design process. The method has a good effect in both long and short tasks.

The invention belongs to the technical field of data processing, and provides a BIM data intelligent matching and conflict detection method, and the method comprises the steps: generating a fusion feature of a component through fusing a geometric topology feature and a semantic constraint feature; through alignment and coupling of the installation state sequence and the compliance state sequence, a space-time state sequence of the component is formed, so that the construction constraint graph can fit the actual dynamic change of construction; the method comprises the following steps: injecting a space-time state sequence into entity nodes by taking components as the entity nodes and a design specification and a construction process as constraint nodes, constructing a construction constraint graph, and predicting a conflict propagation path between the components through a time sequence graph neural network; the prediction result of the conflict propagation path is subjected to actual construction verification, the verified prediction result is subjected to influence tracing and grading, and the comparison data of the actual construction result and the prediction result is combined to adjust the parameters of the time sequence diagram neural network and the edge weight of the construction constraint diagram, so that the prediction accuracy of the conflict propagation path is further improved.

The embodiment of the invention provides a robot driving control method and device, electronic equipment and a storage medium, and belongs to the technical field of nonlinear system control. The method comprises the following steps: constructing a state prediction model according to a first observation state sequence, a second observation state sequence and a historical control input sequence, and calling the state prediction model to perform state error prediction based on the first observation state sequence, the second observation state sequence and the historical control input sequence, performing parameter adjustment on the state prediction model according to the predicted observation state error sequence and a preset dimension reduced-order matrix to obtain an error compensation state space model; based on the first given state data and the first control input data of the current time, an error compensation state space model is called to conduct state prediction, second given state data corresponding to the to-be-predicted time is obtained, and the target robot is driven and controlled according to the second given state data. According to the embodiment of the invention, the driving control accuracy of the robot can be improved.

The invention belongs to the technical field of large language models, and particularly discloses a large language model illusion detection method and system, a terminal and a medium. Comprising the steps of obtaining a middle hidden state of each token in a process of generating an answer by a large language model, and organizing and forming a hidden state sequence according to a generation sequence; performing time sequence preprocessing on the hidden state sequence to complete the preprocessing of the sequence; inputting the pre-processed hidden state sequence and a dynamic feature formed by a difference vector, a trend drift degree or local fluctuation intensity into a dynamic sequence modeling network, and enabling the network to extract a time dependence feature reflecting an internal state change mode based on an evolution relation of a hidden state along with a generation process; and generating an illusion risk score according to the comprehensive time sequence characteristics output by the dynamic sequence modeling network, and determining that the answer contains the illusion content when the score exceeds a preset threshold. The reliability of the output content of the large language model can be improved.

This invention discloses a spoken language understanding method based on multi-view expert fusion and interest lexical selection, belonging to the field of natural language understanding and semantic parsing technology. The method includes: extracting hidden state sequences from the input utterance using a shared encoder; constructing a multi-view expert fusion module containing utterance view experts, block view experts, and lexical view experts to generate multi-granularity expert features; generating intent aggregation features and slot aggregation features through a task decoupling gating mechanism; performing interest lexical selection on the intent aggregation features, calculating lexical importance scores, generating weighted intent representations, and predicting multi-intent labels; and inputting the slot aggregation features into a decoder with diagonal mask constraints to generate a position-aligned slot label sequence. This invention enhances the model's dynamic focusing ability on key intent signals and can be applied to intelligent dialogue systems, virtual assistants, and vertical domain semantic parsing scenarios.

The invention discloses an active intelligent dialogue pushing method and system based on user state perception, and belongs to the technical field of computers. The method comprises the steps of obtaining a trigger event; extracting multi-dimensional state information of the user according to the trigger event and the historical interaction data to obtain a structured snapshot; generating an action sequence according to the plurality of structured snapshots; searching according to the action sequence to obtain candidate contents; and screening high-score contents from the candidate contents for pushing. According to the user behavior analysis and active dialogue decision-making mechanism based on the time sequence state snapshots, the state timeline is formed by continuously generating the user state snapshots containing multi-dimensional information (demands, emotions and trends), the active dialogue opportunities are intelligently recognized based on the change mode of the state sequence, the information is planned, and the action sequence is obtained. And high-quality personalized dialogue content is generated.

The invention relates to the technical field of operator overflow control, in particular to an AI processor operator overflow optimization method and system.The method comprises the steps that a finite automaton is introduced into numerical values for state combination and sorting, a continuous state sequence covering an operator execution stack is generated, and an intermediate result obtains a state number carrying bit width section information and stack position information at the same time; the method comprises the following steps of: setting a section to which an intermediate value of each operator belongs to a discrete identifier in a fixed-point space, executing weighted truncation reduction transformation by introducing a graph neural network, and deducing a clipping weight under topological connection constraints among the operators, so that the intermediate value of each operator obtains differential correction strength in a longitudinal accumulation direction and a transverse link direction at the same time; an operator correction set is generated by guiding an intermediate numerical value through a clipping weight, so that the effective dynamic range utilization rate after right shift and truncation are matched and reduced is improved, and the stability and repeatability of a reasoning process in a long-chain accumulation scene and the tracking capability of an abnormal calculation behavior are enhanced.

The invention provides a disease progress probability prediction method and device based on a Monte Carlo-Markov chain, and relates to the technical field of disease progress prediction. The method comprises the following steps: acquiring multi-modal historical follow-up visit data of a patient to be tested, and constructing a state space with transfer constraint; extracting and checking time series data according to the state space; performing Monte Carlo sampling on the time sequence data, establishing a non-homogeneous transfer model and forming an individualized non-homogeneous transfer kernel set; completing path interpolation and probability filling of missing fragments based on the transfer kernel set to obtain continuous complete state sequence data; a Markov chain Monte Carlo algorithm is used for parameter updating, an individualized transfer model subjected to posteriori updating is obtained, a calibration probability result is obtained, and dynamic prediction and uncertainty evaluation of individualized disease progression are achieved. According to the method, the problems of inaccurate state determination, single transfer modeling, incomplete observation and insufficient prediction calibration in the prior art are solved.

The application discloses an interface fault early warning method based on state sequence statistical characteristics and a server, and relates to the field of operation and maintenance.The method comprises the following steps: collecting the physical layer, IPv4 layer and IPv6 layer states of each interface of a network device, and aligning the states in time sequence to form a multilayer state sequence. Intermittent jitter interfaces with frequent state changes and high availability are identified. Each state change is analyzed one by one, and it is judged whether each layer changes synchronously according to a preset time window, and marked as multilayer linkage or single layer independent change. The linkage change frequency and linkage synchronization rate are calculated according to a statistical unit, and two types of time sequence are generated. When the linkage synchronization rate in the observation window continuously exceeds the threshold value, the linkage change frequency sequence is linearly fitted to obtain a deterioration rate. If the deterioration rate is positive and exceeds the threshold value, the remaining available time window is calculated according to the rate, the current frequency and the unavailable frequency threshold value, and time prediction and fault early warning are output. By implementing the application, the ability to judge the continuous deterioration of the interface health condition can be improved.

The application relates to the technical field of operator overflow control, in particular to an AI processor operator overflow optimization method and system, state combination and sequencing are performed on a numerical value introduction finite automaton, a continuous state sequence covering an operator execution stack is generated, a state number carrying bit width section information and stack position information at the same time is obtained for an intermediate result, a section to which an operator intermediate value belongs is provided with discrete identification in a fixed point space, a graph neural network is introduced to perform a weighted truncation reduction transformation, a clipping weight is derived under a topological connection constraint between operators, each operator intermediate value is simultaneously provided with differentiated correction strength in a longitudinal accumulation direction and a horizontal link direction, the intermediate value is guided to generate an operator correction set through the clipping weight, the utilization rate of an effective dynamic range after right shift and truncation cooperation reduction is improved, and the stability, repeatability and tracking ability for abnormal calculation behavior of an inference process in a long chain accumulation scene are enhanced.

The invention relates to the technical field of information security, in particular to a sequential cipher security analysis method for a composite structure, which comprises the following steps of: acquiring round function parameters of a target cipher algorithm, nonlinear component definition and a key scheduling rule, and generating a standardized configuration data set; dividing the complete algorithm into a plurality of independent low wheel structures along the boundary of the nonlinear component according to the wheel function parameters, and constructing a directed graph data model containing state transition constraints to compress a state space; calling a password component vulnerability feature library to quantify a sub-structure active S box distribution abnormal value and a linear diffusion layer vulnerability path, and generating a risk weight priority ranking list; and inputting the state sequence data into the target application environment simulation platform, and generating a reproducible vulnerability report when the output deviation exceeds a security threshold. Through hierarchical compression, risk guidance and multi-environment verification, the technical defect of insufficient path coverage of a high-round-number lightweight algorithm in a resource-limited scene is solved.

The invention provides a complex system control method, device and equipment based on a diffusion model, and relates to the technical field of control systems. The method comprises the following steps: acquiring a current system state and a target system state of a complex system and initial system noise information sampled in a preset time period; inputting the current system state, the target system state and the initial system noise information into a denoising network in a diffusion model, and denoising the initial system noise information through the denoising network to obtain a plurality of target prediction system states of the complex system in a preset time period; inputting a prediction system state sequence formed by a plurality of target prediction system states into the inverse dynamic model, estimating two adjacent target prediction system states in the prediction system state sequence through the inverse dynamic model, and generating a corresponding control signal sequence; and performing finite time control on the complex system based on the control signal sequence. By adopting the technical scheme provided by the invention, the control effect of a complex system can be improved.

Embodiments of the present application disclose a method and device for stream security prevention and control based on large model generated content. In the embodiments of the present application, the hidden layer state of all intermediate layers when the large model generates the current word element is obtained; the hidden layer state sequence and the current risk state vector of all word elements in any layer of all intermediate layers are determined; the target risk state vector of the current word element is generated according to the hidden layer state sequence and the current risk state vector; and the intervention instruction is generated in response to the target risk state vector being greater than or equal to a set threshold, wherein the intervention instruction is used to instruct the large model to stop generating the current word element or replace the current word element. Through the above method, efficient and accurate risk detection can be performed on the current word element before it is generated.

The invention discloses a task self-adaption-based multi-modal fusion garment matching recommendation method and system, and the method comprises the steps: S1, splicing the image features, original text features and fashion perception text features generated by a multi-modal large-language model of each single item in a set of garment matching, and carrying out the splicing of the image features, original text features and fashion perception text features; a fine-grained multi-modal feature vector is formed; s2, combining the multi-modal feature vectors of all single items in clothing matching with a learnable specific task token to form an input sequence, and encoding the input sequence through a Transform encoder to capture a context dependency relationship among the single items to generate an output sequence containing the hidden state of each single item; s3, introducing a learnable global query vector, and actively selecting and integrating significant information from the hidden state sequences of all single items through a cross attention mechanism to generate a global matching representation; and S4, inputting the global matching representation into the prediction model, and outputting a result of the clothing matching task.

The application provides a mixed supervision language model distillation method based on bridge domain-park domain decomposition, comprising the following steps: constructing a prefix state sequence, and saving context and task identification. The state information is used for subsequent partition discrimination and error statistics. The probability distribution, hard label, confidence and entropy statistics of the teacher model are obtained for each state. The state risk is calculated, and the bridge domain and the park domain are divided to form a partition label. The partition label can be reused for training weighting and slice analysis. Soft supervision and hard supervision are calculated at the state level, and are dynamically weighted. The joint optimization process keeps the single model parameter space solution, and reduces the process complexity. The final model is selected according to the accuracy, stability, time delay and cost, and an auditable evaluation report is formed. In the evaluation stage, unified benchmarks and fixed random seeds are recommended to improve comparability.

The application provides a kind of based on streaming AI intelligent cache and breakpoint continuation method and system, the method includes: through the real-time encoding of the operation sequence of user and corresponding data to be uploaded by encoder, generates latent state sequence;The latent state sequence is input into the preset prediction model, and the prediction result is output, the prediction result includes: user next operation intention and affected data range;According to the latent state sequence, periodically generate and persist first breakpoint state snapshot;When breakpoint continuation is needed, the second breakpoint state snapshot selected by user from the first breakpoint state snapshot is loaded, and the replay operation sequence and replay data corresponding to the second breakpoint state snapshot are replayed based on the corresponding decoder of the encoder.

This application discloses an autonomous driving scene generation method based on a spatiotemporal decoupled world model, belonging to the field of autonomous driving technology. The method includes: firstly, acquiring a multimodal autonomous driving dataset; after preprocessing to construct a state sequence; then, discretizing the state sequence into a pose word sequence and an image word sequence using an equidistant binning strategy and an improved time-aware vector quantization encoder; subsequently, inputting the word sequence into a spatiotemporal multimodal fusion module; and decoupling spatiotemporal information and predicting the potential state features of the next moment by alternately stacking temporal Transformer layers and spatial multimodal Transformer layers; during the training phase, employing a random masking strategy to prevent long-term generation drift, and calculating the cross-entropy loss between predicted words and ground truth words generated by the internal state autoregression module to update the model parameters; finally, generating autonomous driving scene data through binning inverse operation and a time-aware decoder; the method of this invention has the ability to generate long-term, high-fidelity videos.

The invention discloses a phonetic symbol generation method and device based on a deep neural network architecture and a storage medium, and the method comprises the steps: predefining a universal Chinese phonetic symbol system based on Pinyin as a unified output standard, and constructing a modeling architecture comprising a parallel text encoder, an acoustic encoder and a single decoder; the double encoders respectively process text and audio input and generate a pronunciation detail / acoustic feature hidden state sequence consistent in dimension format, the single decoder converts and outputs Chinese phonetic symbols conforming to Chinese cognition based on the sequence and a universal Chinese phonetic symbol system, and the hidden state sequence provides deep features for the decoder to support accurate alignment. According to the phonetic symbol generation method based on the deep neural network architecture, output is visual and easy to understand, cognition of a Chinese user is fitted, the learning threshold is greatly reduced, and error accumulation of a traditional process can be avoided.

The application relates to the technical field of multi-level context intelligent construction based on artificial intelligence, and discloses a multi-level context intelligent construction method based on artificial intelligence applied to a text scene. A complete context intelligent modeling system is constructed, which covers processes such as normalized feature construction, adaptive spatial topology establishment, prototype state division, single-point and edge potential calculation, time transition modeling, energy function construction and improved ICM reasoning. The method starts from original context data and finally realizes state sequence recognition of nodes changing with time, and is strong in systematicness. Through fusion of three types of potential energy of nodes, edges and time, and cooperation of a point-by-point optimization reasoning mechanism, the recognition precision and operation efficiency are improved. The method has good adaptability, interpretability and wide application potential, and is suitable for context perception tasks in multi-source, multi-node and multi-time scene.

The application discloses a kind of state sequence-based relay protection test system test method, belong to power system relay protection technical field, solve how to reduce the error problem when test system is tested to long time test scheme;The state sequence in test scheme is decomposed into multiple fixed time intervals, when running to set time, test data processing module and power signal source carry out data interaction, test data processing module compares the time data contained in received data with the time of self timing, if consistent, continue normal operation, if inconsistent, obtain its time data to control module, if the difference of test data processing module and control module meets the requirement, use the time of test data processing module as the latest running time, otherwise use the time of power signal source as the latest running time, correct in time when time error occurs, reduce the error of test result, and has good application prospect for high-precision test of relay protection equipment.

The application discloses a terminal equipment light-weight operation and maintenance method based on artificial intelligence, relates to the technical field of terminal equipment light-weight operation and maintenance, and comprises the following steps: in the case that state continuation writing data is missing, extracting a state sequence after hot start from the running state record in the current period, constructing a hot start residual piece mutation mapping tensor, and determining a running state change piece through a field stability score and a boundary disturbance gradient model; inputting the running state change piece into a credibility mapping network, generating a credible mapping vector through calculation of a cross-field residual value smoothing factor and a local mutation mode alignment rate, and judging whether the operation and maintenance has credibility based on the credible mapping vector. The application solves the problem that operation and maintenance judgment is inaccurate due to missing state data after terminal equipment hot start, and realizes accurate quantification and dynamic regulation and control of operation and maintenance credibility under the condition that data is incomplete.