Using ESG rating data of Sino-Securities Index Information Service from 2009 to 2020, this paper examines the impact of listed companies' ESG rating on the level of stock price, financial and operational risk in the next year. The study finds that better ESG rating has a significant inhibitory effect on all three risk levels of enterprises in the next year. Specifically, for the risk of stock price crash, ESG rating higher than the benchmark level, as a strong market signal, has a more significant reduction in the risk level of stock price crash. The trading volume of individual stocks, which reflects the attention of investors, has an intermediary effect on ESG to reduce the risk of enterprise stock price crash. ESG of large-scale enterprises that occupy an important position in the market and attract more attention from investors has a stronger inhibitory effect on the risk of stock price crash; In addition, the negative relationship between ESG and the risk of stock price crash is more significant after the implementation of the “Environmental Protection Law”. For financial risk, ESG has a marginal diminishing effect on reducing corporate financial risk, and the improvement of ESG rating from low to medium can improve the level of corporate financial risk. At the same time, enterprises' voluntary disclosure of non-financial information could strengthen the inhibitory effect of ESG on financial risks. For operational risk, ESG rating has a marginal diminishing effect on reducing operational risk; At the same time, the nature of equity has a moderating effect on the reduction of operating risks by ESG rating. Compared with private enterprises, ESG has a stronger inhibition effect on the operation risk of state-owned enterprises. Finally, the sub-sample heterogeneity test results based on the length of enterprise life in this paper show that the inhibitory effect of ESG rating on risk is stronger for enterprises with a long establishment age, but weaker for enterprises with a short establishment age.

Based on investor behavior data from the Eastmoney Guba, this paper employs the effective transfer entropy method to investigate whether the mediating role of stock forums in the stock market better represents investor attention or sentiment. The findings reveal that: 1) Investor attention in stock forums conveys more valuable information flows to the stock market, providing a more accurate representation of investor behavior; 2) The net comment volume, as a key indicator of investor attention, exhibits strong information flow correlations with the stock market, significantly influencing stock volatility, with its effect extending up to one month into the future; 3) Across posts with different sentiments (positive, neutral, and negative), investor attention consistently transmits information flows that predict stock volatility, while investor sentiment lacks significant incremental value in transmitting information flows to the market. This study further underscores the mediating role of stock forums in capturing investor attention, particularly highlighting the unique predictive value of net comment volume, offering investors a novel perspective on analyzing online stock forum data. Future research should focus on integrating more comprehensive datasets and conducting in-depth explorations to enhance the accuracy of market forecasts and the scientific rigor of investment decision-making.

Focusing on the supply side of vehicle charging service, this study proposes two differential game models for charging pile operation decision-making, with the government, operator, and third-party platform (hereinafter referred to as the platform) as the participants. The two differential game models are decentralized (i.e., operators and platforms aim to maximize their own interests) and centralized (i.e., operators and platforms aim to maximize the overall interests of both). The results show that under the equilibrium state with fixed revenue distribution ratios between operator and platform, compared with the decentralized decision-making mode, the centralized decision-making mode can improve the efforts of operator and platform, service quality and social benefit. When the platform is relatively weak and has a lower share of revenue, adopting the centralized decision-making mode can achieve a Pareto improvement in the revenues of both operator and platform; conversely, when the platform is relatively strong and has a higher share of revenue, adopting the decentralized decision-making mode can increase the revenues of both operator and platform. This indicates that as platform develops from weak to strong, the decision-making mode of the charging service market may shift from centralized to decentralized. At this time, the proportion of government policy support will increase, and social benefit and service quality may decrease.

In this paper, the stability and robustness of a recurrent neural network (RNN) controller with saturation function and ReLU function as activation function are analyzed for first-order linear uncertain systems. The necessary conditions for the closed-loop system to converge to the non-zero target and the suffcient conditions for exponential stability are provided. The quantitative relationships between the recurrent neural network controller’s parameters and the robustness of the initial state value, the target value and the unknown parameter of the plants are analyzed. The analysis results show that the RNN controllers with ReLU function as the activation function have stronger robustness.

The registration-based system represents a pivotal reform in China’s capital market development. The inquiry system reform aims to transfer pricing authority more substantially to market participants and enhance IPO pricing effciency. Consequently, systematic research evaluating IPO pricing effciency and the effects of inquiry system reforms under the registration-based framework have attracted considerable scholarly attention. This paper employs a bilateral stochastic frontier model to measure IPO pricing effciency across 2365 listed companies in China’s A-share market from 2016 to 2023, and empirically verifies the systematic impact of inquiry system reforms on IPO pricing under the registration-based system. Research findings indicate that during 2016–2023, underpricing effects dominated overpricing effects in A-share market initial offerings, with overall pricing 7.06% below reasonable levels, exhibiting distinct characteristics across different boards, years, ownership structures, and break-even status. The inquiry system reform generally elevated initial offering prices, primarily driven by the distinctive characteristics of the STAR Market, while other boards demonstrated declining trends. During the registration-based system expansion phase, significant competitive dynamics emerged between the STAR Market and ChiNext Board, while reform effects on the main board remained limited. Furthermore, the study identifies two critical transmission pathways explaining these impacts: The number of inquiry institutions and the effectiveness of price quotations. Based on these conclusions, this paper proposes targeted recommendations for regulatory authorities to guide future inquiry system reforms.

Sparsity is a crucial assumption in high-dimensional modeling, as only a small subset of variables typically exert significant influence on the response in high-dimensional regression analysis. Based on varying coeffcient models, this paper proposes a varying sparse coeffcient mixed-effects quantile regression (VSCMEQ) model for longitudinal data, which incorporates variable selection. In this model, the coeffcient functions are estimated using B-splines, and penalties are imposed on both random and fixed effects to investigate the influence of relevant important factors, including varying effects and constant effects. Finally, the proposed method is applied to the Primary Biliary Cirrhosis (PBC) dataset to analyze disease progression, identifying the influence of significant factors on disease progression (biomarkers) at different quantiles.

This paper considers the supply interruption of a supply chain composed by two suppliers and one retailer, where one supplier is an integrated supply and marketing supplier and the other supplier is a pure supplier, and the retailer makes replenishment from the latter supplier and makes and emergency order from the former when the supply is interrupted. For this system, we mainly investigate the retail price decision and the emergency replenishment from the supply and retailing integrated supplier in the case of a supply interruption with a random end time from the second supplier. Based on maximizing the benefits of each member in the supply chain, we establish an optimization model, and its solution is obtained via a theoretical analysis which gives the optimal decision for each member of the supply chain. Some numerical experiments are made which give the impact analysis of main parameters on the optimal decision of each member of the supply chain and their benefits for the supply interruption period.

Trend filtering is a widely used method for extracting long-term trends and eliminating short-term noise from time series data. In order to accurately capture the global change pattern and local fluctuation of the potential trend, this paper proposes the generalized trend filtering model with composite $\ell_0$ constraint (L0CTF) based on the primitive function representing sparsity, and the optimality theory is analyzed. However, solving the L0CTF model is a challenging task because of the combinatorial property and indivisibility of the composite $\ell_0$ function. Therefore, based on the properties of composite $\ell_0$ function, this paper reformulates the L0CTF model as a mixed integer programming problem with special ordered sets of type 1 and analyzes its equivalence with L0CTF in the sense of global optimal solution. Finally, experimental results on simulated and real data sets show that the proposed method is superior to some mainstream trend filtering methods in extracting potential trends.

In engineering practice, exact failure times of individual components are generally not available. In contrast, only the number of component failures and the system’s cumulative operating time are known, which leads to the aggregate lifetime data. Inference of lifetime distributions based on the aggregate lifetime data is of great challenge. This paper proposes a moment-based point estimation method, and uses the bias-corrected Bootstrap method to construct confidence intervals for quantities of interest. The maximum likelihood method needs the likelihood function of the aggregate lifetime data, however, it is only applicable to a few distributions that have the closure property with respect to the operation of convolution. Differently, the proposed method does not utilize the likelihood function, thus it applies to more distributions. Finally, both the simulation study and the real data analysis are performed for demonstration and illustration.

In order to generate structured representations of conflict events, and identify the logical relations of events, the key events and event evolution patterns, this paper proposes a conflict event ontology models and an automated construction framework of event-centric conflict knowledge graph. Graph neural network incorporating attention mechanism and dependent syntactic analysis are leveraged to solve long-range dependencies, and RoBERTa-based fine-tuning is employed to extract the implicit event relations. The results show that both algorithms of event and event relation detection in this paper outperform the comparing algorithms. Additionally, the MPCCN algorithm identifies key events and critical paths that influence the development of conflict events. The event-centric conflict knowledge graph not only can be used to identify the key events and evolution paths of conflict events, but also provides multi-level analysis of conflict events, and improves the comprehensiveness and scientificity of decision-making.

The efficient and accurate estimation of sensitivity distribution parameters and quantiles are crucial for the design and evaluation of the reliability of pyrotechnic products. The approach developed in this paper employs a Bayesian framework to establish a semiparametric generalized linear model for sensitivity data, using the Hamiltonian Monte Carlo algorithm for posterior inference. Within this framework, the deviance information criterion and the logarithm of the pseudo-marginal likelihood are used in a data-driven manner to select the optimal model. Extensive simulation comparisons demonstrate that the proposed method can accurately estimate sensitivity distribution parameters in the case of small sample sizes. Finally, the new method is applied to two real datasets, validating its effectiveness. The new method provides an alternative and complementary modeling tool for the analysis of sensitivity data.

Duadic codes are an important class of cyclic codes. It is interesting to construct duadic codes whose minimum distance has the square-root lower bound. In this paper, we propose two construction methods of odd-like duadic codes whose minimum distance has the square-root lower bound. Two classes of odd-like duadic codes with the square-root lower bound on the minimum distance are obtained.

Hydraulic actuators are widely used in industrial control systems, where precise displacement control is critical to system performance. Traditional physical modeling methods struggle to accurately capture the nonlinear and time-series characteristics of hydraulic actuators, limiting their application in complex environments. This paper proposes a displacement modeling method for hydraulic actuators based on a long short-term memory (LSTM) neural network. By collecting time-series data of voltage input and displacement output, an LSTM network is employed to characterize the dynamic behavior of hydraulic actuators. The LSTM network effectively captures long-term dependencies in the data, adapting to the nonlinear time-series properties of hydraulic systems. During model training, the mean squared error is used as the optimization objective, and the effectiveness of the model is validated through experiments. The experimental results demonstrate that, compared to traditional methods, the LSTM network achieves lower prediction errors on the validation set, exhibiting stronger modeling capabilities and higher accuracy.

The national independent innovation demonstration zone (NIIDZ), as an important engine leading innovative development, takes institutional and policy reforms as a starting point to radiate and drive the coordinated development of surrounding regions. The gradual improvement of the high-speed rail (HSR) network has opened up a new pattern for the “ual circulation” and expanded the scope of the NIIDZ's innovation spillover effects. Based on data of HSR city pairs from 2008 to 2019 in China, this paper examines the impacts and mechanisms of the improvement in innovation levels of ordinary cities after the opening of HSR connected to NIIDZs by applying a staggered DID model. The empirical results are as follows. Firstly, the opening of HSR connected to NIDDZs significantly improves the innovation levels of ordinary cities. Secondly, the innovation spillover effects are more pronounced for cities in the eastern region, cities with a better innovation environment, and large-scale cities. Thirdly, the innovation spillover effects are realized by utilizing innovation endowment, government-guided innovation and demonstration driving effects. This paper provides empirical evidence and policy insights for innovation-driven development in the context of HSR network. It optimizes the spatial allocation of innovation resources and accelerates the development of new quality productive forces, achieving high-quality economic development.

Multi-source data are commonly encountered nowadays. The analysis of multi-source data is important for unleashing the data potential and realizing data value. However, many multi-source data still exist in the form of “ata silos”. Interconnection between data remains extremely challenging. Meanwhile, the data security issue is a significant concern, making it crucial to achieve secure development of multi-source data while protecting data privacy. To address these challenges, we propose a privacy-protected paradigm for multi-source data analysis. This method is based on the federated learning framework, enabling different data sources to collaborate on data analysis tasks without exposing their raw data. Meanwhile, to further prevent malicious attacks on data, we incorporate differential privacy into federated learning by adding noise to the transmitted data to protect individual-level information. Finally, we demonstrates the practical application of the proposed paradigm using the example of predicting violation risks of enterprises. By combining data from various departments, the prediction accuracy can be well enhanced.

Pre-disaster planning and post-disaster repair are critical strategies for enhancing the resilience of urban road transportation systems. This study comprehensively considers the multidimensional uncertainties in damaged road locations, damage types, and repair durations, as well as the heterogeneity of repair tasks. A two-stage stochastic programming model is developed for the emergency facility location and repair scheduling of urban road transportation systems, aiming to maximize the combined value of emergency facility coverage and overall system resilience. Based on the structural characteristics of the proposed model, a concentration set-based Sample Average Approximation (SAA) algorithm is designed. Experimental results demonstrate that the proposed resilience-optimal recovery strategy outperforms traditional approaches such as random recovery, betweenness-priority recovery, flow-priority recovery, and length-priority recovery. Moreover, the concentration set-based SAA algorithm enables efficient solution of the problem. The findings of this study provide decision-making support and algorithmic guidance for the development of facility location and emergency repair scheduling plans aimed at enhancing the resilience of urban road transportation systems.

In this paper, an unknown system dynamic estimator based sliding mode control scheme is proposed for anti-sway control of overhead cranes with unmatched disturbances and unmodeled dynamics. Through introducing a first-order low-pass filter, the unknown system dynamic estimator is designed to compensate for the unknown system dynamics including unmatched disturbances, such that the disturbance rejection ability can be enhanced. Then, a two-phase power reaching law based sliding mode controller is constructed to obtain the relatively accurate convergence time of the sliding mode variable and guarantee the fast convergence speed. The experimental results show that the proposed method can effectively achieve the satisfactory anti-sway performance and positioning accuracy of the overhead crane.

Determining the effective and efficient lower bounds of the discrepancy criterion in uniform designs is a critical aspect of experimental designs. In this work, we investigate the issue of lower bounds of the newly proposed absolute discrepancy. The lower bound of absolute discrepancy on symmetric multi-level designs is presented and some new sharp lower bounds of this criterion on two- and three-level designs are also displayed. These lower bounds can serve as evaluation metrics for design uniformity and act as benchmarks in the construction of uniform designs. The construction of uniform designs based on absolute discrepancy is also discussed.

This paper uses the OP method and event study method to study the impact of blockchain technology application on enterprise total factor productivity and stock price, and analyzes whether the application of enterprise blockchain technology has effectively promoted the improvement of enterprise total factor productivity, or just created more “foam” for the company’s stock price? The main conclusion of this paper is that the application of blockchain technology mainly promotes the improvement of total factor productivity by reducing financing constraints, and has a greater impact on the improvement of total factor productivity for large enterprises and state-owned enterprises; In addition, after the application of blockchain technology in enterprise announcements, the company’s stock price level has significantly increased, meaning that the company can obtain higher stock premiums from blockchain technology based announcements.

In practical research, the data may come from different distributions, and there are some limitations when using a single distribution model to fit the data. In order to overcome this problem, the mixed model can better adapt to complex data types. Exponential distribution and Rayleigh distribution are important life distributions in reliability analysis, and there are few related mixed models in the context of censored data. In this paper, a mixed model of two-parameter exponential distribution and two-parameter Rayleigh distribution is proposed, and the EM algorithm is used to estimate the parameters of the mixed model with right censored data. Finally, the model is applied to the actual data, and the goodness of fit test is carried out to verify that the proposed model is suitable, which further shows that the model can better adapt to the complex data characteristics and has certain practical significance.

To interpolate the specified Frenet frame, curvature and torsion, a method is proposed for the construction and shape optimization of spatial quintic F³ interpolating curves. F³ continuity of spatial curves is a special k-th order Frenet frame continuity and ensures the satisfaction of G² continuity and torsion interpolation. Firstly, a quintic Bézier curve interpolating the given F³ data is constructed, whose control points are expressed with two parameters denoting the lengths of the curve’s end tangents. Then, the optimal parameter values are determined by minimizing a quadratic energy function. Finally, by defining the objective function as the integral of a weighted sum of squared curvature and torsion, another better optimization method is proposed. Compared to the previous G² interpolation scheme, the new methods can generate curve shapes with more satisfactory curvature and torsion profiles, although using a stricter continuity constraint.

Firstly, the port-Hamiltonian differential-algebraic systems are transformed into the port-Hamiltonian ordinary differential systems with parameter $\varepsilon$. Then, based on the parameteric ordinary differential systems, two structure-preserving model reduction methods are proposed. The first method is parametric moments matching: Constructing the parametric moments based on the frequency parameter $s$ and the embedding parameter $\varepsilon$ of the parametric systems, and then obtaining the reduced-order models of the parametric systems through parametric moments matching. The reduced-order systems match the parametric moments of the original systems. Finally, by taking the embedded parameter $\varepsilon = 0$, the structure preserving reduced-order models of the original port-Hamiltonian differential-algebraic systems are obtained. The second method is low-rank balanced truncation: Using Laguerre functions to construct the low-rank decomposition factors of the controllability and observability Gramians of the parametric ordinary differential systems. The approximate balanced systems are obtained through projection, and finally, the reduced-order models are constructed by truncating the states corresponding to smaller Hankel singular values. This procedure offers adaptability and enables the construction of reduced-order models meeting specified accuracy conditions while maintaining lower computational complexity. Both algorithms use Gram-Schmidt process to construct new projection matrices, thereby preserving the differential structure of the original system. Finally, the effectiveness of the algorithms is demonstrated through a numerical example.

The frequent occurrence of societal events in today's society has a profound impact on people's daily lives and societal development. Prediction of future events helps analysts understand social dynamics, make rapid and accurate decisions as well. This paper proposes a temporal graph model and transforms the target entity prediction into a reasoning task in the temporal event graph. The model first constructs a temporal event graph based on the historical events. In order to explore the influence between different types of events, a dual attention mechanism combining nodes and edges is designed for information aggregation. After encoding the time information through gated recurrent unit, the embedding vectors are input to the fully connected layer to predict the target entity. In addition, given the repeated occurrence of societal events along the historical timeline, the model adopts a copy mechanism to modify the prediction function. Experimental results on multiple datasets demonstrate that the model outperforms other baseline models.

China has introduced a large-scale equipment upgrade policy that can renovate livestock manure collection and processing facilities. However, the impact of this policy on manure management has not yet been explored in existing research. Additionally, there is a gap in the analysis of refined market strategies regarding the collaboration between third-party companies (TPCs) and small to medium-sized livestock farmers (SMS-LFs). To address these issues, this paper constructs an evolutionary game-theoretic model that examines the equipment upgrade strategy of SMS-LFs and the classified pricing strategy of TPCs. The study incorporates prospect theory and mental accounting theory (PT-MA) to explore how farmers decide whether to invest in equipment upgrades, considering their risk preferences. By combining the expected utility function with the value perception function and adhering to the principle of those who invest receive the subsidies, the paper analyzes which party would benefit more from implementing the upgrades in the context of effective policy execution. The study conducts simulation analyses of strategies and summarizes the systemic archetypes for upgrading manure collection and processing facilities. The findings are as follows: 1) Providing large-scale equipment upgrade subsidies to TPCs, allowing them to enhance the manure collection and processing facilities for SMS-LFs, is the most effective strategy for advancing the policy. 2) TPCs should actively implement a classified pricing strategy. 3) The large-scale renewal and upgrading of livestock manure collection and treatment systems exemplify a limits to growth archetype. The solution is removing constraints from balancing loops through a policy mechanism allowing TPCs to obtain equipment renewal subsidies. This subsidy mechanism encourages TPCs to invest in upgrading manure collection and treatment facilities for SMS-LFs. Subsequently, these companies can implement a classified charging strategy to secure higher-quality manure-based raw materials. This creates an incentive mechanism that motivates SMS-LFs to increase their investments in manure treatment. Ultimately, this virtuous cycle enhances the proportion of subsidies received by SMS-LFs through improved environmental performance.

Considering the high volatility and complexity of electricity spot price time series, a combined forecasting model based on wavelet transform and LGBM (light gradient boosting machine, LGBM) is proposed. By introducing rolling time window and wavelet transform, the dynamic multi-scale decomposition of electricity spot price series can be realized, and the frequency characteristics can be extracted to reduce its modal complexity and effectively avoid data leakage. In this study, the proposed model is constructed by utilizing the complex nonlinear feature extraction ability of the LGBM algorithm. The spot market data of Shanxi electric power is used to verify the validity of the proposed model. The results show that the proposed model is superior to the mainstream forecasting methods such as long-term and short-term memory model, support vector machine, elastic network regression model and extreme gradient lifting model in many key performance indexes, such as root mean square error, average absolute error and determination coefficient, among which the $ R^2 $ reaches 0.9792, showing high forecasting accuracy. At the same time, the proposed model shows robustness and adaptability under different market conditions, which shows the proposed model can be seen as a reliable forecasting tool for power market participants and helps to optimize trading strategies and reduce market risks.

Privatization reform is regarded as an effective strategy to reduce waiting times in the public healthcare system. This paper focuses on two modes of privatization reform: One is the competition mode, which allows private hospitals to enter the market and compete with public hospitals; the other is the collaboration mode, where public hospitals and private hospitals cooperate to achieve common goals. This paper employs a queueing model to describe the patient consultation process, analyzes the service rates and prices of public and private hospitals under different privatization reforms, and studies the impact of these reforms on the number of patients covered by medical services, patient waiting times, patient welfare and social welfare. The study finds that the competitive mode can significantly reduce patient waiting time, thereby expanding the number of patients covered by medical services and enhancing patient utility and social welfare. In contrast, while the cooperative mode can also reduce patient waiting time, it exhibits uncertainty in increasing the number of patients, patient utility, and social welfare, and can effectively promote the expansion of the number of patients covered by medical services, patient utility, and social welfare only when the service capacity of public-private partnership hospitals is relatively large or the degree of privatization is high. Finally, when private hospitals choose between the cooperative or competitive mode, it mainly depends on the subsidy rate provided by the government to public hospitals and the level of privatization pursued by public-private partnership hospitals for their own interests. Specifically, when the subsidy rate or the level of privatization is high, private hospitals are more inclined to choose the cooperative mode; conversely, they are more inclined to choose the competitive mode.

Semi-supervised graph-based dimensionality reduction is a kind of meth-od that utilizes data structure graph to deal with semi-supervised dimensionality reduction problem. However, most of these algorithms only take account of data information while ignore class label information; And they don't take account of the differences among samples in the training process, which reduces the robustness of the algorithms in the case of noise or outliers. In this paper, by combining sparse representation with self-paced learning, a self-paced learner is proposed to obtain the linear dimensionality reduction mapping based on sparse discriminant graph. In detail, the proposed method firstly constructs a sparse discriminant graph by integrating the propagation of class labels with sparse representation of data. Then, by considering the distance between each low-dimensional data point and the corresponding class anchor, and the ability of low-dimensional data to maintain the discriminative sparse structure of the original high-dimensional data, this paper proposes a self-paced learning problem for dimensionality reduction. On the one hand, the proposed method constructs a sparse discriminant graph that can extract the discriminative information of data more effectively; On the other hand, the proposed method is based on self-paced learning mechanism, which makes it can automatically calculate the importance values of training data, suppress the negative impact of unreliable data or labels, and improve the robustness of the model to noise or outliers. The results of five experimental data sets demonstrate the effectiveness of the proposed algorithm.

The feedback control method developed by Pyragas (1992) is mainly utilized to add the time-delayed feedback into the middle equation of the three-dimensional continuous chaotic system. But in fact it is also convenient and effective for chaos control if the feedback is added into the last equation. In this paper, Shimizu-Morioka chaotic system is taken as an illustration example and a single time-delayed feedback term is added into its last equation. The local stability and the occurrence of Hopf bifurcation are studied by taking the delay as the bifurcating parameter. The direction and stability of the bifurcating periodic solutions are further investigated by means of the center manifold theorem and the norm form theory. Finally, numerical experiments are presented to illustrate the correctness of the theoretical results as well as the effectiveness of chaos control.

Learning from label proportions (LLP) is a weakly labeled learning problem, where the instance-level label information is abstracted in the form of bags, that is, only the label proportion information of each bag is available. Consequently, LLP can be grouped into learning with bags community, where bags consisted of instances are related. Similar to typical classification, our aim is not only to learn a classifier to greatly recover the instance-level labels in training data, but also to generalize this label prediction to unseen data. However, due to the ambiguous or approximate property in statistic estimation and the existence of label noises, a more realistic situation for this learning framework is prone to conceive an interval-type proportion information, instead of real-valued proportions in LLP. Thus, for these universal scenarios, the standard LLP methods are failed to offer a satisfied label predictor. In this paper, we propose a new learning framework called Bounded Label Proportions (BLP) to tackle this puzzled problem. In addition, we perform a robust algorithm for BLP based on random forest (RF): BLPForest, which is naturally able to deal with multi-class and high dimensional problems. For the purpose of comparison, we divided our experiments into two parts. In the first part, we degenerated BLPForest into standard LLP problem, in order to verify the evolution between these two similar learning problems. Consequently, the results demonstrated BLPForest with a natural advantage even in the case of real-valued proportion information equipped, which mainly benefited from the application of RF algorithm. For the second part, we chose large datasets with multi-class and much higher dimensions. In a meantime, appropriate noise for proportion information in each bag was deliberately added. All experiments showed that BLPForest can yield the best accuracies in the most cases. The final conclusion is that the method proposed in this paper demonstrates the best performance when dealing with multi-class and high-dimensional problems.

Relative errors are widely employed in sales forecasting and other domains to assess prediction accuracy due to its scale invariance and ease of comprehension. In this paper, with a homoscedastic Guassianity assumption, a novel model averaging approach based on multiplicative models and mean squared relative risk is proposed to improve the forecast accuracy measured by mean squared relative error. A theoretical guarantee that the expectation of the weight criterion is asymptotically equivalent to the relative risk is provided. Under some mild conditions, when all candidate models are misspecified the model averaging method is asymptotically optimal, and when correct models exist the sum of the weights placed on these correct models converges to 1. Furthermore, numerical experiments and an empirical analysis on sales forecasting show the superiority and practicability of the proposed method.

In this paper, the linear water wave equation is used to describe the fluctuation of an ideal water body in a two-dimensional bounded rectangular region, and the Craig-Sulem transform is used to transform the water wave equation into a linear development equation with velocity potential function and wave height as state variables. In this paper, we assume that the wave height of water wave is the measured output of the system, and on this basis, we analyze the recognizability of the water depth and velocity potential function, and design a synchronous identification algorithm to estimate the water depth and potential function from the wave level. In this paper, a numerical identification algorithm based on adjoint method is designed, which can effectively estimate both water depth and potential function. Firstly, the traditional quadratic target functional is improved to target functional with system model constraints by introducing Lagrange multipliers. Secondly, the adjoint equation of the water wave equation is derived by the Lagrange functional differential formula, and the gradient of the target functional is obtained by solving the equation. Finally, the gradient descent method is used to estimate the water depth and velocity potential functions iteratively, and the effectiveness and accuracy of the proposed algorithm are verified by numerical simulations.

To enhance supply chain efficiency, it has become a standard production mode for geographically diverse factories to collaborate in completing production tasks. Constructing scheduling models that reflecting real-world conditions and designing simple yet effective optimization algorithms are key to achieving efficient collaboration. Distributed flexible job shop scheduling problem has emerged as a promising tool for modeling and optimizing such problems. However, existing researches rarely account for sequence-dependent setup time for machines, instead simplifying the time as constant. This approximation can lead to inferior schedules, thereby impeding system efficiency. This paper establishes a mixed-integer programming model to describe the distributed flexible job shop scheduling problem with sequence-dependent setup time. A Q-learning based iterated greedy algorithm is proposed to solve it, wherein the Q-learning mechanism is utilized to dynamically select the appropriate perturbation magnitude, effectively overcoming the decline in search performance caused by unreasonable perturbation in conventional iterative greedy algorithms. By introducing the correlation between machines' setup time and operation sequences into benchmark instances for distributed flexible job shop scheduling with machine eligibility constraints, 207 instances are constructed. The proposed algorithm is compared with three iterative greedy algorithms with fixed perturbation magnitudes, simulated annealing, scatter search, backtrack search-based hyper-heuristic and random permutation descent-based hyper-heuristic. Experimental results demonstrate that the proposed Q-learning based iterative greedy algorithm achieves higher search quality and faster convergence speed.

While the franchise-based express delivery industry has developed rapidly, there are also issues of unstable cooperation caused by conflicting interests and low service quality, which makes it difficult to satisfy the increasing demand for high-quality and high-service from consumers. To this end, this paper uses the method of evolutionary game to dynamically analyze the evolutionary stability of the courier company's regulatory strategy, the production behavior of terminal franchisees and the government's regulatory rewards and punishments strategy, and reveals the impact of different decision parameters on evolutionary stability, demonstrating the conditions for evolutionary stability. The numerical analysis results indicate that when the risk cost and profit-sharing ratio are in different threshold intervals, the game system between express delivery companies and franchisees presents four different evolutionary stability results. In addition, when formulating reward and punishment policies, the government should ensure that the sum of rewards and punishments for all parties is greater than their speculative gains, in order to ensure the standardized operation and cooperative stability of express service enterprises. The results of this paper are of great significance to the establishment of a suitable default punishment system, risk identification and early warning mechanisms, and enhancing the government's regulatory functions, while creating a favorable market operating environment.

In recent years, there have been more frequent disasters occurred in China, which pose significant threats to the lives and property of the people. To cope with the increasing complexity and severity of disasters, decision-makers need to store and dispatch emergency supplies rationally based on the real situation. Current studies on regional dispatch considering multiple warehouses and demand points are insufficient, and the problems such as ‘who/how/how much to dispatch’ have not been well-answered. To solve these problems, this paper proposes three regional dispatching strategies (including strict administrative hierarchy supply dispatch, cross-administrative hierarchy supply dispatch and free and nearest supply dispatch strategies) based on a comprehensive summary of relevant case studies, then builds a multi-agent simulation model based on deprivation cost. A simulation experiment is conducted in Mengcheng County, Bozhou City, Anhui Province, and the result shows that when the regional demand is large in a short time, the free proximity strategy can minimize the total social logistics cost. On the contrary, when the regional demand is small, the differences of the total social cost among three strategies are small. In conclusion, our research suggests that, when facing severe disasters and catastrophes, governments should cooperate and coordinate on the dispatching of relief supplies. However, when facing normal disasters without the risk of life, the demand can be satisfied with the strict administrative strategy.

Identifying key segments within road networks is crucial for selecting repair objectives and optimizing repair sequences during the post-disaster recovery phase. Traditional methods for identifying key segments have not fully explored the interactions between multiple segments, particularly the significance of studying road network vulnerability under simultaneous disruptions of multiple links. To tackle this issue, this study introduces a machine learning model called transportation graph attention networks for criticality analysis (TGAT) to identify key road segments when facing multiple disruptions. This model is trained on data samples that include scenarios of multiple segment failures, utilizing the graph attention network to evaluate the influence weights between segments and calculating the criticality of each segment based on these weights. The model, trained using mean squared error as the loss function, is capable of identifying segments that play a crucial role in the performance of the road network. Taking the Kunshan City road network as an example, this paper compares the effectiveness of the TGAT method with three other methods:Degree centrality, weighted betweenness centrality, and eigenvector centrality, in optimizing repair sequences during the post-recovery phase. Experimental results indicate that the TGAT method is more effective in identifying key segments within the road network compared to the other three methods, and the repair sequence optimized using TGAT further enhances the repair performance of the road network.

Currently, one of the main challenges in practical modeling lies in the fact that training and testing data come from different distributions. Stable learning addresses this issue by decorrelating all covariates through sample reweighting, thereby achieving stable predictive performance. While machine learning methods such as stable learning show good results in experiments, there are still theoretical gaps, such as the lack of metrics for model stability under testing data and explanations for why stable learning maintains stable predictions across multiple environments. This paper proposes a new metric of stability, compares stable learning methods with ordinary least squares and explores the reasons why stable learning maintains stability across multiple environments. Finally, the paper validates the theory through simulated experiments. This research contributes to refining the theory of stability in stable learning, enhancing the understanding of stability in stable learning, and guiding the selection of practical modeling methods.

Based on the data of some "illegal" public accounts from 2012 to 2020, this paper studies the impact of "illegal" public account recommendations on Chinese stock market and investors. The paper finds that the stocks recommended by "illegal" public accounts experience significant increases in absolute excess returns and trading volume after the information is released, but the long-term cumulative excess returns are significantly lower than other stocks. Empirical results also show that a significant increase in net inflows of individual investors and a significant decrease in net inflows of institutional investors among the recommended rising stocks. Furthermore, the channel tests find that "illegal" public accounts mainly attract investors' attention by creating false private information, and mainly attract users who rely on mobile devices.

The issue of employee behavior in the process of intelligent manufacturing is not only a problem in the field of cognitive science, but also a management problem in production. This topic focuses on two dimensions, behavioral cognition and lean, which are interrelated and have their own characteristics. The core issue is to explore the factors and mechanisms that affect labor behavior in the continuous promotion of intelligent manufacturing based on the theoretical perspectives of cognitive bias and lean, and analyze their impact on Jidoka systems. The focus of this study is on the imperfect production process of assembled products, where labor may overlook inspection items due to various factors during the production process. Taking into account Type I and Type II errors, an extended EOQ model is used to optimize the order quantity in order to minimize total costs. Proved that positive lean strategies can eliminate overlooked occurrences, reduce total system costs, and measure the degree of reduction. This paper provides some numerical examples, sensitivity analysis, and graphical tables to illustrate the model.

Salary incentives are an important means for enterprises to stimulate employees' work passion, creative potential and improve corporate performance. On the one hand, the internal pay gap has a positive motivating effect and promotes the effort of employees. On the other hand, it can also lead to a sense of unfairness and cause some employees to feel “flat”. This paper constructs a mathematical model including core executives, non-core executives, and ordinary employees to analyze the impact of internal salary gaps on corporate performance, and conducts empirical research using data from privately-owned listed companies in Shanghai and Shenzhen from 2008 to 2020. Both theoretical and empirical results show that the relationship between pay gap within management, executive-employee pay gap, the degree of compensation incentives of non-core executives and corporate performance all show an inverted U shape. Further empirical research shows that non-core executive-employee pay gap has the strongest effect on corporate performance, while core executive-employee pay gap has the smallest effect on corporate performance. This research shows that non-core executive-employee pay gap is the most important compensation relationship within the company and core executive-employee pay gap is of least importance. In addition, in the salary incentive design of private enterprises, the “constraint” of operating profit is greater than the “constraint” of operating income, which reflects that private enterprises pay attention to seizing the key points.

To quantify the impact of the stability of power supply on the development of the digital economy, a system dynamics model is constructed, including variables such as power supply installed capacity and the scale of the digital economy in the three industries. The model explores the mechanism of how the stability of power supply affects the development of the digital economy in scenarios such as power outages and electricity rationing. The results show that the scale of China's digital economy will maintain a high growth trend during the “14th Five-Year Plan” period and is expected to exceed 70 trillion yuan by 2025. The scale of the digital economy will decrease with the reduction of power supply. If the average daily electricity generation time or the number of working days per year decreases by 1%, 5%, and 10%, the digital economy scale will decrease by an average of 9.28%, 14.24%, and 20.43% respectively. By promoting technological innovation to improve the value-added coefficients of various industries, the impact of power outages or generator failures on the development of the digital economy can be reduced. Finally, policy recommendations are proposed to enhance power supply stability in China.