|
Category | Method | Descriptions | Applications |
|
Standard | Principal component analysis (PCA) | PCA summarizes the variation in a correlated multiattribute data to a set of uncorrelated components, a linear combination of the original variables. | Pattern recognition [100], dimension reduction [101], feature extraction [102], process monitoring [103]. |
Partial least squares (PLS) | PLS can find the fundamental relations between two data matrices, and latent variables are needed to model the covariance structure in these spaces. | Power load forecasting [104], performance evaluation of power companies [105], etc. |
Linear discriminant analysis (LDA) | LDA finds a linear combination of features that characterizes or separates two or more classes of objects or events. | Face recognition [106], feature selection for power system security assessment [107]. |
Subspace identification methods (SIM) | SIMs are powerful tools for identifying the state space process model directly from data. | Power oscillatory state space model [108], power system stability analysis [109], etc. |
|
Time-varying | Recursive PCA | RPCA is a generalization of PCA to time series; the eigenvector and eigenvalue matrices are updated with every new data sample. | Voltage stability moitoring [110], power system fault location detections [111]. |
Dynamic PCA | DPCA includes dynamic behavior in the PCA model by applying a time lag shift method while retaining the simplicity of model construction. | Industrial monitoring [112, 113], dynamic economic evaluation of electrical vehicles [114]. |
|
Nonlinear | Kernel PCA/PLS | KPCA is first to map the input space into a feature space via nonlinear mapping and then to compute the PCs in that feature space. | Power equipment assessment [115], real-time fault diagnosis [116], power system monitoring [117], etc. |
Neural networks | Neural networks are computational models that can be used to estimate or approximate unknown nonlinear functions. | Dimension reduction [118, 119], voltage stability assessment [120], fault location detection [121], etc. |
|
Non-Gaussian | Independent component analysis (ICA) | ICA decomposes multivariate signals into additive subcomponents which are independent non-Gaussian signals. | Fault detection [122], power quality monitoring [123], and estimation [124]. |
Gaussian mixture models (GMM) | GMM describe an industrial process by local linear models using finite GMM and Bayesian inference strategy. | Power flow modeling [125], power load modeling [126]. |
Support vector data description (SVDD) | SVDD defines a boundary around normal samples with a small number of support vectors. | Classification, process monitoring [127], oscillation modes detection [128], etc. |
|