Method for the computer-assisted modeling of a technical system
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06F-015/18
G06N-003/08
G06N-003/04
출원번호
US-0992799
(2011-11-16)
등록번호
US-9489619
(2016-11-08)
우선권정보
DE-10 2010 062 832 (2010-12-10)
국제출원번호
PCT/EP2011/070243
(2011-11-16)
§371/§102 date
20130710
(20130710)
국제공개번호
WO2012/076306
(2012-06-14)
발명자
/ 주소
Düll, Siegmund
Hans, Alexander
Udluft, Steffen
출원인 / 주소
SIEMENS AKTIENGESELLSCHAFT
대리인 / 주소
Slayden Grubert Beard PLLC
인용정보
피인용 횟수 :
1인용 특허 :
13
초록▼
A method for computer-assisted modeling of a technical system is disclosed. At multiple different operating points, the technical system is described by a first state vector with first state variable(s) and by a second state vector with second state variable(s). A neural network comprising a special
A method for computer-assisted modeling of a technical system is disclosed. At multiple different operating points, the technical system is described by a first state vector with first state variable(s) and by a second state vector with second state variable(s). A neural network comprising a special form of a feed-forward network is used for the computer-assisted modeling of said system. The feed-forward network includes at least one bridging connector that connects a neural layer with an output layer, thereby bridging at least one hidden layer, which allows the training of networks with multiple hidden layers in a simple manner with known learning methods, e.g., the gradient descent method. The method may be used for modeling a gas turbine system, in which a neural network trained using the method may be used to estimate or predict nitrogen oxide or carbon monoxide emissions or parameters relating to combustion chamber vibrations.
대표청구항▼
1. A method for the computer-assisted modeling of a technical system, comprising: for a plurality of operating points, describing the operation of the technical system by a first state vector with one or more first state variables and by a second state vector with one or more second state variables;
1. A method for the computer-assisted modeling of a technical system, comprising: for a plurality of operating points, describing the operation of the technical system by a first state vector with one or more first state variables and by a second state vector with one or more second state variables;modeling the technical system training a neural network comprising at least one feed-forward network based on training data from known first and second state vectors for a plurality of operating points;wherein the at least one feed-forward network contains a plurality of neural layers comprising an input layer, at least one hidden layer and an output layer, wherein the neural layers are interconnected by connectors with respective weights and the input layer is linked with at least one first state vector and the output layer with at least one second state vector;wherein at least one connector of the at least one feed-forward network is a bridging connector which connects a neural layer with the output layer while bridging at least one hidden layer; andwherein the at least one feed-forward network comprises a plurality of hidden layers and each hidden layer not connected directly with the output layer is connected with the output layer by a bridging connector,wherein the neural network furthermore contains a recurrent neural network, coupled with the at least one feed-forward network, comprising an input layer and a recurrent hidden layer, wherein the input layer of the recurrent neural network comprises first state vectors at chronologically sequential operating points of the technical system comprising one current and one or more past operating points,wherein each first state vector at the respective operating point of the input layer of the recurrent neural network is connected by way of a connector with corresponding weight with a hidden state vector at the respective same operating point of the recurrent hidden layer of the recurrent neural network, andwherein the hidden state vector at the current operating point represents the input layer of the at least one feed-forward network and the output layer of said feed-forward network represents the second state vector at the current operating point. 2. The method as claimed in claim 1, wherein the input layer of the at least one feed-forward network is connected with the output layer by a bridging connector. 3. The method as claimed in claim 1, wherein the neural network is trained by a gradient descent method using error back-propagation. 4. The method of claim 1, wherein the input layer of the recurrent neural network furthermore comprises one or more first state vectors at chronologically sequential future operating points of the technical system and each first state vector at a future operating point is connected by way of a connector with corresponding weight with a hidden state vector at the future operating point of the recurrent hidden layer of the recurrent neural network, wherein each hidden state vector at a future operating point forms the input layer of the at least one feed-forward network for the future operating point, andwherein the output layer of each feed-forward network for a future operating point represents the second state vector at the future operating point. 5. The method of claim 4, wherein the connectors of all feed-forward networks which connect neural layers corresponding to one another have the same weights. 6. The method of claim 1, wherein the connectors between the input layer of the recurrent neural network and the recurrent hidden layer of the recurrent neural network at past operating points have the same weights. 7. The method of claim 4, wherein the connectors between the input layer of the recurrent neural network and the recurrent hidden layer of the recurrent neural network at the current and at future operating points have the same weights. 8. The method of claim 1, wherein the connectors of the hidden recurrent layer of the recurrent neural network, which extend out of hidden state vectors at past operating points, have the same weights. 9. The method of claim 4, wherein the connectors of the hidden recurrent layer of the recurrent neural network, which extend into hidden state vectors at future operating points, have the same weights. 10. The method of claim 1, wherein a gas turbine is modeled as a technical system. 11. The method of claim 10, wherein the first state vectors of the gas turbine comprise one or more of the following state variables of the gas turbine: one or more temperature values at or in the gas turbine, in particular one or more fuel gas temperature values, one or more fuel gas pressure values at or in the gas turbine, andone or more control values for setting one or more of the partial fuel flows fed to the gas turbine. 12. The method of claim 10, wherein the second state vectors comprise one or more of the following state variables: one or more emission values for nitrogen oxides, one or more emission values for carbon monoxide; and one or more parameters which describe vibrations of the combustion chamber of the gas turbine. 13. A method for computer-assisted estimation of states of a technical system, wherein the technical system is modeled by: for a plurality of operating points, describing the operation of the technical system by a first state vector with one or more first state variables and by a second state vector with one or more second state variables;modeling the technical system by training a neural network comprising at least one feed-forward network based on training data from known first and second state vectors for a plurality of operating points;wherein the at least one feed-forward network contains a plurality of neural layers comprising an input layer, at least one hidden layer and an output layer, wherein the neural layers are interconnected by connectors with respective weights and the input layer is linked with at least one first state vector and the output layer with at least one second state vector;wherein at least one connector of the at least one feed-forward network is a bridging connector which connects a neural layer with the output layer while bridging at least one hidden layer;wherein the at least one feed-forward network comprises a plurality of hidden layers and each hidden layer not connected directly with the output layer is connected with the output layer by a bridging connector; anddetermining one or more second state vectors of the technical system using the trained neural network based on one or more first state vectors of the technical system. 14. A computer program product having program code stored on a non-transitory computer-readable medium and executable by a processor for modeling a technical system by: for a plurality of operating points, describing the operation of the technical system by a first state vector with one or more first state variables and by a second state vector with one or more second state variables;modeling the technical system by training a neural network comprising at least one feed-forward network based on training data from known first and second state vectors for a plurality of operating points;wherein the at least one feed-forward network contains a plurality of neural layers comprising an input layer, at least one hidden layer and an output layer, wherein the neural layers are interconnected by connectors with respective weights and the input layer is linked with at least one first state vector and the output layer with at least one second state vector;wherein at least one connector of the at least one feed-forward network is a bridging connector which connects a neural layer with the output layer while bridging at least one hidden layer; andwherein the at least one feed-forward network comprises a plurality of hidden layers and each hidden layer not connected directly with the output layer is connected with the output layer by a bridging connector.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (13)
Dixon, Kristian Robert; Düll, Siegmund; Egedal, Per; Esbensen, Thomas; Sterzing, Volkmar, Control of a wind turbine, rotor blade and wind turbine.
Alonso, Jose L.; Bogdoll, Dieter; Dull, Siegmund; Sancewich, Glenn E.; Sterzing, Volkmar, Method and apparatus for deriving diagnostic data about a technical system.
Sterzing, Volkmar; Udluft, Steffen; Singh, Jatinder; Brummel, Hans-Gerd; Sancewich, Glenn E., Method for computer-aided closed-loop and/or open-loop control of a technical system.
Schneegaβ, Daniel; Udluft, Steffen, Method for computer-aided control and/or regulation using two neural networks wherein the second neural network models a quality function and can be used to control a gas turbine.
Schäfer, Anton Maximilian; Udluft, Steffen; Zimmermann, Hans-Georg, Method for the computer-assisted control and/or regulation of a technical system where the dynamic behavior of the technical system is modeled using a recurrent neural network.
Hans, Alexander; Schneegaβ, Daniel; Schäfer, Anton Maximilian; Sterzing, Volkmar; Udluft, Steffen, Method for the computer-assisted exploration of states of a technical system.
Hans, Alexander; Udluft, Steffen, Method for the computer-assisted learning of a control and/or a feedback control of a technical system using a modified quality function and a covariance matrix.
Brummel, Hans-Gerd; Düll, Siegmund; Singh, Jatinder P.; Sterzing, Volkmar; Udluft, Steffen, Method for the computerized control and/or regulation of a technical system.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.