Treffer: 農家自作型 IoT システムによる CO2 遠隔監視の経営的 実用性と状態空間モデルを用いた異常値検出法の検討

Our first aim was to clarify the effectiveness of using a CO2 sensor with a farmer-built IoT (Internet of Things) system that had proven to be of practical use in rice nursery growing. We conducted an experiment at kikurage mushroom and strawberry growing sites to observe whether the system worked and asked the property owners to evaluate the system. At both sites the system provided CO2 concentration data with almost the same accuracy as the existing detection devices that the farmers had been using. Because our system uses an IoT study kit and Raspberry Pi, which are provided cheaply, the farmers evaluated it as practical and lowcost. We put one CO2 sensor in the strawberry greenhouse, with an annual cost of 25,620 yen; this was much lower than the cost of the existing devices that the farmer had used in the past. In the experiment in the kikurage mushroom house, we placed four CO2 sensors, with an annual cost of 72,480 yen; this was higher than the cost of the handy-type detection device that the farmer had been using. However, our system sends CO2 concentration data to cell phones automatically, and it detected high concentrations at times when the farmer was engaged elsewhere. The farmer therefore considered that the annual cost was not high. The CO2 sensor has tendency to bring value shifts upwards; it takes place when micro dusts adhere the sensor tip. We are therefore unable to write a threshold CO2 concentration into the program to detect anomalous values, such as those occurring from changes in temperature with seasonal change. Our second aim was to clarify a method of using a state-space model to detect anomalous values with the farmer-built IoT system. We created a calculation method and used the variance in lag between the predicted and measured values as an evaluation criterion to judge the anomalous values. The method worked in a test using CO2 concentration data from the strawberry greenhouse. [ABSTRACT FROM AUTHOR]

本稿は, これまでに筆者らが市販キットを応用して開発した農家自作型 IoT システムを, 施設型の作物におけ る実用的なシステムとすることを目的に, 新たに接続した CO2 センサが実用に耐える機能を有するか, および費 用に対して効果が上回ると経営者が判断できるかを確認することを第一の課題とした．また, 第二の課題として, 測定誤差やセンサの劣化の影響, 季節的な変化を含む計測値に対して, 計算能力の低い同システムを前提としな がら状態空間モデルを用い, 温度や CO2 の値の分散により異常値を検出できることを明らかにする．この課題解 明のため, キクラゲ施設とイチゴハウスにおいて実用試験に供した．試験の結果, キクラゲ施設, イチゴハウス とも, 実用的と認められる CO2 濃度の計測および値の送信ができた．また効果は費用を上回ると経営者に評価さ れた.計測値にセンサの劣化や季節的変化の影響を含むイチゴハウスにおける異常値検出の方法として,Python の機械学習ライブラリを用いたパラメータの推定を作業用 PC で実施し, 計測した値をもとにした簡単な計算の みを IoT システムに計算させる二段階の計算方法を開発した.これにより,効果的に異常値を検出できることを 明らかにした. [ABSTRACT FROM AUTHOR]