Single-step retrosynthesis prediction aims to identify reactants for synthesizing a target molecule and is crucial for pathway planning. Despite improved prediction accuracy, existing methods often struggle with performance disparities between high- and low-resource reaction classes. This limitation hampers the overall effectiveness of the retrosynthesis prediction methods. Here, we introduce two novel strategies─Retrosynthetic Mutual Distillation (Retro-MD) and Retrosynthetic Self-Distillation (Retro-SD)─leveraging distillation learning techniques to bridge this performance gap. Retro-MD uses dual sampling temperatures and cross-model knowledge transfer, while Retro-SD leverages a fixed temperature and self-distillation from prior iterations. Evaluations on Transformer-based models show a state-of-the-art performance among template-free approaches. Ablation studies further validate the rationale of reaction-class-aware task partitioning, demonstrating the robust effectiveness of our proposed distillation strategies.

The green production mode centered on energy-saving has become increasingly important in the manufacturing industry. As a classic problem in green production, the energy-saving hybrid flow shop scheduling problem (EHFSP) has attracted extensive attention from scholars. Considering the variable processing speed and worker constraint in EHFSP, this paper studies an energy-saving hybrid flow shop scheduling problem with variable speed and worker constraint (EHFSPVS-WC), and establishes a mixed-integer linear programming model with the optimization objectives of minimizing the maximum completion time (makespan) and total energy consumption (TEC). Firstly, a heuristic method named NEH with variable speed characteristics (NEHVS) is proposed, which adjusts the processing speed of jobs whose processing time is shorter than the average at that stage. Secondly, a genetic algorithm based on adaptive crossover selection (GACS) is designed. In GACS, by analyzing the factors that may increase the objective function of EHFSPVS-WC, four crossover operators based on machine idle time, machine utilization, machine processing time and speed level are designed, and a framework based on Q-learning for adaptive selection of crossover operators is developed. In addition, a neighborhood search based on worker assignment (NSWA) is designed, which reduces the idle time of processing machines by adjusting worker assignment at a specific stage to further improve the quality of solution. Finally, through comparative analysis on multiple instance experiments, the results demonstrate that the GACS outperforms the other comparison algorithms.

A practical hybrid flow shop scheduling problem under uncertain processing stages, which derives from the steelmaking continuous casting (SCC) process, is investigated in this paper. Firstly, a mixed-integer mathematical model of the SCC scheduling problem with uncertain processing stages (SCCSPUPS) is established, and the feature of which is analyzed. Secondly, a novel heuristic approach based on the feature of uncertain processing stages is designed, and the better solution can be obtained. Thirdly, a two-dimensional discrete artificial bee colony algorithm based on multiple neighborhood swaps (DABCMNS) is proposed to solve this SCCSPUPS, where a two-dimensional encoding scheme is developed for the solution representation, a four-dimensional tournament selection strategy is explored to select, and the multiple neighborhood swaps operator is designed to conduct the neighborhood searches for the better individuals. Finally, a comparison of DABCMNS algorithm with other six notable metaheuristic algorithms is carried out, and the findings indicate that the DABCMNS algorithm exhibits superior performance in comparison to other metaheuristic methods.

Driven by global competition and energy efficiency demands, intelligent scheduling is critical for sustainable distributed manufacturing. This paper addresses the distributed group scheduling problem with preventive maintenance (DFGSP_PM) by establishing a threshold-based maintenance-triggered mathematical model. To effectively solve this problem, a novel Deep Q-network-based Cooperative Multi-objective Optimization Evolutionary algorithm (DQN-CMOEA) is proposed. The innovations lie in a DQN-driven adaptive strategy selection mechanism, a multi-population co-evolution framework for enhanced exploration, a maintenance-aware multi-phase energy-saving strategy for reducing idle-time energy waste, and a composite convergence-diversity indicator for promoting a well-distributed and high-quality Pareto front. Extensive experiments on 405 benchmark instances show that DQN-CMOEA significantly outperforms four state-of-the-art algorithms across multiple metrics, demonstrating its effectiveness and robustness in solving complex distributed scheduling problems.

Energy-efficient utilization and emissions reduction have become increasingly important for modern industrial manufacturing. At the same time, many industries have adopted the multi-factory production mode due to rapid economic development. Therefore, this paper investigates the energy-efficient scheduling of a distributed permutation flow shop with sequence-dependent setup times (EEDPFSP/SDST) aiming to simultaneously minimize the total flowtime (TF) and total energy consumption (TEC). Firstly, the initial sequencing is determined according to the job's completion time, and a multi-objective NEH_B heuristic with an inner-block swap operator (NEH_BIBSO) is proposed. Secondly, an iterative greedy algorithm based on neighborhood search (IGBNS) is proposed according to the problem characteristics. Two operators (inner-block swap, block-to-block swap) are designed to reduce the TF, the energy-saving operator is developed to diminish the TEC in the neighborhood search, and a self-adaptive local search strategy with Pareto front diagram judgment is introduced in IGBNS. Finally, through experimental comparisons based on 612 instances, the IGBNS algorithm is verified to outperform the other five famous multi-objective optimization algorithms.

The hybrid flowshop scheduling problem (HFSP) and its variants are common problems in industrial production. Evolutionary algorithms based on fixed rules and heuristic strategies have proven effective in solving such problems. However, these algorithms often strictly follow predefined scheduling rules, with limited consideration of the real-time scheduling process per and the solution state, thereby leading to insufficient scheduling flexibility. Therefore, this paper employs a Deep Q-Network (DQN) to adjust scheduling strategies in real-time, and accordingly proposes a Dual-Agent Neighborhood Search Algorithm (DANSA). Specifically, a two-dimensional vector encoding solution is adopted to facilitate scheduling strategy adjustments across all stages, and two agents are designed based on the DQN. The first agent determines the optimal scheduling decision at each decision point by leveraging features such as job arrival time, waiting time, and processing time. The second agent selects the optimal neighborhood search strategy based on the workshop state under the current solution, dynamically adjusting the search direction to accelerate the algorithm’s convergence. The performance of DANAS was evaluated on instances of varying scales, and the results demonstrate that DANSA exhibits excellent efficiency and robustness in solving HFSP, and significantly outperforms mixed-integer linear programming and five other state-of-the-art algorithms.

Amid increasing demands for sustainable and flexible production, this paper addresses the multi-objective distributed blocking flow shop group scheduling problem, formally denoted as DFm|group, sdst,blocking|(Cmax, TEC). To address it, this paper proposes a Q-learning-assisted collaborative multi-objective evolutionary algorithm QACMOEA. Specifically, a sequence-based mathematical model is formulated and validated via Gurobi. To boost search quality, a multi-feature fusion strategy leverages group processing time, average setup time, and skewness. A Q-learning-based mechanism dynamically selects local search strategies by maximizing cumulative rewards across Pareto-derived states. A hybrid convergence-diversity selection combined with a metrical-driven reference update maintains elite solutions and diversity, preventing premature convergence and adapting to varied scenarios. Additionally, a dynamic machine speed adjustment with hierarchical evaluation expands the Pareto front, balancing economic and ecological goals. Extensive computational experiments and comparative analyses on 405 test instances demonstrate that the proposed QACMOEA outperforms existing state-of-the-art algorithms in terms of solution quality, convergence stability, and scalability for this complex multi-objective scheduling problem.

In the context of rapidly evolving intelligent manufacturing systems, efficient coordination between production and intralogistics is crucial for productivity and flexibility. The flexible job shop scheduling problem with automated guided vehicles (FJSP-AGVs) poses a significant challenge due to tight coupling among operations, machines, and transportation resources, limiting the effectiveness of traditional heuristics and learning-based methods. To address this, we propose a multi-policy proximal policy optimization (MPPPO) unified framework featuring a heterogeneous graph attention-based encoder. By representing operations, machines, and AGVs as nodes in a heterogeneous graph with multiple relation types, the encoder captures high-dimensional interactions and global system dependencies. In addition, a composite reward mechanism that jointly considers makespan minimization, machine load balancing, and AGV utilization is designed to enhance solution stability and operational efficiency. Extensive experiments on generated instances and public benchmark datasets demonstrate that the proposed method consistently outperforms representative heuristic rules and state-of-the-art deep reinforcement learning approaches. The results highlight its potential for real-world deployment in large-scale intelligent manufacturing systems.

Distributed Flexible Job Shop Scheduling with Transportation Constraints (DFJSP-T) is widely used in manufacturing workshops across various industries, as it more accurately represents real-world production scenarios. However, many existing studies do not fully leverage critical problem features or domain knowledge, limiting the ability of current algorithms to find high-quality solutions for complex large-scale problems. To address these challenges, this paper proposes a Self-Adaptive Differential Evolution Enhanced Quality-Diversity Optimization (SADE-QD). The SADE-QD incorporates domain knowledge in three main ways: (1) it models the problem using machine idle and job transportation features, allowing the algorithm to retain solutions with diverse behaviors; (2) it introduces a knowledge-guided heuristic search strategy based on critical path to discover more high-quality solutions; and (3) it applies an self-adaptive differential evolution search method that leverages machine idle and transportation features to explore a broader range of solutions. This helps valuable information from the feature space contribute directly to the search process. Experiments on small, medium, and large scale benchmarks show that SADE-QD achieves an average makespan reduction of 12% compared to several recent state-of-the-art algorithms.

This paper investigates the Hybrid Flow Shop Scheduling Problem with Batch Processing Machines and Variable Sublots (HFSP-BVS), considering sequence-dependent setup times and transportation times, with the objective of minimizing total tardiness. The complexity of HFSPBVS lies in the tight coupling among lot sequencing, lot splitting, and machine assignment, making it highly challenging in modern manufacturing environments. To address this problem, a Mixed-Integer Linear Programming (MILP) model is formulated and validated using the Gurobi solver. Subsequently, a hybrid algorithm, MADDQN_IG, is proposed by integrating the MultiAgent Double Deep Q-Network (MADDQN) with Iterated Greedy (IG). The algorithm incorporates four key components: (1) a triple two-layer initialization strategy; (2) a dual-layer destruction-reconstruction parameter selection agent; (3) a local search strategy selection agent; and (4) a multi-agent DDQN construction and training process. These elements are embedded within a unified framework to enhance search efficiency and optimization depth. Extensive computational experiments on 100 benchmark instances demonstrate that MADDQN_IG consistently outperforms existing advanced algorithms (NCIG, QABC, vCCEA, GA), achieving superior solution quality and robustness within limited computation time. Specifically, under three termination criteria (delta = 100, 200, 300), MADDQN_IG improves the ARDI by 78.57%-98.57% and ranks first in the Friedman test, confirming the effectiveness and adaptability of the proposed framework.

This study develops an optimized hybrid renewable energy system (HRES) for Islamabad, Pakistan, that simultaneously addresses the integrated electrical demand of conventional loads, electric vehicle (EV) charging, and hydrogen refueling requirements for fuel cell electric vehicles (FCEVs). A novel multi-modal, multi-objective quantum-inspired particle swarm optimization (MMOQPSO) algorithm is employed to solve the tri-objective optimization problem, balancing the levelized cost of energy (LCOE) as the economic objective, carbon emissions as the environmental sustainability indicator, and loss of power supply probability (LPSP) as a measure of system reliability. The system architecture integrates photovoltaic arrays, hydrogen storage tanks, a biogas generator, and fuel cells to establish a resilient multi-energy supply framework. Through annual simulations using real-world meteorological and load profile data, the optimized configuration achieves a minimum LCOE of $0.2146/kWh, a 67% reduction in annual carbon dioxide (COQ) emissions, and a sustainable decrease in LPSP to below 3%, ensuring high energy reliability. The hydrogen subsystem produces up to 6000 kg/year, with storage levels peaking at 120.02 kg, enabling consistent fuel cell dispatch. Additionally, the system attains a hydrogen environmental advantage to levelized cost of energy (HEA/LCOE) ratio above $6 per kg of COQ avoided, underlining its strong combined environmental and economic performance. Furthermore, the MMOQPSO algorithm exhibits faster convergence and enhanced solution diversity, enabling effective exploration of trade-offs. These findings underscore the system's robustness against demand fluctuations and renewable intermittency, offering a practical pathway toward decarbonized, integrated energy infrastructure in developing urban contexts such as Islamabad. This work advances multi-objective optimization methodologies and their application in supporting the transition toward low-carbon transportation and integrated HRES.

The reentrant hybrid flow shop scheduling problem (RHFSP) with total energy consumption constraints poses significant challenges for sustainable manufacturing. To address this issue, this study proposes a multi-objective evolutionary algorithm combining clustering and entropy-guided unsupervised learning (CEUL-MOEA). In CEULMOEA, an exploration-exploitation co-evolution population framework is employed to balance global search and local refinement. Then, four lightweight internal evaluation indicators are introduced to assess convergence, boundary coverage, and distribution uniformity, further supporting the design of six adaptive indicator-driven local search strategies. Furthermore, a clustering and entropy-guided neighborhood search is devised, which leverages density-based clustering and entropy-based weighting to dynamically allocate search resources based on solution density. Additionally, two theorems are proposed concerning the energy-driven backward adjustment (ED-BA) strategy, along with their formal proofs. Based on these theorems, single-lap and multi-lap ED-BA strategies are designed to reduce idle energy consumption. Extensive experiments on 275 RHFSP instances demonstrate that the proposed algorithm outperforms state-of-the-art methods in terms of convergence, diversity, and energy optimization, achieving an average reduction of 90% in AGD, 84% in AIGD, and an average improvement of 55% in AHV.

The Flexible Job Shop Scheduling problem with Automated Guided Vehicles (FJSP-AGVs) is commonly encountered in modern manufacturing systems, with complexity increasing exponentially as the problem scales, making traditional methods inadequate. So far, a limitation of meta-heuristic algorithms incorporating machine learning for solving FJSP-AGVs is that machine learning is primarily used for heuristic selection (e.g., dispatching rules or local search operators) rather than actively guiding solution evolution based on the diverse features of solutions, leading to incomplete exploration and premature convergence. To tackle these issues, this paper first presents a Q-Learning Region Selection Enhanced Quality-Diversity (Q-QD) algorithm, which integrates a pairwise selection scheme to identify potential collaborative solutions within the feature space for two practical Chinese manufacturing scenarios: a coal machinery structural parts production workshop and a machining workshop in Nanjing. Unlike traditional approaches that focus solely on optimizing the objective, Q-QD leverages the customized feature space incorporating behaviors such as job transportation times and machine idle times. By combining Q-learning-based region selection with heuristic rule-based local search for AGV transportation, the proposed method effectively discovers high-performing and diverse solutions. The results obtained from two real-world cases and 21 benchmark instances indicate that the proposed Q-QD algorithm surpasses five advanced counterparts, achieving a makespan reduction of 12.2-30.8% relative to the strongest baseline. These substantial performance gains highlight the robustness and efficiency of QD-based optimization for solving FJSP-AGVs and suggest a valuable avenue for future research in intelligent job shop scheduling.

Over the past decade, Pakistan has grappled with a severe power deficit, primarily due to a generating capacity of only 22,003 MW against a demand of 29,000 MW. This shortfall has led to frequent electricity outages, severely impacting both consumers and businesses, and undermining the credibility of governing bodies. In urban areas of Pakistan, power shutdowns and load shedding extend up to 14 h per day, escalating to 20 h in remote rural regions. This study explores a framework for designing a hybrid renewable energy system (HRES) tailored for rural electrification in Bannu district, Pakistan. The proposed system integrates concentrated solar power (CSP), photovoltaic (PV) panels, a biogas generator (BG), and a backup storage system to ensure a stable power supply. Our findings show that the cost of energy (COE) for the HRES is $0.391/kWh, with a total net present cost (TNPC) of $155,395. Financial analyses, including projections of cash flows and evaluations of metrics such as net present value and payback period, were conducted. Furthermore, a Monte Carlo simulation was employed to assess the risks associated with the HRES, enhancing the reliability of project decisions. The results confirm the commercial viability of the hybrid system through robustness, sensitivity, and breakeven analyses, establishing its potential to provide reliable, round-the-clock electricity to the community. Additionally, the study evaluates the performance of the proposed system in comparison to current leading-edge research, suggesting a roadmap for sustainable energy transition and offering strategic guidance for decision-makers on prioritizing renewable resources and addressing both immediate and long-term challenges.

Quality-Diversity (QD) optimization is a paradigm of evolutionary computation (EC) that extends the classic approaches, aiming to generate a collection of solutions that are both diverse and high-performing. Unlike traditional evolutionary algorithms (EAs), QD methods emphasize the illumination (or coverage) of a user-defined feature space, while simultaneously aiming for local optimization within each discovered region of the feature space. Over the past decade, QD has rapidly developed and proven effective in areas such as evolutionary robotics and video games. However, a systematic review of this growing field remains lacking. To date, the most recent review article on QD was published in 2021. Therefore, to offer a more comprehensive overview of the latest QD research, this paper provides a thorough survey of QD optimization, covering its foundational principles and representative algorithmic frameworks such as Novelty Search with Local Competition (NSLC), MAP-Elites, the unified modular QD framework, and RIBS. In addition, we divide the algorithm improvement part into three modules for discussion: containers, selection, and mutation. Then, the evaluation metrics widely used in QD optimization are listed for researchers. We further explore its diverse applications across domains such as evolutionary robotics, video games, scheduling, software testing, and engineering design. Finally, we discuss the current challenges in the field and outline promising directions for future research.

In modern manufacturing under economic globalization, the Distributed Hybrid Flowshop Scheduling Problem with Consistent Sublots (DHFSP_CS) is critical for improving production efficiency. This problem requires optimizing job sequences and machine assignments across multiple stages and distributed factories. A Mixed Integer Linear Programming (MILP) model is formulated for DHFSP_CS and validated using Gurobi across 27 instances. To solve large-scale instances, a novel Q-learning-enhanced Reconstructed Discrete Particle Swarm Optimization (QRDPSO) algorithm is proposed. QRDPSO incorporates a Q-learning mechanism to adaptively adjust particle discretization, enhancing search efficiency. A global search strategy based on dispersion variation improves exploration, while seven neighbourhood structures tailored to the DHFSP_CS objective of minimizing makespan increase search depth. Extensive experiments on 400 instances show that QRDPSO consistently outperforms state-of-the-art algorithms in terms of average relative percentage increase.

Incorporating the concept of delivery time windows into the distributed blocking flowshop group scheduling (DBFGSP) contributes to ensuring the timely delivery of products and enhancing customer satisfaction. Therefore, this paper focuses on the study of DBFGSP with delivery time windows. However, its several unresolved challenges, i.e., establishing the problem model and reducing the algorithmic complexity, persist. Therefore, we first construct a mixed-integer linear programming model for this problem, and verify its correctness by Gurobi solver. Second, we define, for the first time, the concept of the ’critical machine’, and propose the theoretical analysis and methods of a critical machine-based rapid evaluation for the total weighted earliness and tardiness (TWET) criteria. Then, we integrate the proposed strategies, i.e., two rapid evaluation methods, a modified idle time insertion method based on family chunks, and two idle time insertion-driven rapid local search strategies, into iterated greedy (namely RAP_IG_ITI) to coincide better with delivery time windows. Through statistical analysis, it becomes evident that our computational efforts have significantly decreased in computing TWET objective, and the proposed RAP_IG_ITI can effectively solve DBFGSP within an acceptable time.

The reentrant hybrid flow shop scheduling problem (RHFSP) poses significant challenges due to its complex structure and limited research on model accuracy, solution space exploration, and adaptive strategy selection. To address these gaps, this paper proposes a novel feature-driven double deep Q-network with iterated greedy (FD3QNIG) algorithm. First, a five-dimensional mixed-integer linear programming (MILP) model is developed to significantly improve modeling accuracy. Second, a feature-driven initialization strategy (FDNEH) enhances the quality of initial solutions, while an adaptive historical information-driven destruction and reconstruction strategy (AHDDR) effectively balances exploration and exploitation during the search process. Third, multi-scale local search strategies are employed, including the first application of comprehensive exploration-driven mandatory operations local search (CED_MOLS) to RHFSP, substantially deepening the search capability. Fourth, an adaptive strategy selection mechanism based on double deep Q-Network (DDQN_ASS) dynamically guides the algorithm toward more efficient decision-making. Extensive experiments on 285 benchmark instances demonstrate that FD3QNIG achieves a 53 %-94 % improvement in average relative percentage increase (ARPI) over state-of-the-art methods, confirming its effectiveness and robustness.

In response to growing demands for low-carbon and flexible manufacturing, this study addresses the Multi-Objective Flexible Job Shop Scheduling Problem with Automated Guided Vehicles (MFJSP-AGVs), optimizing makespan and Total Energy Consumption (TEC). We first develop a novel mixed-integer linear programming model using sequence-based modeling approach, rigorously validated by the Gurobi solver. The innovations of the proposed Improved Multi-Objective Evolutionary Algorithm (IMOEA) are that: 1) A three-layer encoding mechanism uniquely decouples machine selection, AGV assignment, and operation sequencing subproblems, overcoming traditional methods' limitations in handling interdependencies; 2) Ternary crossover and adaptive dual mutation operators synergistically enhance global exploration while maintaining solution diversity, addressing convergence-robustness trade-offs in existing algorithms; 3) A hybrid critical path optimization integrates Tabu Search for critical operations and energy-aware local search for non-critical paths, enabling simultaneous makespan-TEC optimization. The proposed IMOEA algorithm is applied to 55 test sets, and compared with the state-of-the-art four algorithms. Comprehensive experiments demonstrate IMOEA's superiority over EHA, EMOEA, and recent problem-specific algorithms, achieving 91.8 % and 227.6 % average improvements in IGD and HV respectively. These advancements establish a new benchmark for multi-objective AGV scheduling in energy-conscious production systems.