An Improved Bidirectional Gate Recurrent Unit Combined with Smoothing Filter Algorithm for State of Energy Estimation of Lithium-ion Batteries

The accurate estimation of state of energy (SOE) is the key to the rational energy distribution of lithium-ion battery based energy storage equipment. This paper proposes an improved bidirectional gate recursive element combined with a time-varying bounded layer based smooth variable structure filtering algorithm. First, based on the solid temporal nature of the estimated parameters, a BiGRU neural network structure is constructed to strengthen further the influence of past and future information on the current estimates. Then, based on the traditional variable structure filtering, a time-varying bounded layer smoothing mechanism with saturation restriction (TS-VBL) is proposed to smooth the output of BiGRU to obtain a more accurate estimate. Finally, the test was conducted under 15℃ hybrid pulse power characterization (HPPC) and 35℃ Beijing bus dynamic stress test (BBDST). Compared with other algorithms, the BiGRU-TSVSF algorithm has a minor maximum estimation error of 0.00495 and 0.00722, respectively. The experimental results show that the algorithm has high precision and robustness and is of great value to the energy storage research of energy storage equipment.