Fig. 1. Structural diagram of BP network with hidden layer

Fig. 2. Schematic of overall inverse design process

Fig. 3. Training error curve

Fig. 4. Output power spectra obtained by TWM numerical simulation and neural network for sample data

Fig. 5. Comparison of output power spectra obtained by TWM simulation and neural network for test data

Fig. 6. Schematic of searching process in PSO algorithm

Fig. 7. Comparison between inverse design power spectra and target power spectrum. (a) Comparison among two sets of inverse design power spectra, numerical inverse power spectrum, and target power spectrum; (b) deviation of two sets of inverse design power spectra and numerical inverse power spectrum compared with target power spectrum

Table 1. Comparison of training error between different network structures

Table 2. Comparison of partial output data between TWM simulation and neural network for sample data

Table 3. Comparison of partial output data between TWM simulation and neural network for test data

Table 4. Comparison of partial data between target power spectrum and inverse design power spectra

Table 5. Comparison of parameters between target power spectrum and two sets of inverse design power spectra