Search

Item classification is an issue among inventory managers who want to achieve key-point management of items with different emphases, thus prompting managers and researchers to pursue efficient and effective classification algorithms. In processing multiple-criteria inventory classifications, back-propagation neural network (BPNN) shows its superiority in balancing items’ multiple competing attributes. However, because the training processes and the final results of BPNN rely on the initial connection weights and thresholds, finding reasonable values of the two parameters is a challenge. This paper introduces biogeography-based optimization (BBO) and an adaptive mutation particle swarm optimizer (AMPSO) to BPNN to optimize the training parameters, i.e., the global initial connection weights and thresholds, of BPNN. On the basis of this, two hybrid classification algorithms—BBO-BPNN and AMPSO-BPNN—are presented. Real-life data from three cases are adopted to verify the effectiveness and feasibility of the two proposed hybrid algorithms. Experimental results demonstrate that BBO-BPNN and AMPSO-BPNN show higher classification accuracy than other hybrid models, i.e., BPNN, PSO-BPNN, DE-BPNN, and GA-BPNN, in the multiple-criteria inventory classification problem.

In this paper, by assuming a B2C e-business company with several regional distribution centers (DCs), an extension of multi-item joint replenishment-distribution problem (JRD) is raised with stochastic lead-time and demands to investigate their mutual impacts to the JRD system. For the proposed JRD, the objective comprising four types of costs, namely, the ordering cost, the inventory holding cost, the lost sale penalties and the transportation cost is minimized by jointly deciding the basic cycle time, the replenishment frequencies and safety stock factors of all items. A hybrid differential artificial bee colony (DABC) algorithm, which combines the superiorities of the differential evolution (DE) algorithm in global search and the artificial bee colony (ABC) algorithm in fine search, is presented to solve the proposed JRD model. Numerical experiments and parameter sensitivity analyses are conducted. The computational results have testified that DABC is faster than that of DE (16%) and two hybrid ABC-DEs (18% and 9%), more effective than that of genetic algorithm (GA, 0.15%), ABC (0.03%) and ABC-DE1 (0.015%) and robust than that of GA (99%), DE (71%), ABC (0.97%) and two hybrid ABC-DEs (86% and 88%). Most importantly, different uncertainty significance of lead-time and demands to JRD are revealed and analyzed. Furthermore, management insights, such as the magnitudes of the problem-related parameters and the mutual effects of the two stochastic variables to the system cost, are indicated in the replenishment process.