DEVELOPMENT OF GPU-BASED EDGE DEVICE FOR BIG DATA ANALYSIS ON SMART BATTERY MANAGEMENT SYSTEM APPLICATION

• Main Author: Raihan Dhaifullah, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68535
• Institution: Institut Teknologi Bandung

Smart Battery Management System (SBMS) is an important component as a support for integrated smartgrid Battery Energy Storage System (BESS) that can estimate battery operating and performance parameters such as State-of-Health (SoH) and State-of-Charge (SoC). With the development Graphics Processing Unit (GPU)- based edge device can make BMS with faster performance compared to BMS that only uses Central Processing Unit (CPU)-based computing. The development of edge device platform for big data analysis of BESS using two architectures such as, Cross Industry Standard Process for Data Mining (CRISPDM) as a framework for development flow for big data analysis and Smart Grid Architectural Model (SGAM) as an architectural system to represent the intelligent network. The BESS object used is a battery module containing fifteen LFP (Lithium Iron Phosphate) battery cells with a total voltage value of 48 VDC and a capacity of 100 Ah. platform starts from data acquisition to visualization of SoH and SoC estimations. The results of the research show that the estimated SoC value using Artificial Neural Network (ANN) method with 4,87% Mean Absolute Percentage Error (MAPE) error and the estimated SoH value using Recurrent Neural Network (RNN) method with 0,68% MAPE error. Algorithms for SoC and SoH estimation values have been implemented on edge devices with GPU computing having a faster time efficiency of 31-35% on SoC and 3-5% on SoH compared to CPU computing. When using parquet format files for data analysis, the fastest time is 0-3% of csv files and 6-12% of excel files. Data processing for data training, which is a sequential computation, with a CPU-based computation has a faster time and more efficient of energy usage than GPU-based computation. While for data estimation, GPU-based computation has faster time and mor efficient of energy usage than CPU-based because GPUs have advantages in parallel computation.