24 results about "Time series representation" patented technology

Systems and methods for monitoring and evaluating time-series real-time environment data to create a high-resolution, high-fidelity actual (e.g., nowcast) and predicted (e.g., forecast) representation of an environment of interest. In some aspects, the system comprises instructions to obtain a set of real-time environment measurements stored in a data repository corresponding to a time-series capture of environment data across an observational time period, identify one or more precursory signals within the set of real-time environment measurements, determine at least one anomalous precursory signal from the one or more precursory signals that exceeds the corresponding signal threshold, generate a time-series representation of an actual environment state across the observational time period based on the at least one anomalous precursory signal and the set of real-time environment measurements, and display, at a user interface, the generated time-series representation of the actual environment state.

The invention provides a time sequence prediction method and device for performing multi-level text alignment by using a large model, and belongs to the technical field of time sequence prediction of a rail transit system based on the large model. Comprising the following steps of splitting multivariable time sequence input into a plurality of univariate time sequences according to feature dimensions, performing additive decomposition on each univariate time sequence, and performing fragmentation processing on each decomposed time sequence component; embedding the fragments into a text embedding space of a pre-training language model and aligning the fragments with the text embedding space; combining the structured prompt with the aligned time sequence representation to form input of a large model; and feeding the input combining the prompt and the alignment representation into the frozen large language model, obtaining an output representation of the model, and mapping the output representation into a final prediction result through a linear projection layer. According to the method, the time sequence data and the natural language modality are effectively aligned and fused, and the prediction accuracy and interpretability are remarkably improved.

The application discloses a steam turbine fault diagnosis method, comprising the following steps: S1, collecting multi-dimensional state parameters of the steam turbine, forming a time series data set, and preprocessing the data; S2, constructing a time series encoder based on a Transformer, which is used for learning the global dependence relationship and dynamic time series characteristics of the time series, and outputting a time series representation; S3, constructing a double residual branch, which is used for comprehensively capturing trend anomalies and sudden anomalies in equipment operation; S4, through a self-supervised contrast learning technology, the feature discrimination ability of the time series encoder to normal working condition samples is enhanced; S5, calculating the current comprehensive anomaly score of the steam turbine; S6, setting a fault diagnosis threshold according to the comprehensive anomaly score distribution of historical normal data. Through the innovative architecture of constructing a prediction and reconstruction double residual branch, multi-dimensional monitoring of the equipment operation state is realized.

The invention discloses a cattle herd oestrus prediction method based on time sequence deep learning, which comprises the following steps: collecting multi-source time sequence data of body temperature, motion steps, behavior tags and abnormal events, and constructing a first channel value sequence and a second channel event sequence; on the basis of a double-channel cross attention mechanism, time-dependent modeling is carried out on the two channels through parallel GRU, and behavior event information from the second channel is fused with the first channel as a query vector; based on a memory enhanced time sequence network, performing attention matching on the current fused time sequence representation and a historical estrus cycle vector set, and extracting a key historical memory fragment; current input and historical memory are fused through a gating mechanism, and finally an estrus probability prediction sequence is output through a full-connection neural network. According to the method, accurate identification and intelligent early warning of the oestrus state of the cattle herd are realized, and the multi-source data fusion capability and the time sequence modeling precision are improved.

The invention discloses an AI modeling data cleaning and obtaining method and system based on a probability strategy, and relates to the technical field of industrial time series data processing and artificial intelligence modeling, and the method comprises the steps: obtaining target time series data; constructing a time sequence statistical model containing a trend term; performing parameter estimation on the time sequence statistical model based on a probability strategy, and calculating a probability weight corresponding to each data point; performing weighted modeling on the time sequence statistical model to obtain cleaned time sequence representation; based on the cleaned time sequence representation, trend characteristic parameters reflecting the system operation state corresponding to the target time sequence data are extracted; according to the trend characteristic parameters, the dynamic state and the steady state of the target time sequence data are judged, cleaned dynamic data and cleaned steady data are obtained, and classified acquisition of the dynamic data and the steady data of the cleaned data is achieved. The precision, efficiency and adaptability of data processing can be improved.

The application provides a time series prediction method and device for multi-level text alignment using a large model, belonging to the technical field of time series prediction based on large models in rail transit systems. The method comprises the following steps: inputting a multivariate time series into a plurality of single-variable time series according to feature dimensions, performing additive decomposition on each single-variable time series, and performing fragmentation processing on each time series component after decomposition; aligning the fragments with the text embedding space of a pre-trained language model; combining the structured prompts with the aligned time series representation to form the input of the large model; feeding the input combined with the prompt and the aligned representation into a frozen large language model to obtain the output representation of the model, and mapping the output representation to the final prediction result through a linear projection layer. The application effectively aligns and fuses time series data and natural language modalities, significantly improving the accuracy and interpretability of the prediction.

The invention relates to a time sequence anomaly detection method based on dynamic memory perception. The method comprises the following steps: (1) carrying out data preprocessing on a photovoltaic power generation multivariate time sequence; (2) obtaining the preprocessed standardized multivariate time sequence data from the step (1); (3) obtaining a multivariate time sequence expression Se embedded in the step (2); a U-Net model is constructed, the U-Net model is processed by an encoder-decoder based on Transform and a dynamic adaptive memory module, and an original multivariate time sequence S is reconstructed. (4) constructing a photovoltaic power generation anomaly detection model based on dynamic memory perception, and training the model by using a photovoltaic power generation multivariate time sequence; and (5) evaluating the performance of the model based on three evaluation indexes including the accuracy rate, the recall rate and the F1 score. According to the scheme, the accuracy of photovoltaic power generation anomaly detection is improved.

The invention discloses a time sequence characterization method based on a time-frequency domain and comparative learning, and the method comprises the steps: obtaining time sequence data, and taking the time sequence data as a sample data set; a TS-TFC model is constructed, the TS-TFC model comprises an encoder, a linear projection layer, a time domain comparison module and a frequency domain comparison module, and the encoder comprises a random mask layer and an expansion convolution layer; the time domain comparison module comprises time comparison and instance comparison; performing training optimization on the constructed TS-TFC model to obtain a target TS-TFC model; and inputting time sequence data to be represented into the target TS-TFC model to obtain time sequence representation. According to the method, the TS-TFC model is constructed, time-frequency two-domain features are fused, a representation space structure is enriched, and the generalization ability of the model is enhanced; multi-level discrete wavelet decomposition is adopted to replace a traditional Fourier method, and the modeling effect on the non-stationary time sequence is improved.

The application discloses a kind of based on equipment operation memory curve Building mechanical and electrical abnormality prediction method and system, comprising: first, collect and preprocess equipment historical multi-source operation data, utilize time series representation learning to generate the benchmark memory curve of representation health state.Succeed in constructing space-time memory network model, learn the prototype of health operation mode by comparison learning strategy training model and form semantic representation space.Furthermore, the real-time operation data sequence is input into the model, and the deviation degree of the real-time operation data sequence and the health mode in the semantic representation space is calculated.Based on dynamic threshold and multi-scale prediction framework, the abnormal state, predictive warning level and confidence of the equipment are output according to the deviation degree.Finally, the benchmark memory curve and model memory bank are dynamically updated through online learning mechanism, and the interpretable abnormal report is generated by using causal reasoning model.The application realizes early and accurate prediction and root cause tracing of equipment abnormality, and improves the intelligent level of building mechanical and electrical system operation and maintenance.

The application belongs to the technical field of Internet, and particularly relates to an industrial Internet cross-domain metadata trusted fusion method and system, comprising the following steps: collecting structured metadata, time series metadata and relationship metadata of each industrial domain node, and constructing an intra-domain metadata graph based on field attributes and call relationships; performing semantic coding and time series coding on the intra-domain metadata graph, obtaining corresponding semantic representations and time series representations, and calculating inter-field consistency; identifying stable field clusters, drift field clusters and abnormal nodes according to the inter-field consistency in each industrial domain node; and performing segmented aggregation on the stable field clusters and the drift field clusters by each industrial domain node to generate anchor field summaries and drift trajectory summaries. The application combines metadata graph coding, field clustering and federal dynamic weighting, and realizes industrial cross-domain metadata private trusted fusion through abnormal compensation correction.

Software for the analysis of accumulation phenomena (1A) in an environment to be monitored (10); the software for the analysis of accumulation phenomena (1A) being storable in and executable by electronic processing resources (2), and designed to cause, when executed, these electronic processing resources (2) to become configured to receive a time sequence of representations of the environment to be monitored (10); where each representation of this time sequence represents the environment to be monitored (10) at a different time instant. In addition, the software for the analysis of accumulation phenomena (1A) is designed to cause, when executed, these electronic processing resources (2) to become configured to implement a segmentation model (6) configured to locate, or segment, one or more target areas occupied by one or more substances deposited on, or covering, a surface locatable in an input representation. The software for the analysis of accumulation phenomena (1A) is designed to cause, when executed, these electronic processing resources (2) to become configured to compute (block 8) an evolutionary scenario of a deposition, accumulation, or coverage phenomenon, based on the time sequence of representations of the environment to be monitored (10) and based on the segmentation model (6) implemented. The evolutionary scenario of the phenomenon is a time sequence of representations, of different future time instants compared to the current time instant, representing one or more substances deposited on, or covering, a surface present in an environment to be monitored (10). In addition, the accumulation phenomena analysis software (1A) is designed so that, when executed, these electronic processing resources (2) become configured to provide an output (9) based on this evolutionary scenario of the phenomenon.

The application discloses a kind of multi-element coupling under coastal area estuary water level prediction method, comprising: the time series representation of short-term dynamics and long-term rhythm is established, and spatial hydrological dependence information transmission mechanism is established by time series representation, to output estuary response prediction;The application models the spatial topology and physical coupling relationship between monitoring stations through "heterogeneous correlation body network structure", realizes the multi-level hydrological dependence expression from precipitation driving to water level response.This mechanism significantly improves the explainability and propagation accuracy of spatial characteristics, enabling the model to truly reproduce the transmission and diffusion behavior of hydrological quantities in natural basins.The proposed "periodic spectrum mapping mechanism" projects precipitation and water level time series from time domain to frequency spectrum space, and through principal frequency energy decomposition and feature reconstruction, realizes the extraction of periodic hydrological law.

The application belongs to the technical field of artificial intelligence and high-performance computing, and specifically discloses a financial time sequence representation method fusing dynamic sparse topology and parallel computing power optimization. Through the application, a multi-dimensional pattern capturer is used to identify typical market nonlinear states; an asymmetric dynamic sparse feature fusion unit based on content perception is used to generate self time sequence dependent feature tensors, and high-order hidden layer interconnection representation tensors are generated in combination with coarse-grained correlation feature tensors intercepted; a multi-objective decoding based model is used to generate financial sequence prediction results of target assets. Through the above method, the multi-dimensional pattern capturer is introduced, the feature tensors are strictly aggregated, the endogenous sparse self-attention is used to construct fine-grained time sequence dependence, the exogenous sparse cross-attention is used to perform coarse-grained feature dimension reduction and key information filtering, and the joint optimization model trained by the multi-objective joint optimization functional is introduced for prediction, so that the accuracy of representing financial time sequences can be effectively improved.

The invention discloses a space-time fusion photovoltaic power generation prediction method, system and device and a medium, and relates to the technical field of power systems, the method comprises the following steps: carrying out space-time and time conversion on obtained photovoltaic power generation data, generating a fusion embedding representation, and carrying out block processing to obtain a block embedding representation; inputting the block embedding representation into a space-time attention layer for transformation to obtain a target representation vector of each node, and outputting an initial power generation prediction result; calculating the Euclidean distance between each target node and each class cluster center represented by the fusion space-time embedding and carrying out dynamic tuning to determine the class cluster center corresponding to each target node; aggregating all nodes in each cluster center to obtain a class cluster time sequence representation; in cooperation with the class cluster time sequence representation and the time sequence representation vector, distributing weights for the prediction experts so as to output a target power generation prediction result; the optimal power generation plan of the target power system is determined according to the target power generation prediction result, and the scheduling reliability is improved.

The invention discloses a coastal area estuary water level prediction method under a multi-element coupling effect, and the method comprises the steps: building a short-term dynamic and long-term rhythm time sequence representation, building a spatial hydrological dependence information transmission mechanism through the time sequence representation, and outputting estuary response prediction; according to the invention, modeling is carried out on a spatial topology and physical coupling relationship between monitoring stations through a heterogeneous association body network structure, and multi-level hydrological dependency expression from rainfall driving to water level response is realized. According to the mechanism, the interpretability and the propagation precision of the spatial features are remarkably improved, and the model can truly reproduce the transmission and diffusion behaviors of the hydrological volume in the natural drainage basin. A proposed periodic spectrum mapping mechanism projects a rainfall and water level time sequence from a time domain to a spectrum space, and refining of a periodic hydrological law is realized through main frequency energy decomposition and feature reconstruction.

The application discloses a kind of crop precision irrigation method and system based on algorithm fusion, it is related to intelligent agriculture and precision irrigation control technical field, method includes: the time series data of meteorology, soil, crop physiology and irrigation process are collected, carry out exception elimination, missing completion and standardization processing and construct derivative feature;The historical characteristic sequence is input to the bidirectional time series network with attention to obtain time series representation, input neural fuzzy inference network to output future water demand prediction value and corresponding prediction interval, obtain confidence from prediction interval;Water demand prediction value, confidence and crop stress are input to the first layer fuzzy inference to obtain basic irrigation duration;Soil water potential and post-irrigation feedback are input to the second layer fuzzy inference to obtain valve fine tuning amount;Fusion generates irrigation instruction and drives execution equipment.By outputting water demand prediction interval and constructing confidence, the prediction reliability is improved, and by double-layer fuzzy inference, soil water potential and post-irrigation feedback are combined to realize executable and safety protection of irrigation instruction.

The invention discloses a time sequence anomaly detection method and device for association difference mining, equipment and a medium, and relates to the technical field of time sequence analysis. The method comprises the following steps: S1, obtaining an original time sequence, and carrying out feature representation to obtain time sequence representation; and S2, performing time dimension correlation difference calculation on the time sequence representation to obtain a time correlation difference. And S3, carrying out spatial dimension correlation difference calculation on the time sequence representation to obtain a feature correlation difference. And S4, reconstructing the time sequence representation to obtain a reconstructed time sequence, and calculating a reconstruction error. And S5, calculating an abnormal score of each time point by combining the time correlation difference, the feature correlation difference and the reconstruction error. And S6, determining the time point when the abnormal score exceeds the abnormal threshold value as an abnormal point. According to the method, the time-dimensional difference module and the space-dimensional difference module are fused, abnormal points appearing in the time sequence can be accurately detected, and help is provided for a time sequence anomaly detection task.

The application discloses a time series data prediction method and device, equipment and a storage medium. The method comprises the following steps: obtaining a time series to be processed; the time series to be processed is used for representing running data of a to-be-predicted object at a plurality of detection times; performing feature extraction processing on the time series to be processed through a prediction model to obtain local feature data, global feature data and fusion feature data of the time series to be processed; performing feature coding processing on the local feature data, the global feature data and the fusion feature data by using the prediction model to obtain a coding result, and performing classification prediction on the coding result to obtain a prediction result of the to-be-predicted object. The scheme can capture short-range periodic information and long-range periodic information of the time series to be processed in a more fine-grained manner, thereby combining more comprehensive time series representation information to perform classification prediction on a downstream task, and greatly improving the prediction accuracy.

The invention provides a cross-modal fusion time sequence prediction method, device and equipment based on a large language model, and relates to the technical field of time sequence prediction. The method comprises the following steps: acquiring numerical time sequence data in a historical time window and associated text information; respectively processing the numerical time sequence data and the text information through parallel processing modal preprocessing branches to obtain a time sequence feature representation, a semantic reasoning prediction result generated by a large language model and a text semantic feature representation; performing hierarchical interactive fusion on the time sequence feature representation, the semantic reasoning prediction result and the text semantic feature representation through a multi-stage cross-modal attention fusion network to obtain enhanced time sequence representation; generating an adaptive fusion weight for each time step in the prediction time window based on the enhanced time sequence representation; and carrying out weighted fusion on a numerical prediction result generated based on the time sequence feature representation and a semantic reasoning prediction result by using the adaptive fusion weight to obtain a final time sequence prediction result.

The invention relates to the technical field of computers, can be applied to the fields of medical health, financial science and technology and the like, and discloses a cloud native system fault analysis method, device and equipment and a storage medium, and the method comprises the steps: obtaining multi-modal observation data of a plurality of service entities in a cloud native system; performing feature fusion on the multi-modal observation data through a multi-modal data fusion tool to generate time sequence representation; based on the time sequence representation, a root cause entity causing the fault is positioned through a root cause positioning tool; based on the root cause entity, determining a fault type of the root cause entity through a fault classification tool; integrating the root cause entity, the fault type and pre-stored system context information into a text prompt according to a preset format; and inputting the text prompt into a big language model expert agent, and generating a root cause analysis report. By means of the method, the data size input into the large language model is reduced, the analysis accuracy is improved, reasoning illusion is effectively avoided, and the reliability of the positioning result is guaranteed.

The invention relates to a multivariable time sequence prediction system based on deep learning and a language model, and the system carries out the instance normalization of input multivariable time sequence data, and obtains standardized data; dividing the normalized data into local segments by using a sliding window and mapping the local segments into embedded vectors; dynamically aligning the time sequence fragments with language clues through a multi-head cross attention mechanism on the basis of a text prototype library, and generating a time sequence representation vector fusing text semantics; constructing a prompt text containing domain knowledge, task instructions and statistical characteristics, and splicing the prompt text with the time sequence representation vector to form a language model input sequence; the sequence is input into a language model obtained through knowledge distillation, efficient context-aware reasoning is achieved by freezing main body parameters and only finely adjusting prompt token parameters, and output representation is obtained; and the output representation is mapped into a future multi-step prediction value through the multi-layer perceptron, so that the calculation resource consumption is reduced, and meanwhile, the multivariable time sequence prediction precision and the cross-domain generalization ability are remarkably improved.

According to one embodiment, first input features are extracted from received past time-dependent inputs. The first input features are represented at least in part by a first plurality of input time series. A first encoder output array is generated by a first encoder with a first cross-attention mechanism based at least in part on the first input features. The first encoder output array is provided as query, key and value inputs to a pretrained core model with a self-attention mechanism to generate a core model output array. Forecasting results in a forecasting time period are generated by a decoder based at least in part on the core model output array. The forecasting results are represented by one or more output time series.

The invention discloses a user behavior anomaly detection method. The method comprises the steps of obtaining behavior counting sequences of a target user under multiple time granularities to form multivariable time sequence input data; learning time correlation in the TFT model to obtain time sequence representation, starting Bayesian inference to carry out multiple forward inference to obtain a prediction mean value and a prediction standard deviation at the next moment, and calculating a prediction confidence interval; a real behavior count of a target user at a current moment is obtained, a current prediction error is calculated through a TFT model deployed at an edge side, a standardized deviation index Z-score is calculated based on a prediction standard deviation, a cumulative offset is constructed based on the prediction error, a CUSUM statistical magnitude is calculated, the Z-score and the CUSUM statistical magnitude are fused to obtain a combined abnormal variable, and the combined abnormal variable is calculated. And comparing with two risk thresholds, and judging three risk levels of normal transmission, probe exposure and direct interception. According to the method, millisecond-level prediction of user behaviors is realized, abnormity is dynamically judged, and abnormal user behaviors are intercepted before bidding.

The invention discloses a high-dimensional hierarchical time sequence modeling method for predicting the residual life of industrial rotating equipment, which comprises the following steps of: constructing global representation based on information bottleneck constraint by using an information bottleneck theory, and fusing and mapping multi-source time sequence data to a low-dimensional feature space so as to capture low-dimensional global representation with rich information; constructing a block interaction mechanism to extract local features and perform hierarchical feature fusion to realize global-local collaborative enhancement representation fusion; time sequence representation based on manifold learning non-diagonal covariance priori constraint is constructed, and time sequence degradation features are embedded in global representation. The method solves the key problem that the existing method is difficult to capture the degradation trend in the residual service life prediction task under the high-dimensional data challenge, obviously reduces the residual service life prediction error and parameter quantity under the high-dimensional multi-source data, and provides accurate and lightweight technical support for the application of the equipment residual service life prediction.