US 7,610,252 B2Granted

Method for developing a process model

AssigneePowitec Intelligent Technologies GmbH

InventorsFranz Wintrich, Thomas Vesper

Life of this patent

Estimating the term…

Loading patent…

Abstract

A neuronal network is trained using measurement data of state variables comprising measurement data from input channels and measurement data from at least one output channel. The neuronal network is tested using measurement data from the input channels and measurement data from the output channel, and a first standard deviation is determined from the deviations of the predicted values for the output channel from the measurement data of the output channel. The measurement data of at least one input channel are replaced by a distribution. Values for the output channel are again calculated using the distribution or parts thereof and a second standard deviation of the calculated values for the output channel is determined from the associated measurement data. In the case of an increase in the second standard deviation compared with the first standard deviation, the input channel is significant for the neuronal network.

Claims20

Dependent claims

A method for developing a process model for regulating a combustion process in a plant in which, while at least oxygen is supplied, material is converted by way of the combustion process with at least one flame being formed, and a state of a system in the plant is described by state variables, the method comprising:

setting up a neuronal network;
training the neuronal network using at least some measurement data of the state variables, wherein the measurement data comprises measurement data from a plurality of input channels and measurement data from at least one output channel;
performing a first evaluation of the neuronal network, wherein the performing of the first evaluation comprises
(a) calculating first values for the output channel using the neuronal network and at least some of the measurement data from the plurality of input channels, and
(b) calculating a first standard deviation from deviations of the first values from at least some of the measurement data of the output channel;
providing a distribution for use in place of measurement data from at least one input channel of the plurality of input channels;
performing a second evaluation of the neuronal network, wherein the performing of the second evaluation comprises
(a) calculating second values for the output channel using the neuronal network and predetermined data, wherein the predetermined data comprises at least a portion of the distribution, and
(b) calculating a second standard deviation from deviations of the second values from at least some of the measurement data of the output channel;
determining whether the at least one input channel is significant for the neuronal network, wherein the at least one input channel is significant if the second standard deviation is larger than the first standard deviation.

7 more elements

A control loop apparatus for regulating combustion in a plant in which the combustion converts material and forms at least one flame, the control loop comprising:

a plurality of sensors for obtaining measurement data for state variables, wherein the plurality of sensors is adapted so that the measurement data comprises measurement data from a plurality of input channels and measurement data from at least one output channel, and wherein the plurality of sensors comprises
(a) at least one observation device, which is for imaging the flame, and
(b) further sensors;
at least one adjustment device for carrying out actions to be taken, wherein the adjustment device is for adjusting supplying of material, oxygen, air or any combination thereof;
and
a computer adapted for at least
(a) training a neuronal network using at least some of the measurement data,
(b) performing a first evaluation of the neuronal network, wherein the performing of the first evaluation comprises
(1) calculating first values for the output channel using the neuronal network and at least some of the measurement data from the plurality of input channels, and
(2) calculating a first standard deviation from deviations of the first values from at least some of the measurement data of the output channel,
(c) providing a distribution for use in place of measurement data from at least one input channel of the plurality of input channels,
(d) performing a second evaluation of the neuronal network, wherein the performing of the second evaluation comprises
(1) calculating second values for the output channel using the neuronal network and predetermined data, wherein the predetermined data comprises at least a portion of the distribution, and
(2) calculating a second standard deviation from deviations of the second values from at least some of the measurement data of the output channel,
(e) determining whether the at least one input channel is significant for the neuronal network, wherein the at least one input channel is significant if the second standard deviation is larger than the first standard deviation, and
(f) determining the actions to be taken by evaluating at least some of the measurement data via the neuronal network.

13 more elements

Dependencies are inferred from the claim text.

Facts

Application: 11/837,950Filed 13 Aug 2007
Priority: EP 0601710517 Aug 2006
Published: US 2008/0046391 A121 Feb 2008
Granted: 27 Oct 2009
Term ends: Not estimatedThe source records don’t support an estimate.

Classification7 codes

G05B 13/042
Involving the use of models or simulators: in which a parameter or coefficient is automatically adjusted to optimise the performance
F23N 1/022
Conjointly with air supply: using electronic means
F23N 5/082
Using light-sensitive elements: using electronic means
F23N 2223/00
Signal processing

Family

US 2008/0046391 A1Pre-grant publication · published 21 Feb 2008
US 7,610,252 B2This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key