1 An Intelligence-Based State of Charge Prediction for VRLA Batteries DeShaunna Scott, Jide Lu, Haneen Aburub, Aditya Sundararajan, and Arif I. Sarwat Department of Electrical and Computer Engineering Florida International University Miami, Florida asarwat@fiu.edu Abstract— A battery management system (BMS) has three main functions, voltage monitoring, current discharge monitoring and remaining life monitoring. This paper primarily focuses on remaining life monitoring through the estimation of battery’s state of charge (SOC). An Experimental set-up was prepared to measure the Valve-Regulated Lead-Acid (VRLA) battery’s SOC under different operating conditions. Backpropagation (BP) neural network to estimate the battery’s SOC using the experimental data. The results showed a satisfactory estimation of battery’s SOC with a small (4.25%) root mean square perdition error (RMS). Keywords- SOC, state of charge, SOC estimation, Neural Network I. INTRODUCTION VRLA batteries are widely used in many systems and devices because of its energy density, rechargeability, and low cost. Due to the multitude of applications for VRLA batteries, they need to be managed properly [1-3]. A huge part of the management system is in predicting the battery’s SOC and adapting accordingly. An accurate prediction of a battery’s SOC is still a concerning issue within in a BMS. The definition of a battery’s SOC is the ratio of available capacity within the battery to the maximum capacity [4].  ܥΨൌ ஼ ೡ೔೗೗೐ ஼ ೘ כͳͲͲ (1) Without a proper estimation of the battery’s SOC the battery can fall into harmful and irreversible situations that will ultimately diminish its performance and life expectancy [5]. Situations that include overcharging which could lead to explosions and over-discharging that will over exhaust the battery. Conventional techniques such as coulometric counting and open circuit voltages are used to estimate the battery’s SOC, however, these techniques cause great inaccuracy. Common models for VRLA batteries include electric circuit models, electrochemical models, and neural networks [5]. The method presented in [6] suggested that in order to accurately estimate battery data, such as SOC, a model to understand the internal workings of the battery would be essential. However, the internal resistance is not an intrinsic value; the internal resistance model requires a lot of experimental data. Such as the maximum capacity of the battery at different temperatures, the output voltage of the battery at different current magnification, the internal resistance of the battery at different temperatures. In addition, it is difficult to find a suitable function to describe the battery model due to the very complex chemical reactions inside the battery. Therefore, the authors in [4], proposed the use of neural network with a radial basis function as the data training method to estimate SOC. A simple structure was used for the neural network with three layers, and with limited number of inputs such as voltage, current , and temperature [4]. This paper proposed a method of neural networking with backpropagation and increased inputs. Regarding previous work, this method of this paper eliminates the need for battery modeling by collecting enough parameters as inputs to represent the battery’s state. It also disregards the action of data reduction presented in [7]. This neural network encompasses a set of increased inputs in the form of time and battery age in addition to common data such as voltage, current, and temperature. The work in this paper is based on a long term research in the back up battery system for communication systems. The different characteristics of our research provides a prediction model with wide varieties. Compared to the work done in [4] and [5], the application in our research has the relatively stable conditions and regularisation for the charging and discharging periods. The rest of this paper is organized as follows, Section II will explain experimental setup and the data collected to perform the training of the neural network explained in Section III. Finally, Section IV will provide a discussion for the neural network results and SOC values based on the method’s effectiveness through error analysis and comparison. II. DATA COLLECTION AND ANALYSIS This section describes the experiment that was conducted to collect the data for SOC prediction. This paper focuses on 2017 IEEE Transportation Electrification Conference (ITEC-India) 978-1-5386-2668-9/17/$31.00 ©2017 IEEE