Simulation and optimization

The Modelling group of the Home ribbon tab has two buttons: Simulate and Optimize.

• When the Simulate button is selected, AVEVA Unified Supply Chain runs each process unit, going downstream from the feeder unit. If a process unit is not fully defined, for example because it is missing an operating parameter, any process units downstream of this process unit are not simulated.

• When the Optimize button is selected, AVEVA Unified Supply Chain attempts to optimize the whole flowsheet, including the purchase of materials, sale of products and unit performance.

In simulation mode, the input of material must be fixed to calculate the performance of each unit. In contrast, in optimization mode the amount of material may vary, although it can be constrained or fixed. AVEVA Unified Supply Chain optimizes the purchase of material and blending of product to maximize the objective function.

To remove all solution information from a supply chain model, follow these steps:

1. In the Supply Chain Models tab of the Model Explorer, select the model from which you want to remove results.

2. Click Clear Results in the Model Maintenance group of the Tools ribbon tab.

Optimization in AVEVA Unified Supply Chain

The model of a plant includes operating parameters known as variables. These may be the ratio of Arab Light in the CDU feed, the riser temperature for the FCC, or the proportion of ethanol in the gasoline blend. The model also has constraints, such as the maximum flow of material that can be processed by a unit, the minimum proportion of a component in a blend, or a fixed sale amount of a product. Optimization is used to find the combination of variables which results in all the constraints being met and has the highest objective function. The objective function measures the profit from the plant operations.

More in detail:

AVEVA Unified Supply Chain uses successive non-linear programming (SNLP) optimization to find the solution to the problem. SNLP works by analyzing the possible combinations of operating variables to see if they meet all the constraints, and then finding the combination of variables which meets the constraints and has the greatest objective function.

At the end of one pass of the optimizer, a solution is found. This solution is then used as the starting point of another pass of the optimizer. The solution from this run is then passed to a subsequent run, and so on. When the solutions between two subsequent runs are the same, or within a certain tolerance threshold, the problem is said to have converged, and the set of values for the model variables is reported as the optimal solution.

Post-optimization simulation

An optimization run in AVEVA Unified Supply Chain consists of several steps:

3. An initial simulation is performed to generate a starting point, unless a previous case is set as the starting point. This generates stream flows and property values for all pipes in the flowsheet.

4. The problem and starting point are transferred to the optimizer and optimized.

5. The optimizer returns a solution which defines the value for every degree of freedom in the model.

6. The optimizer solution is used to configure the simulator and a simulation is run. This double-checks the objective function and completes all stream flows and property values, which may not have been sent to the optimizer if not required.

Starting points

Choosing the correct starting point is an important part of solving optimization problems. Well-chosen starting points can shorten the time taken to find the optimum solution, as well as helping ensure that the final solution is not a local optimum.

Local optima

When using successive linear programming, an initial estimate of the solution must be made to define the starting point of the optimization. This is particularly important when optimizing complex problems where many of the intermediate values in the problem are not known at the start of the optimization. For example, to optimize a refinery's fuel oil blending, the density and viscosity of the fuel oil components must be known. However, these depend on the choice of crudes and operating parameters of the process units, which are not known until the problem has been solved. Therefore, on the first pass of the optimization an estimate of the density and viscosity must be made to solve the problem.

As optimizers trial many different values of the input variables, the choice of these intermediate values may have an effect on the final solution value. As several passes of the optimizer are used to solve the problem, usually errors in the original estimates are not significant, and the values converge to reach their correct value.

However, sometimes the original choices may affect the final solution value and lead to the incorrect optimum being found. This can be particularly significant if there are local optima in the solution.

Local optima are a consequence of the method in which linear programming optimizers find the optimum solution. As the optimizer examines different solutions, it may locate an area of solutions which seem to be moving toward an optimum, and eventually concentrate its investigations to find the optimal solution of this set of possibilities. However, it is possible that this area of solutions does not represent the best choice of the entire set of solutions, but only the best choice in a small area.

Multi-start

To minimize the chance of encountering local optima, it is possible to run the optimizer with several starting points. This is known as multi-start optimization. At the end of the optimization for each run, the results are compared and the solution found. Even if there is a local optimum, the majority of runs will likely have converged to find another better optimum, and so this can be treated with confidence as the true optimum.

If you set AVEVA Unified Supply Chain to use multi-start optimization with 12 different starting points, a case is optimized 12 times. The same starting mathematical problem is always used, so it only needs to be built once. The difference between the runs is the starting point.

When running multiple start points, you can use the Multi-Start Metrics window to review the solution of each optimization. When multiple runs are performed, each run will likely converge to a different solution. However, most of these solutions will be functionally identical, that is, the variable solution values and objective function will be the same within a small tolerance.

Where multi-start runs do not lead to a set of similar answers, this suggests poor problem formulation within the model, leading to many local optima.

Warm start

It may be desirable to use a previous solution as the starting point for a new optimization. This is common where only small parts of the problem have changed and the new solution is expected to be similar to the original solution. This is useful in analytics, for example where for each analytic run using the solution from the base case can help speed up the optimization (as the start point is already close to the solution) and lead to stability in solution values.

Infeasibility breakers

An optimization problem in AVEVA Unified Supply Chain may not have any solution. In this situation the problem is said to be infeasible. Infeasible problems can be created by mistakes in setting up the original problem, or through a choice of input and output constraints that combine to create a situation that has no solutions.

Example: You might set up a problem where five different crudes can be purchased in amounts up to 100 kbbl/day. You might then set a constraint that the CDU capacity must be between 100-120 ktonne/day. This problem is infeasible, as it is not possible with the crude purchase constraints to meet the CDU capacity constraints, the maximum total mass of all the crudes being less than the minimum CDU capacity. In this situation, the CDU capacity has been entered in the wrong unit of measure and should be corrected to kbbls.
In another problem you might limit the amount of natural gas you can purchase to manufacture hydrogen, and also set very low sulfur limits on the diesel produced by the refinery along with a large minimum sale amount of diesel. When using sour crudes, the limited natural gas purchase could result in a hydrogen production too low to hydrotreat all the refinery's diesel, resulting in insufficient volumes of low sulfur diesel to meet the minimum sale constraint. In this situation the formulation of the problem is correct, but the choices made are preventing the problem from being solved. To optimize the problem it is necessary to relax one of the constraints.

In historical optimizer systems, when the constructed problem could not be solved the optimizer would fail and report the problem as infeasible. However, the reason for the infeasibility could often not be deduced easily from the problem. It would take a significant amount of time to find the problem, often through a slow trial and error approach of relaxing constraints until a feasible solution was found. AVEVA Unified Supply Chain includes infeasibility breakers, which allow constraints to be violated if necessary to produce a feasible solution, but then report the violations for easy identification of the original problem.

Infeasibility breakers can be applied to each case in a supply chain model. If the optimizer finds that the original problem cannot be solved, it may break one or more of the original problem's constraints in order to find a solution.

Calculation Hub

Standard desktop PCs and laptops can be used to run the majority of AVEVA Unified Supply Chain analyses, including analytics which may contain many hundreds of cases. However, when analyzing large case stacks it may be useful to take advantage of a Calculation Hub.

The Calculation Hub is a remote server capable of distributing AVEVA Unified Supply Chain calculations across a number of calculation nodes. The Calculation Hub may be hosted within a corporate intranet, or may be accessible across the internet using a secure connection. The Calculation Hub runs on a server shared by many users; however, the Hub distributes the calculations efficiently, meaning that calculations are always performed faster than on a local PC, even if multiple users are connected at the same time.

No changes are necessary to supply chain models in order to use a Calculation Hub, but you must have configured a connection in your profile.

To use a Calculation Hub for optimization, use the drop-down menu next to the Optimize button to change the calculation location from Local to the name of the Calculation Hub.

Note: The number of available calculation nodes depends on your license agreement and the Hub infrastructure available to you. It is not possible to change the number of calculation nodes from the AVEVA Unified Supply Chain interface.