Method for the computer-assisted control and/or regulation of a technical system
AssigneeSiemens Aktiengesellschaft
InventorsAnton Maximilian Schäfer, Steffen Udluft
Life of this patent
Estimating the term…
Abstract
A method for the computer-assisted control and/or regulation of a technical system is provided. The method is used to efficiently reduce a high-dimensional state space describing the technical system to a smaller dimension. The reduction of the state space is performed using an artificial recurrent neuronal network. In addition, the reduction of the state space enables conventional learning methods, which are only designed for small dimensions of state spaces, to be applied to complex technical systems with an initially large state space, wherein the conventional learning methods are performed in the reduced state space. The method can be used with any technical system, especially gas turbines.
Claims20
A method for computer-aided control and/or regulation of a technical system, comprising:
- receiving from a technical system a plurality of state variables having a first state space with a first dimension for a plurality of points in time, the first dimension comprising a high-dimensional state space;
- reducing the high-dimensional state space with a recurrent neural network by modeling the state of the first state space as a training data using the recurrent neural network with an aid of a plurality of known states, the recurrent neural network comprises:
- an input layer, a recurrent hidden layer, and an output layer,
- wherein the input layer and the output layer are formed from the high-dimensional state space in the first state space for the plurality of points in time, and
- wherein the recurrent hidden layer is formed by a plurality of hidden states with a plurality of hidden state variables, in which a hidden state at time t is coupled via a matrix to a next hidden state at a subsequent point in time t+1, the recurrent hidden layer having a second state space with a second dimension such that the second dimension is lower than the first dimension;
- and
- performing a learning and/or optimization method for a regulation of the technical system on the plurality of hidden states in the second state space having a lower dimensional state space than the first state space by executing a plurality of actions on the technical system.
A non-transitory computer readable medium storing a program code, when the program executes on a processor of a computer, the program comprising:
- receiving from a technical system a plurality of state variables having a first state space with a first dimension for a plurality of points in time, the first dimension comprising a high-dimensional state space;
- reducing the high-dimensional state space with a recurrent neural network by modeling the state of the first state space as a training data using the recurrent neural network with an aid of a plurality of known states, the recurrent neural network comprises:
- an input layer, a recurrent hidden layer, and an output layer,
- wherein the input layer and the output layer are formed from the high-dimensional state space in the first state space for the plurality of points in time, and
- wherein the recurrent hidden layer is formed by a plurality of hidden states with a plurality of hidden state variables, in which a hidden state at time t is coupled via a matrix to a next hidden state at a subsequent point in time t+1, the recurrent hidden layer having a second state space with a second dimension such that the second dimension is lower than the first dimension;
- and
- performing a learning and/or optimization method for a regulation of the technical system on the plurality of hidden states in the second state space having a lower dimensional state space than the first state space by executing a plurality of actions on the technical system.
Dependencies are inferred from the claim text.