Solar irradiation prediction using deep learning with end-to-end training
AssigneeSiemens Aktiengesellschaft
InventorsTi-chiun Chang, Patrick Reeb, Joachim Bamberger, Kuan-Chuan Peng, Jan Ernst
Life of this patent
Estimating the term…
Abstract
Deep learning is used to train a neural network for end-to-end prediction of short term (e.g., 20 minutes or less) solar irradiation based on camera images and metadata. The architecture of the neural network includes a recurrent network for temporal considerations. The images and metadata are input at different locations in the neural network. The resulting machine-learned neural network predicts solar irradiation based on camera images and metadata so that a solar plant and back-up power source may be controlled to minimize output power variation.
Claims12
A system for solar irradiation prediction, the system comprising:
- a camera positioned to capture a sequence of images of a sky;
- an image processor configured to receive the images and predict the solar irradiation from sequential input of the images to a machine-learned neural network, the machine-learned neural network having a convolutional layer concatenated with a following recurrent layer;
- and
- a controller configured to control supply of energy from a solar plant and a back-up power source based on the predicted solar irradiation,
- wherein the image processor is configured to predict the solar irradiation as a list of values for different times in a future, the image processor configured to weight the values of the list based on a cloud speed.
A method for solar irradiation prediction for a power system, the method comprising:
- capturing images of a sky region;
- determining a time of day, date, or time of day and date;
- predicting the solar irradiation on a solar plant for future times in response to the input of one of the images with a machine-trained deep network architecture, the deep network architecture having a convolutional network to which the images are input in sequence, a recurrent network receiving an output of the convolutional network, and a dense layer receiving an output of the recurrent network and the time of day, the date, or the time of day and the date for the predicting;
- controlling the solar plant based on the predicted solar irradiation;
- and
- estimating cloud speed, and wherein predicting comprises predicting with the future times limited by the cloud speed.
Dependencies are inferred from the claim text.