Paws Proactive (PawsPro)a two-stage framework that combines time-series modeling with reinforcement learning jointly enhance failure prediction and maintenance decision-making. It centers on the idea that.
- Hard disk state migration: Based on the hard disk sequence data, the state change of the hard disk is studied.
- Sequence Decision Making: Using reinforcement learning to build decision models to take different measures in different states of the hard disk.
The dataset used in this study is publicly available from Backblaze: https://www.backblaze.com/cloud-storage/resources/hard-drive-test-data
dataset/ # Experimental data set
├── data1_train_data.npy ├── data1_train_label.npy ├── data1_test_data.npy ├── data1_test_label.npy ├── data2_train_data.npy ├── data2_train_label.npy ├── data2_test_data.npy └── data2_test_label.npy
GMMHMM/ # GMMHMM model implementation ├── trainHMM.py # Training the GMMHMM model └── network.py # Network model
RL/ # DQN approach implementation ├── loadData.py # Load data ├── utils.py # Auxiliary functions, including creating catalogs and plotting convergence curves ├── env.py # Reinforcement Learning Environment ├── buffer.py # Experience replay pool ├── DQN.py # deep Q network implement ├── train.py # Train agents └── test.py # Test
figure/ # Plotting Graphics ├── PCA_raw.py # Principal component analysis of raw labels ├── PCA_decoded.py # Principal component analysis after HMM classification ├── trans.py # Plotting the state transfer matrix ├── reward_function.py # Plotting the reward function └── rewardConvergence.py # Plotting convergence curves for reward values