50 results about "Time series representation" patented technology

A method of effectively representing and processing data sets with time series is disclosed. The method may comprise representing time series as a virtual part of data in a data store layer of a user system, thereby allowing processing of time-series related queries in said data store layer of said user system.

A method for reconstructing a high quality image from undersampled image data is provided. The image reconstruction method is applicable to a number of different imaging modalities. Specifically, the present invention provides an image reconstruction method that incorporates an appropriate prior image into the image reconstruction process. Thus, one aspect of the present invention is to provide an image reconstruction method that requires less number of data samples to reconstruct an accurate reconstruction of a desired image than previous methods, such as, compressed sensing. Another aspect of the invention is to provide an image reconstruction method that produces a time series of desired images indicative of a higher temporal resolution than is ordinarily achievable with the imaging system. For example, cardiac phase images can be produced with high temporal resolution (e.g., 20 milliseconds) using a CT imaging system with a slow gantry rotation speed.

A navigation script includes time and point information for navigation and information for guidance, and describes an instruction sequence which can represent these information in time series in a mark-up language. According to the structured data generated from the navigation script, an instruction corresponding to a current time or point is executed, so that information for guidance to be presented is output.

Illness signatures are mathematically characterized by entrainment relationships among multiple time series representations of physiological processes. Such characteristics include time and phase lags, window lengths for optimum detection, which time series are most entrained with each other, the degree of entrainment relative to the rest of the large database, and the concordance or discordance of the time-varying changes. These optimum disease-specific characteristics can be determined, for example, from large, clinically well-annotated databases of time series representations of physiological processes during health and illness. These characteristics of the entrainment relationships among multiple time series representations of physiological processes are used to make mathematical and statistical predictive models using multivariable techniques such as, but not limited to, logistic regression, nearest-neighbor techniques, neural and Bayesian networks, principal and other component analysis, and others. These models are quantitative expressions that transform measured characteristics to the probability of an illness, or p(illness).

The invention discloses a time sequence and CNN-based unsafe behavior identification method and system. The method comprises the steps of inputting original data of a training set video of unsafe behaviors to a CNN for performing training and learning, and outputting a spatial eigenvector from the last pooling layer of the CNN; taking the spatial eigenvector as an input of a time recurrent neuralnetwork, learning a time sequence dependency relationship contained in the spatial eigenvector by using the time recurrent neural network to learn a time sequence representation of spatial behaviors,and obtaining an average pooling eigenvector; inputting the average pooling eigenvector to a softmax layer, thereby obtaining a deep mixed learning model of the CNN and the time recurrent neural network, namely, a softmax classifier; and performing online identification on a monitoring video of a construction site by utilizing the softmax classifier to obtain the unsafe behaviors of the construction site. The support can be provided for real-time investigation and correction of the unsafe behaviors in a whole building engineering construction process.

A computer-implemented method for employing deep learning for time series representation and retrieval is presented. The method includes retrieving multivariate time series segments from a plurality of sensors, storing the multivariate time series segments in a multivariate time series database constructed by a sliding window over a raw time series of data, applying an input attention based recurrent neural network to extract real value features and corresponding hash codes, executing similarity measurements by an objective function, given a query, obtaining a relevant time series segment from the multivariate time series segments retrieved from the plurality of sensors, and generating an output including a visual representation of the relevant time series segment on a user interface.

Future media viewership is forecast based on time ordered analysis of historical viewership information from an individual or combination of a plurality of data sets. Forecast models having coefficients derived from comparisons of time series representations of data sets across a plurality of time periods and data sources join together disparate data sets. Individual data sets from disparate data sources may be compared to identify possible untrustworthy data or data that requires further investigation. Organizing viewership information in a time series allows for imputing missing data in a respective data set.

A method of predicting the performance of a financial variable having corresponding financial parameters represented by time series of adjacent-in-time series terms. Differential series are calculated from adjacent series terms in the financial parameter series and term trends of upward, downward and unchanged trends are then located. Cumulative variations of the term trends are calculated and are then used to identify a series of sign state progressions. A time position for a select financial parameter series term is then selected and terms in the differential series, cumulative variation series and sign state series corresponding to the time position are located. The series terms are then searched to locate prior time positions where similar term values and trends occurred. Once the prior time positions are located, the series values at time positions subsequent to the located time positions are used as a forecast for the performance of the financial variable.

A time sequence mode representation-based weighted directed complicated network construction method comprises the steps of adopting a zero-mean normalization method to normalize an original time sequence; dividing a new time sequence into n sections in an equal probability manner, using the characters in a set character string to represent the sections, and representing the new time sequence into a character string sequence; moving a sliding window of which the length is 1 from left to right from the first character of the character string sequence, every time the sliding window moves one step, dividing the character string sequence into ((n-1)+1) fragments of which the lengths are all 1, and regarding each fragment as a mode; taking the different modes as the nodes of a complicated network, determining the connection edge weights and directions between the nodes of the complicated network according to the conversion frequency and the conversion directions between the nodes, and mapping the character string sequence into the weighted directed complicated network; and calculating the network topology statistical characteristics of the weighted directed complicated network. The method of the present invention enables the classification or identification precision of the time sequence signals to be improved remarkably.

A method for implementing automated information technology (IT) system failure recommendation and mitigation includes performing log pattern learning to automatically generate sparse time series for each log pattern for a set of classification logs corresponding to a failure, performing multivariate log time series extraction based on the log pattern learning to generate a failure signature for the set of classification logs, including representing the sparse time series as a run-length encoded sequence for efficient storage and computation, calculating a similarity distance between the failure signature for the set of classification logs and each failure signature from a failure signature model file, determining a failure label for the failure corresponding to a most similar known failure based on the similarity distance, and initiating failure mitigation based on the failure label.

For every business interaction with customers consists of cases and each case consists of sequence of events: First_Contact_Customer, . . . intermediate events, . . . Case_Closed. The most important characteristics are frequencies of transitions between events and mean time between events (MTBE, TBE) for each type of cases. Type of cases could be type of customer, group of products, branch of enterprise, geographical area, etc. Existed methods of visualization (the most popular of them are MS Excel pivot charts) could not visualize two characteristics (Frequency and MTBE) simultaneously to locate business problems.

Our method combines standard SPC run chart for time series representation with three new types of charts for cross-sectional representation: “matrix bar chart” for portraying types of cases, “flower bed chart” for displaying Frequencies and MTBE. and “Tower Chart” that can be element of “Flower Bed Chart” and “Matrix Bar Chart” when we need detailed visualization of distribution of TBE.

This new method is applicable for any customer service—help desks, stores, doctor offices, banks and gives the user ability to identify immediately the most business important factors

The invention relates to a support vector machine-based mine fiber grating monitoring system missing data compensation method and belongs to the monitoring system missing data compensation method. The method includes the following steps that: step 1, a mine fiber grating monitoring system data sensing device performs data acquisition; step 2: monitoring data collected by the fiber grating sensing device are represented by time sequences and are preprocessed; step 3, the kernel function of a support vector machine is selected and constructed; step 4, the hyper-parameters of the support vector machine are selected and determined; step 5, a kernel matrix is built according to the determined kernel function, a loss function and penalty parameters, an optimization algorithm is adopted to get the optimal parameter value of the support vector machine; and step 6, the regression function model of the support vector machine is established, fitting regression analysis is performed on the data, so that the compensation result of missing data is obtained, and compensation work for the missing data is performed. With the support vector machine-based mine fiber grating monitoring system missing data compensation method of the invention adopted, the monitoring missing data can be compensated, and therefore, the monitoring data can be improved, and the data are closer to real and reliable data, and the safety production, construction and stability of a mine can be ensured.

The mobile application program identification method of K-means clustering and a random forest algorithm comprises the following steps: firstly, discretizing an encrypted data stream in a time periodinto a plurality of data streams according to the characteristics of a TCP session, and representing each data stream by adopting an input grouping time sequence, an output grouping time sequence andan input and output grouping time sequence; Performing mathematical statistics on the three time sequences corresponding to each data stream to obtain statistical characteristics of the data packet; Afterwards, Carrying out statistical characteristic clustering analysis on the encrypted data flow by using a K-means clustering algorithm; Scoring the purity of each clustering cluster obtained by clustering analysis through an entropy calculation method, and filtering samples in the clustering cluster with lower purity; And finally, carrying out modeling on the filtered cluster serving as a dataset through a random Sendon algorithm, so that identification of the encrypted Liu mobile application type is realized. According to the method, supervised learning and unsupervised learning are combined, and different mobile application types can be accurately identified in encrypted traffic with various application types.

The invention discloses a sequence data prediction system of a novel multi-scale attention mechanism. The sequence data prediction system comprises a sequence data coding module combined with time feature coding, a multi-scale time feature extraction module and a long sequence rapid prediction module; the sequence data coding module combined with time characteristic coding codes sequence data andtime characteristics of an input sequence to obtain a new input vector; the multi-scale time feature extraction module is used for segmenting the new input vector, inputting the segmented new input vector into a corresponding feature extraction structure, and combining the extracted sequence features of different time scales to obtain stable and effective time sequence representation; and the long-sequence rapid prediction module constructs an initial input sequence, performs data encoding and then performs fusion to obtain a prediction output result.

The invention discloses a CDN flow anomaly detection device and method based on an improved hierarchical time memory network. The device comprises a data acquisition module, a data preprocessing module, a data storage module, a system scheduling module, an anomaly detection module and a display module. The method comprises the steps that a data collection module collects data of a native log, converts the data into a json format and sends the data to a data preprocessing module; feature extraction is carried out to obtain CDN flow time sequence representation, and the log data of the data acquisition module and the CDN data of the data preprocessing module are stored by the data storage module; the anomaly detection module acquires flow time series data through the system scheduling module, inputs the flow time series data into the time series anomaly detection model based on the improved hierarchical time memory network for online learning, completes anomaly possibility calculation and outputs a detection result of anomaly possibility judgment, and the display module visually displays a key process. The method has the advantages of high detection speed and high accuracy.

The invention provides a network fault prediction method which comprises the following steps: S1, preprocessing network fault data, wherein the network fault data comprises fault types and fault occurrence time and is converted into time sequence data, and each time sequence represents all fault types occurring in the current time period; and S2, constructing a neural network model based on gated multi-head attention, wherein the neural network model comprises an embedded layer, an attention memory network layer, a multi-head attention layer and a gated fusion layer, inputting the time sequence data into the neural network model based on gated multi-head attention, and predicting faults in the network fault data by the neural network model based on gated multi-head attention. The invention aims to solve the problem that a traditional prediction method cannot predict the influence of different network faults on other faults due to the complexity and randomness of the faults, so that the faults cannot be predicted, thereby providing a technical scheme for accurately predicting the network faults.

Change points of a system represented by a plurality of time series are detected more appropriately. An information processing system includes means for learning, with respect to each of a plurality of time series, models that approximate partial time series respectively and are defined by parameters of the partial time series respectively, the partial time series being obtained by dividing a corresponding time series into a plurality of segments at change point candidates; and means for detecting, with respect to each of the change point candidates for the plurality of time series, a global change point that is a change point for the plurality of time series based on a difference between a parameter of a first partial time series starting from a time point of a corresponding change point candidate and a parameter of a second partial time series before the corresponding change point candidate, and outputting the global change point.

In accordance with the present invention, a method for obtaining frequency information about a given data set is realized. The method comprises the steps of providing a processing unit; inputting a raw data set into the processing unit; optionally removing at least one trend from the raw data; ordering the raw data; estimating power spectral density using an eigenanalysis approach and the inputted raw data and the ordered raw data; simultaneously estimating the power spectral density using the raw data and a periodogram; generating a time-series representation of the raw data to which curve fitting is applied; comparing the results from the power spectral density estimating steps and the time-series representation generating step to determine if any frequencies suggested by the eigenanalysis approach estimating step are valid; and generating an output signal representative of each valid frequency.

The invention discloses a time series symbol aggregation approximate representation method fusing trend features. The time series approximate representation method fusing trend features comprises thefollowing steps: acquiring time series data; preprocessing the time series data; performing time sequence feature segmentation; performing time sequence statistical feature extraction and symbolic representation; performing trend feature extraction and symbolic representation of the time sequence; fusing time series symbol representation and similarity measurement of trend features. According to the method, the trend characteristic information and the statistical characteristic information of a time sequence are combined to form a new symbol aggregation approximate representation method capable of considering the statistical characteristics and the trend characteristics of the time sequence, and the time sequence is mapped from a high-dimensional space to a low-dimensional space on the premise of not losing the sequence characteristic information. Compared with a traditional time sequence representation method, the method not only has better lower bound sealing performance, but also can obtain better classification and clustering effects, thereby better representing time sequences with different morphological characteristics.

A system and method for individual quantitative identification by means of human dynamic rhythmic electric activity spectra is provided. The method for distinguishing an individual, comprising; contacting the individual with an electrical probe; measuring, with the electrical probe, an electrical signal associated with the individual; processing the electrical signal to produce a time-series representation of the electrical signal and a frequency-domain representation of the electrical signal; identifying a distinct pattern in the time-series representation in a range of about 30 kHz to about 50 kHz; and identifying a distinct pattern in the frequency-domain representation in a range of about 500 kHz to about 1.5 MHz.

Techniques are disclosed relating to the modification of data in a time-series data lake. For example, in various embodiments, the disclosed techniques include a cloud-based service that maintains a time-series data lake that includes, for an organization, a time-series representation of data from one or more of the organization's data sources. The cloud-based service may receive a request to modify data associated with a particular user of the organization. As a non-limiting example, this request may correspond to a “Right to Be Forgotten” request from the particular user. This request may include one or more search parameters and an indication of one or more modifications to be performed. Based on the request, the cloud-based service may parse the time-series data lake to identify a subset of data that matches the one or more search parameters and perform the requested modifications on the subset of data in the time-series data lake.