Stirred tank reactors, as core equipment in the industrial field, face the critical challenge of balancing the maximization of stirring efficiency and the minimization of energy consumption under the preferences of different decision-makers. This study proposes a multi-stage optimization framework that integrates Computational Fluid Dynamics (CFD) models, data prediction models, and a multi-objective optimization model considering decision-maker preferences. Firstly, a dual-layer genetic algorithm optimized data prediction model GA-GABP is developed against the backdrop of stirred tank reactors. It is verified that GA-GABP exhibits advantages in prediction accuracy and structural optimization compared to traditional BP networks. Subsequently, focusing on key parameters such as stirring speed, impeller diameter (D), baffle width (Wb), and impeller bottom height (C), different optimal results are obtained by analyzing optimization objectives with varying subjective weights. Utilizing the TOPSIS decision-making method, stable and reasonable values for the subjective weights are determined. The mapping relationship between each optimization parameter and the objective is established using GA-GABP, and its accuracy and reliability are validated through rigorous fourfold cross-validation. Based on these mappings, an optimization model with energy consumption (Np), fluid mixing degree (Nq), and suspension uniformity (σ) as objectives is constructed. The Non-dominated Sorting Genetic Algorithm II (NSGA II) optimizes the predictions of GA-GABP, ultimately resulting in a final optimization solution based on the Pareto frontier. The TOPSIS algorithm is then used to determine reasonable subjective weights, yielding the final optimized solution. When decision-makers prioritize energy savings, with a subjective weight of 0.4 for Np, energy consumption is reduced by an average of 86.5%. When decision-makers emphasize fluid mixing degree, the ideal value for Nq is 0.23476. Lastly, when decision-makers prefer a more uniform solid suspension degree, σ is reduced to 9.93% of the base case.

ANN-CFD,ellipsoidal bottom stirred tank,double-layer genetic optimization algorithm,BP neural network