Data Analytics for Real-time Prediction of Battery Charge
Since the advent of electric and hybrid electric vehicles, it has become very important to know how much charge is left in the battery running the vehicle. This gives us an idea about how far will we be able to go with the available charge. Although it is quite important, the state of charge (SOC) of a battery is not easy to find out due to nonlinear behavior of discharge and charge characteristics.
One of the approaches is to conduct experiments by charging and discharging batteries and collecting data. This data can be used to derive a mathematical model empirically. Often engineers tend to use a higher order polynomial to fit the data. However, the right approach is to make use of fundamental laws of physics and derive equation that represents the charge / discharge behavior. Due to the nonlinear behavior, using simple voltage and current relationship does not yield accurate results.
Our approach was to start from the first principles of defining what charge is. This equation needs to be looked at differently by using integral equations. Even such equations need to be modified to include some extra parameters based on current. While juggling with such modifications one should not lose sight of the fact that the equations should still be simple to compute on a real time basis. Such modifications led to results that are more accurate and it could exhibit the nonlinearity quite well.
Every type of battery has different charge/discharge characteristics. Often researchers model each one differently. However, with our method, we could keep the same equations for all types of batteries by changing the values of parameters rather than changing the equations in entirety. This made our equations battery agonistic.
This method of SOC and SOH is pending a patent grant. The invention disclosure is filed at the US Patent Office and the US Patent Number is US2014/0232411 A1.
This method is in use by two major OEMS in their passenger vehicles.