Desenvolvimento de um sistema de gerenciamento de baterias em nuvem

Abstract

Batteries are vital in many modern applications, from portable electronic devices to electric vehicles and energy storage systems. Their correct operation is essential to ensure safety, optimized performance, energy e!ciency, and cost reduction over time. To achieve this, the BMS (Battery Management System) plays a fundamental role, performing cell monitoring, state estimation, charge balancing, thermal management, and charge/discharge control activities. However, the low processing and data storage capacity of these devices is a significant limitation. To address this problem, a new hardware and software approach has been developed and utilized in management systems, the cloud BMS. In this new architecture, the availability of processing and data storage capacity grows exponentially. This work presents the development of a hardware and software solution for a cloud-based battery management system, using the ESP32 IoT module and the AWS IoT platform. The designed system aims to overcome the limitations of traditional BMS by leveraging cloud computing for greater processing power and data storage capabilities. The proposed solution includes a system composed of a specially designed BMS (Battery Management System) board, embedded software that enables monitoring, state estimation, charge balancing, thermal control, and charge/discharge management, a dashboard hosted on a remote server, and a Python application running in the cloud that executes a neural network algorithm for battery state-of-charge (SoC) estimation. For system validation, an initial test was conducted by connecting a 48V battery pack, consisting of 13 lithium-ion cells, to the input of the designed BMS, while a resistive load of approximately 50 ω was connected to the output to generate current consumption. The monitoring of pack voltage, individual cell voltage, pack current, cell temperature, and the temperature of the control FETs during the test is carried out via the dashboard. The estimation of the pack’s state of charge (SoC) is performed using two distinct methods: Coulomb counting (local SoC) and a previously trained neural network (remote SoC). At the end of the test, the comparison between the results of the two estimation methods returned an MSE value of 7.82, a MAPE of 6.57%, and an MAE index of 2.33, indicating that the remote SoC estimation by the neural network closely approximates the local SoC value with good accuracy.