Xiaoqiu Shi

There are an increasing number of manufacturing service resources appeared on the cloud manufacturing (CMfg) service platform recently, which leads to a serious information overloading problem to the enterprises that need these resources. To tackle this problem, a graph neural network-based recommendation method for CMfg service resources is proposed, which effectively overcomes some limitations of the traditional recommendation methods. Specifically, we first use different similarity calculation methods (e.g., Cosine similarity, Pearson correlation coefficient, etc.) to calculate the similarities between different resources based on the feature information of CMfg service resources. A resource graph dataset is accordingly established. A graph neural network is then used to perform representation learning of nodes in these graphs, obtaining the vector representations of these nodes. Finally, new links that may appear in a graph are predicted by performing dot product calculations on these nodes’ vector representations. And these links can be used to recommend suitable resources. Experiments mainly show that (i) the proposed method obtains better link prediction accuracy compared with that of other link prediction algorithms; (ii) when the network density used for training is relatively high, the predictive performance of the proposed method is improved significantly. Our method can shed light on how to choose suitable CMfg service resources from the CMfg service platform.

In recent years, an increasing number of population-based metaheuristic algorithms have been proposed to solve the flexible job-shop scheduling problem (FJSP) according to its practicality and complexity. Most of these algorithms are single-population-based and hence are very susceptible to becoming trapped in local optimal values. Thus, multipopulation methods are widely used to improve these algorithms, resulting in multipopulation algorithms. These multipopulation algorithms have been widely studied in the context of single-layer complex networks recently. However, coupled networks used to control two (or more) algorithms simultaneously to get a better algorithm are always ignored in literature. Therefore, in this paper, using coupled scale-free networks (with different scaling exponents) to control the genetic algorithm (GA) and the invasive weed optimization (IWO) simultaneously, a multipopulation GA/IWO with coupled scale-free networks (MPGAIWO-SF) is proposed to solve FJSP. Then, we study how some parameters (e.g., subpopulation size, subpopulation number, and scaling exponent) of MPGAIWO-SF affect its performance, which is also ignored in literature. The simulation results illustrate that (1) the performance of MPGAIWO-SF is significantly improved compared with that of GA and IWO; (2) as the subpopulation size increases, the performance of MPGAIWO-SF first becomes better and then remains almost unchanged; (3) as the subpopulation number increases, the performance of MPGAIWO-SF first becomes better and then decreases rapidly; and (4) the performance of MPGAIWO-SF becomes worse slightly as the scaling exponent decreases. Finally, solving more FJSP instances, the MPGAIWO-SF with optimized parameters is compared with other related algorithms to verify its effectiveness.

Local disruptions can be propagated from one firm to another in a supply network (SN) and eventually influence the whole SN. Therefore, numerous studies on SN resilience considering the ripple effect have been reported recently. However, previous studies paid less attention to this phenomenon from a network structure perspective: if a firm is facing the risk of failure, then its partners may help it to mitigate the risk of failure by collaboration during the process of disruption propagation. Specifically, how SN structures (e.g. characterised by different scaling exponents) and other parameters (e.g. redundancy) influence the effectiveness of collaboration on improving SN resilience considering the ripple effect is not clear. Accordingly, we propose a ripple effect with collaboration (REC) model to consider the aforementioned phenomenon. We also present three new SN resilience metrics to evaluate SN resilience. Then, using both generated (by a novel SN generating model) and real-life SNs, we simulate the SN resilience considering REC under random and targeted disruptions. Our results demonstrate that the effectiveness of collaboration can be affected by SN structures and other parameters, and collaboration can even negatively affect SN resilience in some cases. We also summarise managerial implications and give future research directions.

The width of convolutional neural networks (CNNs) is crucial for improving performance. Many wide CNNs use a convolutional layer to fuse multiscale features or fuse the preceding features to subsequent features. However, these CNNs rarely use blocks, which consist of a series of successive convolutional layers, to fuse multiscale features. In this paper, we propose an approach for improving performance by fusing the low-level features extracted from different blocks. We utilize five different convolutions, including 3×3, 5×5, 7×7,5×3 ∪ 3×5 and 7×3 ∪ 3×7, to generate five low-level features, and we design two fusion strategies: low-level feature fusion (L-Fusion) and high-level feature fusion (H-Fusion). Experimental results show that the L-Fusion is more helpful for improving the performance of CNNs, and the 5×5 convolution is more suitable for multiscale feature fusion. We summarize the conclusion as a strategy that fuses multiscale features in the preceding stage of CNNs. Furthermore, we propose a new architecture to perceive the input of CNNs by using two self-governed blocks based on the strategy. Finally, we modify five off-the-shelf networks, DenseNet-BC (depth = 40), ALL-CNN-C (depth = 9), Darknet 19 (depth = 19), Resnet 18 (depth = 18) and Resnet 50 (depth = 50), by utilizing the proposed architecture to verify the conclusion, and these updated networks provide more competitive results.

With recent industrial upgrades, it is essential to transform the current forward supply networks (FSNs) into closed-loop supply networks (CLSNs), which are formed by the integration of forward and reverse logistics. The method chosen in this paper for building reverse logistics is to add additional functions to the existing forward logistics. This process can be regarded as adding reverse edges to the original directed edges in an FSN. Due to the limitation of funds and the demand for reverse flow, we suppose that a limited number of reverse edges can be built in a CLSN. To determine the transformation schemes with excellent robustness against malicious attacks, this paper proposes a multi-population evolutionary algorithm with novel operators to optimize the robustness of the CLSN, and this algorithm is abbreviated as MPEA-RSN. Then, both the generated and realistic SNs are taken as examples to validate the effectiveness of MPEA-RSN. The simulation results show that the index R, introduced to evaluate the robustness of CLSNs, can be improved by more than 95%, and this indicates that (1) the different schemes for adding reverse routes to an FSN can lead to different robustness values, and (2) the robustness of the transformed CLSN to malicious attacks can be significantly improved after optimization by MPEA-RSN. When an FSN is to be transformed into a CLSN, this paper can provide a frame of reference for building a CLSN that is robust to malicious attacks from a network structural perspective.

Populations of multipopulation genetic algorithms (MPGAs) parallely evolve with some interaction mechanisms. Previous studies have shown that the interaction structures can impact on the performance of MPGAs to some extent. This paper introduces the concept of complex networks such as ring-shaped networks and small-world networks to study how interaction structures and their parameters influence the MPGAs, where subpopulations are regarded as nodes and their interaction or migration of elites between subpopulations as edges. After solving the flexible job-shop scheduling problem (FJSP) by MPGAs with different parameters of interaction structures, simulation results were measured by criteria, such as success rate and average optimal value. The analysis reveals that (1) the smaller the average path length (APL) of the network is, the higher the propagation rate will be; (2) the performance of MPGAs increased first and then decreased along with the decrease of APL, indicating that, for better performance, the networks should have a proper APL, which can be adjusted by changing the structural parameters of networks; and (3) because the edge number of small-world networks remains unchanged with different rewiring possibilities of edges, the change in performance indicates that the MPGA can be improved by a more proper interaction structure of subpopulations as other conditions remain unchanged.

A genetic algorithm (GA) cannot always avoid premature convergence, and multi-population is usually used to overcome this limitation by dividing the population into several sub-populations (sub-population number) with the same number of individuals (sub-population size). In previous research, the questions of how a network structure composed of sub-populations affects the propagation rate of advantageous genes among sub-populations and how it affects the performance of GA have always been ignored. Therefore, we first propose a multi-population GA with an ER network (MPGA-ER). Then, by using the flexible job shop scheduling problem (FJSP) as an example and considering the total individual number (TIN), we study how the sub-population number and size and the propagation rate of advantageous genes affect the performance of MPGA-ER, wherein the performance is evaluated by the average optimal value and success rate based on TIN. The simulation results indicate the following regarding the performance of MPGA-ER: (i) performance shows considerable improvement compared with that of traditional GA; (ii) for an increase in the sub-population number for a certain TIN, the performance first increases slowly, and then decreases rapidly; (iii) for an increase in the sub-population size for a certain TIN, the performance of MPGA-ER first increases rapidly and then tends to remain stable; and (iv) with an increase in the propagation rate of advantageous genes, the performance first increases rapidly and then decreases slowly. Finally, we use a parameter-optimized MPGA-ER to solve for more FJSP instances and demonstrate its effectiveness by comparing it with that of other algorithms proposed in other studies.

Supply networks (SN) must maintain operations and connectedness under disruptions to remain competitive; this is referred to as SN resilience. Building a resilient SN is an underlying challenge in supply chain management. In this paper, SN resilience is examined from the complex network topology perspective to understand how supply chain managers construct resilient networks. The proposed growth model considers enterprises leaving the network, which previous studies have ignored. Considering the heterogeneous roles of enterprises, new metrics based on a new proposed sub-network concept are presented to evaluate resilience. Using a computer simulation, the resilience of the SN generated by the model proposed in this paper is compared with that of other models, and the results indicate that (i) the proposed model can be tuned to generate a desired resilient network; (ii) the proposed metrics capture the resilience requirements of the SN very well; (iii) the more uniform the distribution of the enterprises, the more resilient the corresponding SN; and (iv) the higher the values of α and β, the lower the SN resilience, and β affects the resilience more than α does.