Deep Forest-Based Fault Diagnosis Method for Chemical Process

Table 7

Performance comparison of different faults diagnosis methods.

FDR (%)	(a)	(b)	(c)	(d)	(e)	(f)
Fault 1	99.87	96.37	100	100	99.88	99.17
Fault 2	97.87	97.62	99	99	99.13	100
Fault 3	2.37	20.62	6	95	6.13	42.33
Fault 4	100	82.75	100	98	100	98.83
Fault 5	99.87	96	100	86	99.88	90.97
Fault 6	99.5	100	100	100	100	100
Fault 7	100	100	100	100	100	100
Fault 8	96.62	96.87	99	78	97.00	96
Fault 9	3.37	12.12	3	57	5.75	51.50
Fault 10	82.25	88.25	84	98	92.13	75.83
Fault 11	64.75	73.5	82	87	74.00	82.00
Fault 12	99	93.62	100	85	99.75	93.50
Fault 13	95	72.25	95	88	95.63	94.83
Fault 14	100	95.87	100	87	100	96.83
Fault 15	9.75	21.12	17	0	10.25	7.17
Fault 16	81.62	78.12	89	0	90.00	78.00
Fault 17	84.87	80.25	96	100	97.13	94.50
Fault 18	89.5	86.37	90	98	90.25	99.67
Fault 19	76.12	96.12	52	93	92.00	89.50
Fault 20	66.37	86.75	88	93	85.25	76.00
Average	77.44	78.73	80	82.1	80.53	83.33

Note: (a) optimized variable selection-based PCA [34]; (b) supervised local multilayer perceptron [35]; (c) Bayesian method [36]; (d) DBN-based model [30]; (e) residual subspace associated with PCA [37]; (f) WCForest-based model (proposed in this paper).