Multiobjective hybrid evolutionary path planning with adaptive pareto ranking of variable-length chromosomes

Pareto genetic path planning hybridized with multi-objective Dijkstra's algorithm

“…The experiments are done on a population of 1000 individuals, randomly generated according to the configurations #1, #5, #7 and #9 presented in Table II, for ߜ ∈ (0,1). The results of MOO-CP1&2 (Table III) are compared with those given by the single objective correction (SOO-CP1&2) introduced in [12], which, during the 2 nd step of CP, finds the repaired sub-paths via the minimization of L, only. Special attention is paid to the diversity and the quality of the resulted feasible paths; the diversity is monitored in both genotypic and phenotypic space.…”

“…More precisely, a chromosomal string includes the float Cartesian coordinates of the inner vertices, stated as decision variables according to (1). The GA is supplemented with all the mechanisms required for handling variable-length chromosomes: i) the construction of the initial paths with different number of turning points (N < N max ) according to a uniform distribution of the vertices within WS; ii) a crossover operator which extends the shorter parent [12] before mating, without altering its phenotype (by duplicating some randomly selected inner vertices). Feasibility constraint is solved by means of the external, non-evolutionary CP described in Section III.…”

“…GA employs the adaptive MOO ranking scheme introduced in [12]. This ranking ensures a partial sorting of solutions, useful for controlling the selection of parents and the insertion of offspring.…”

“…Recent improvements of this algorithm refer to: strategies which enforce the exploration around key points of the first Pareto-front [8], [9]; agreeing between objectives according to preset or adaptively set priorities [10]; splitting/ merging of ranks via adaptive insertion/ deletion of proxy objectives [7]; enhancement of solutions' diversity in support of a more relevant ranking [11]. The ranking scheme used in this paper integrates an application-specific preference mechanism based on a preliminary grouping of individuals [12]. The grouping implicitly adapts the importance of the objectives in relation to objective space's landscape and permits the detection of diversity loss for the most useful solutions.…”

“…Additionally, graph-based correction is followed by a reduction step which shortens the paths; the probability of reduction is adapted by GA, depending on the diversity of the current population. This MOO-driven correction of paths is the main improvement added to [10] and [12]. This paper is organized as follows.…”