Large and extensive manufacturing systems consume a large proportion of manufacturing energy. A key component of energy efficiency management is the accurate prediction of energy efficiency. However, the nonlinear and vibration characteristics of machining systems’ energy consumption (EC) pose a challenge to the accurate prediction of system EC. To address this challenge, an energy consumption prediction method for machining systems is presented, which is based on an improved particle swarm optimization (IPSO) algorithm to optimize long short-term memory (LSTM) neural networks. The proposed method optimizes the LSTM hyperparameters by improving the particle swarm algorithm with dynamic inertia weights (DIWPSO-LSTM), which enhances the prediction accuracy and efficiency of the model. In the experimental results, we compared several improved optimization algorithms, and the proposed method has a performance improvement of more than 30% in mean absolute error (MAE)and mean error(ME).

Forecasting emergency drug demand (FEDD) during earthquakes is essential for optimizing logistics and improving disaster response efficiency. Traditional forecasting methods mostly rely on indirect forecasting, that is, predicting the number of casualties and then extrapolating drug demand from that. However, the number of casualties in an earthquake is influenced by numerous factors, which introduces significant computational complexity for existing methods that attempt to handle these multi-faceted variables. Furthermore, drug demand forecasts must also consider seasonal and regional variations, which existing methods fail to adequately incorporate. To address these issues, this paper proposes a novel FEDD approach that integrates XGBoost and AFT-LSTM. First, a method employs XGBoost-based multi-feature extraction techniques to identify and prioritize key factors influencing earthquake casualties. Next, an AFT-LSTM model is developed to predict casualties during earthquakes, capturing temporal dynamics and the interactions among factors. Finally, a mathematical model is established to predict total drug demand, considering the number of casualties, seasonal fluctuations, and regional characteristics. Empirical analysis demonstrates that the FEDD model performs excellently in casualty prediction, offering significant advantages over traditional BP, LSTM, and Transformer models. Moreover, the model accurately predicts total drug demand based on casualty estimates. The FEDD model provides a scientific foundation for disaster management, facilitating the efficient allocation of rescue resources during earthquake disasters and enhancing overall response efficiency.

Production scheduling and machine tool intermittent state control separately influence a workshop’s machining and intermittent energy consumption. Effective scheduling decisions and intermittent state control are crucial for optimizing the overall energy consumption in the workshop. However, the scheduling scheme determines the machine tool intermittent durations, which imposes strong constraints on the decision-making process for intermittent state control. This makes it difficult for intermittent state control to be used in providing feedback and optimizing scheduling decisions, significantly limiting the overall energy-saving potential of the workshop. To this end, a workshop energy-saving scheduling method is proposed integrating machine tool intermittent state control. Firstly, the variation characteristics of workshop machining energy consumption, machine tool intermittent durations, and intermittent energy consumption are analyzed, and an energy-saving optimization strategy is designed. Secondly, by incorporating variables such as intermittent durations, intermittent energy consumption, and variable operation start time, a multi-objective integrated optimization model is established. Thirdly, the energy-saving optimization strategy is integrated into chromosome encoding, and multiple crossover and mutation genetic operator strategies, along with a low-level selection strategy, are introduced to improve the NSGA-II algorithm. Finally, the effectiveness of the proposed method is verified through a machining case. Results show that the generated Gantt chart reflects both production scheduling and intermittent state control decision outcomes, resulting in a 1.51% reduction in makespan, and 3.90% reduction in total energy consumption.

Carbon efficiency, measuring effective output per unit of carbon emissions, is vital for managing low-carbon workshops and advancing sustainable manufacturing. However, production processes often face concurrent discrete events (e.g., equipment failures, parameter adjustments) and numerous emission factors with complex relationships, making it hard to identify dominant factors and event impact degrees, thus lacking direction for operation and maintenance decisions. This paper proposes a deductive monitoring model to analyze carbon efficiency changes under event concurrency. First, for traceability, a multi-resolution enhanced carbon efficiency information transfer network is proposed. It classifies emission factors into time-driven and event-driven accounting, and under the Parallel Discrete Event System Specification framework, adopts a multi-resolution approach with high- and low-resolution models for hierarchical aggregation from equipment to workshop, establishing a traceability path to specific factors. Second, for unclear impact degrees, a dynamic monitoring model for concurrent events is designed. A state-driven dynamic carbon efficiency accounting method automatically settles upon equipment state switching, and a rule-driven priority deduction strategy enables independent accounting of each event’s impact degrees in a determined order. A case study on a machine tool spindle production workshop validates the proposed model. Under baseline conditions, the relative accounting errors for 8 h cumulative carbon emissions and effective output are approximately 1.05% and 0.89%, respectively. In concurrent event scenarios, the model achieves deterministic trajectory reproducibility across 30 independent deduction runs and enables independent impact-degree decomposition, whereas traditional discrete event simulation exhibits trajectory ambiguity. Furthermore, testing under 42 multi-parameter perturbation combinations demonstrates traceability path integrity and accurate root-cause localization, delivering a transparent and reliable quantitative basis for low-carbon maintenance decisions.

The digital transformation of supply chain operations demands manufacturing systems capable of autonomous decision-making in response to dynamic production fluctuations. However, existing inventory management approaches suffer from limited adaptability to plan variations and delayed response times, primarily due to the absence of tight coupling between inventory control and real-time production demand. To address these challenges, this study proposes AISDBM-Agent, an LLM-enabled autonomous decision-making prototype system for intelligent supply chain inventory optimization. The framework is structured into four functional modules: the perception module receives and translates external multi-source information; the memory module integrates, categorizes, and stores the processed information to mitigate the risks arising from outdated knowledge or contextual inconsistencies; the strategy module utilizes the processed input from the perception module and the knowledge provided by the memory module to perform reasoning and decision-making, automatically generating optimal procurement decisions, employing the Chain of Thought (CoT) method to enhance the transparency and interpretability of these decisions; the execution module interacts with external systems and obtains execution feedback to ensure information timeliness, thereby resolving the issue of delayed response in inventory management. A case study on steel scrap inventory management is conducted to validate the effectiveness of the proposed framework. The results demonstrate that AISDBM-Agent significantly enhances the system’s ability to respond to fluctuations in demand, enabling the smooth management of stock levels.