为了对反映回转支承性能退化状况的健康指标进行准确预测，提出了一种基于改进时间卷积网络（Temporal convolution network, TCN）的退化趋势预测模型——密集时间卷积网络（Densely temporal convolution network, DTCN）。该模型借鉴Dense-Net网络中的Dense-block模块对网络结构进行改进，以解决时间卷积网络在训练中损失函数下降缓慢，以及网络不易收敛、收敛效果差的问题；随后，使用回转支承全寿命实验数据，借助卷积自编码网络（Convolutional Auto-Encoders, CAE）与隐马尔可夫模型（Hidden Markov Model, HMM）建立健康指标，验证该改进算法的有效性；最后，将DTCN与其他序列预测模型如长短时记忆网络（long short-term memory neural networks, LSTM）、门控循环单元网（gated recurrent unit neural networks, GRU）络等对比，结果表明该模型在预测效果上具有优越性，能够更准确地预测健康指标的变化情况，可用于回转支承的退化趋势预测任务。

Here, to accurately predict the health indicator reflecting performance degradation of slewing support, a degradation trend prediction model based on the improved temporal convolution network (TCN) called the densely temporal convolution network (DTCN) was proposed. DTCN drew lessons from the Dense-block module in Dense-Net network to improve its own network structure, and solve problems of the loss function of TCN dropping slowly in training, its network being not easy to converge and poor convergence effect. Then, the whole life-cycle test data of slewing support were used, the health indicator was established with help of the convolutional auto-encoders (CAE) and the hidden Markov model (HMM), and the effectiveness of this improved algorithm was verified. Finally, DTCN was compared with other series prediction models, such as, the long-short term memory (LSTM) network and the gated recurrent unit (GRU) network. The results showed that the proposed model has advantages in prediction effect; it can more accurately predict changes of the health indicator; it can be used to predict degradation trend of slewing support.