Inhalt des Dokuments
Modelling and optimization of operational processes in industrial plants
The accurate forecasting of time series is a vital part in the decision-making process in the chemical industry, facilitating the optimization of many operational processes within a company.
For example, by accurately predicting industrial aging processes (IAPs), it is possible to schedule maintenance events further in advance, thereby ensuring a cost-efficient and reliable operation of the plant. So far, these degradation processes were usually described by mechanistic models or simple empirical prediction models.
In recent years, recurrent neural network (RNN) models have emerged as the most successful methods for modelling sequential data in a wide range of applications. However, their effectiveness many real-world industrial applications is often limited by small amounts of data due to high data acquisition costs and covariate shifts due to changes in the real-world environment.
Our research cooperation aims to combine machine learning expertise from the field of time-series forecasting with the domain knowledge by chemical engineers at BASF to examine the effectiveness of sequential models such as long short term memory (LSTM) and echo-state networks (ESN) on the industrial processes forecasting scenario, but more importantly develop new machine learning models that can deal with many of the aforementioned challenges arising in real-world industrial processes and are capable of learning accurate forecasting models from a small number of samples, as well as generalizing well in the face of changing process conditions.