Big-Data Based Online State of Charge Estimation and Energy Consumption Prediction for Electric Vehicles 2016-01-1200

Whether the available energy of the on-board battery pack is enough for the driver’s next trip is a major contributor in slowing the growth rate of Electric Vehicles (EVs). What’s more, the actual capacity of the battery pack depend on so many factors that a real-time estimation of the state of charge of the battery pack is often difficult. We proposed a big-data based algorithm to build a battery pack dynamic model for the online state of charge estimation and a stochastic model for the energy consumption prediction. And the good performance of sensors, high-bandwidth communication systems and cloud servers make it convenient to measure and collect the related data, which are grouped into three categories: standard, historical and real-time data.

First a resistance-capacitance ( RC )-equivalent circuit is taken consideration to simplify the battery dynamics. And the nonlinear relationship between the open-circuit voltage (V_oc ) and the state of charge ( SOC ) is described by five linear piecewise functions, achieving good fitting effect and accuracy. The moving window recursive least square algorithm is utilized to identify the parameters of the battery model online with the historical and real-time data. A Luenberger state observer is used for online estimation of SOC, the estimation accuracy of which is not dependent on the actual capacity’s value. Afterwards, giving more insight into the relationship between the received data from different sources and the estimated SOC, we take the objective law of energy consumption as a Gaussian distribution, which can provide the confidence interval analysis to confine decision-making about whether to have the trip to the driver.