A Neural Learning Approach for a Data-Driven Nonlinear Error Correction Model

Table 3

Neural learning results.

No.	No. of nodes		No. of parameters	Goodness of fit statistical indicators				Chi-square test value compared with simplified model
	First hidden layers	Second hidden layers		Loss	MSE
1	0	0	5	1.85	0.0156	0.2068	157.68
2	1	1	12	1.72	0.0145	0.2558	166.33	0.02 (compare with no. 1)
3	1	3	18	1.72	0.0145	0.2556	166.29
4	1	5	24	1.72	0.0145	0.2561	166.38	1.00 (compare with no. 2)
5	3	1	18	1.72	0.0145	0.2558	166.33	1.00 (compare with no. 2)
6	3	3	28	1.75	0.0147	0.2449	164.41
7	3	5	38	1.72	0.0145	0.2560	166.36	1.00 (compare with no. 2)
8	5	1	24	1.73	0.0145	0.2541	166.04
9	5	3	38	1.76	0.0148	0.2422	163.99
10	5	5	52	1.72	0.0145	0.2559	166.35	1.00 (compare with no. 2)