本文へジャンプ

成果報告書詳細
管理番号20190000000104
タイトル*2018年度中間年報 次世代人工知能・ロボットの中核となるインテグレート技術開発 人工知能技術の社会実装に向けた研究開発・実証 人工知能技術の風車への社会実装に関する研究開発
公開日2019/4/16
報告書年度2018 - 2018
委託先名株式会社日立製作所 国立研究開発法人産業技術総合研究所 国立大学法人東京大学
プロジェクト番号P18002
部署名ロボット・AI部
和文要約
英文要約Title: Adopting Artificial Intelligence for Wind Farm Control. (FY2018-FY2019) FY2018 Annual Report

I. Development of AI Algorism for wind turbine inflow wind prediction using LiDAR data
In order to prepare teacher data for the AI module that will be developed in this project, a field experiment was conducted using Vertical Azimuth Display LiDAR at an actual wind turbine site. The terrain is almost flat and there is a downwind type 5.2MW large wind turbine of with rotor diameter D: 127m at that site. The LiDAR was installed at the position about 1D away from the wind turbine toward in the upwind direction, and the measurement started from December 27, 2018.
A prototype system of the AI module for wind condition estimation was developed in this year. The system conducts time-series forecasting using existing wind speed/direction data measured by a cup anemometer. Long short-term memory, which is one of the most powerful and useful type of neural network, was applied as a prediction model. The prototype system was evaluated on the experiment of short-period wind direction forecasting and its effectiveness was confirmed.

II. Development of algorithm for power maximization of windfarm by the AI
To address the power losses induced by wakes in the wind farm, yaw-based wake steering is utilized to maximize the power production. There are mainly three achievements reached this year.
Firstly, a new multiple wake model applicable to large wind farm was developed, in which the local effective turbulence and wake mixing effects were considered. Comparisons with numerical simulation data showed that the proposed model can accurately predict both mean and turbulent flow field in the wind farm.
Secondly, the yaw-based power production maximization program was developed by utilizing the gradient-based optimization algorithm. The yaw angle range of ±15°was demonstrated as the optimal one, where a favorable balance between the safety and power maximization was achieved.
Finally, the wind farm layout was systematically varied and effects of turbine density on the energy improvement from yaw-based optimization were quantitatively evaluated.

III. Development of wind turbine control using the above algorithm
Implementation method of "wind prediction algorithm using AI" and "wind farm power production improvement algorithm (wake steer)” was examined. There are mainly three achievements reached this year.
The relationship between each related algorithm and the wind turbine control algorithm was organized. Overall control logic in this development was constructed.
The allowable command value in wake steering control was evaluated. The moment of the yaw bearing and the tower base was simulated, and set to ± 15 deg.
The behavior of the wind turbine control when the wake steer was implemented was evaluated by simulation. Wind turbine control could follow wake steer control by using wind prediction results and improve the power production of the wind farm using these algorithms.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る