Abstract: In response to the problems of multi-source influences, high quality feature extraction and selection, complexity and nonlinearity in identification methods for energy consumption processes, a method for identifying and predicting energy consumption of machine tools was proposed by combining integrated models and deep learning. Taking CNC milling as an example, an energy consumption model was established based on different cutting periods, and signals were preprocessed by wavelet transform. The preprocessed signals were used to train and predict the energy consumption of the model combining RF and LSTM neural network(RF-LSTM model). Meanwhile, the RF was used to identify the cutting stages and realize the energy consumption classification prediction. The effectiveness and superiority of the proposed method were demonstrated through practical cases, and the RF-LSTM model was used to compare with the other four schemes, which verify that this recognition method may accurately predict different operating states and energy consumption of the machine tools.

Key words: energy consumption model, random forest(RF), long short term memory(LSTM) neural network, energy prediction

摘要： 针对能耗过程识别方法存在的多源影响、高质量特征提取与选择以及复杂性和非线性等问题，提出了一种融合集成模型与深度学习的机床能耗识别与预测方法。以数控铣削加工为例，建立基于不同切削时段的能耗模型，并通过小波变换进行信号预处理。利用预处理信号对融合随机森林(RF)与长短时记忆(LSTM)神经网络的模型(RF-LSTM模型)进行训练和能耗预测，同时利用RF识别切削阶段，实现能耗分类预测。通过实际案例来证明所提方法的有效性和优越性，利用RF-LSTM模型对比其他四种方案，验证了该识别方法能够精准预测机床不同运行状态以及能量消耗。

关键词: 能耗模型, 随机森林, 长短时记忆神经网络, 能耗预测