Neuro-fuzzy controller for battery equalisation in serially connected lithium battery pack

This study presents a non-linear, dynamic control method for equalising battery cell voltages in a serially connected lithium-ion battery system based on an adaptive neuro-fuzzy inference system. By using a combination of neuron networks and fuzzy logic, the optimal control method is obtained by self-learning capability to equalise the current between battery cells. The duty cycle used to control the metal-oxide-semiconductor field-effect transistors in individual battery cell equalisers are changed based on the dynamic equalising and system status. While energy is transferred from higher voltage cells to lower voltage cells, online measurement is utilised to collect data for tracking. Therefore the duty cycle control has an optimal response in this battery system. The state of the optimal control output is presented in simulation results. To demonstrate the effectiveness of the proposed control scheme and robustness of the acquired neuron-fuzzy controller, the controller was implemented in a serially connected lithium battery system model using a microprocessor. The proposed system achieved a learning accuracy error of 1.8 × 10^-5, and the equalising time was approximately 3000 s for a 0.25-V voltage gap.