Abstract:

Because some raw material consumption in complex chemical production process is difficult to measure directly, a soft sensor method based on the deep learning is proposed. Based on a period of the historical data, the proposed method extracts multi-scale information from the historical data using stationary wavelet decomposition. Then the observable data at every point of time are combined to get a complete dataset which is divided into the training dataset and the testing dataset. Moreover, the soft sensor model is trained and obtained by using the depth learning algorithm based on the attention mechanism. Finally, the proposed method is applied to the soft measurement of acetic acid consumption in a terephthalic acid (PTA) production unit. Compared with the extreme learning machine (ELM), multi-layer perceptron (MLP) and common long short-term memory (LSTM) method, the result analysis shows that the validity and the applicability of the proposed model is verified. Meanwhile, the consumption of acetic acid in the PTA production plant is predicted and analyzed to improve the production capacity and reduce the energy consumption.

Key words: long short-term memory, neural networks, stationary wavelet decomposition, algorithm, multiscale, terephthalic acid

摘要：

针对复杂化工生产过程中的一些原材料消耗量难以直接测量的问题，提出了一种基于深度学习的软测量方法。该方法基于一段时间的历史数据，利用平稳小波变换提取历史数据中的多尺度信息，然后与每一个时间点的可观测数据进行合并得到完整的数据集，再划分出训练集和测试集，用带有注意力机制的深度学习算法进行训练和泛化，进而建立软测量模型。最后将提出的方法应用到对苯二甲酸（PTA）生产装置乙酸消耗的软测量中。通过与极限学习机 (extreme learning machine，ELM)、多层感知器 (multi-layer perceptron，MLP)以及普通长短期记忆网络(long short-term memory，LSTM)方法比较，结果表明，该模型的预测准确度较高，具有一定的有效性和适用性，同时对PTA生产装置的乙酸消耗量进行预测分析，从而提高产能和降低能耗。

关键词: 长短期记忆网络, 神经网络, 平稳小波变换, 算法, 多尺度, 对苯二甲酸