DEVELOPMENT OF STATE OF CHARGE ESTIMATION OF LITHIUM ION BATTERIES BASED ON TEMPERATURE EFFECT USING DEEP NEURAL NETWORK

• Main Author: Ramadan, Rainda
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75580
• Institution: Institut Teknologi Bandung

<p align="justify"> Lithium-ion batteries are the energy storage that is widely used in electric vehicles and other needs such as battery energy storage system. One thing that is very crucial in battery management system is State of Charge (SOC). State of Charge describes the remaining charge in the battery cell which is related to the battery capacity However, the capacity value of the battery is not constant over time, so a method is needed to estimate State of Chage (SOC). One method state of charge estimation that is currently developing is a data driven method. State of charge historical data has a large amount of data, so it requires a model that can handle the data with high accuracy. Among data-based methods, deep learning methods can handle large amounts of data with high accuracy. Data driven method especially deep learning method are currently widely used for prediction and estimation. The Supervised learning method is used because it is more accurate and faster in the case of input and output achieved prediction. One of the deep learning method used in this study is deep neural network (DNN). In this study, the parameter analysis used is State of Charge (SOC) with the temperature effect. This research was conducted using the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology. In business understanding phase defines the goal that will achieved that is state of charge estimation with the temperature as the additional parameter input in battery energy storage system (BESS). The battery energy storage system (BESS) used is 1 module consisting of 15 LiFePO4 battery cells 3,2 VDC 100 Ah. This study used two different types of datasets, the first one which is 1 LFP battery module testing under discharge conditions, and the second one is testing under idle conditions. In data understanding it describes the specification of the battery, structure and profile of each dataset used. After that, feature selection process, capacity calculation, and SOC calculation will be done. Then modeling is carried out using deep neural network (DNN) to estimate state of charge. The input targets used are voltage, current, capacity and temperature. Therefore, this study is using deep neural network by adding temperature, and also carried out in several scenarios, each scenario divided into 4 scenarios such as voltage capacity temperature (VCT), voltage capacity current temperature (VCIT), voltage current temperature (VIT), and voltage temperature (VT). The best result obtained in datasets 1 is the first scenario voltage capacity temperature (VCT) with the value of RMSE = 0,0026, MAE = 0,0017, MAPE = 3,2289%. Meanwhile is datasets 2, the value of metrics model is exactly the same as the other scenarios, RMSE = 0,0120, MAE = 0,0152, MAPE = 0,0152%. In datasets 1, the fastest training time is obtained in VIT scenario with 28,29 s. However, the fastest estimation time is in voltage, capacity, current, temperature (VCIT) & voltage temperature (VT) scenario with 0,23 s. In datasets 2, the fastest training rime is obtained in voltage, current, temperature (VIT) scenario, if it is sorted based on training time and estimation time from the fastest to the longest in datasets 2, it will be VIT<VT<VCIT<VCT.