Research Article
Remaining Useful Life Prediction Techniques of Electric Valves for Nuclear Power Plants with Convolution Kernel and LSTM
Table 4
The architecture definition of LSTM.
| Name | Definition | Value |
| Network structure | Convolution kernel + LSTM network | None | Sliding window size | Size of sliding window for data preprocessing | 40 | Normalization mode | Normalization of data | Z-score | Layers of convolution kernel | Layers of convolution kernel | 3 | Number of convolution kernels | Number of convolution kernels in each layer | 64 | Convolution kernel size | Kernel size of convolution calculation | 3 × 3 | Pooling size | Size of average pooling | 2 × 2 | LSTM_size | LSTM layer unit | 256 | num_layer | Number of stacked LSTM layers | 2 | Keep_prob | Percentage retained in dropout operations | 0.99 | Init_learning_rate | Initial learning rate | 0.01 | Init_epoch | Iterations using the initial learning rate | 5 | Max_epoch | Total training times | 200 | Batch_size | The amount of data used in small batches | 64 | Loss function | None | MSE | Optimization method | Optimization algorithms for backpropagation | SGD |
|
|