论文部分内容阅读
石油化工过程系统及其现场数据复杂,基于数据驱动的任何研究、设计、运行工作首先都需要进行数据滤波。本文研究了用自联想神经网络对化工过程数据进行滤波的方法。自联想神经网络通过使输入节点的信息压缩在隐层节点上,从网络输入的高维参数空间中提取反映系统结构的最具代表性的低维子空间,同时有效地滤去了测量数据中的噪声和测量误差,再通过输出层实现数据的解压缩,将前面压缩的信息还原到各个参数值,从而实现各测量数据的重构。通过对测试函数的应用和误差比较验证了该方法可以达到比较理想的滤波效果,并采用该方法对某企业精对苯二甲酸(PTA)工业数据进行滤波后BP建模,该模型的预测效果要大大好于没有进行数据滤波建立的模型,从而进一步说明了用自联想神经网络对工业数据滤波不但是可行且有效的,同时也提高了模型预测的准确性。
Petrochemical process systems and their field data are complex, and any data-driven research, design, and operation needs data filtering in the first place. This paper studies the method of filtering chemical process data with self-associated neural network. Self-organizing neural network extracts the most representative low-dimensional subspace that reflects the system structure by compressing the information of input node on the hidden layer node and extracting the high-dimensional parameter space input from the network, and effectively filter out the measured data Of the noise and measurement error, and then through the output layer to achieve data decompression, the previously compressed information is restored to each parameter value, in order to achieve the reconstruction of the measurement data. Through the application of the test function and the comparison of the error, it is verified that this method can achieve the ideal filtering effect and this method can be used to filter the BP industrial data of PTA. The forecasting effect of this model Which is much better than that of the model without data filtering. It further shows that it is not only feasible and effective to filter industrial data by self-associated neural network, but also improves the accuracy of model prediction.