Prediksi Sisa Umur Pakai Baterai Li-ion Menggunakan Denoising Self-retention Retentive Encoders (De-SerRE)

Nugraha, Gifary Divanda (2024) Prediksi Sisa Umur Pakai Baterai Li-ion Menggunakan Denoising Self-retention Retentive Encoders (De-SerRE). Other thesis, Institut Teknologi Sepuluh Nopember.

	Text BukuTA_Gifary Divanda Nugraha.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (7MB) | Request a copy
	Text 5002201092-Undergraduate_Thesis.pdf Restricted to Repository staff only until 1 October 2026. Download (7MB) | Request a copy

Abstract

Penggunaan baterai Li-ion telah menjadi pilihan utama yang terkemuka dalam green industry era, mulai dari penyediaan daya untuk perangkat elektronik portabel, kendaraan hybrid dan electric serta penyimpanan energi terbarukan. Prediksi Remaining Useful Life (RUL) atau sisa umur pakai baterai Li-ion yang akurat menjadi tantangan utama yang sekaligus memiliki peran penting dalam pemeliharaan proaktif dan keselamatan perangkat elektronik. Untuk mengatasi tantangan tersebut, penelitian Tugas Akhir ini mengusulkan model Denoising Self-retention Retentive Encoders (De-SerRE), yang memanfaatkan kemampuan beberapa modul denoising pada arsitektur Denoising Autoencoder (DA) untuk mengatasi jenis noise tertentu yang telah ditambahkan pada data input. Kemudian, dikombinasikan dengan arsitektur Retentive Network (RetNet) Encoders yang memanfaatkan modul multi-scale retention untuk menghasilkan prediksi remaining useful life (RUL). De-SerRE dilatih menggunakan dataset sekunder baterai CALCE. Hasil eksperimen menunjukkan bahwa model De-SerRE dengan Denoising Autoencoder (DA) berdasarkan uniform noise memiliki performa terbaik dalam menangkap pola data dibandingkan dengan De-SerRE dengan DA berdasarkan noise lainnya, dengan error MAE sebesar 0.0156, error RMSE sebesar 0.0182, error RE sebesar 0.0041. Kemudian, model yang diusulkan mampu menghasilkan prediksi RUL yang lebih akurat dan dapat mengungguli model De-SaTE dalam memprediksi RUL untuk DA berdasarkan Gaussian noise dan uniform noise. De-SerRE berdasarkan Gaussian noise menghasilkan error RMSE sebesar 0.058, error MAE sebesar 0.0262 dan error RE sebesar 0.0151. De-SerRE berdasarkan poisson noise menghasilkan error RMSE sebesar 0.0721, error MAE sebesar 0.0461 dan error RE sebesar 0.0566. De-SerRE berdasarkan speckle noise menghasilkan error RMSE sebesar 0.0718, error MAE sebesar 0.0684 dan error RE sebesar 0.1215. De-SerRE berdasarkan speckle noise menghasilkan error RMSE sebesar 0.0721, error MAE sebesar 0.0461 dan error RE sebesar 0.0566. Kemudian, diperoleh hasil terbaik pada model De-SerRE berdasarkan uniform noise, yaitu error MAE sebesar 0.0156, error RMSE sebesar 0.0182, error RE sebesar 0.0041.
==================================================================================================================================
The usage of Li-ion batteries has become the leading choice in the green industry era, from powering portable electronic devices to hybrid and electric vehicles, as well as renewable energy storage. Accurately predicting the Remaining Useful Life (RUL) of Li-ion batteries is a significant challenge that plays a crucial role in proactive maintenance and the safety of electronic devices. To address this challenge, this final project proposesthe Denoising Self-retention Retentive Encoders (De-SerRE) model, which leverages the capabilities of several denoising modules in the Denoising Autoencoder (DA) architecture to handle specific types of noise added to the input data. Subsequently, it is combined with the Retentive Network (RetNet) Encoders architecture, utilizing multi-scale retention modules to generate predictions for the remaining useful life (RUL). De-SerRE is trained using the CALCE battery dataset. Experimental results show that the De-SerRE model with Denoising Autoencoder (DA) based on uniform noise performs best in capturing data patterns compared to De-SerRE with DA based on other types of noise, achieving an MAE error of 0.0156, RMSE error of 0.0182, and RE error of 0.0041. Furthermore, the proposed model is able to generate more accurate RUL predictions and outperform the De-SaTE model in predicting RUL for DA based on Gaussian noise and uniform noise. De-SerRE based on Gaussian noise achieved an RMSE error of 0.058, MAE error of 0.0262, and RE error of 0.0151. De-SerRE based on Poisson noise achieved an RMSE error of 0.0721, MAE error of 0.0461, and RE error of 0.0566. De-SerRE based on speckle noise achieved an RMSE error of 0.0718, MAE error of 0.0684, and RE error of 0.1215. The best results were obtained by the De-SerRE model based on uniform noise, with an MAE error of 0.0156, RMSE error of 0.0182, and RE error of 0.0041.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Baterai Li-ion, Li-ion battery, RUL, De-SerRE, Multi-Scale Retention
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Gifary Divanda Nugraha
Date Deposited:	03 Sep 2024 07:48
Last Modified:	03 Sep 2024 07:48
URI:	http://repository.its.ac.id/id/eprint/114254

Actions (login required)

View Item