ABNORMALITY DETERMINATION DEVICE, ABNORMALITY DETERMINATION MODEL GENERATION METHOD AND ABNORMALITY DETERMINATION METHOD

To provide an abnormality determination device, an abnormality determination model generation method and an abnormality determination method capable of generally detecting abnormality of a facility having nonlinear characteristics with high accuracy without individually monitoring each facility.SOLUTION: An abnormality determination device 1 includes time series signal segmentation means, normal vector registration means, abnormality determination model registration means, and abnormality determination means. When an abnormality determination flag is a first type, the abnormality determination means calculates distances to registered normal vectors, extracts the prescribed number of normal vectors as neighboring data in an ascending order of the distances, calculates a distance between the gravity center vector of the neighboring data and an M-dimensional vector of an object of abnormality determination, and performs abnormality determination of a facility on the basis of the distance. When the abnormality determination flag is a second type, the abnormality determination means calculates deviation from a main component on the basis of a conversion coefficient of the main component calculated in advance, and performs abnormality determination of the facility on the basis of the deviation.SELECTED DRAWING: Figure 1 【課題】各設備を個別に監視することなく、非線形な特性を有する設備の異常を汎用的かつ高精度に検知することができる異常判定装置、異常判定モデル生成方法および異常判定方法を提供する。【解決手段】異常判定装置１は、時系列信号切り出し手段と、正常ベクトル登録手段と、異常判定モデル登録手段と、異常判定手段と、を備え、異常判定手段が、異常判定フラグが第一型である場合、登録した正常ベクトルとの距離を演算し、距離が小さい順に所定数の正常ベクトルを近傍データとして抽出し、近傍データの重心ベクトルと異常判定の対象のＭ次元ベクトルとの距離を演算し、距離に基づいて設備の異常判定を行い、異常判定フラグが第二型の場合、予め演算した主成分の変換係数に基づいて、主成分からの偏差を演算し、偏差に基づいて、設備の異常判定を行う。【選択図】図１