Method for the computer-assisted modeling of a wind power installation or a photovoltaic installation with a feed forward neural network

Patent metadata

Field	Value
Publication number	US-10133981-B2
Application number	US-201214239313-A
Country	US
Kind code	B2
Filing date	Jul 24, 2012
Priority date	Aug 18, 2011
Publication date	Nov 20, 2018
Grant date	Nov 20, 2018

Abstract

Disclosed is a method for the computer-assisted modeling of a technical system. One or more output vectors are modeled dependent on one or more input vectors by the learning process of a neural network on the basis of training data of known input vectors and output vectors. Each output vector comprises one or more operating variables of the technical system, and each input vector comprises one or more input variables that influence the operating variable(s). The neural network is a feedforward network with an input layer, a plurality of hidden layers, and an output layer. The output layer comprises a plurality of output clusters, each of which consists of one or more output neurons, the plurality of output clusters corresponding to the plurality of hidden layers. Each output cluster describes the same output vector and is connected to another hidden layer.

First claim

The invention claimed is:

1. A method for predicting one or more operating parameters of a technical system, wherein one or more input variables are supplied, via an input layer (I), to a neural network (NN), the method for predicting comprising a modeling method for the computer-assisted modeling of the technical system, the modeling method comprising:
modeling one or more output vectors on the basis of one or more input vectors by a learning process of the neural network (NN) based on training data comprising known input vectors and output vectors, a respective output vector comprising one or more operating variables of the technical system and a respective input vector comprising one or more input variables which influence the operating variable(s); wherein
the neural network (NN) is a feed-forward network comprising the input layer (I), a plurality of hidden layers (H 1 , H 2 , H 3 ) and an output layer (O), the input layer (I) containing a number of input neurons for describing the input vector(s), and each respective hidden layer (H 1 , H 2 , H 3 ) containing a number of hidden neurons, and the output layer (O) containing a number of output neurons for describing the output vector(s); and wherein the input layer, the plurality of hidden layers and the output layer are connected to one another and wherein the output layer (O) comprises a plurality of output clusters (O 1 , O 2 , O 3 ) each comprising one or more output neurons, and each output cluster being assigned and connected to one of the hidden layers only, and each output cluster (O 1 , O 2 , O 3 ) describing the same output vector as the other output clusters,
wherein the technical system is a wind power installation or a photovoltaic installation;
wherein a first input vector comprises, as input variables, one or more predicted environmental conditions for a future time from a plurality of future times, and
wherein the predicted environmental condition(s) are weather data comprising at least one of the following variables:
ambient temperatures; humidity values; wind speeds; wind directions: values relating to the cloud cover of the sky; and solar radiation values;
wherein a first output vector comprises, as operating variables, amounts of energy (ES) generated by the energy generation installation for a plurality of successive future times, an amount of energy (ES) being the amount of energy generated between two successive future times;
the neural network (NN) determines, using the first input vector, the first output vector having one or more operating parameters of the technical system for at least one output cluster (O 1 , O 2 , O 3 ) of the output layer (O); and
adjusting an amount of energy sold on an energy market, wherein the amount of energy sold is generated by the wind power installation or the photovoltaic installation based on the first output vector; or
supplying predicted generated energy based on the first output vector as control energy in an energy network.

2. The method for predicting as claimed in claim 1 , wherein the input layer (I) of the neural network (NN) is connected to each of the hidden layers (H 1 , H 2 , H 3 ).

3. The method for predicting as claimed in claim 1 , wherein a respective output vector comprises one or more operating variables for a plurality of successive future times within a future period.

4. The method for predicting as claimed in claim 1 , wherein a respective input vector comprises one or more predicted input variables for a future time of successive future times within a future period.

5. The method for predicting as claimed in claim 1 , wherein the input layer (I) also comprises one or more input neurons for describing one or more further input vectors, the further input vector(s) comprising one or more of the operating variables of the technical system which are determined using an analytical model.

6. The method for predicting as claimed in claim 1 , wherein at least 10 hidden layers are provided or each hidden layer comprises between 20 and 30 hidden neurons.

7. The method for predicting as claimed in claim 1 , wherein during the learning process of the neural network, the difference between the output vector described by the output cluster (O 1 , O 2 , O 3 ) and the output vector according to the training data is minimized, as the target variable, for each output cluster (O 1 , O 2 , O 3 ).

8. The method for predicting as claimed in claim 1 , wherein the learning process of the neural network takes place on the basis of error back-propagation.

9. The method for predicting as claimed in claim 1 , wherein the corresponding output vectors are determined for a plurality of output clusters and for all output clusters (O 1 , O 2 , O 3 ), the operating variables of which output vectors are then averaged.

10. A computer program product having a program code which is stored on a non-transitory machine-readable data storage medium and is intended to carry out a method as claimed in claim 1 when the program code is executed on a computer.

11. The method for predicting as claimed in claim 1 , wherein the electrical energy generation installation is a regenerative electrical energy generation installation.

12. The method for predicting as claimed in claim 3 , wherein the future period comprises one or more days and the times have an interval of one hour.

13. The method for predicting as claimed in claim 4 , wherein the future period comprises one or more days and the times have an interval of one hour.

Assignees

• Cleve Jochen
• Grothmann Ralph
• Heesche Kai
• Tietz Christoph
• Zimmermann Hans Georg
• Siemens Ag

Inventors

• Cleve Jochen
• Grothmann Ralph
• Heesche Kai
• Tietz Christoph
• Zimmermann Hans-Georg

Classifications

• G06N3/10 Primary

  Interfaces, programming languages or software development kits, e.g. for simulating neural networks · CPC title

• G06N3/04 Primary

  Architecture, e.g. interconnection topology · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title