US6701195B2 — Sensor prediction system utilizing case based rea…

Claim 1

Independent Claim 236 words

A method for control of a technical process which generates a plurality of sensor values which are recorded and stored in a sensor data archive, comprising cases generated from recorded sensor data representing a behavior of the technical process; extracting a case from the sensor data archive containing at least a subset of all cases generated previously; generating a new case representing a current state of the technical process for which a prediction is to be made; comparing the new case with previous cases as a basis of prediction for future behavior of the technical process; further comprising recording time-stamped sequences of sensor values and storing this data in the sensor data archive; generating cases representing a time-interval within the data archive for all sensor values which is defined by a case time marking the boundary between a preceding and a consecutive projected period of said time-interval; determining a case similarity value for each previous case to the new case by comparing sensor value sequences of preceding periods of said new case and said previous cases; carrying out case retrieval by ordering previous cases with respect to descending similarity to the new case; synchronizing a case time of a one or more most similar case to a latest recorded sensor value of a new case; and using a projected period of the most similar case as a prediction for the future behavior of the technical process.

Claim 2

Independent Claim 236 words

A method for control of a technical process which generates a plurality of sensor values which are recorded and stored in a sensor data archive, comprising cases generated from recorded sensor data representing a behavior of the technical process; extracting a case from the sensor data archive containing at least a subset of all cases generated previously; generating a new case representing a current state of the technical process for which a prediction is to be made; comparing the new case with previous cases as a basis of prediction for future behavior of the technical process; further comprising recording time-stamped sequences of sensor values and storing this data in the sensor data archive; generating cases representing a time-interval within the data archive for all sensor values which is defined by a case time marking the boundary between a preceding and a consecutive projected period of said time-interval; determining a case similarity value for each previous case to the new case by comparing sensor value sequences of preceding periods of said new case and said previous cases; carrying out case retrieval by ordering previous cases with respect to descending similarity to the new case; synchronizing a case time of a one or more most similar case to a latest recorded sensor value of a new case; and using a projected period of the most similar case as a prediction for the future behavior of the technical process.

Claim 3

Dependent Claim 36 words

The method according to claim 1 , further comprising displaying the prediction for the future behavior of the technical process to provide a basis on which a control decision can be made by a human expert.

Claim 4

Dependent Claim 34 words

The method according to claim 1 , further comprising selecting a relevant subset of all sensors which are of interest and using only value sequences of relevant sensors for determining a case similarity value.

Claim 5

Dependent Claim 119 words

The method according to claim 3 , wherein determination of a case similarity value further comprises selecting a corresponding pair of sensors from the previous case and the new case; extracting a number of sensor values in accordance with a case index, said case index specifying which sensors are worthy of consideration during case retrieval and how many values for each pair of sensors need to be compared for a reliable similarity determination; computing a numeric similarity value to determine a correspondence between the sensor value sequences of a selected pair of sensors according to the case index; and summing up numeric similarity values for all pairs of sensors according to the case index and/or a case similarity value.

Claim 6

Dependent Claim 64 words

The method according to claim 4 , further comprising extraction of a case base from the sensor data archive comprising computing a numeric interest value for each time point of the sensor values; determining whether the numeric interest value of each time point exceeds a given threshold; creating a case when the threshold has been exceeded; and inserting the case into the case base.

Claim 7

Dependent Claim 74 words

The method according to claim 4 , further comprising extraction of a case base from the sensor data archive comprising employing a probabilistic distribution of cases throughout the sensor data archive, wherein a probability of creating a case at a given time point depends on a period of time since a last case was created, and a metric of an amount of information in the sensor values in a vicinity of said time point.

Claim 8

Independent Claim 169 words

The method according to claims 4 , 5 or 6 , further comprising extraction of a case index comprising generating on top of the sensor data archive a training case base used to evaluate predictions, and a test case base for which predictions must be made, said training and test case base being disjointed from one another; creating a case index, containing all sensors and substantially all values for each sensor within the preceding periods of the cases; determining an ideal retrieval result representing a best possible prediction for each test case on a basis of known behavior of the test cases in their projected periods; refining the case index by selecting a sensor; reducing the case index's of values for said sensor; determining if the reduction in sensor values leads to an improved retrieval result with respect to the ideal retrieval result; accepting the reduced number of sensor values in the case index, if the determination is positive; and reversing said reduction if the determination is not positive.

Claim 9

Dependent Claim 104 words

The method according to claim 7 , wherein determining whether a reduction in.sensor values leads to an improved retrieval result comprises selecting a test case of the test case base; carrying out a case retrieval from the training case base based on the test case and case index and creating thereby an actual retrieval result; computing a numeric value for a degree of correspondence of an actual retrieval result with respect to an ideal retrieval result; adding numeric values of all test cases to a combined evaluation value; and converting the combined evaluation value into a decision as to whether an evaluation is positive.

Claim 10

Dependent Claim 54 words

The method according to claim 8 , wherein the extraction of a case index is carried out in a long-term maintenance cycle; the extraction of a case base is carried out in a medium-term maintenance cycle; and the prediction for the future behavior of the technical process is performed in a normal prediction cycle.

Claim 11

Dependent Claim 37 words

The method according to claim 8 , wherein the long-term maintenance cycle is once in the installation and rarely thereafter; the medium-term maintenance cycle is daily; and the normal prediction cycle is as regular as every minute.

Claim 12

Dependent Claim 64 words

An apparatus for implementing the method according to claim 1 , comprising a data base containing substantially all sensor data, a data base containing all individual cases used by a prediction and control system; a prediction unit that generates a set of predictions; a unit that generates a new case; and a retrieval unit that determines case similarity values and carries out case retrieval.

Claim 13

Dependent Claim 31 words

The apparatus according to claim 11 , further comprising a unit that uses predictions as a basis of a graphical display of the predictions in order to assist a human controller.

Claim 14

Independent Claim 28 words

The apparatus according to claims 11 or 12 , further comprising a unit that allows a subset of all sensor data to be determined as relevant for retrieval.

Claim 15

Dependent Claim 61 words

The apparatus according to claim 13 , further comprising a unit that selects each corresponding pair of sensor data from a previous and new case and then extracts an appropriate number of values in accordance with a case index; a unit that computes a numeric similarity value; and a unit that adds a result of unit to a case similarity value.

Claim 16

Dependent Claim 80 words

The apparatus according to claim 14 , further comprising a unit that extracts the case base from the sensor data; said case extraction unit comprising a unit that computes a numeric interest value for each time point of the sensor data archive; a unit that determines whether the interest value of each time point exceeds a given threshold; and a unit that creates a case when the threshold has been exceeded and that inserts said case into the case base.

Claim 17

Dependent Claim 131 words

The apparatus according to claim 15 , further comprising a training unit for extracting a case index; said training unit comprising a training case base and a test case base; a unit that creates an initial state of the case index; a unit that determines a best possible prediction of each test case; a unit that selects a sensor; a unit that reduces a number of values of a corresponding sensor that are included in the case index; an evaluation unit that determines if said reduction in sensor values leads to an improvement in prediction performance with respect to a best possible prediction; a unit that makes said reduction permanent in case of a positive evaluation by unit; and a unit that reverses said reduction in case of a negative evaluation.

Claim 18

Dependent Claim 78 words

The apparatus according to claim 16 , further comprising a unit that selects each of the test cases; a unit that carries out a case retrieval to create an actual retrieval result; a unit that computes a numeric measure of a difference in a prediction by an actual retrieval result with respect to an ideal retrieval result; and a unit that converts a combined numeric evaluation into a decision as to whether or not an evaluation was positive.

Claim 19

Dependent Claim 14 words

The method according to claim 1 , wherein the technical process is cement kiln.

Annuity Stage	Window Opens	Payment Deadline	Surcharge Period Until	Estimated Fee	Status
3.5 Years (Year 4)	2007-03-02	2007-09-01	2008-03-02	$2000.00	Window Closed / Lapsed
7.5 Years (Year 8)	2011-03-02	2011-09-01	2012-03-02	$3760.00	Window Closed / Lapsed
11.5 Years (Year 12)	2015-03-02	2015-09-01	2016-03-02	$7700.00	Window Closed / Lapsed

Sensor prediction system utilizing case based reasoning

Abstract

Inventors

Bibliographic Details

Claims Analysis

References Cited (Backward Citations)

USPTO Annuity & Maintenance Windows

Assigned Categories & Tags

Assign Category Tag