@article{oai:ipsj.ixsq.nii.ac.jp:00219017, author = {大島, 悠 and 石曽根, 毅 and 樋口, 知之 and Hisashi, Oshima and Tsuyoshi, Ishizone and Tomoyuki, Higuchi}, issue = {3}, journal = {情報処理学会論文誌デジタルプラクティス(TDP)}, month = {Jul}, note = {近年,スマートメータからの電力需要値を用いた在宅推定の需要が高まっている.時刻ごとの世帯の在宅状況を把握することは,災害時の避難計画の策定,配送業と営業活動の経路最適化などといった応用可能性がある.先行研究として,15分や60分間隔の電力需要値の平均値や標準偏差といった説明変数を用いて,在不在を判別する教師あり学習の2値判別モデルを構築することが行われてきた.そこで本稿では新たな教師あり学習の判別モデルを構築した.先行研究との違いとしては2つある.1つ目は,国内のスマートメータの規格に従い,測定時間頻度が30分間隔であり3相交流の和として定義される電力需要値を用いたことである.これは服部・篠原による研究を除く先行研究との違いである.2つ目は電力比という1変数を主たる説明変数としたことである.短い時間間隔における電力需要値の絶対量とその付近の変動だけでなく,在宅時と比較した生活パターンの逸脱度が在宅状況の判別に重要な情報であることを示した.結果として新たに構築した2つの判別モデルの精度が,先行研究と同程度以上のものであることが示された., Recently, a demand for occupancy detection is increased, since it is beneficial to society in many ways. Knowing whether a household is occupied by its residents in real time contributes to removing absent delivery, route optimization for work activities. The existing supervised classification models can perform at from about 70 percent accuracy to 90. On the other hand, There are two problems. First, The previous classification models always require high-resolution electricity data such as 1 second interval measurement and each data from three phases of smart meter. The requirement is too difficult to get in Japan, due to the installation methods of the sensor. Second, the previous models only use electricity data's absolute and near-absolute variation for short time intervals as explanatory variables. It is not enough to explain occupancy, since it will lead to a misclassification in the situation where a household is occupied regularly without using electricity. In this paper, we propose new classification models (method (a) and method (b)) which has two differences to overcome the previous problems. First, our models only require electricity data measured at 30 minutes intervals as main trunk electricity data, which has been measured from smart meter in general household. Second, we propose a new variable called an electricity data ratio. The electricity data ratio at given time can be regarded as the degree of deviation from the lifestyle pattern seen from the time when the person is at home. It can reduce misclassification when a household is occupied regularly without appliances. The method (a) is to set a threshold on electricity data ratio. The numerator of the ratio at given time is electricity data that is observed at 30 minutes intervals, and the denominator is electricity data at the same time when a household may be occupied. Definition of when a household may be occupied is when electricity data is maximum value for 2 weeks before and after and also 30 minutes before and after. Given that i is number of days and j is the observation point which is measured at 30 minutes intervals in the i, electricity data ratio is defined by<inline-formula><tex-math notation="TeX"><![CDATA[$R_{i,j} = \frac{y_{i,j}}{y_{i_*,j_*}},i_*,j_* = \mathrm{argmax}_{k = i - 14,i - 13,...,i + 14,l = j - 1,j,j + 1}(y_{k,l})$]]></tex-math></inline-formula>. The method (b) is based on Random Forest with eight explanatory variables. The ratio is added to the representative seven explanatory variables of the previous studies. As a result, method (a) and method (b) detected occupancy status with equally or more accurately compared with previous studies.}, pages = {53--65}, title = {家庭用スマートメータへの適用を目的とする世帯の在宅推定手法の開発}, volume = {3}, year = {2022} }