Abstract: Nickel-based superalloys are key materials in the aero-engine manufacturing industry due to their high hardness and good thermal stability, however, they are also difficult to machine. Therefore, improving its machinability by establishing a predictive model is of considerable significance for the aerospace manufacturing industry. The study employed orthogonal experimental method to determine the process parameters of cylindrical grinding GH3044 stepped shaft, including grinding wheel speed, workpiece speed, reciprocating speed and grinding depth. On this basis, an improved particle swarm optimization-genetic algorithm (PSO-GA) was used to optimize the back-propagation neural network model (BPNN) yielding a more accurate prediction model, which includes the number of nodes in the hidden layer, activation function, initial weights and bias. The optimized BPNN model achieves correlation coefficients of 0.959 38 for surface roughness R_a and 0.905 92 for removal efficiency e, which are the main machinability of the workpieces, and the optimized BPNN model is expected to provide important guidance for the process of grinding nickel-based superalloys, which can help to maintain high-precision machining, improve productivity, and reduce the cost of production.