Three cylindrical Li-ion batteries are simulated using COMSOL Multiphysics in this study. A circular PCM layer is used to cover the batteries. A battery pack connected to a solar system is tested to store energy in a residential building. The PCM is 100% melted at the beginning, and the frozen PCM volume fraction (PVOF) is studied over time. In addition, the temperatures of individual batteries in three different configurations are analyzed at air velocities 1 and 3 mm/s. The three battery configurations included a positive slope, a negative slope, and no slope. The finite element method (FEM) is utilized to derive algebraic equations. The test results are given to the network as training data, and different neural network architectures are tested in terms of the number of layers and the number of neurons in each layer, and finally a structure that has a minimum mean squared error as a network. Battery behavior predictor is selected. Finally, this model is given to the optimization algorithm as an optimization function to investigate the point at which the maximum voltage is obtained. The results showed that the positively sloped battery configuration decreased the temperatures of the batteries, while the non-sloped configuration increased the temperatures. In addition, the positively sloped configuration resulted in more frozen PCM than non-sloped and negatively sloped configurations at a given time. The minimum frozen PCM was observed in the non-sloped configuration. A change in the air velocity from 1 to 3 mm/s increased the frozen PCM and improved the cooling of the batteries.