Wiratama, Maulana Andra (2026) Sistem Prediksi Remaining Useful Life dan Estimasi State of Health Baterai Berbasis Deep Learning pada Simulator Battery Energy Storage System. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2042221075_Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2042221075_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy |
Abstract
Peningkatan pemanfaatan energi terbarukan mendorong penggunaan *Battery Energy Storage System* (BESS), namun degradasi baterai menyebabkan penurunan *State of Health* (SoH) dan *Remaining Useful Life* (RUL), sehingga diperlukan sistem pemantauan kesehatan baterai. Penelitian ini bertujuan merancang dan mengimplementasikan sistem estimasi SoH dan prediksi RUL berbasis *deep learning* yang terintegrasi pada simulator BESS serta mengevaluasi performanya menggunakan data kondisi operasional. Sistem dikembangkan menggunakan *battery pack* LiFePO₄ konfigurasi 8S dengan variasi kapasitas 12 Ah, 6 Ah, dan 3 Ah, *Smart BMS Jikong*, sensor PZEM-016, mikrokontroler ESP32, komunikasi Modbus RS485, serta perangkat lunak BattIQ. Model menggunakan pendekatan dua tahap berbasis *Bidirectional Long Short-Term Memory* (BiLSTM), yaitu estimasi SoH pada Fase 1 dan prediksi RUL pada Fase 2. Dataset publik LiFePO₄ dengan histori degradasi jangka panjang digunakan untuk pengembangan model, sedangkan data *charging–discharging* aktual digunakan untuk mengevaluasi implementasi sistem. Hasil pengujian menunjukkan bahwa model estimasi SoH memperoleh RMSE sebesar 9,42 poin persentase, MAE sebesar 9,26 poin persentase, dan MAPE sebesar 12,75%, sedangkan model prediksi RUL pada pengujian temporal menghasilkan RMSE sebesar 346,64 siklus dan MAE sebesar 332,97 siklus. Sistem juga mampu mengakuisisi dan mengolah data tingkat siklus serta menjalankan estimasi SoH dan prediksi RUL setelah kebutuhan histori data terpenuhi. Meskipun *pipeline* sistem berhasil diimplementasikan, masih ditemukan kecenderungan *overestimation* dan keterbatasan generalisasi model pada beberapa kondisi *battery pack*. Selain itu, dataset eksperimen belum mencakup degradasi hingga *End of Life* (EoL), sehingga prediksi RUL pada *battery pack* aktual belum dapat divalidasi secara langsung. Dengan demikian, sistem terintegrasi untuk pemantauan, estimasi SoH, dan prediksi RUL pada simulator BESS telah berhasil dikembangkan, sedangkan peningkatan performa model masih memerlukan dataset eksperimen dengan histori degradasi yang lebih panjang dan representatif.
===================================================================================================================================
The increasing utilization of renewable energy has encouraged the implementation of *Battery Energy Storage Systems* (BESS); however, battery degradation leads to a decline in *State of Health* (SoH) and *Remaining Useful Life* (RUL), creating the need for a battery health monitoring system. This study aims to design and implement a deep learning-based SoH estimation and RUL prediction system integrated into a BESS simulator and to evaluate its performance using operational condition data. The system was developed using an 8S LiFePO₄ battery pack with nominal capacity variations of 12 Ah, 6 Ah, and 3 Ah, a Jikong Smart BMS, a PZEM-016 sensor, an ESP32 microcontroller, Modbus RS485 communication, and BattIQ software. The model applies a two-stage approach based on *Bidirectional Long Short-Term Memory* (BiLSTM), consisting of SoH estimation in Phase 1 and RUL prediction in Phase 2. A public LiFePO₄ dataset with a long-term degradation history was used for model development, while actual charging–discharging data were used to evaluate the system implementation. The test results showed that the SoH estimation model achieved an RMSE of 9.42 percentage points, an MAE of 9.26 percentage points, and a MAPE of 12.75%, while the RUL prediction model in temporal testing produced an RMSE of 346.64 cycles and an MAE of 332.97 cycles. The system was also able to acquire and process cycle-level data and perform SoH estimation and RUL prediction after the required historical data had been accumulated. Although the system pipeline was successfully implemented, overestimation and limited model generalization were still observed under several battery pack conditions. In addition, the experimental dataset did not yet cover battery degradation up to the *End of Life* (EoL), preventing direct validation of the predicted RUL against the actual RUL of the battery packs. Therefore, an integrated system for battery monitoring, SoH estimation, and RUL prediction on a BESS simulator was successfully developed, while further improvement of model performance requires a longer and more representative experimental degradation history.
Actions (login required)
 |
View Item |