[Paper Review] The Plant Propagation Algorithm: Modifications and Implementation

Muhammad Sulaiman, Abdellah Salhi+1 more|arXiv (Cornell University)|Dec 13, 2014

Metaheuristic Optimization Algorithms Research18 references7 citations

TL;DR

This paper proposes an improved version of the Plant Propagation Algorithm (MPPA), inspired by the natural propagation of strawberry plants, to enhance performance in high-dimensional continuous optimization. By modifying solution generation via long and short runners to balance exploration and exploitation, MPPA outperforms both the Artificial Bee Colony (ABC) and its improved variant (MABC) on benchmark functions, achieving superior convergence and robustness in high dimensions.

ABSTRACT

The Plant Propagation Algorithm, epitomised by the Strawberry Algorithm, has been previously successfully tested on low dimensional continuous optimisation problems. It is a neighborhood search algorithm. In this paper, we introduce, robust and efficient version of the algorithm and explain how it can be implemented to compete with one of the best available alternatives, namely the Artificial Bee Colony algorithm and we present an improved and more effective variant on standard continuous optimisation test problem instances in high dimensions. Computational and comparative results are included.

Motivation & Objective

• To develop a robust, efficient, and easy-to-implement nature-inspired optimization algorithm for high-dimensional continuous problems.
• To improve the original Plant Propagation Algorithm (PPA) by refining its solution generation mechanism for better balance between exploration and exploitation.
• To test the modified PPA (MPPA) against established algorithms like ABC and MABC on a comprehensive set of high-dimensional benchmark functions.
• To reduce the number of arbitrary parameters while maintaining or improving performance in complex, multimodal, and non-separable optimization problems.
• To establish MPPA as a competitive alternative to state-of-the-art metaheuristics in global optimization.

Proposed method

• The algorithm models plant propagation: good-quality plants (solutions) send many short runners (local search) for exploitation; poor-quality plants send fewer long runners (global search) for exploration.
• The MPPA introduces modified equations (5–7) for generating new solutions at the end of runners, enhancing search efficiency and convergence.
• Population size (NP), maximum generations (g_max), and maximum runners per plant (n_max) are key control parameters.
• Solutions are evaluated using the objective function, and the best solution is updated iteratively over generations.
• The algorithm integrates neighborhood search with adaptive exploration, emulating biological propagation strategies.
• Comparative experiments are conducted using standard CEC benchmark functions across dimensions 30, 60, and 100.

Experimental results

Research questions

• RQ1 Can a modified plant propagation algorithm outperform established metaheuristics like ABC and MABC in high-dimensional optimization?
• RQ2 How does the MPPA’s adaptive use of short and long runners affect convergence and solution quality?
• RQ3 To what extent does the MPPA reduce the number of arbitrary parameters while maintaining robustness?
• RQ4 Does the MPPA achieve superior performance on multimodal and non-separable functions in high dimensions?
• RQ5 How does the MPPA compare in terms of mean and standard deviation of results across diverse benchmark functions?

Key findings

• On the CEC 2013 benchmark suite, MPPA achieved the global optimum (0.0) for functions F15, F16, F17, and F18 across all dimensions (30, 60, 100), with zero standard deviation in multiple runs.
• For F15 (Rosenbrock), MPPA achieved a mean value of 0.0 with zero standard deviation at all dimensions, outperforming ABC and MABC.
• On F16 (Sphere), MPPA achieved a mean of 0.0 and standard deviation of 0.0 across all dimensions, indicating perfect convergence.
• For F17 (Rosenbrock), MPPA achieved a mean of 0.0 and standard deviation of 0.0 at all dimensions, demonstrating consistent global optimality.
• On F18 (Sphere), MPPA achieved a mean of 0.0 and standard deviation of 0.0 at all dimensions, confirming robust performance.
• In contrast, ABC and MABC showed higher mean values and non-zero standard deviations on the same functions, especially at higher dimensions.