Deep Learning Berbasis CNN Untuk Pengenalan Pola Partial Discharge Isolasi Silicone Rubber

Partial discharge (PD) activity measurements have been carried out by selecting noise signals (de-noising) using Support Vector Machine (SVM)and then recognized Convolutional Neural Network (CNN). CNN testing was various models such as activation methods: Sigmoid, Softmax, Relu, Tanh, Swish. Number of layers used is 1, 2, 3, 4 with filter sizes 32, 64, 128, 256 and kernel 3x3, 2x2, 1x1, 1x2, 1x3 in the MaxPooling AveragePooling pooling methods. The results obtained, On sigmoid method 1 having a low accuracy around 14.40% but other configurations gets high 98.99% both has done or without de-noising. In Softmax method, an 84.94% de-noising 90.66%. 65.25% 75.29% de-noised. result shows that SVM increases 11.12% method. Swish methods, level obtained average 14.40%, doesn’t increase accuracy, so CNN-based deep learning more suitable Sigmoid Softmax.