Abstract:  Considering the complexity of technological processes， the serious coupling between multiple processes and the difficulties in optimizing process parameters during the process manufacturing， a multi-process technological parameter fusion optimization method was proposed based on LSTM neural network， XGBoost algorithm and IPSO algorithm. A data preprocessing model was established based on LSTM neural network， and the time series characteristics of processing data were extracted through LSTM neural network， which realized the processing of outlier in process data. And a PSO-XGBoost quality prediction model was constructed by fitting the nonlinear relationship between processing parameters and quality indexes with XGBoost and combining with particle swarm optimization algorithm. Then the output of the quality prediction model was taken as the fitness， and the improved particle swarm algorithm was used for trolling the global optimal processing parameters， which realized the fusion optimization of the quality of process manufacturing. A process production line of an enterprise was taken as an example to verify the effectiveness of the multi-process technological parameter fusion optimization model. 

Key words:  , process manufacturing, multi-process parameter optimization, improved particle swarm optimization(IPSO), long and short term memory(LSTM) neural network, extreme gradient boosting（XGBoost）

摘要： 针对流程制造过程中工艺过程复杂、多工序耦合严重、工艺参数优化困难等问题，提出一种基于长短期记忆（LSTM）神经网络、极限梯度提升（XGBoost）算法和改进粒子群优化（IPSO）算法的多工序工艺参数融合优化方法。基于LSTM神经网络建立了数据预处理模型，通过LSTM神经网络提取流程工艺数据的时序特征，进而实现了对工艺数据中异常值的处理。在此基础上，通过XGBoost算法拟合工艺参数与质量指标间的非线性关系，并结合粒子群算法构建了PSO-XGBoost质量预测模型，再将预测模型的输出作为适应度，调用改进粒子群算法反向搜索全局最优工艺参数，得到各工序的最优工艺参数组合，从而实现了流程制造加工质量的融合优化。以某企业的一条流程生产线为例，验证了多工序工艺参数融合优化模型的有效性。

关键词: 流程制造, 多工序工艺参数优化, 改进粒子群优化算法, 长短期记忆神经网络, 极限梯度提升