50 results about "Time-series segmentation" patented technology

Learning-type motion control is performed using a hierarchical recurrent neural network. A motion pattern provided through human teaching work is automatically time-serially segmented with the hierarchical recurrent neural network, and the motion control of a machine body is carried out with a combination of the segmented data, whereby various motion patterns can be produced. With the time-serial segmentation, local time-serial patterns and an overall pattern as a combination of the local time-serial patterns are produced. For those motion patterns, indices for static stability and dynamic stability of the machine body, e.g., ZMP stability criteria, are satisfied and hence control stability is ensured.

The present invention relates to the field of traffic safety, in particular to a judgment and prediction method of driving behavior intention based on the implicit Markov model (HMM), aiming to overcome the defect that the existing driving behavior intention recognition and prediction technology does not take into account the dynamics and continuity of driving behavior, as well as complex behaviors such as lane changing, car following and braking and the like. The judgment and prediction method of driving behavior intention obtains time series segmentation data from cluster analysis of dynamic driving data, the linear direction HMM, lateral HMM and speed classification model are trained respectively, and the obtained identification results are regarded as the observation sequence of behavior recognition layer; Then, off-line training is performed to deal with normal or emergency braking, normal or emergency lane change, normal or dangerous driving behavior, and multi dimensional discrete HMM model; according to the model parameters and the observation sequence, the next time step driving behavior can be predicted. The judgment and prediction method of driving behavior intention takes the complexity and continuity of driving behavior into account and can dynamically judge and predict the driving behavior intention and warn the dangerous behavior, and accordingly can be applied to driving behavior evaluation and driving assistance system.

Initially, customer time series credit files are acquired. The credit files are organized in a data mart environment for supporting a query system. Time series utilization attributes are created and a neural network time series segmentation process is applied and N.times.N dimension segments are generated for analysis. The chart may be modified to more accurately depict profitable credit revolvers. Credit data from each potential new customer is processed in a similar fashion by the neural network segmentation process. Profitable credit revolvers are identified by having credit utilization patterns belonging to profitable segments previously identified.

A method for segmenting a signal into segments having similar spectral characteristics is provided. Initially the method generates a table of previous values from older signal values that contains a scoring value for the best segmentation of previous values and a segment length of the last previously identified segment. The method then receives a new sample of the signal and computes a new spectral characteristic function for the signal based on the received sample. A new scoring function is computed from the spectral characteristic function. Segments of the signal are recursively identified based on the newly computed scoring function and the table of previous values. The spectral characteristic function can be a selected one of an autocorrelation function and a discrete Fourier transform. An example is provided for segmenting a speech signal.

The invention discloses a time series trend feature extraction method based on an important point double evaluation factor, comprising: regarding a time series piecewise linear representation as basis, defining an important point as an alternative set of time series segmentation points, calculating an important point distance factor and a trend factor, using the distance factor to measure the relative degree of difference and using the trend factor to measure its impact on the overall trend globally, using a comprehensive evaluation model to evaluate the importance of each important point to the overall trend to select the segmentation points, and finally connecting adjacent segmentation points to obtain a segmented trend representation of the time series. The invention proposes the concept of a time series important point distance factor, and combines the two evaluation factors to evaluate the important points of the time series, which overcomes the shortcomings of single evaluation functions and locality of the existing piecewise linearization method, may effectively weaken the noise interference, retain the time series change trend feature, and has fast processing speed and higher extraction precision than the existing method when the number of segments is the same.

Data processing method for analysis of the energy consumption of a site on the basis of measurements including:

• • selection of data time series,
  • segmentation of the time series into segments,
  • projection of the segments,
  • display of the projections of the data segments,
  • e) selection of at least one first group of data segments,
  • establishment of a digital model of classification characterising the first group of data segments, selected on the basis of calendar metadata
  • establishment of a digital model of classification characterising the first group of data segments, selected on the basis of the segments' profiles (FIG. 5B).

A method for accelerating time series data base queries includes segmenting an original time series of signal values into non-overlapping chunks, where a time-scale for each of the chunks is much less than the time scale of the entire time series, representing time series signal values in each chunk as a weighted superposition of atoms that are members of a shape dictionary to create a compressed time series, storing the original time series and the compressed time series into a database, determining whether a query is answerable using the compressed time series or the original time series, and whether answering the query using the compressed time series is faster. If answering the query is faster using the compressed representation, the query is executed on weight coefficients of the compressed time series to produce a query result, and the query result is translated back into an uncompressed representation.

Systems and methods for managing one or more physical systems, including determining system behavior switching based on time series data from one or more sensors in the system. Time series is divided into a plurality of segments, and each of the segments represents a system behavior. A fitness model is generated for each of the segments to determine whether to select each of the segments as invariants, and an ensemble of local relationship models are built for each of the time series for each invariant to identify local behavior switching points over time. The identified local behavior switching points of each invariant are aggregated by aligning the local switching points of all invariant segments, computing a density distribution of the aligned switching points, and extracting local maximas of the density distribution to determine the global switching points. System operations are controlled based on the determined system behavior switching.

In order to improve accuracy for detecting presence or absence of a target object. A time-series segmentation unit 102 creates first time-series data by segmenting processing target data into each frame of “n” time zones. Each of first determination units 103 creates “m” second time-series data by determining each frame of the first time-series data using “m” models having different characteristics. A second determination unit 104 creates a second determination result as a presence probability of the target object for a set of second time-series data including n×m data.

The invention belongs to the technical field of displacement detection, and discloses a short-term displacement detection method based on GNSS dynamic positioning time sequence segmentation. The method comprises the steps: firstly obtaining an original coordinate time sequence through GNSS dynamic positioning; carrying out time sequence segmentation by utilizing a detection window to obtain a setof possible change points; for each possible change point, obtaining a statistical value of the possible change point, and judging whether the possible change point is a change point or not accordingto the statistical value and a preset threshold value; finally, for each change point, obtaining a previous displacement estimation value corresponding to the change point, and obtaining a final displacement estimation value based on the previous displacement estimation value. According to the method disclosed in the invention, the problem that the displacement cannot be effectively detected and extracted in a short time in the prior art is solved, and the displacement can be effectively detected and extracted in a short time.

The invention provides a structural health monitoring multivariate data anomaly diagnosis method based on a convolutional neural network and transfer learning, and the method comprises the steps: carrying out the time series segmentation data visualization processing of the multivariate monitoring data of a certain large-scale structure A, converting the data into a time domain response image, carrying out the manual marking according to the time domain response image data corresponding to a data segment, selecting samples of various abnormal types with manual marks to form a data set A; inputting the data set A into a convolutional neural network model A for anomaly detection, and training the model A; visualizing and manually marking multivariate monitoring data of a certain large structure B to form a data set B; and adding a data set B on the basis of the model A, carrying out transfer learning training, so that the generalization performance of a classification model is improved, a convolutional neural network model can adapt to data of different distributions, and the model trained through transfer learning serves as a multivariate data anomaly detector; the method can solve the problem that there is no detection method for structural health monitoring multivariate data at present.

The invention discloses a financial time series segmentation distribution feature computing method. According to the method, based on TICK data of the bargain price, a differential direct-current removal data preprocessing method is adopted, a new segmentation series data structure is configured, continuous segmentation is conducted on the difference series, statistics is conducted on prior probability distribution of the segmentation series, and financial transaction data trend distribution computing is achieved. The invention further provides a financial time series segmentation distribution feature computing system, compared with other financial time series feature extraction algorithms, the system is of a more concise data processing structure, more excellent in recognition performance and good in data consistency; meanwhile, the series distribution feature obtained through the data processing method is obvious, and the performance is more excellent on the aspect of fuzzy estimation compared with other similar algorithms.

The invention discloses a time sequence segmentation method based on an LAC-FLOSS algorithm and an IER algorithm, and relates to a time sequence data segmentation method. The invention provides an improved time sequence segmentation algorithm LAC-FLOSS for limiting arc spanning based on Matrix Profile, the algorithm adds a weight to an arc to form a weighted arc, and then the problem of cross-state sub-sequence mismatching of the arc is solved by setting a matching distance threshold. In addition, the shape features of the CAC sequence are utilized to extract the minimum value from the trough, and then an improved segmentation point extraction algorithm IER is provided. According to the algorithm, the situation that an existing segmentation point extraction algorithm ER uses a window to take segmentation points at non-inflection points can be avoided, the accuracy of segmentation result extraction is improved, and through comparison with the ER algorithm, it is verified that the segmentation point extraction effect of the IER is better than that of the algorithm ER.

The invention relates to a multi-mode time series anomaly detection method, which comprises the following steps of: constructing a multi-mode sub-time series clustering model based on time series segmentation and sub-time series clustering modes according to a historical time series; iteratively optimizing the multi-mode sub-time-sequence clustering model based on a method of initializing, splitting, combining and removing sub-time-sequence clustering modes, and solving to obtain each sub-time-sequence clustering mode; and based on the sub-time sequence clustering mode, extracting each sub-time sequence from the to-be-detected time sequence to perform anomaly detection. According to the method, various sub-time sequence modes with different lengths can be adaptively extracted from the timesequence, the situation of insufficient mode expression or interference of mode expression caused by extracting the sub-time sequence with a fixed length at present is avoided, and the time sequenceabnormity detection precision based on the situation is higher. The invention further relates to a storage medium and equipment.

The invention discloses a multifunctional radar state sequence clustering method based on time sequence segmentation, and the method can achieve the clustering of a complex multifunctional radar working mode pulse sequence with a variable pulse input form, a variable mode number, a variable mode length and a variable mode modulation parameter on each PDW parameter, pulse-level unsupervised time sequence segmentation clustering is performed; the inter-pulse modulation type of each PDW parameter is utilized to construct a corresponding time sequence parameterization model and give corresponding model parameter estimation; each parameterized model can represent time sequence characteristics of different inter-pulse modulation types; the clustering result is marked to the pulse level, each pulse belonging to different radar working modes in the long pulse sequence can be clustered, meanwhile, the clustering result can give model parameters of each parameterized model, and conditions are provided for fine analysis and processing of the subsequent pulse sequence.

The invention discloses an unsupervised non-intrusive load identification method based on power theme discovery. The method comprises the following steps: collecting total active power signals at a home entry to form a load power time sequence; filtering power peaks of a non-fluctuation section from an original active power time sequence by adopting a median filtering preprocessing method; detecting the key points (an important extreme point and an important trend turning point) of the load power time sequence, and marking in the time sequence; segmenting the time sequence into a plurality of subsequences with different lengths based on the key points; and storing subsequently extracted subsequences in a subsequence list, matching the extracted subsequences with templates in an electric appliance load mark template library in a template matching mode, and finally achieving unsupervised non-intrusive load recognition. According to the invention, subsequences which frequently appear in a time sequence and have any duration can be autonomously found, load events corresponding to the start-stop process of the electric appliance can be detected, and a complex power change mode in the operation process of the electric appliance can be found.