Deep Multiscale Soft-Threshold Support Vector Data Description for Enhanced Heavy-Duty Gas Turbine Generator Sets’ Anomaly Detection

Table 3

Extraction of hidden features and hypersphere optimization.

---

Algorithm 2: The algorithmic procedure for the second phase

Input: The training set , batch size B, the learning rate a, iteration count M, encoder ψ(·;W), regularization parameter λ

Output: Encoder ψ(·;W), the center of the hypersphere a

Initialization: Encoder ψ(·;W), model parameters W

(1) The center a of a hypersphere can be calculated by

(2) for m = 1 to M do:

(3)  Randomly select a mini-batch of data from the training set

(4)  Randomly select a mini-batch of data

(5)  Input to the encoder ψ(·; W) to obtain the hidden features Z

(6)  

(7)  

(8) end for

(9) The updated center a of the hypersphere is obtained by

---