Operating revenue is a core metric for firm value chain analysis, impacting profit forecasting, resource allocation, and strategy. Traditional methods often leads to high error rates, exceeding 20% in oil-price or operation revenue-related forecasts for cyclical industries like energy. We propose a novel hybrid framework integrating dynamic feature engineering with an enhanced SARIMA model. Our key innovations are: (1) A dynamic pre-processing pipeline significantly improving data stationarity; (2) An intelligent parameter search combining grid search with Bayesian criteria reduces error rates by 18.5%; (3) Empirical results demonstrate superior performance: for energy revenue forecasting, the model achieves a 20-step ahead forecast error standard deviation within ± 5%. Future work will explore integrating LSTM-Transformer modules for nonlinear features, reducing grid search time consumption by enabling intelligent optimization to automatically skip invalid parameter combinations. On the other hand, embedding predictive models into ERP applications will deliver enhanced data value to users.