【摘要】 现有热误差预测多基于温度数据建模，特征维度单一，且热误差非线性和耦合性特点，导致预测模型适应性较弱，预测精度较低。针对上述问题，设计了一种多源异构数据采集方案，建立基于多维温度、能耗数据和深度神经网络的立式机床热误差预测模型。搭建了实验平台，进行了支持向量和随机森林回归模型预测精度的对比。对比分析可知：DNN模型相比于传统回归模型，适应性较强，预测精度较高，Z向热误差平均绝对误差为0.973μm,在提高预测模型适应性的同时，显著提高了热误差预测精度。更多还原

【Abstract】 The existing thermal error prediction is mostly based on temperature data modeling, with a single characteristic dimension, and the characteristics of nonlinearity and coupling of thermal error, resulting in weak adaptability and low prediction accuracy of the prediction model.Aiming at above problems, a multi-source heterogeneous data acquisition scheme is designed, and a vertical machine tool thermal error prediction model based on multi-dimensional temperature, energy consumption data and artificial neural network is established.Build an experimental platform to compare the prediction accuracy of support vectors and random forest regression models.Compared with the traditional regression model, the DNN model has strong adaptability and higher prediction accuracy, and the average absolute error of Z-direction thermal error is 0.973 μm, which significantly improves the thermal error prediction accuracy while improving the adaptability of the prediction model.更多还原