2019 Fiscal Year Final Research Report
Local and short-time PV generation prediction from All sky image.
| Project/Area Number |
17H01922
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Design and evaluation of sustainable and environmental conscious system
|
|---|
| Research Institution | Shizuoka Institute of Science and Technology |
|---|
Principal Investigator |
Kato Takekazu 静岡理工科大学, 理工学部, 准教授 (30362859)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
延原 章平 京都大学, 情報学研究科, 准教授 (00423020)
|
|---|
| Project Period (FY) |
2017-04-01 – 2020-03-31
|
| Keywords | 太陽光発電予測 / 電力マネージメント / 深層学習 / 画像認識 |
|---|
| Outline of Final Research Achievements |
Our project aims to predict the local and short-term fluctuations of solar power generation by using the whole sky images.
First, we developed the whole-sky observation system that continuously observes and records whole-sky images 24 hours, 236 days capturing a high dynamic range image every second, and its calibration algorithm. Second, we developed two basic methods. One is a CV based sky image analysis model, a sky model that separates the sun, clouds, and blue-sky areas from the sky image. The other one is a machine learning based prediction method that predicted the movement of clouds and CNN and RNN from image sequence of the past few seconds. Finally, we developed a method to improve the prediction accuracy by performing image processing in advance and DNN by combination of the basic method.
|
| Free Research Field |
機械学習、画像認識、電力マネージメント
|
| Academic Significance and Societal Importance of the Research Achievements |
気象データや衛星画像を使って行う技術はすでに開発されているが，時間分解能，空間解像度が荒く，局地的な変動をピンポイントでかつ秒単位の変動を予測することはできていない．このような問題に対して，本研究はピンポイントで時間分解能が高い（秒単位）変動予測を目指している点で独創的である．
このような局地的かつ秒単位の予測が可能になれば，電力消費機器を太陽光発電の変動に連動して制御することで，太陽光発電の電力変動を吸収しつつ有効活用することができ，電力グリッドの局地的な安定性維持，自家消費型の電力管理，さらに電力網の整っていない地域における独立型電力網制御など，その恩恵は広く社会的な意義が高い．
|
