Multi-layer hybrid model power generation prediction method and computing system

Patent metadata

Field	Value
Publication number	US-2021287310-A1
Application number	US-202016818177-A
Country	US
Kind code	A1
Filing date	Mar 13, 2020
Priority date	Mar 13, 2020
Publication date	Sep 16, 2021
Grant date	—

Abstract

Example implementations described herein involve a multi-layer hybrid model power generation prediction method and computing system. The example implementations described herein are configured to predict power generation that can be utilized for short-term planning by utilities to optimize operation planning to ensure economic and stable operation of the electricity.

First claim

What is claimed is:

1 . A method for generating a multi-layer hybrid model power generation prediction, comprising:
extracting features from historical data of one or more renewable energy systems; generating a layer 1 prediction model from the extracted features; generating a plurality of layer 2 prediction models based on one or more of an output from the layer 1 prediction model, an output from layer 3 prediction models, and the extracted features; generating the at least one layer 3 hybrid prediction model based on an output of a selection of layer 2 prediction models from the plurality of layer 2 prediction models; and executing the at least one layer 3 hybrid prediction model on new data received from the one or more renewable energy systems to provide prediction on power generation of the one or more renewable energy systems.

2 . The method of claim 1 , wherein the generating the at least one layer 3 hybrid prediction model comprise generating multiple layer 3 hybrid prediction models based on different combinations of the selected layer 2 models, wherein an output of the at least one layer 3 hybrid prediction model is provided to generate additional layer 2 models.

3 . The method of claim 1 , wherein the generating the layer 1 prediction model from the extracted features comprises:
extracting the extracted features from historical data of the one or more renewable energy systems; building the layer 1 prediction model from the extracted features, executing a training and validation process on the layer 1 prediction model.

4 . The method of claim 1 , wherein the generating the plurality of layer 2 prediction models based on one or more of the output from the layer 1 prediction model, the output from layer 3 prediction models, and the extracted features comprises:
extracting additional features from one or more of the output from the layer 1 prediction model and the extracted features; generating, from the layer 1 prediction model and the output from the layer 1 prediction model, the plurality of layer 2 prediction models based on the extracted features and the additional features; and executing a training and validation process on the plurality of layer 2 prediction models.

5 . The method of claim 1 , wherein the generating the plurality of layer 2 prediction models based on one or more of the output from the layer 1 prediction model, the output from layer 3 prediction models, and the extracted features comprises:
extracting additional features from one or more of the output from the layer 1 prediction model and the extracted features; generating, from the layer 3 prediction models and the output from the layer 3 prediction models, the plurality of layer 2 prediction models based on the extracted features and the additional features; and executing a training and validation process on the plurality of layer 2 prediction models.

6 . The method of claim 1 , wherein the generating the at least one layer 3 hybrid prediction model based on the output of the selection of the layer 2 prediction models from the plurality of layer 2 prediction models comprises:
generating the at least one layer 3 hybrid prediction model from the selection of the layer 2 prediction models and the output of the selection of the layer 2 prediction models; and executing post-processing comprising a smoothing process and a hyper-parameters tuning process on the at least one layer 3 hybrid prediction model.

7 . A non-transitory computer readable medium, storing instructions for generating a multi-layer hybrid model power generation prediction, comprising:
extracting features from historical data of one or more renewable energy systems; generating a layer 1 prediction model from the extracted features; generating a plurality of layer 2 prediction models based on one or more of an output from the layer 1 prediction model, an output from layer 3 prediction models, and the extracted features; generating the at least one layer 3 hybrid prediction model based on an output of a selection of layer 2 prediction models from the plurality of layer 2 prediction models; and executing the at least one layer 3 hybrid prediction model on new data received from the one or more renewable energy systems to provide prediction on power generation of the one or more renewable energy systems.

8 . The non-transitory computer readable medium of claim 7 , wherein the generating the at least one layer 3 hybrid prediction model comprise generating multiple layer 3 hybrid prediction models based on different combinations of the selected layer 2 models, wherein an output of the at least one layer 3 hybrid prediction model is provided to generate additional layer 2 models.

9 . The non-transitory computer readable medium of claim 7 , wherein the generating the layer 1 prediction model from the extracted features comprises:
extracting the extracted features from historical data of the one or more renewable energy systems; building the layer 1 prediction model from the extracted features; and executing a training and validation process on the layer 1 prediction model.

10 . The non-transitory computer readable medium of claim 7 , wherein the generating the plurality of layer 2 prediction models based on one or more of the output from the layer 1 prediction model, the output from layer 3 prediction models, and the extracted features comprises:
extracting additional features from one or more of the output from the layer 1 prediction model and the extracted features; generating, from the layer 1 prediction model and the output from the layer 1 prediction model, the plurality of layer 2 prediction models based on the extracted features and the additional features; and executing a training and validation process on the plurality of layer 2 prediction models.

11 . The non-transitory computer readable medium of claim 7 , wherein the generating the plurality of layer 2 prediction models based on one or more of the output from the layer 1 prediction model, the output from layer 3 prediction models, and the extracted features comprises:
extracting additional features from one or more of the output from the layer 1 prediction model and the extracted features; generating, from the layer 3 prediction models and the output from the layer 3 prediction models, the plurality of layer 2 prediction models based on the extracted features and the additional features; and executing a training and validation process on the plurality of layer 2 prediction models.

12 . The non-transitory computer readable medium of claim 7 , wherein the generating the at least one layer 3 hybrid prediction model based on the output of the selection of the layer 2 prediction models from the plurality of layer 2 prediction models comprises:
generating the at least one layer 3 hybrid prediction model from the selection of the layer 2 prediction models and the output of the selection of the layer 2 prediction models; and executing post-processing comprising a smoothing process and a hyper-parameters tuning process on the at least one layer 3 hybrid prediction model.

13 . An apparatus communicatively coupled to one or more renewable energy systems, the apparatus, comprising:
a processor, configured to:
receive data from the one or more renewable energy systems;
extract features from the data from the one or more renewable energy systems;
execute a layer 1 prediction model to generate a first output;
execute a plurality of layer 2 prediction models on the output from the layer 1 prediction model, the extracted features, and the additional features; and
provide output from the plurality of

Assignees

• Hitachi Ltd

Inventors

• Chongfuangprinya Panitarn
• Yang Bo
• Ye Yanzhu
• Abe Masanori
• Gao Jian

Classifications

• G06N20/20

  Ensemble learning · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• Y02E40/70

  Smart grids as climate change mitigation technology in the energy generation sector · CPC title

• G06Q50/06 Primary

  Energy or water supply · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title