On the Robustness of Forward and Inverse Artificial Neural Networks for the Simulation of Ethylene/1‐Butene Copolymerization

Abstract

Forward and inverse artificial neural network (ANN) models are used to describe ethylene/1‐butene copolymerization with a model catalyst having two site types. The forward ANN predicts number and weight average molecular weights, average comonomer content, and polymer yield as a function of a set of polymerization conditions, while the inverse model estimates polymerization conditions needed to produce copolymers with desired microstructures. The forward model is found to be robust and resilient to random noise introduced into the datasets. The inverse model, however, leads to multiple solutions (several polymerization conditions can produce polymers with similar microstructures) and is sensitive to random noise in the data. Although the polymerization conditions estimated from inverse ANN are different from the model data, the estimated polymerization conditions are found to provide similar microstructures even with the random noise.

Using Artificial Intelligence Techniques to Design Ethylene/1‐Olefin Copolymers

“…The range of polymerization conditions investigated in the optimizations discussed in this article is summarized in Table 6. [16,17] The kinetic rate constants are based on values reported in the literature ( Table 7). [16,17,[23][24][25] As explained above, the optimization targets were divided in three groups: Even though all three groups include polymer yield, Y, they differ in how to account for the polymer microstructures.…”

“…[16,17] The kinetic rate constants are based on values reported in the literature ( Table 7). [16,17,[23][24][25] As explained above, the optimization targets were divided in three groups: Even though all three groups include polymer yield, Y, they differ in how to account for the polymer microstructures. Group 1 includes the full microstructural distributions, MWD and CCD, Group 2 accounts only for its main averages, M n , M w , and CC, and Group 3 for a complete distribution (MWD) and an average (SCB) across that distribution.…”

“…The polymer microstructures calculated with the kinetics model are free of experimental error, but experimentallymeasured distributions contain experimental errors resulting from sampling and polymer analysis. [17] To evaluate how sensitive the modified artificial bee colony optimization is to "experimental" error, we introduced uniform ±5% random noise into the microstructures predicted for Groups 1 and 3, and used the modified artificial bee colony method to find possible solutions in the range of low MSE values (MSE ≤ average MSE of modified artificial bee colony for the corresponding group). Figure 21 shows the modified artificial bee colony optimization results for Group 1 ±5% random noise.…”

“…[13,14] Artificial neural networks (ANN)-an AI technique that imitates nervous systems-have already been used to determine the polymerization conditions needed to make polyolefins with specified microstructures. [15][16][17][18] Global optimization tools based on AI techniques have also been applied to solve large complex problems. Multiple solutions may result when these methods are applied to ethylene/ 1-olefin copolymerizations, but this may be considered an advantage, not a shortcoming.…”

“…Multiple solutions may result when these methods are applied to ethylene/ 1-olefin copolymerizations, but this may be considered an advantage, not a shortcoming. [17] If the same target polymer microstructure can be made via multiple polymerization conditions, the most viable one, from operational and economic points of view, may be selected.…”