A Driving-Behavior-Based SoC Prediction Method for Light Urban Vehicles Powered by Supercapacitors

Abstract

Range anxiety is one of the problems that hinders the large-scale application of electric vehicles (EVs). We propose a driving-behavior-based State of Charge (SoC) prediction (DBSP) algorithm to overcome this problem. This algorithm can determine whether drivers can reach their destinations while also predicting the SoC if drivers were to return the trip. First, two supercapacitor equivalent circuit models are established, with one based on the historical average power and the other based on the equivalent current, which is proposed in this algorithm. Then, based on the equivalent transformation of the two models, an analytical expression relating the historical average power and the predicted SoC is derived by using the equivalent current as a 'bridge'. Therefore, the predicted SoC can be dynamically adjusted in response to recorded historical data, including the output power, speed and distance of EVs powered by supercapacitors. The simulation results demonstrate that the total prediction error is less than 0.5% of the real SoC at different initial SoC and temperature, which represents idealized behavior-based driving. In contrast, in actual driving experiments, the total prediction error is less than 3% of the real SoC at different initial SoC and temperature.

A Simplified Historical-Information-Based SOC Prediction Method for Supercapacitors

“…We can solve formulas ( 16) and (17) by using the bisection method [25] and achieve current () can be calculated discretely [18].…”

“…Since cannot be modeled exactly, the value of inaccessible by calculation. Then, we will calculate  supercapacitor in [18],…”

“…The test routine consists of uphill sections, downhill sections, flat roads, and bends. For average power estimation at a new initial SOC, the interpolation method can be used based on historical information [18]. The sampling frequency of all the data is 10Hz.…”

“…SHSP and DBSP are executed based on the pseudocode in Tab. 1 and the pseudocode in [18] respectively. Setting parameters of LSTM are the same in [17], and we keep training window size/predicted window size=60 in this paper.…”

“…However, machine learning-based methods are usually time-consuming. We also proposed a driving-behavior-based SOC prediction method (DBSP) based on a Rint model with a liner capacitor [18].…”