README.md

RUL-Prediction-for-Li-ion-Batteries

With its use seen in critical areas of safety and security, it is essential for lithium-ion batteries to be reliable. Prediction of the Remaining Useful Life (RUL) can give insights into the health of the battery. Variations of Recurrent Neural Networks (RNN) are employed to learn the capacity degradation trajectories of lithium-ion batteries. Using several regressor models as the baseline, an ensemble of RNNs is created to overcome the shortcomings of one RNN over the other. The critical point approach and the data-driven approach for regressor models and neural network models respectively help predict the RUL.

Report

Project Report

Results

Various RNN Model Results

Experiment	Model	Training RMSE	Testing RMSE	Validation RMSE
Experiment 1	LSTM	0.0312	0.0304	0.0311
	BiLSTM	0.287	0.2792	0.3259
	GRU	0.0278	0.0342	0.0356
	BiGRU	0.0901	0.0945	0.1059
Experiment 2	LSTM	0.019	0.0173	0.0122
	BiLSTM	0.5521	0.1376	0.4871
	GRU	0.0962	0.0868	0.1957
	BiGRU	1.568	1.3482	1.7741
Experiment 3	LSTM	0.0183	0.0336	0.0542
	BiLSTM	0.107	0.1108	0.1237
	GRU	0.029	0.0425	0.0516
	BiGRU	0.034	0.0454	0.053
Experiment 4	LSTM	0.0248	0.0236	0.0232
	BiLSTM	0.2583	0.2232	0.1943
	GRU	0.0152	0.0242	0.0452
	BiGRU	0.2186	0.2282	0.1775
Experiment 5	LSTM	0.0145	0.0981	0.1329
	BiLSTM	1.2602	1.1037	0.9943
	GRU	0.0253	0.0811	0.1761
	BiGRU	0.3437	0.4544	0.4535
Experiment 6	LSTM	0.0123	0.0189	0.0277
	BiLSTM	0.7338	0.6262	0.6111
	GRU	0.0967	0.1094	0.2436
	BiGRU	0.138	0.2562	0.2627
Experiment 7	LSTM	0.0132	0.0253	0.0245
	BiLSTM	0.2486	0.3564	0.3278
	GRU	0.0578	0.0645	0.0689
	BiGRU	0.1896	0.2486	0.2156
Experiment 8	LSTM	0.0226	0.0356	0.0312
	BiLSTM	0.227	0.2792	0.4123
	GRU	0.0156	0.0236	0.0384
	BiGRU	0.0689	0.1562	0.1047
Experiment 9	LSTM	0.0196	0.0265	0.0241
	BiLSTM	0.3568	0.3956	0.4256
	GRU	0.0452	0.0546	0.0514
	BiGRU	0.0918	0.1256	0.1298

Ensembling Results

Experiment	Model	Train RMSE	Validation RMSE	Test RMSE
Experiment 1	Experiment1_CatBoostRegressor	0.007525	0.027265	0.020191
	Experiment1_ExtraTreesRegressor	2.05E-15	0.032114	0.019708
	Experiment 2_LGBMRegressor	0.015742	0.01869	0.121217
Experiment 2	Experiment 2_RandomForestRegressor	0.008446	0.016777	0.118291
Experiment 3	Experiment 3_ExtraTreesRegressor	1.97E-15	0.027225	0.050309
	Experiment 3_XGBRegressor	0.001102	0.032008	0.052786
Experiment 4	Experiment 4_ExtraTreeRegressor	0	0.038955	0.096077
	Experiment 4_LGBMRegressor	0.064483	0.076645	0.094264
Experiment 5	Experiment 5_DecisionTreeRegressor	0	0.036995	0.027503
	Experiment 5_RandomForestRegressor	0.023868	0.033271	0.033192
Experiment 6	Experiment 6_HuberRegressor	0.012126	0.018583	0.020066
	Experiment 6_LinearRegression	0.01206	0.018489	0.019934
Experiment 7	Experiment 7_HuberRegressor	0.015001	0.026221	0.013487
	Experiment 7_LinearRegression	0.014968	0.026394	0.013531
	Experiment 7_LinearSVR	0.016352	0.029895	0.01475
Experiment 8	Experiment 8_DecisionTreeRegressor	0	0.075652	0.339136
	Experiment 8_LinearSVR	0.042245	0.108166	0.342323
Experiment 9	Experiment 9_CatBoostRegressor	0.006819	0.043317	0.031779
	Experiment 9_ExtraTreesRegressor	1.59E-15	0.039875	0.031513
	Experiment 9_LGBMRegressor	0.04493	0.124974	0.031858

References

Time Series

1. Time Series Prediction: How Is It Different From Other Machine Learning? [ML Engineer Explains]
2. A Comprehensive Guide to Time Series Analysis
3. A RoadMap to Time-Series Analysis

Papers

1. A review of lithium-ion battery state of charge estimation and management system in electric vehicle applications: Challenges and recommendations
2. Remaining Useful Life Prediction and State of Health Diagnosis of Lithium-Ion Battery Based on Second-Order Central Difference Particle Filter
3. Data-Driven Remaining Useful Life Prediction for Lithium-Ion Batteries Using Multi-Charging Profile Framework: A Recurrent Neural Network Approach
4. Novel Statistical Analysis Approach for Remaining Useful Life Prediction of Lithium-Ion Battery

RUL Prediction Code

1. RUL Prediction for Li-ion Batteries using Critical Point
2. Estimation of the Remaining Useful Life (RUL) of Lithium-ion batteries using Deep LSTMs.
3. Transformer Network for Remaining Useful Life Prediction of Lithium-Ion Batteries