top of page

Is your LP planning software truly optimizing?

Updated: Jan 31, 2024

From Excel Solver to LP planning software, many optimization tools frequently fail to optimize. Many people have experienced this firsthand. For example, some have spent countless hours making changes to Excel cells before the Solver actually provides a decent solution, while others keep updating the initial values in their manufacturing optimization software so it can converge. However, the issue goes beyond merely finding a solution. Our analysis indicates that, for a standard refinery, these tools provide solutions that are >$0.5/bbl worse than the true optimal solution 70% of the time.


The main reason these software fails to optimize is that they rely on “local optimization” algorithms such as SLP (XPRESS) and GRG (Excel solver). While these algorithms can be effective for certain types of problems, they are notoriously ineffective for models that involve blending streams. In fact, even in relatively straightforward blending examples like the one shown below, these methods fail to find the optimal solution 95% of the time



Local vs. global optimization.

Most manufacturing optimization technology relies on local optimization algorithms. These methods find the best possible solution that is closest to the starting point, rather than searching the entire solution space for the absolute best solution, and work well with simple (convex) problems. However, for more complex problems such as blending and manufacturing models, the effectiveness of these algorithms heavily relies on the choice of initial solution. A bad (or outdated) guess can cause long solve times, bad solutions, and even failures to converge.


In contrast, generic global optimization algorithms can find the best possible solution across the entire search space, but they are more complex and computationally expensive compared to local optimization algorithms, limiting their practical use.


At Prometheus.ai, we have developed a proprietary algorithm specifically designed to solve blending and manufacturing problems. Our algorithm uses knowledge from the physical system, such as mass balances and natural network structures, to create search patterns that generic solvers cannot identify. It achieves global optimality for real instances of blending and manufacturing problems in seconds, and consistently outperforms local solvers in the quality of the solution and often in the solution time.


Whether you use Excel or LP planning software, our algorithm can help increase your company's profits and reduce costs. Contact us to learn more about how we can help you optimize your manufacturing plans processes.

93 views0 comments

Comments


bottom of page