ABSTRACT Nowadays, with the growing penetration of renewable generation, economic dispatch is increasingly important in short‐term power system operation. In this paper, a deep renewable scenario generation model combining Multi‐Scale Decomposition mixer and Wasserstein Generative Adversarial Network with Gradient Penalty is proposed to achieve novel decision‐oriented forecasting, thus realizing effective characterization of renewable temporal dynamics and economic performance. From the perspective of wind and solar generation, the validity of the proposed method is demonstrated on a real‐world dataset with power station at regional level. Experimental results confirm the superiority of model performance through statistical indicators and power system scheduling test, compared with a number of scenario generation and time series forecasting benchmarks.

Rong Cheng, Jinyi Liu, Yan Zheng, Fei Ni, Jiazhen Du, Hangyu Mao, Fuzheng Zhang, Bo Wang, Jianye Hao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Role-based collaboration (RBC) theory is a promising paradigm for problem-solving in complex systems. Multigroup role assignment (MGRA) specifically tackles the task of assigning roles for multigroup collaboration. However, due to the constraint that an agent can only play a role in one group, the current MGRA models are incapable of handling when required agents outnumber the available supply. Group multirole assignment (GMRA) resolves the problem by permitting an agent to be assigned multiple roles, but it cannot address the assignment involving multiple environments-classes, agents, roles, groups, objects (E-CARGO) groups. Therefore, this article presents a comprehensive overview of the GMRA problem in multiple E-CARGO groups under various conditions, generalized as the multigroup multirole assignment (MGMRA) problem. The MGMRA problem primarily revolves around two key factors: the maximum number of roles that an agent can undertake within an E-CARGO group, and the maximum number of different roles across all E-CARGO groups, which have a significant impact on the sufficiency or necessity conditions of the algorithm as well as its performance. Therefore, a unified model and its special cases are proposed to solve the concrete assignment problems under different conditions. The effectiveness of models is verified through comprehensive experiments.

Portfolio selection stands as a paramount concern within the realm of decision-making and management engineering. However, owing to the inherent intricacies of capital markets and the presence of irrational investor behaviors, the attainment of predefined investment objectives by investors remains a formidable challenge. In order to comprehensively depict investor behavior patterns and to provide investment guidance in highly uncertain and volatile markets, this study introduces a novel fuzzy model for representing prospect theory and based on this, develops a novel portfolio selection optimization framework. In addition, a new particle swarm optimization consists of adaptive and cooperative strategy is proposed to find the optimal solution of this model. The effectiveness of this model is validated through two case study utilizing real-market data, while the efficiency of the solution algorithm is confirmed through a test fitness functions-based case study.

Role-based collaboration (RBC) is an emerging and advanced methodology for problem-solving. A critical aspect of RBC theory is agent evaluation, which aims to assess agents’ abilities through a qualification value derived from a comprehensive analysis of their characteristics. This evaluation directly impacts the quality of role assignments. Existing research typically assumes that the qualification value is either predetermined, based on multiscale criteria, or following a predefined distribution. These assumptions, however, are overly idealistic and difficult to generalize, failing to capture the inherent volatility of the qualification value. To address this challenge, this article introduces a Wasserstein-based ambiguity set to model potential fluctuations in the qualification value, drawing on empirical distributions derived from historical sample data. Building upon the RBC framework and its abstract model environments, classes, agents, roles, groups, and objects (E-CARGO), we propose two data-driven models: distributionally robust group role assignment (DRGRA) and group multirole assignment (DRGMRA). These models aim to achieve more robust and optimal role assignments under uncertainty in agent evaluation. Leveraging strong duality, we reformulate DRGRA and DRGMRA as tractable finite mixed 0–1 convex problems, providing an approximation framework that reduces computational complexity. Notably, these models are adaptable to other problems with no uncertainty in agent evaluation, highlighting their modeling scalability. Experimental results demonstrate the effectiveness and robustness of the proposed models.

Role-based collaboration (RBC) is a new problem-solving paradigm that uses model environments-classes, agents, roles, groups, and objects (E-CARGO) to facilitate modeling. Task decomposition is widely adopted to reduce the difficulty of execution, resulting in multigroup collaboration problems. Multigroup role assignment (MGRA) has been proposed to solve the assignment of multiple groups. In many actual scenarios, there are dependencies between the decomposed subtasks, which is neglected by existing MGRA methods. Moreover, they disregard the fact that subtasks have diverse significance in the development of a project, which is of paramount importance to ensure the proper allocation of resources. To solve the complicated problem, the structured subtasks are formalized based on the emerging and promising RBC theory and E-CARGO model. Then, the assignment is abstracted into a complicated single-objective multiconstraint problem, named adaptive-equalized MGRA (AE-MGRA). In the formulated AE-MGRA problem, to improve the utilization of limited resources, the performance of each E-CARGO group needs to be equalized according to the corresponding subtask’s weight. As the optimal solution is difficult and time consuming to obtain, a tolerable deviation is utilized to achieve a near-optimal solution. Extensive experiments are conducted to sufficiently demonstrate the efficiency and stability of the proposed practical solution. In addition, the experimental results on static assignment and dynamic assignment further prove the effectiveness of the solution.

Contrastive learning (CL) continuously achieves significant breakthroughs across multiple domains. However, the most common InfoNCE-based methods suffer from some dilemmas, such as uniformity-tolerance dilemma (UTD) and gradient reduction, both of which are related to a $\mathcal{P}_{ij}$ term. It has been identified that UTD can lead to unexpected performance degradation. We argue that the fixity of temperature is to blame for UTD. To tackle this challenge, we enrich the CL loss family by presenting a Model-Aware Contrastive Learning (MACL) strategy, whose temperature is adaptive to the magnitude of alignment that reflects the basic confidence of the instance discrimination task, then enables CL loss to adjust the penalty strength for hard negatives adaptively. Regarding another dilemma, the gradient reduction issue, we derive the limits of an involved gradient scaling factor, which allows us to explain from a unified perspective why some recent approaches are effective with fewer negative samples, and summarily present a gradient reweighting to escape this dilemma. Extensive remarkable empirical results in vision, sentence, and graph modality validate our approach’s general improvement for representation learning and downstream tasks.

Graph based methods have been widely used in incomplete multi-view clustering (IMVC). Most recent methods try to fill the original missing samples or incomplete affinity matrices to obtain a complete similarity graph for the subsequent spectral clustering. However, recovering the original high-dimensional data or complete n X n similarity matrix is usually time-consuming and noise-sensitive. Besides, they generally separate the cluster indicator learning into an individual step, which may result in sub-optimal graphs or spectral embeddings for clustering. To address these problems, this paper proposes a robust Spectral Embedding Completion based IMVC (SEC-IMVC) method, which incorporates spectral embedding completion and discrete cluster indicator learning into a unified framework. SEC-IMVC performs completion on spectral embeddings, and the embedding noise is eliminated to reduce the negative influence of original data noise. The discrete cluster indicator matrix is seamlessly learned by using spectral rotation, and it can explore the first-order feature consistency among different views. To further improve the completion robustness, the second-order correlation consistency is also captured by pairwise relations alignment. We compare our method with some state-of-the-art approaches on several datasets, and the experimental results show the effectiveness and advantages of our method.

Considering nonstatistical uncertainties and/or insufficient historical data in security return forecasts, fuzzy set theory has been applied in the past decades to build portfolio selection models. Meanwhile, various risk measurements such as variance, entropy, and Value-at-Risk have been proposed in fuzzy environments to evaluate investment risks from different perspectives. Sharpe ratio, also known as the reward-to-variability ratio, which measures the risk premium per unit of the nonsystematic risk (asset deviation), has received great attention in modern portfolio theory. In this study, the Sharpe ratio in fuzzy environments is introduced, whereafter, a fuzzy Value-at-Risk ratio is proposed. Compared with Sharpe ratio, Value-at-Risk ratio is an index with dimensional knowledge that reflects the risk premium per unit of the systematic risk (the greatest loss under a given confidence level). On the basis of the two ratios, a multi-objective model is built to evaluate their joint impact on portfolio selection. Then, the proposed model is solved by a fuzzy simulation based multi-objective particle swarm optimization algorithm, where the global best of each iteration is determined by an improved dominance times based method. Finally, the algorithm superiority is justified via comparing with existing solvers on benchmark problems, and the model effectiveness is exemplified by using three case studies on portfolio selection.

In this study, we consider a portfolio-optimization-incorporated budget investment problem under managers' risk tolerance ambiguity. In order to capture the decision dynamics driven by the risk tolerance ambiguity, a two-stage adaptive optimization model is developed. The budget allocation is the first-stage decision, which is made before knowing each manager's actual risk tolerance level, and the portfolio selection conducted by each manager is the second-stage decision, which adapts to the manager's risk tolerance. We introduce the concept of risk-neutral budget threshold (RNBT) that is modeled by a fuzzy set granule, and upon which the ambiguous risk tolerance curve is constructed, which can realistically capture the managers' risk-averse and/or risk-seeking attitudes. Due to the (realistic) nonconvex/nonconcave structure of the risk tolerance curve, and the existence of the ambiguity, the resulting problem is essentially a nonconvex adaptive optimization problem under uncertainty. To achieve a robust modeling and an efficient solution, we first restructure and robustize the information of fuzzy RNBTs and then transform the developed model into a mixed integer linear programming (MILP), which can be handled efficiently by off-the-shelf mixed integer program solvers. Leveraging the derived MILP structure, we can use the Benders decomposition to further enhance the scalability of the model. Furthermore, some model extensions on robustizing the probability estimations are discussed. Finally, computational studies are performed to demonstrate the effectiveness and insights of the model.

Based on portfolio selection theory, this study pro poses an improved fuzzy multi-objective model that can evaluate the invest risk exactly and increase the probability of obtaining the expected return. In building the model, fuzzy Value-at-Risk (VaR) is used to evaluate the exact future risk, in term of loss. The VaR can directly reflect the greatest loss of a selection case under a given confidence level. On the other hand, variance is utilized to make the selection more stable. This model can provide investors with more significant information in decision-making. To better solve this model, an improved particle swarm optimization algorithm is designed to mitigate the conventional local convergence problem. Finally, the proposed model and algorithm are exemplified by some numerical examples. Experiment results show that the model and algorithm are effective in solving the multi-objective portfolio selection problem.

The conventional prediction of future power demands are always made based on the historical data. However, the real power demands are affected by many other factors as weather, temperature and unexpected emergencies. The use of historical information alone cannot well predict real future demands. In this study, the experts' opinions from related fields are taken into consideration. To deal the uncertainty of historical data and imprecise experts' opinions, we employ fuzzy variables to better characterize the forecasted future power loads. The conventional unit commitment problem (UCP) is updated here by considering the spinning reserve costs in a fuzzy environment. As the solution, we proposed a heuristic algorithm called local convergence averse binary particle swarm optimization (LCA PSO) to solve the UCP. The proposed model and algorithm are used to analyze several test systems. The comparisons between the proposed algorithm and the conventional approaches show that the LCA-PSO performs better in finding the optimal solutions.

Based on fuzzy value-at-risk (VaR), this paper proposes a new portfolio-selection model (PSM) called the VaR-based fuzzy PSM (VaR-FPSM). Compared with the existing FPSMs, the VaR can directly reflect the greatest loss of a selected case under a given confidence level. In this study, when the security returns are taken as trapezoidal, triangular, and Gaussian fuzzy numbers, several crisp equivalent models of the VaR-FPSM are derived, which can be handled by any linear programming solvers. In general situations, an improved particle swarm optimization algorithm on the basis of fuzzy simulation is designed to search for the approximate optimal solutions. To illustrate the proposed model and the behavior of the improved particle swarm optimization algorithm, two numerical examples are provided, and the results are discussed. Furthermore, the proposed algorithm is compared with some existing approaches to fuzzy portfolio selection, such as the genetic algorithm and simulated annealing.

In this paper, using value-at-risk, a new fuzzy portfolio selection model named VaR-FPSM is proposed. The value-at-risk is the measure of risk, which describes the greatest loss of an investment with some confidence level. When security returns are same kind of fuzzy variable, we derive two crisp equivalent forms of the VaR-FPSM. Furthermore, in general situations, we designed a fuzzy simulation based particle swarm optimization (PSO) algorithm to find an approximately optimal result. To illustrate the proposed model and hybrid PSO algorithm, a numerical example is provided and some discussions on the results are given.