Abstract: A dispatch-oriented forecast input correction method for industrial integrated energy systems is proposed. A load-profile-driven forecasting route is constructed, feedforward correction parameters are extracted from directional residual quantiles on the validation set, and candidate dispatch input schemes are evaluated under a unified rolling dispatch model, thereby linking the directionality of forecast errors with their operational consequences. A case study of a tobacco factory industrial park shows that load-profile-driven model selection can improve overall forecasting accuracy, whereas symmetric error metrics, such as mean absolute error (MAE) and root mean square error (RMSE), cannot adequately characterize the risks associated with demand-side underestimation and supply-side overestimation. The risk-budget selective correction scheme C4 reduces directional risk by 20.55%, with increases of 10.74% in operating cost and 10.52% in carbon emissions, and achieves favorable overall performance under the 20% risk-budget constraint. The results indicate that directional residual calibration can provide a basis for dispatch input correction while balancing operational safety, economic performance, and low-carbon performance.