Abstract:

A networked performance assessment method based on multi-space locally weighted projection to latent structures (Ms-LWPLS) is proposed to solve the nonlinear relationship between input and output data of chemical process. This method divides the historical training datasets into different sets of performance grades, extracting the process changes of different performance grade of training datasets by Ms-LWPLS method. This method obtains the latent structures accurately by matching the training datasets and performance grade labels through the non-linear networked structure. The“off-line modeling”is achieved by the trained neural network. With the model obtained, the sliding window is used as the assessment unit, working as the input data into the trained neural network model. The current performance grade is identified according to the network output and the transition performance coefficient is constructed. The steady-state performance grades and the transition performance grades are recognized and distinguished. Finally, the method is applied to the online performance assessment of ethylene cracking process, which shows the effectiveness and accuracy of the performance assessment method proposed.

Key words: multi-space, locally weighted projection to latent structures, nonlinearity, neural network, transition, steady state, online assessment

摘要：

针对化工过程输入输出数据间非线性关系问题，提出一种基于多数据空间局部加权潜结构映射(multi-space locally weighted projection to latent structures，Ms-LWPLS)的网络化性能分级评估方法。该方法将历史数据分成不同性能等级的集合，利用Ms-LWPLS方法提取不同性能等级训练数据的过程变化，获得训练数据与性能等级标签之间的非线性映射结构，实现输入数据与性能等级之间的网络化“离线建模”。得到模型后，以数据滑动时间窗为评估单元，将滑动窗口数据输入到训练好的神经网络模型中，根据网络输出划分过程当前性能等级，并构造过渡性能系数，将稳态性能等级和过渡性能等级进行识别和区分。最后，将该方法应用到乙烯裂解过程在线性能评估中，说明此性能评估方法的有效性和准确性。

关键词: 多数据空间, 局部加权潜结构映射, 非线性, 神经网络, 过渡, 稳态, 在线评估