FACTOR ANALYSIS DEVICE, FACTOR ANALYSIS METHOD, AND PROGRAM

• Main Authors: KONISHI YOSHICHIKA, MATSUBARA TAKASHI, INAGAKI KAZUYA, KUMADA SHO
• Format: Patent
• Language: English; Japanese
• Published: 23.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; CONTROL OR REGULATING SYSTEMS IN GENERAL; CONTROLLING; COUNTING; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS ORELEMENTS; PHYSICS; REGULATING

To provide a factor analysis device, a method, and a program for accurately estimating a factor of an abnormality of a facility.SOLUTION: A factor analysis device generates a classifying model which acquires input data indicating a time series change of a physical quantity of a facility in a prescribed period, and outputs a classification label indicating a result of estimating whether the acquired input data is normal data or abnormal data by using a neural network. The factor analysis device generates a classification model by adjusting parameters in the neural network on the basis of teacher data containing normal data and abnormal data for each factor, calculates losses of each of the plurality of classification models on the basis of differences between a classification label obtained by inputting input data into a classification model, and a correct answer label indicating whether normal data or abnormal data, and generates, for each of the plurality of classification models, loss variation information indicating a relation between a differential value obtained by differentiating the loss by the physical quantity and indicating a magnitude of the variation of loss to the change of the physical quantity, and each of time points in the period.SELECTED DRAWING: Figure 2 【課題】設備の異常の要因を高精度に推定する要因分析装置、方法及びプログラムを提供する。【解決手段】要因分析装置は、設備に関する物理量の所定の周期内における時系列変化を示す入力データを取得し、ニューラルネットワークを利用し、取得した入力データが正常データか異常データかを推定した結果を示す分類ラベルを出力する分類モデルを生成する。要因毎に、正常データと異常データとを含む教師データに基づいてニューラルネットワークにおけるパラメータを調整し分類モデルを生成し、入力データを分類モデルに入力して得られる分類ラベルと、正常データであるか異常データであるかを示す正解ラベルとの差分に基づいて、複数の分類モデル夫々について損失を算出し、損失を物理量で微分した、物理量の変化に対する損失の変化の大きさを示す微分値と、周期内における各時点と、の関係を示す損失変動情報を、複数の分類モデルの夫々について生成する。【選択図】図２