The Prediction of Metro Shield Construction Cost Based on a Backpropagation Neural Network Improved by Quantum Particle Swarm Optimization

Table 7

The topological structures of the different network models.

Model	Number of hidden layers	Number of hidden layer nodes		(%)
BP	1	12	0.8531	37.89	263.8791	130.13
		15	0.8867	26.16	153.3810	78.31
		18	0.9152	11.84	54.9819	53.56
	2	12–4	0.8993	20.37	139.3117	58.48
		12–6	0.9205	10.48	48.5281	34.76
		12–8	0.9241	9.38	46.3189	32.18
GA-BP	1	12	0.9532	9.37	31.3819	28.37
		15	0.9648	6.39	27.9042	23.56
		18	0.9761	4.75	22.3047	20.68
	2	12–4	0.9713	5.17	27.3134	16.49
		12–6	0.9848	4.31	20.8443	15.01
		12–8	0.9910	3.48	17.3611	16.31
PSO-BP	1	12	0.9798	8.31	37.3428	26.44
		15	0.9831	4.48	23.189	15.31
		18	0.9920	2.83	12.8864	12.76
	2	12–4	0.9847	5.30	30.3197	17.50
		12–6	0.9945	4.31	20.5610	12.89
		12–8	0.9921	2.30	9.3819	6.15
QPSO-BP	1	12	0.9894	2.06	9.150	8.367
		15	0.9931	1.47	7.679	6.890
		18	0.9980	1.29	6.436	6.078
	2	12–4	0.9942	1.37	7.037	6.401
		12–6	0.9982	1.25	5.936	5.568
		12–8	0.9985	1.23	5.789	5.278