本文回顧國內外與工具機溫升熱變型位移補償相關之研究文獻，探討(1)移動平均方法(2)自迴歸移動平均方法(3)前饋式神經網路(4)遞迴式神經網路，四種方法所建立之熱變形位移預測模型。使用國產某公司所生產之工具機為實驗機臺，量測機臺之溫度變化與主軸端點位移，預測與綜合比較，探討熱變形位移軟體補方法的預測精度與可行性。四種模型之預測結果，均可有效地降低熱變形位移，除前饋式神經網路方法較差之外，其餘三種預測模型可降低熱變形位移至10μm內。

This study reviews the study and experience of thermal deformation in machine tools. Discussing the thermal deformation models, which build by (1)moving average, (2)auto regression moving average, (3)feed-forward neural network and (4)recurrent neural network, also measuring the temperature and thermal deformation of the machine tools, take those four ways to build the thermal deformation model. Comparing predict results of four models to give the feasibility of the models of thermal deformation. The predicted results of four models, which can reduce thermal deformation effective, and beside FNN model, the other can reduce thermal deformation under 10μm.