Abstract
In this paper, the battery inconsistency equalisation strategy is investigated and a novel fusion model based on equivalent circuit models is proposed. The three equivalent circuit models, 1RC, 2RC and PNGV, are weighted and fused by BP neuron network, which realizes the complementary advantages of the three equivalent circuit models. Even though the estimated values of all three models are lower than the true value, they can still be close to the true value after reasonable weight allocation. With reference to the open-source DST dynamic operating test data from the Advanced Battery Association of America, a comparison is made with the three common equivalent circuit models of 1RC, 2RC and PNGV. It is found that the proposed novel fusion model has the highest estimation accuracy with a maximum error of only 0.00947; the RMSE is 0.00217. The 1RC equivalent circuit model has the worst accuracy with a maximum error of 0.10145 and an RMSE of 0.02153. The 2RC and PNGV models have accuracies between the two. Then an active equalisation system is constructed to realise the charge equalisation of multiple unbalanced single batteries by distributed power supply. The PSO-PID strategy is used to control the topology circuit for adaptive equalisation. The equalisation effect of different sized battery packs is verified by simulation. Compared with the traditional logic control strategy, the method is simple and effective, and the equalisation effect is better as the number of inconsistent battery cells increases. In a battery pack with five initial SOC inconsistencies, the inter-cell variability is quickly equalised. When dynamic disturbances are introduced into the system, the algorithm also keeps the battery packs within the equalisation range with an average variance of 0.0016.