Abstract: The finish milling of integral impeller is governed by at least five critical parameters: cutter diameter, milling allowance, spindle speed, feed per tooth, and cutting depth, requiring short machining time and low surface roughness. For this multi-parameter and bi-objective optimization, 32 blades from 4 impellers were fabricated using varied parameters to obtain machining time, surface roughness, and blade thickness data for training neural network prediction models. A weighted optimization speed-prioritized optimization with weighting (SPOW) strategy was proposed. That is, when the changing trends of two targets are inconsistent, this strategy holds that if the weighted optimization speed of one target is higher than the weighted deterioration speed of the other target, then the former optimization target is more important. Based on this, quantitative comparisons can be made between two non-dominated solutions in the Pareto frontier, and the optimal solution can be determined, avoiding the subjectivity of manual screening. An impeller machining process optimization algorithm was developed based on this strategy and Matlab. The optimization results and the actual processing results show that the SPOW strategy can obtain a unique optimal solution. By adjusting the weight coefficients, the optimization direction can be guided. One impeller was processed using the optimization results. The average processing time, surface roughness, and average thickness of the 8 blades were 22.3 min, 0.56 μm, and 1.05 mm respectively. Compared with the processing data before optimization with the same specification milling cutter (diameter 5 mm), this result is close to the optimal level (20 min) in terms of processing time, and both the surface roughness and dimensional accuracy have been improved.