Source-linked AI summary
Multi-objective Flower Algorithm for Optimization
Xin-She Yang, M. Karamanoglu, Xingshi He
TL;DR
Multiobjective engineering optimization must estimate Pareto fronts despite conflicting objectives and computational and distributional challenges. The paper extends the Flower Pollination Algorithm using randomly weighted sums, evaluates it on diverse test functions, and applies it to a bi-objective disc brake problem, reporting better results for almost all four benchmark cases and efficient handling of complex problems.
Problem
Engineering optimization involves conflicting objectives, while accurately and uniformly mapping Pareto fronts is time-consuming and difficult.
Method
The paper extends the Flower Pollination Algorithm to multiobjective optimization by combining objectives with non-negative randomly chosen weights.
Results
MOFPA obtained better results for almost all four benchmark cases and was reported efficient on nonlinear problems with complex constraints and diverse Pareto-optimal sets.
Takeaways & Limitations
The proposed flower algorithm can solve multiobjective design problems and was reported to converge almost exponentially in numerical experiments and design benchmarks.
Takeaways & Limitations
The study uses preliminary parametric studies, with fixed MOFPA parameters based on that preliminary analysis.
Abstract
from arXiv · showhide
Flower pollination algorithm is a new nature-inspired algorithm, based on the characteristics of flowering plants. In this paper, we extend this flower algorithm to solve multi-objective optimization problems in engineering. By using the weighted sum method with random weights, we show that the proposed multi-objective flower algorithm can accurately find the Pareto fronts for a set of test functions. We then solve a bi-objective disc brake design problem, which indeed converges quickly.
1 Introduction
Engineering design optimization often involves conflicting objectives, making accurate and evenly distributed Pareto-front estimation challenging. The paper extends the single-objective Flower Pollination Algorithm to address multiobjective optimization.
- Conflicting objectives make engineering design optimization challenging and require compromises or approximations to estimate the true Pareto front.
- Multiobjective optimization adds challenges involving time complexity, inhomogeneity, dimensionality, and uneven solution distribution on the Pareto front.
- The paper extends the single-objective Flower Pollination Algorithm to solve multiobjective optimization problems.
2 Nature-Inspired Flower Pollination Algorithm
The Flower Pollination Algorithm models pollination through global and local search behaviors, switching between them probabilistically. Its multiobjective extension combines objectives with randomly weighted sums to seek accurate, uniformly distributed Pareto fronts.
- 2.1 Pollination of Flowering Plants: The Flower Pollination Algorithm is inspired by pollination behaviors in flowering plants, including biotic, abiotic, self-, and cross-pollination.
- 2.2 Flower Pollination Algorithm: Global pollination models biotic cross-pollination with pollinator movements following Lévy flights.
- 2.2 Flower Pollination Algorithm: Local pollination uses abiotic and self-pollination, while flower constancy is modeled as reproduction probability proportional to flower similarity.
- 2.2 Flower Pollination Algorithm: A switch probability p ∈[0, 1] controls interaction between global and local pollination, with a slight bias toward local pollination.
- 2.2 Flower Pollination Algorithm: The algorithm initializes a population of flowers or pollen gametes, identifies the best solution, and iterates until a generation or accuracy stopping criterion is met.
- 2.3 Multi-objective Flower Pollination Algorithm (MOFPA): MOFPA combines multiple objectives into a composite single objective using non-negative weights, choosing random weights to obtain accurate and uniformly distributed Pareto fronts.
3 Simulation and Results
MOFA is evaluated on four test functions spanning convex, non-convex, and discontinuous Pareto fronts, then compared with established multiobjective algorithms. The proposed method obtained better results for almost all four cases.
- Test functions: The evaluation uses four test functions with convex, non-convex, and discontinuous Pareto fronts, followed by comparisons with established multiobjective algorithms.The selected functions are intended to provide diverse Pareto-front properties.
- Experimental settings: MOFPA uses a population size n = 50 and runs for t = 1000 iterations, with p = 0.8, λ = 1.5, and γ = 0.1.These parameters were fixed based on a preliminary parametric study.
- Test functions: 100 Pareto points generated by MOFPA were compared with the true ZDT1 front f2 = 1 −√f1.The comparison is shown in Fig. 2.
- Evaluation measures: The generalized distance evaluates the estimated Pareto front relative to the true front, while the absolute measure depends on the number of points.The passages introduce distance or error measures for assessing convergence.
- Performance comparison: MOFPA obtained better results for almost all four cases in the comparison reported in Table 1.Table 1 compares Dg for n = 50 and t = 500 iterations.
4 Design of a Disc Brake With Two Objectives
The paper applies MOFPA to a bi-objective disc brake design problem that minimizes mass and braking time under engineering constraints. The reported benchmark results suggest that MOFPA can handle nonlinear problems with complex constraints and diverse Pareto-optimal sets.
- Design problem: The disc brake problem minimizes overall mass and braking time by selecting the inner radius, outer radius, engaging force, and friction-surface count.The design is subject to torque, pressure, temperature, and length constraints.
- Design problem: The design variables satisfy 55 ≤r ≤80, 75 ≤R ≤110, 1000 ≤F ≤3000, and 2 ≤s ≤20.The objective expressions are given for f1(x) and f2(x).
- Constraint handling: The friction-surface count s is discrete, so the paper uses a simple branch-and-bound method because only one discrete variable is present.The authors note that mixed-integer problems generally require constraint-handling extensions.
- Reported outcome: The benchmark and test-function results suggest that MOFPA is efficient for nonlinear problems with complex constraints and diverse Pareto-optimal sets.The disc brake Pareto front is presented in Fig. 3.
5 Conclusions
The paper extends the flower pollination algorithm to multiobjective design problems and reports efficient convergence. However, its parameter analysis remains preliminary, leaving theoretical and performance-related follow-up work.
- MOFPA extends the single-objective flower algorithm to solve multiobjective design problems.
- Numerical experiments and design benchmarks report that MOFPA is very efficient, with an almost exponential convergence rate.The paper bases this observation on comparisons with other algorithms for multiobjective optimization problems.
- FPA uses one key parameter p together with a scaling factor γ, making the algorithm easier to implement.The paper also notes similarities between FPA, cuckoo search, and other algorithms with Lévy flights.
- The paper reports only preliminary parametric studies, while detailed performance links and theoretical analyses remain future research topics.Suggested directions include dynamic-systems or Markov-chain analysis of the linear main updating formulas.