关键词: 大语言模型, 混合微调, 变工况机械加工, 能耗预测

Abstract: To address the challenges in energy consumption prediction for modern machining,caused by strong nonlinear coupling among equipment,environment and parameters,as well as highly imbalanced data sample distributions,a hybrid fine-tuning large model-based method for variable conditions machining energy consumption prediction was proposed.A data preprocessing approach involving data cleaning,reconstruction,expansion and sampling was designed to establish comprehensive and transient conditions energy consumption datasets.A two-stage hybrid fine-tuning technique with Full Parameter Fine-Tuning(FPFT) and Parameter Efficient Fine-Tuning(PEFT) strategies was proposed.The FPFT training of the large model by full-operating-condition energy consumption data improved the generalization ability and interpretability of the model,while the PEFT of the large model by using transient-operating-condition energy consumption data solved the difficult problem of predicting the processing energy consumption under variable operating conditions.Experimental results demonstrated that the proposed hybrid fine-tuning method significantly outperformed individual techniques,reducing MSE by over 70.42% and MAE by 42.45%.Compared to existing deep learning-based energy consumption prediction methods,the proposed approach more accurately reflected actual energy consumption in variable machining conditions,exhibiting superior performance,which could support the application of large models in the machining field,indicating promising prospects.

Key words: large language model, hybrid fine-tuning, variable conditions machining, energy consumption prediction