(426f) Real Time Optimization of a Complex Industrial Gas Network | AIChE

(426f) Real Time Optimization of a Complex Industrial Gas Network


Puranik, Y. - Presenter, The Optimization Firm
Sahinidis, N., Carnegie Mellon University
Li, T., Air Liquide
Feather, D., American Air Liquide
Besancon, B., Air Liquide

Air Liquide operates a number of industrial gas pipeline networks around the world. Each of them connects several air separation plants and a number of customers. In each of the plants, there can be one or more air separation units (ASU), which separates air into pure oxygen, nitrogen or argon in either gas or liquid phase. The gaseous oxygen and nitrogen supply different pipelines at different pressure levels after being pressurized either in the ASU or by compressors downstream of the ASU. Such a network usually consists of several pipelines. However, the production capabilities of different products are determined by the design of each of the air separation units and the plant layout.

During normal operating conditions, the plant operation connected to a network is subject to at least two rapidly fluctuating factors: the electricity price and customer demand. In such an environment, real time optimization (RTO) can help find the most profitable way of operating the plants by taking into account the two factors above and other process constraints. In order to do RTO, Air Liquide built a model which includes both plant and pipeline equations. This model is a mixed integer nonlinear programming (MINLP) problem due to the plant and pipeline characteristics and the discrete decisions that must be made, such as selecting the most efficient compressors from a series of available ones. The size of the problem is considered as a fairly large one among such type of problems, and it is not surprising that there exist a number of local optima. Finding the global optimum is challenging, especially when the fact is taken into account that such a global optimal solution must be applicable in practical operation environment. For example, it is subject to such constraints as certain equipment not being able to be switched on and off too frequently and certain process variables not being able to be moved too far from their current values in next step. In order to be practically applied, the large problem must be solved rapidly enough to be able to implement changes in real time.  Due to the problem size and complexity, the implementation of RTO within Air Liquide has utilized a more sequential solution approach, whereby discrete variables are first calculated (e.g. compressors to be used), and then the optimization performed with this fixed discrete variables set.

In this work, we will present the collaboration between Carnegie Mellon University and Air Liquide in addressing the above challenges, addressing the full optimization problem and not utilizing a sequential approach.  The work includes optimization solver selection and tuning for global optimization within required time. Preliminary results of certain case studies will be also be discussed.