Deep learning models are widely used in sectors like finance and healthcare for forecasting complex time series patterns. However, their "black box" nature raises explainability concerns. To address this, we propose two methods for generating counterfactual explanations: GENOTOPSIS, combining a genetic algorithm with the TOPSIS method, and NSGA-II, offering faster execution with similar results. We validate these counterfactuals using the CEVO ontology, applying SWRL rules and SPARQL queries to meet domain-specific constraints. Our study focuses on estimating the State of Charge (SOC) for Lithium Iron Phosphate (LFP) cells, demonstrating the effectiveness of our approach in real-world applications.