An enhanced connected banking system optimizer with multiple strategies for numerical optimization problems

Smarter Search for Tough Design Problems

Modern engineering and data science are full of hard “tuning” tasks: choosing the best shape of a bridge, the most efficient power grid setting, or the right parameters for a machine-learning model. These problems are like vast mountain ranges with many peaks and valleys, where finding the highest peak by hand is impossible. This paper introduces a new computer search method, the Enhanced Connected Banking System Optimizer (ECBSO), which uses a banking-network metaphor to explore these landscapes more intelligently and reliably than many existing approaches.

From Bank Transfers to Better Designs

ECBSO builds on an earlier algorithm called the Connected Banking System Optimizer (CBSO). CBSO imagines a network of banks sending money to each other along different routes. Each possible route stands for a candidate solution to a design problem, and the time it takes for the transfer reflects how good that solution is. By trying out many routes and adjusting them over time, the algorithm searches for faster, more efficient paths – in other words, better solutions. CBSO had one major attraction: it needed almost no hand-tuned settings, which made it easy to apply. But it also had weaknesses: it did not share information well across its “banks” and moved too abruptly from broad searching to fine tuning, often getting stuck on second-rate solutions.

Three New Tricks: Feedback, Renewal, and Patterns

The authors upgrade CBSO into ECBSO by adding three complementary ideas. First, a feedback selection strategy lets each “bank” react to its own recent performance. If a particular way of updating a route makes things better, that style is kept; if not, the bank switches to a different style next time. This simple feedback loop replaces rigid, pre-planned stages and allows the search to adapt on the fly to whatever landscape it encounters. Second, a regenerative population strategy occasionally refreshes only part of the population instead of jostling everyone at once. It measures how spread out the solutions are and how fast they are improving, then selectively regenerates some candidates when the search appears to be slowing or converging too narrowly.

Learning the Shape of the Search Space

The third idea, called distribution estimation, looks for patterns in the best solutions found so far. Instead of nudging candidates blindly one coordinate at a time, ECBSO builds a simple statistical picture of how good solutions are arranged in space. It then samples new candidates that follow this learned pattern, paying special attention to directions where variables seem strongly linked. This helps the search move more intelligently through difficult landscapes, covering large areas early on and then zooming in near promising regions. New candidates from this learned pattern are mixed with those from the original banking process, and only the best survivors move on, maintaining both variety and pressure toward improvement.

Putting the Method to the Test

To judge whether these ideas actually help, the authors run ECBSO on a large, well-known collection of artificial test problems known as CEC-2017. These problems mimic many real-world difficulties: some have a single clear best answer, while others have many deceptive peaks or stitched-together landscapes. ECBSO is compared to ten advanced competitors drawn from different families of search methods, including evolutionary, swarm-inspired, physics-inspired, and mathematically inspired algorithms. Using standard statistical tests, the study shows that ECBSO usually finds better answers, and does so consistently, especially on higher-dimensional problems where simpler methods often fail.

From Benchmarks to Real Engineering

Beyond synthetic tests, the authors apply ECBSO to ten real engineering design tasks, such as creating springs, pressure vessels, gear trains, and robotic parts under multiple safety and cost limits. These are constrained problems, meaning that many seemingly good designs are actually forbidden because they break stress, size, or performance rules. ECBSO handles this by penalizing broken rules while its three strategies guide the search toward feasible, high-quality designs. Across these cases, it frequently ranks at or near the top against the same set of rival algorithms, and it delivers solutions that are not only good but also stable from run to run.

What This Means for Everyday Technology

In plain terms, the paper shows that adding simple feedback, selective renewal, and pattern learning to a banking-inspired search can markedly improve a computer’s ability to solve tough design problems. ECBSO does require somewhat more computing time than its parent method, but the payoff is better and more reliable solutions, particularly in complex and high-dimensional settings. As optimization quietly underpins fields from energy systems to transportation, finance, and medical imaging, methods like ECBSO promise to make the invisible “tuning” behind everyday technologies more efficient, robust, and adaptable.

Citation: Yin, Y., Liu, H., Cai, S. et al. An enhanced connected banking system optimizer with multiple strategies for numerical optimization problems. Sci Rep 16, 12564 (2026). https://doi.org/10.1038/s41598-026-38261-9

Keywords: metaheuristic optimization, engineering design, global search algorithms, numerical optimization, banking network models