3.1. Traditional Methods Based on Experiments
3.1.1. Open Circuit Voltage
The Open Circuit Voltage method, also known as the Voltage measurement method, is based on the corresponding relationship between Open Circuit Voltage (OCV) and SOC.
the experimental methods help us obtain Uoc and f (soc) can be obtained through the charge and discharge fitting experiment of the battery.
When determining the above relationship, it is necessary to establish the PNGV model and refer to the PNGV battery experiment manual for the HPPC experiment to obtain the SOC-OCV relationship curve.
The advantage of the OCV method is that the measured quantity is relatively simple, and the accuracy obtained is higher. Moreover, its shortcomings are also very obvious, for measuring the terminal voltage, it needs to be placed for a long period of time (a few hours or even longer), so the amplifier cannot be used for continuous dynamic on-line estimation.
the open-circuit voltage method is always together with other methods at the initial or final stage of charge, and rarely used alone such as the SOC estimation method combining the open-circuit voltage method with the ampere-hour integral method, and the SOC estimation method combining the open-circuit voltage method with the Kalman filtering method.
3.1.2. Ampere-Hour Integral Method
The Ampere-hour (Ah) integral method is the most commonly used method for electric charge accumulation at present. It is based on the calculation of electric quantity during charging and discharging.
Q is the rated capacity of the battery, unit: Ah. Q0 is the initial charge of the battery, unit: Ah. η is the charging efficiency. S is the electric quantity of self-discharge, unit: Ah. ic is the charging current, unit: A. id is the discharge current, unit: A.
The Ah integral method has the advantages of simple calculation, stable algorithm and online measurement.
The disadvantage of this method is also obvious.
1. The quantity needs to be measured is so large that untimely, there is always a chance of errors.
2 The battery will be aged in use, and the loss capacity cannot be compensated. Moreover,
3. The battery’s discharge capacity in the process can be recovered, but the charging capacity is not easy to eliminate the effect.
4. The measurement error will accumulate with time going by.
3.1.3. Internal Resistance Method
Theoretically, the internal resistance method is relatively simple, mainly because it only considers the discharge current and internal resistance of the battery. However, in practice, the relationship between battery parameters and SOC is quite complex. At the initial stage of discharge, the discharge efficiency of the internal resistance is stable and almost has no fluctuation. However, at the later stage of discharge, the internal resistance will obviously increase and will show obvious fluctuations. Therefore, the internal resistance method is always used for SOC estimation in the later stage of discharge.
However, due to the great influence of temperature, it has always been controversial whether it should be estimated in the open-circuit state or in the process of charging and discharging. Therefore, this method is rarely used in real cars.
3.1.4. Discharge Test Method
Its principle is that in the constant current continuous discharge, the SOC equals that the discharge current multiples time. It is mainly used in the laboratory as the standard of SOC accuracy, as well as in the research on charging and discharging characteristics and battery maintenance.
3.2. Modern Methods Based on Control Theory
3.2.1. Neural Network Method
The advantages of this method lie in that it can accurately estimate the SOC quickly and easily. The disadvantages of this method are also obvious, mainly because it requires a large amount of training data as a support to complete the training system.
3.2.2. Kalman Filter Method
The principle of this method is shown in Figure 7. It describes the battery as a system composed of an equation of state and observation equation, considers SOC as an internal state of the system, establishes a state-space model, and makes the minimum variance estimation for SOC.
Its advantage is that it eliminates the error of the ampere-hour integral method accumulated over time. At the same time, it does not have a high requirement for the accuracy of the initial SOC. In other words, even if the initial value has a certain deviation it can well converge to the real value. Even if there is noise, it can have a good correction effect.
Its disadvantage is that the accuracy largely depends on the establishment of the battery equivalent model, and the error mainly comes from three aspects: the time variability of the model, the non-linearity of the model and the approximate treatment of noise.
3.2.3. Particle Filter Algorithm
The idea of the Particle Filter (PF) is based on Monte Carlo methods, which use particle sets to represent probabilities and can be used in any form of the state space model. The core idea is to express its distribution by extracting random state particles from the posterior probability. It is also a sequential importance sampling. In simple terms, the particle filter method refers to the process of finding the state’s minimum variance distribution by finding a set of probability density functions of a set of random samples propagating in the state space, then approximating them, and using the sample mean instead of the integral operation. The sample here is the particle, and when the number of samples (N) is near infinity, it can approximate any form of probability density distribution. However, this means that in order to get high precision, a large number of samples are needed for support. The PF is an optimal filtering technique for estimating non-Gaussian nonlinear systems without any restrictions on process noise and observed noise.
The particle filter algorithm can obtain higher precision, but the algorithm is complex and the testing time is longer.
3.3. Other Methods Based on the Innovative Ideas
Principal Component Analysis (PCA) is an algorithm based on multivariate statistics, which can analyze and process data by simplifying and compressing various data and extracting important elements. The main methods used are zero mean method and least square method.
The basic idea of Supports Vector Regression (also known as SVR) is based on the idea of minimizing structure empirical risk and model complexity. It is very important to select kernel function in SVR, and proper kernel function can make this method have better generalization ability.