Application of PRPD Spectrum Characteristic in the Recognition of the Aging State of EPDM Cable Insulation

In order to identify the aging state of EPDM cable insulation, EPDM samples in different aging states were prepared, and a partial discharge test platform was built. The PRPD spectrum of EPDM samples in different aging states were obtained through experiments, and 19 characteristic parameters were extracted from the partial discharge spectra. In this paper, 19 characteristic parameters were reduced by using principal component analysis method, 10 feature parameters obtained after dimension reduction and 19 feature parameters without dimension reduction are used as inputs of the neural network respectively. The results show that for standard BP algorithm, additional momentum algorithm and variable learning rate algorithm, the training times and convergence time of the non-dimension reduction network are obviously more than that of the dimension reduction network, while for conjugate gradient algorithm and L-M optimization algorithm, the convergence time of the non-dimension reduction network is obviously more than that of the dimension reduction network. L-M optimization algorithm, dimension reduction network training times and convergence time are not significantly improved, but the convergence accuracy has been greatly improved; for the same algorithm, dimension reduction processing of feature data recognition effect is better; for the same feature data, L-M optimization algorithm recognition accuracy is the highest.