SoC Depletion Estimation for Urban-City Driving Using Long Short-Term Memory and Global False Nearest Neighbor Approach

Abstract

An accurate estimation of State of Charge (SoC) depletion is crucial for Electric Vehicle (EV) users. It alleviates range anxiety by ensuring drivers are confident in reaching their destinations with the available SoC, thereby enhancing the widespread acceptance of EVs. However, algorithms estimating SoC depletion using battery cell voltage, current, and temperature may exhibit lower accuracy when applied to EVs with different types of batteries. Thus, this paper proposes a driving behavior-based SoC depletion estimation algorithm for EV users undertaking urban city trips. The proposed algorithm uses real-time velocity, acceleration/deceleration, and distance as inputs, offering greater practical applicability to EVs with various battery types. It employs the Global False Nearest Neighbor (GFNN) method to determine an optimal sliding window length to accommodate unpredictable variations in driving behavior across different trips. The output from GFNN is then utilized in a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) to estimate SoC depletion. The proposed methodology improves the algorithm's resiliency, adaptability, and learning capability of long-term dependencies towards abrupt driving behaviors of EV drivers. This estimation framework is trained and validated on EV drivers' real-world urban city driving behaviors from “Blu Smart Mobility” cabs in India, ensuring the algorithm's reliability. The results show that it outperforms other state-of-the-art algorithms, achieving an overall accuracy of 99.75%, calculated using three metrics: Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE).