中国科学院数学与系统科学研究院期刊网

25 December 2024, Volume 44 Issue 12
    

  • Select all
    |
  • JI Kangxian, XU Jian, LIU Xiaoting, SUN Jialu, XIA Yan
    Systems Engineering - Theory & Practice. 2024, 44(12): 3765-3776. https://doi.org/10.12011/SETP2022-2222
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    The international economic circulation affects China's economic growth through the production process of the product and the market demand of the product. In terms of production process, the mutual substitution of imported intermediate products and domestic intermediate products affects economic growth; in terms of market demand, foreign demand for China intermediate and final products affects China economic growth. Based on the structural decomposition analysis method, this paper decomposes the change of the Leontief inverse matrix into technology level change and import substitution, and decomposes the final demand change into domestic final demand change and export change, so as to measure the impact of international economic circulation on China economic growth from two aspects. The results show that: 1) Import substitution is an important channel for the international cycle to affect China economic growth, and it shows periodic characteristics. From 2000 to 2005, imported intermediate products replaced domestic intermediate products, which had a negative impact on economic growth; From 2005 to 2014, domestic intermediate goods substituted imported intermediate goods, and China gradually took control of more intermediate goods production processes. From 2015 to 2021, the share of imported intermediate goods again increased. 2) Compared to domestic final demand, the contribution of exports to China's economic growth has been continuously decreasing, and China's dependence on the final demand of international circulation has been gradually declining.
  • YUE Ting, ZHOU Jing, LONG Ruyin, ZHANG Yingkai, WANG Qianru, CHEN Hong
    Systems Engineering - Theory & Practice. 2024, 44(12): 3777-3792. https://doi.org/10.12011/SETP2024-0015
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Promoting carbon emission reduction of urban residents is of great significance for mitigating climate problems. Based on the panel data of 288 cities above prefecture level in China from 2009 to 2019, this paper calculated the living carbon emissions of urban residents, and combined population and economic characteristics to cluster cities into four types for analysis, and analyzed the influencing factors of living carbon emissions of urban residents. And BP neural network and scenario analysis were used to predict the carbon reduction potential of various urban residents. The results show that: 1) The total carbon emission of urban residents in China is increasing year by year, and the proportion of carbon emission from electricity is the highest, and the growth rate of carbon emission from heating is the highest. 2) Urbanization level, per capita disposable income, energy structure and total population size all have positive effects on the carbon emissions of urban residents, while energy intensity and consumption tendency of urban residents have negative effects, and the influencing factors of carbon emissions of various cities have certain differences. 3) All kinds of cities have great carbon reduction potential in residents' life, and there are great differences. The carbon reduction potential of the second type of cities is significantly higher than that of other cities. The first type of cities has the lowest carbon reduction potential overall. The change degree of carbon reduction potential of the third and fourth types of cities is similar, showing a trend of first increasing and then decreasing. All localities may formulate and implement carbon reduction measures for residents according to local conditions.
  • Lü Dan
    Systems Engineering - Theory & Practice. 2024, 44(12): 3793-3810. https://doi.org/10.12011/SETP2024-0525
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Improving firm ESG performance is an important measure to achieve sustainable economic development. This study takes the implementation of the “Broadband China” strategy released in 2013 as a quasi-natural experiment and uses differences-in-differences method to evaluate the impact of digital infrastructure on firm ESG performance. The study finds that digital infrastructure has a significant promoting effect on firm ESG performance. The mechanism analysis shows that the impact of digital infrastructure on firm ESG performance is mainly achieved through pathways such as increasing government environmental concerns, incentivizing firms to fulfill social responsibilities, and improving firm information transparency. Heterogeneity analysis reveals that the promoting effect of digital infrastructure on firm ESG performance is more significant in large-scale firms, firms with high customer concentration, high-polluting industries, and firms with strong green innovation capabilities. This study evaluates the practical role of digital infrastructure from a sustainable development perspective, providing new empirical evidence for understanding the influencing factors of firm ESG performance and offering policy recommendations for strengthening digital infrastructure construction and promoting economic green transformation.
  • BEI Honghan, HU Jingyi, YANG Wanyu, GAI Zhaoyi
    Systems Engineering - Theory & Practice. 2024, 44(12): 3811-3828. https://doi.org/10.12011/SETP2024-0002
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    With the escalating impact of climate change leading to more frequent extreme precipitation events, effectively mitigating the risks and economic losses associated with this uncertainty is increasingly critical. This paper introduces a novel “Markov-Gumbel” theoretical model to measure precipitation index, which, in conjunction with risk-neutral theory, forms the basis for a new pricing model for precipitation index derivatives. We apply this model using daily precipitation data from regions such as Zhengzhou City in Henan Province and Xuzhou City in Jiangsu Province, China, to validate and analyze its efficacy. Research indicates that this new method for measuring the precipitation index offers enhanced flexibility and better captures seasonal variations. Moreover, the proposed pricing model for precipitation derivatives, grounded in risk-neutral valuation methods, results in more focused and stable pricing outcomes, significantly improving pricing accuracy. The findings of this research not only provide an innovative framework for the measurement and application of precipitation index derivatives but also offer valuable theoretical and practical insights for effectively hedging against risks associated with precipitation uncertainty.
  • CHEN Rongda, YU Jingjing, CUI Miaosen, JIN Chenglu, WANG Shengnan, CHEN Yiyang
    Systems Engineering - Theory & Practice. 2024, 44(12): 3829-3850. https://doi.org/10.12011/SETP2023-2150
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Analysts' past performance and their employment status in the securities industry jointly affect the information disclosure quality of listed companies tracked and analyzed. This article uses the analyst-company dataset from 2011 to 2021, and constructs an analyst network reputation through 16 relevant indicators at the individual analyst and securities firm levels, and explores its impact on the information efficiency of the stock market. The study found that analyst network reputation increases competitive information, widens opinion divergence, and divides investors' attention, thereby reducing the information efficiency of the stock market. In particular, analyst recommendations may contain invalid information. When investors face numerous information with limited attention, trading activity decreases, stock liquidity declines, and market information efficiency is weakened. In addition, media coverage has a diminishing effect on analyst network reputation, and market investor sentiment has an amplifying effect on analyst network reputation, which reduces information efficiency.
  • ZHAO Er'long, SUN Shaolong, WANG Fenghu, WANG Shouyang
    Systems Engineering - Theory & Practice. 2024, 44(12): 3851-3861. https://doi.org/10.12011/SETP2022-2601
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    The information conveyed by analysts' research reports has a guiding effect on investors' decision-making behavior, which in turn synergistically affects stock price volatility. To quantify the relationship between the impact of analysts' research reports on investor returns under the complex online platform, this study first obtains the headline text data of 39{,}786 analysts' research reports for the period of January 1, 2017 to December 31, 2021 on Eastmoney. Second, the sentiment dictionary of analysts' research reports is constructed by text mining technology, and the corresponding stock sentiment values are obtained based on the SESTM model, and stocks with sentiment values greater than a certain threshold are screened out. Finally, simulated backtesting is conducted by equal weighting, and the results show that this deep learning-based analyst forward-looking report headline construction of a complete quantitative investment trading strategy has a high investment return in trading backtesting. This study is of great theoretical and practical significance for understanding the forward-looking analysis of analysts' research reports and effectively guiding rational investment behavior.
  • LI Xiao, LIU Yilian
    Systems Engineering - Theory & Practice. 2024, 44(12): 3862-3876. https://doi.org/10.12011/SETP2024-1010
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Existing literature mainly focuses on the impact of investor attention on stock return predictability and asset dynamics. However, less research interest is paid on the impact of investor attention allocation on asset prices. In this paper, we employ the COVID-19 event as the attention-distraction shock and empirically investigate the impacts of investor attention allocation on the return comovement. This attention-distraction shocks are characteristics by have been salient nationwide, repeated, and suitable for the stock market dominated by individual investors. The empirical results mainly reveal that: 1) return comovement is significantly increased when investors are distracted; 2) analyst following, market openness, information transparency could significantly alleviate the magnitude of return comovement. We mainly contribute to the literature in two aspects: First, with the novel attention-distraction event, we confirm that substitution effect of investor attention allocation; second, we also contribute to literature on the determinants of return comovement by documenting a new determinant of return comovement, i.e., investor attention allocation. All these findings should be of great interest to investment professionals interested in the investor attention allocation on financial markets. Much work needs to be done on the impacts of investor attention allocation on micro-behavior, e.g., order imbalance, probability of informed trading, anomalies, as well as cross-assets and cross-markets movements. We leave these for future research.
  • FANG Xia, TAN Longxin, WU Jie
    Systems Engineering - Theory & Practice. 2024, 44(12): 3877-3895. https://doi.org/10.12011/SETP2023-1635
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Digital finance accelerates the flow of cross-regional factors, improves the efficiency of regional capital allocation, and provides new impetus for regional income convergence. This article uses the 2014—2021 district and county data in the Yangtze River Delta region as a sample to explore whether digital finance development center areas can drive low-income regions converge towards high-income levels, realizing that the first-rich areas drive the development of later-rich areas. Research has found that low-income areas can catch up with the income growth rate by strengthening the digital financial spatial connection with central areas. From the perspective of the mechanism, the learning effect, improving the level of innovation and entrepreneurship is an effective mechanism for digital financial spatial correlation to achieve regional income convergence. Further analysis found that digital financial development center areas can break through terrain constraints to achieve spatial linkage and alleviate uneven income distribution, but the coverage of this effect is limited; the opening of high-speed rail will also be beneficial to the spatial correlation of digital finance in promoting regional income convergence. The research conclusions of this article provide useful reference for the Yangtze River Delta region to further alleviate regional income differences and solidly promote common prosperity.
  • ZHANG Liang, REN Yeyao, HOU Tianyu, FENG Haofang
    Systems Engineering - Theory & Practice. 2024, 44(12): 3896-3916. https://doi.org/10.12011/SETP2023-0390
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Although the decoupling response can temper the conflicts between different stakeholders, it may create some negative consequences for firms. However, little research explores whether and how firms adjust their giving to mitigate or avoid such negative consequences. Drawing on stakeholder literature and social comparison theory, this study explores the relationship between the decoupling of corporate giving and its subsequent increase and investigates the moderating effects of managers' social comparison on this relationship. Using a sample of Chinese listed firms from 2003 to 2017, we find that firms adopting a decoupling response in corporate giving tend to increase their giving in the subsequent stage. This study proposes and verifies that managers find reference points in two ways: Reference groups (industry peers or community peers) and stakeholder expectations. The results show that: 1) firms tend to make more donations in the subsequent year when they adopted the decoupling of corporate giving and the level of their giving is below the giving of the reference groups (industry peers or community peers); 2) the higher the expectation of stakeholders on firms' donation (i.e., firms with high financial ability or firms in heavy pollution industry), the stronger the relationship between the decoupling of corporate giving and its subsequent increase. Hence, by investigating the relationship between the decoupling of corporate giving and its subsequent adjustment, this study enriches the literature on the consequences of organizational decoupling and fills the research gap in corporate giving adjustment literature. The findings regarding the moderating effects of managers' social comparison are helpful for managers to solve the donation dilemma, which suggests that they can adjust the level of corporate giving in the next stage through social comparison, avoid the situation whereby firms donate too little or too much, thus extending the literature on corporate giving adjustment.
  • ZHOU Qing, WU Zhengyi, CHEN Wenchong
    Systems Engineering - Theory & Practice. 2024, 44(12): 3917-3931. https://doi.org/10.12011/SETP2022-2270
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    The diffusion of a manufacturing technology standard (MTS) which is transformed and upgraded by a local standard alliance is a significant method for a Chinese corporation to occupy the market of “Belt and Road” and in turn promote the transformation of local industry chain. Considering the two dimensional attributes of a MTS (i.e., MTS-based product diffusion and technology diffusion), this paper proposes a new decision-making model for the two-sided networks coordination problem to accelerate the MTS diffusion along “Belt and Road”. Based on the decision-making processes of customers, standard alliance, and local manufacturing enterprises, a multi-objective leader and multi-follower interactive optimization is formulated after introducing a contract design and adoption mechanisms to reveal the processes of such two-sided network coordination. The new product pricing problem and component procurement contract design problem with standard competition is defined as the leader optimization problem to maximize the degree of new product diffusion, the quantity of local manufacturing enterprises adopted new technologies, and the benefit of standard alliance simultaneously. The contract acceptance and component supply problem of each local manufacturing enterprise is defined as the follower to maximize the net revenue of each one. The nested NSGA-II algorithm is designed to solve the interactive optimization problem. The effectiveness of the model and the algorithm is validated by a practical case study.
  • DAI Qianzhi, WANG Yihong, XIE Qiwei, LEI Xiyang
    Systems Engineering - Theory & Practice. 2024, 44(12): 3932-3946. https://doi.org/10.12011/SETP2024-0439
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    The evaluation result of energy and environmental efficiency is directly related to the interests of the decision-making units (DMU), thus it would be helpful to improve the acceptance of the evaluation result if the non-cooperative game relationship between DMU is considered. According to the typical two-stage structure of “energy utilization-environmental protection” in the energy environment system, this paper proposes an approach based on two-stage DEA and non-cooperative game. We first provide an algorithm based on the non-cooperative game of subsystems and obtain the efficiency scores of the subsystems. Then we obtain the efficiency of the total system by integrating the efficiency scores of the subsystems. We prove that the algorithm converges to a unique Nash equilibrium point. The method was further applied to the energy and environmental efficiency evaluation of 30 Chinese provinces in 2019. The findings include that: 1) The efficiency of the total system and its subsystems show regional imbalances, decreasing sequentially as “east-middle-west-northeast”; 2) Many provinces are in a “low energy utilization-low environmental protection” development mode. The key to transforming this mode is to improve the environmental protection subsystem efficiency in the eastern region, while both subsystems need improvement in the central, western, and northeastern regions. Based on the aforementioned findings, this paper suggests specific improvement directions for each province according to their different development modes, which has positive guiding significance for China to improve the level of energy and environment efficiency.
  • ZHU Qingyuan, LIU Chang, PAN Yinghao, WU Jie, LI Feng
    Systems Engineering - Theory & Practice. 2024, 44(12): 3947-3962. https://doi.org/10.12011/SETP2023-0774
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Under the background of the “dual credit” policy of promoting the healthy development of new energy vehicles and energy saving and emission reduction of fuel vehicles, a competitive game model including the government, fuel vehicle manufacturers, new energy vehicle manufacturers and consumers is established. The emission reduction R&D investment of fuel vehicle manufacturers is included in the model, and the impacts of the dual credit policy and the government's gradually declining subsidies are theoretically studied. The findings are as follows: 1) Under certain conditions, the gradual decline of government subsidies will be more conducive to the increase of R&D investment of fuel vehicle manufacturers; 2) The impacts of the credit transaction price in the dual credit policy on the emission reduction R&D investment of fuel vehicles and the automobile market demand are non-monotonic. Therefore, under the background of the decline of government subsidies, a low credit transaction price should be set to stimulate the R&D investment of fuel vehicles and stimulate the market demand for new energy vehicles; 3) The impact of the decline of government subsidies on carbon emissions is non-monotonous, and the decline of subsidies will reduce carbon emissions in the automobile market.
  • WANG Zhiyuan, GUO Xian, RAN Lun, YAO Zhaosheng
    Systems Engineering - Theory & Practice. 2024, 44(12): 3963-3978. https://doi.org/10.12011/SETP2024-0115
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    This paper addresses the location and capacity planning of battery swapping stations of electric vehicles, combining the charging and swapping operations in the stations. The charging and swapping operations within the swapping station are a crucial link connecting the swapping demand with the decisions on station location and capacity planning. However, previous research has overlooked providing a detailed characterization of this process. This study models the internal operations of the swapping station as a multi-period optimization problem and provides insights into the structural properties of the optimal solution to this problem. Building upon this foundation, considering the uncertainty in swapping demands, we integrate the internal operational aspects with the station location and capacity planning to construct a distributionally robust optimization model and a robust satisficing model. To deal with the hard multistage problem in the model, we utilize the linear decision rule to approximately solve the two models and extend the lifting technique by incorporating auxiliary variables into multistage scenario-wise robust optimization models. The theoretical analysis establishes the relationship between the models before and after lifting. Finally, numerical experiments are conducted to validate the effectiveness of the proposed model and lifting techniques.
  • ZHANG Yu, GUO Renyong
    Systems Engineering - Theory & Practice. 2024, 44(12): 3979-3996. https://doi.org/10.12011/SETP2023-1227
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    In this paper, an optimization model of the one (driver)-to-many (riders) mode is proposed for the ride-sharing problem with continuous timeframes. The objective of the proposed optimization incorporates both the sharing rates at the current timeframe and the predicted demand densities at trip destinations. This approach characterizes the instantaneous utility and potential utility of trips, thereby improving the overall service rate and quality during the operation period. We then design a demand prediction algorithm. In the feature engineering phases, time-indexed features and spatial-indexed features are extracted, and sequential features at various time intervals are generated. Based on deep learning method, we construct the spatio-temporal auto-sequence net (STAS-Net) capable of processing both spatio-temporal and sequential features, thereby providing highly accurate demand prediction. Finally, the proposed model and algorithm are applied to a real-world scenario to verify their effectiveness. The results indicate that compared with the traditional model, the matching solution generated by the proposed model improves the service rate by 9.57% and reduces the travel distance by 11.54%. Meanwhile, the designed prediction algorithm, STAS-Net, outperforms other prediction algorithms with a mean absolute error of 2.84 and a mean square error of 20.35.
  • ZHANG Yanlu, CHAO Zhuoyi, YANG Naiding, YANG Jiaqi
    Systems Engineering - Theory & Practice. 2024, 44(12): 3997-4010. https://doi.org/10.12011/SETP2023-0790
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Most existing studies evaluate node importance from the perspective of network topology, but ignore the impact of network dynamic characteristics on the node importance evaluation results. But the studies based on the perspective of network dynamics mainly focuse on the abstract networks to identify key nodes, and have some certain limitations to be applicated in real networks. Therefore, this paper identifies key node enterprises of the new energy vehicle research and development network based on the perspective of cascading failures. Firstly, this paper builds a real new energy vehicle research and development network by using new energy vehicle cooperation patent data. Then, a cascade failure model is proposed from three aspects: Defining the initial load of the node enterprise, determining the capacity of the node enterprise, and establishing load propagation rules. Next, numerical simulation method is used to reveal the cascade failure process and rules of the new energy vehicle R&D network under deliberate and random attack strategies. Finally, a simulation analysis of a real new energy vehicle research and development network is conducted to identify the key node enterprises of the new energy vehicle research and development network. The research results have important reference value for preventing large-scale cascading failures caused by the failure of some key node enterprises and ensuring the normal cooperative development of new energy vehicle node enterprises.
  • ZHANG Qian, WANG Zhongbin, LI Yongjian
    Systems Engineering - Theory & Practice. 2024, 44(12): 4011-4025. https://doi.org/10.12011/SETP2023-2160
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    In recent years, China's food delivery industry has undergone substantial growth, driven by the rapid expansion of the platform economy and the influence of the COVID-19 pandemic. Food delivery services have not only lessened customers' sensitivity to delays associated with in-person dining but have also generated increased market demand for merchants. It is noteworthy that the majority of merchants employ a centralized operational mode, which combines food delivery and dine-in services within a single establishment. However, certain merchants opt for a decentralized approach, wherein they establish dedicated food delivery outlets exclusively handling food delivery orders while maintaining an offline restaurant. To examine the impact of food delivery channels on merchant decision-making, this study establishes a dual-channel service system operating within a congestion-prone environment. It characterizes the equilibrium strategy of customers under the two operational policies and investigates how the quality of food delivery services affects merchant profits. Furthermore, the research reveals the optimal operational approach based on varying levels of delivery quality. The key findings of the study are as follows. 1) In the case of decentralized operations, the service capacity allocated to the food delivery channel by the merchant exhibits a non-monotonic relationship with its quality. This implies that higher food delivery quality may gradually prompt the merchant to shift its focus toward the offline channel. 2) Despite the fact that higher food delivery quality has the potential to attract more customers, the study surprisingly finds that improving food delivery quality may actually reduce merchant profits in both centralized and decentralized scenarios. 3) While decentralized operations may lead to decreased order processing efficiency, adopting this approach can effectively mitigate the cannibalization effect of the food delivery channel and result in higher profits, particularly when food delivery quality is high. Consequently, centralized mode is recommended only when the food delivery quality falls within an intermediate range. Additionally, we further validated the robustness of this conclusion from various perspectives, including marginal costs and delivery fees.
  • SONG Yanan, LIU Lu, LI Tingting, YAN Xiangbin, ZHAO Enlong
    Systems Engineering - Theory & Practice. 2024, 44(12): 4026-4044. https://doi.org/10.12011/SETP2023-2339
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Based on the exponential smoothing model and Logit choice model, this paper describes the multi-period purchase behavior of consumers, constructs the revenue model of two heterogeneous retailers under different money-back guarantees, and discusses the impact of money-back guarantees on experiential-learning consumers' purchase likelihood and retailers' revenues. We further investigate the changes brought about by the introduction of AI-assisted decision-making. We find that, due to the fluctuation of product quality and consumers' experiential-learning, consumers' beliefs about the average quality of products are lower than the real average quality. Different money-back guarantees will affect consumers to form different beliefs about product quality. When the focal retailer provides money-back guarantees while the competing retailer not, consumers' quality belief of the focal retailer is the closest to the real average quality. The retailers do not always benefit from offering money-back guarantees. When the unit return cost is low, it is the dominant strategy to offer money-back guarantees. As the return cost increases, the equilibrium of the two retailers' money-back guarantees gradually changes from providing money-back guarantees to not. Numerical analysis shows that the equilibrium reached by two retailers has a phenomenon of Prisoner's dilemma. Specifically, when the return cost is low, both retailers provide the money-back guarantees in equilibrium, but the equilibrium revenue of both retailers is lower than the revenue when no money-back guarantee is provided. This suggests that competition induces both retailers to choose to offer money-back guarantees, although it does not achieve the optimal retailer's payoff optimal. After introducing AI-assisted decision-making, retailers with large quality fluctuations obtain an increase in long-term profits. Offering money-back guarantees remains a dominant strategy when the return cost is low, but AI-assisted decision-making changes the threshold for strategy formulation.
  • YANG Lian, LIU Wenxiu, ZHANG Zhipeng, SHI Baofeng
    Systems Engineering - Theory & Practice. 2024, 44(12): 4045-4063. https://doi.org/10.12011/SETP2023-1977
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    The existing credit evaluation models have weak applicability to imbalanced samples and weak scalability to datasets with different credit characteristics. In this study, the principle of Taylor expansion is used to transform the Cross Entropy function into a linear combination of polynomials, and the perturbation factor $\varepsilon$ is added to the first polynomial coefficient. Based on this, the BPNN-TaylorLoss default prediction model is constructed. We use 4 real credit data, 7 comparative models, and 5 model evaluation criteria to verify the performance of the model. The results show that the proposed model helps to reduce the losses caused by misjudgment of defaulting customers to financial institutions and prevent the loss of high-quality customers caused by misjudgment of non-defaulting samples. The proposed model exhibits robust default prediction performance in most credit data sets, thus exhibiting good model scalability. The Innovation of this study is we use the perturbation factor $\varepsilon$ to modify the Taylor expansion of the Cross Entropy function, and constructs the BPNN-TaylorLoss default prediction model. It can change the current situation that the existing credit evaluation models are not suitable for unbalanced data sets and have weak scalability for data sets with different credit characteristics by adjusting only one hyperparameter $\varepsilon$. This study provides a new research perspective for credit risk assessment of imbalanced samples.
  • ZHANG Zhongliang, GONG Shengchen, WANG Yi, LUO Xinggang
    Systems Engineering - Theory & Practice. 2024, 44(12): 4064-4083. https://doi.org/10.12011/SETP2023-0686
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Federated learning is a distributed machine learning technique which enables clients with limited resources to collaboratively train models without sharing private data, effectively protecting the data privacy of the clients. Classic federated learning systems lack a strict mechanism for selecting clients, but generally use an average strategy to aggregate the local model parameters, which may lead to the inclusion poor-quality clients in the training process of the federated learning, consequently affecting the overall performance of the final models. To address the above issues, a federated learning client selection method based on dynamic programming is proposed (FedWeight). The proposed method uses the Shapley Value method to measure the contribution of each client in different communication rounds, addressing the inherent difficulty of evaluating clients' data quality directly. Using the Shapley Value as an important measurement to dynamically select high-quality clients by the server, and then the server improves the overall performance of model by aggregating these high quality clients. To construct different federated learning scenarios, MNIST, CIFAR-10, Fashion-MNIST, EMNIST and KMNIST datasets are used in our experiments. The experimental results demonstrate that the proposed method can effectively identify high-quality clients, and the performance of the obtained final federated model is almost unaffected by poor-quality clients. Furthermore, our method exhibits significant advantages in terms of convergence speed and model stability.
  • TENG Chenmei, XIANG Yin, LI Shanliang
    Systems Engineering - Theory & Practice. 2024, 44(12): 4084-4096. https://doi.org/10.12011/SETP2023-0668
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    Facing the twin challenges of an aging population and a high incidence of chronic diseases, the rational distribution of healthcare resources to adapt to the dynamically changing healthcare demands is critically important. This study introduces an innovative cross-sector collaborative allocation model aimed at enhancing the dynamic response to healthcare needs at various stages. Through analyzing the phased evolution of healthcare demand and its correlation with resource distribution, the research uncovers pivotal points for optimizing healthcare services. To solve this model effectively, the study has developed and refined a hierarchical genetic algorithm, introducing discriminant operators and a novel encoding strategy to boost algorithm performance and the suitability of solutions. Case studies and sensitivity analysis have verified the model's heightened efficiency in response, particularly when the budget is ample, revealing the model's capacity to fulfill diverse healthcare needs cost-effectively. P-value statistical analysis indicates that, in comparison to existing methods, our proposed algorithm demonstrates superior precision and efficiency in tackling practical problems, showing its real-world application value in future healthcare resource management.
  • GUO Shujuan, PENG Kangzhen, GUAN Zekun, LIU Yizhuo, JIN Zhihong
    Systems Engineering - Theory & Practice. 2024, 44(12): 4097-4112. https://doi.org/10.12011/SETP2023-0234
    Abstract ( ) Download PDF ( )   Knowledge map   Save
    In this paper, a dynamic inter-terminal truck routing optimization model is formulated based on the time-space graph combining the inter-terminal with railway transportation, considering the uncertainties such as truck travel time and maximum loading capacity of railway trains. A Q-learning hyper-heuristic (QHH) algorithm is designed to solve the problem of truck routing, train departure problem, and train container loading problem. The QHH algorithm employs the Q-learning algorithm as the high-level selection strategy to guide the hyper-heuristic algorithm to select the appropriate low-level heuristics (LLHs) in different environments. A random strategy is proposed to improve the diversity of the population and accelerate the convergence speed under the condition of changing environments. The experimental results indicate that the proposed QHH algorithm in this paper, compared with the genetic algorithm, decreases the total cost of the optimization scheme by 8.7% on average in small-scale cases and 27.4% in the large-scale. This proves the effectiveness of the proposed algorithm in addressing this problem, which can compensate for the shortcomings of the traditional meta-heuristic algorithm in the simplex mechanism and problem-oriented customization. When the environment changes at high, medium, and low frequencies, the QHH algorithm outperforms the sample random hyper-heuristic, which proves that Q-learning can adapt to environmental changes and flexibly select the LLHs. The results of this study can provide decision support for managing inter-terminal truck scheduling operations in large container ports.