This paper investigates the incorporation of a PV-ESS linked, all-in-one power conversion system (PCS) at a small factory, in an attempt to improve the factory’s resilience against the increase in load due to facility expansion. The aforementioned factory was installed with 210kWh PV and 550kWh ESS, both of which were simultaneously managed by 250kW hybrid power conversion system. This independent operation of the factory through the PCS was proven to be efficient and economical, because of the reduced power usage and the sale of surplus electricity.

Nowadays, lithium-ion batteries are the most promising candidate as the power source for electric vehicles and energy storage system. Unfortunately, the performance of lithium-ion batteries tends to degrade and their capacity declines after a number of charging and discharging process. Most of the lithium-ion batteries are considered to replaced or discard when the capacity or voltage drops 20% of its initial state. In order to predict the state, prognosis algorithms are necessary to screen out highly degraded batteries. In this paper, life prediction system equipped with prognosis algorithm is proposed for estimating the battery health and predicting it remaining useful life based on the non-linear slope of charge-voltage curve. In particular, the prediction of state-of-charge(SOC) is modeled by Backpropagation Neural Network algorithm, which is conducted by 6 coefficients and the prediction of state-of-health (SOH) is modeled by particle-filter algorithm in form of a probability distribution, which is carried out by Arrhenius electrochemical modeling. 10 sets of lithium-ion batteries were initially prepared with 18650 cells and degraded by repeating the 100 times of charging and discharging process in order to compare the experimental data with the modeled data. Both SOC and SOH error rate tend to decline when the gather input experimental data is increased. The average SOC and SOH error rate of 10 samples at 100th cycle with 4 input cycle data was 6.75% and 3.30% respectively, which correspond to the algorithm efficiency above 95%. For conclusion, the particle filtering method for predicting the residual life (SOL) is not only more accurate than the prediction by simple curve fitting, but also is very effective because it reflects the lack of data and uncertainty due to measurement in the confidence interval. In addition, the remaining useful life prediction of SOC and SOH by particle filtering and backpropagation neural network algorithm is a method of continuously updating the deterioration model coefficient in real time based on the measured data of the battery and modeling data. Figure 1