Fuzzy-learning-based extraction of time-series behavior
First Claim
1. A computer-implemented method of extracting time-series behavior based on time-series information, the method comprising:
- loading into a computer time-series data, wherein the loaded time-series data comprises time-series data points comprising an input-component-part and an output-component-part, the input-component-part comprising one or more input components and the output-component-part comprising one or more output components, the one or more input components collectively representing a value from an input space and the one or more output components collectively representing a value from an output space, wherein loading time-series data further comprises incorporating at least one time-related input component of at least one data point into the loaded time-series data;
dividing into fuzzy regions a range of possible values for each component of the time-series data;
assigning with one or more computer processors a fuzzy membership function to each fuzzy region;
generating with one or more computer processors non-conflicting fuzzy rules that are determined at least in part by the fuzzy membership functions and at least in part by how a plurality of the time-series data points are clustered;
determining a mapping from the input space to the output space based on defuzzification of the fuzzy rules; and
displaying the mapping in a manner that allows a user to perform time-series prediction or time-series-trend recognition.
25 Assignments
0 Petitions
Accused Products
Abstract
Systems and methods for extracting or analyzing time-series behavior are described. Some embodiments of computer-implemented methods include generating fuzzy rules from time series data. Certain embodiments also include resolving conflicts between fuzzy rules according to how the data is clustered. Some embodiments further include extracting a model of the time-series behavior via defuzzification and making that model accessible. Advantageously, to resolve conflicts between fuzzy rules, some embodiments define Gaussian functions for each conflicting data point, sum the Gaussian functions according to how the conflicting data points are clustered, and resolve the conflict based on the results of summing the Gaussian functions. Some embodiments use both crisp and non-trivially fuzzy regions and/or both crisp and non-trivially fuzzy membership functions.
-
Citations
30 Claims
-
1. A computer-implemented method of extracting time-series behavior based on time-series information, the method comprising:
-
loading into a computer time-series data, wherein the loaded time-series data comprises time-series data points comprising an input-component-part and an output-component-part, the input-component-part comprising one or more input components and the output-component-part comprising one or more output components, the one or more input components collectively representing a value from an input space and the one or more output components collectively representing a value from an output space, wherein loading time-series data further comprises incorporating at least one time-related input component of at least one data point into the loaded time-series data; dividing into fuzzy regions a range of possible values for each component of the time-series data; assigning with one or more computer processors a fuzzy membership function to each fuzzy region; generating with one or more computer processors non-conflicting fuzzy rules that are determined at least in part by the fuzzy membership functions and at least in part by how a plurality of the time-series data points are clustered; determining a mapping from the input space to the output space based on defuzzification of the fuzzy rules; and displaying the mapping in a manner that allows a user to perform time-series prediction or time-series-trend recognition. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A computer-implemented method of extracting time-series behavior based on time-series information, the method comprising:
-
loading into a computer time-series data, wherein the loaded time-series data comprises time-series data points comprising an input-component-part and an output-component-part, the input-component-part comprising one or more input components and the output-component-part comprising one or more output components, the one or more input components collectively representing a value from an input space, the one or more output components collectively representing a value from an output space, and the one or more input components comprising at least one time-related input component; dividing into fuzzy regions a range of possible values for each component of the time-series data; assigning with one or more computer processors a fuzzy membership function to each fuzzy region, wherein assigning a fuzzy membership function to each fuzzy region comprises assigning a triangular fuzzy membership function to at least one fuzzy region; constructing with one or more computer processors a Fuzzy Associated Memory (FAM) bank comprising fuzzy rules each having an antecedent clause and a consequent clause, wherein the consequent clause of a given fuzzy rule of the FAM bank is determined by the output space value at which a sum of functions associated with the antecedent clause of the given fuzzy rule achieves a greatest relative maximum, wherein the sum of functions associated with the antecedent clause of the given fuzzy rule comprises a sum of Gaussian functions on the output space, each such Gaussian function achieving a greatest relative maximum at an output space value determined by a time-series data point whose input-component-part corresponds to the antecedent clause of the given fuzzy rule; determining a mapping from the input space to the output space based on defuzzification of the FAM bank; and displaying the mapping in a manner that allows a user to perform time-series prediction or time-series-trend recognition.
-
-
20. A computer-implemented method of extracting time-series behavior based on time-series information, the method comprising:
-
loading into a computer a first increment of time-series data, wherein the loaded first increment of time-series data comprises time-series data points comprising an input-component-part and an output-component-part, the input-component-part comprising one or more input components and the output-component-part comprising one or more output components, the one or more input components collectively representing a value from an input space and the one or more output components collectively representing a value from an output space; dividing with one or more computer processors into fuzzy regions a range of possible values for each component of the time-series data; assigning with one or more computer processors a fuzzy membership function to each fuzzy region; using the first increment of time-series data to build a Fuzzy Associated Memory (FAM) bank comprising fuzzy rules, the fuzzy rules of the FAM bank being determined at least in part by the fuzzy membership functions and at least in part by the way a plurality of the time-series data points of the first increment of time-series data are clustered; and providing with one or more computer processors a model of the behavior of the first increment of time-series data in a manner that allows a user to perform time-series prediction or time-series-trend recognition. - View Dependent Claims (21, 22, 23, 24, 25)
-
-
26. An apparatus for analyzing time-series behavior comprising:
-
a first computer system; and a second computer system, wherein a first code module that is loaded into a memory of the first computer system communicates with a database and a second code module that is loaded into a memory of the second computer system communicates with the first code module so as to retrieve from the database data relating to a time series, extracts a model of the behavior of the time series via fuzzy learning based in part on how conflicting data points of the retrieved data are clustered, where two data points conflict if they give rise to conflicting fuzzy rules, and makes the extracted model accessible to a user such that the user can use the model to perform time-series analysis, wherein the second code module further incorporates at least one time-related input component of at least one data point into the retrieved time-series data. - View Dependent Claims (27)
-
-
28. An apparatus for analyzing time-series behavior comprising:
-
a first computer system; and a second computer system, wherein a first code module that is loaded into a memory of the first computer system communicates with a database and a second code module that is loaded into a memory of the second computer system communicates with the first code module so as to retrieve from the database data relating to a time series, extracts a model of the behavior of the time series via fuzzy learning based in part on how conflicting data points of the retrieved data are clustered, where two data points conflict if they give rise to conflicting fuzzy rules, and makes the extracted model accessible to a user such that the user can use the model to perform time-series analysis, wherein each data point of the retrieved data has one or more output components that collectively represent a value from an output space and, for each set of conflicting data points, the second code module further clusters output components of the conflicting data points;
defines a function for each resulting cluster, the characteristics of each cluster determining at least one relative maximum value of its associated function;
selects a cluster whose associated function has a relative maximum value equal to the greatest relative maximum value of all the functions associated with a cluster; and
creates a fuzzy rule using the antecedent clauses of the fuzzy rules generated by the data points in the set of conflicting data points and the consequent clause corresponding to the fuzzy region for which an output value substantially representing the center of the selected cluster has the highest degree of membership. - View Dependent Claims (29)
-
-
30. A system for analyzing time-series behavior comprising:
-
a computer system, wherein a first code module that is loaded into and is executable on the computer system, and that is in communication with a training database, comprises; a second code module that reads data based on a time series from the training database, the data being comprised of data points, and a third code module that communicates with the second code module and that extracts via fuzzy learning a model of the behavior of the time series based in part on how conflicting data points are clustered; and a user interface in communication with the third code module, the user interface providing a user with access to models extracted by the third code module such that the user can perform time-series analysis, wherein the third code module resolves conflicts between data points based on summing Gaussian functions that are substantially determined by the output values of conflicting data points, where the sums are calculated according to how the output values are clustered.
-
Specification