Lithium-ion (Li-I) batteries have recently become pervasive and are used in many physical assets. For the effective management of the batteries, reliable predictions of the end-of-discharge (EOD) and end-of-life (EOL) are essential. Many detailed electrochemical models have been developed for the batteries. Their parameters are calibrated before they are taken into operation and are typically not re-calibrated during operation. However, the degradation of batteries increases the reality gap between the computational models and the physical systems and leads to inaccurate predictions of EOD/EOL. The current calibration approaches are either computationally expensive (model-based calibration) or require large amounts of ground truth data for degradation parameters (supervised data-driven calibration). This is often infeasible for many practical applications. In this paper, we introduce a reinforcement learning-based framework for reliably inferring calibration parameters of battery models in real time. Most importantly, the proposed methodology does not need any labeled data samples of observations and the ground truth parameters. The experimental results demonstrate that our framework is capable of inferring the model parameters in real time with better accuracy compared to approaches based on unscented Kalman filters. Furthermore, our results show better generalizability than supervised learning approaches even though our methodology does not rely on ground truth information during training.

The parallel machine scheduling problem (PMSP) involves the optimized assignment of a set of jobs to a collection of parallel machines, which is a proper formulation for the modern manufacturing environment. Deep reinforcement learning (DRL) has been widely employed to solve PMSP. However, the majority of existing DRL-based frameworks still suffer from generalizability and scalability. More specifically, the state and action design still heavily rely on human efforts. To bridge these gaps, we propose a practical reinforcement learning-based framework to tackle a PMSP with new job arrivals and family setup constraints. We design a variable-length state matrix containing full job and machine information. This enables the DRL agent to autonomously extract features from raw data and make decisions with a global perspective. To efficiently process this novel state matrix, we elaborately modify a Transformer model to represent the DRL agent. By integrating the modified Transformer model to represent the DRL agent, a novel state representation can be effectively leveraged. This innovative DRL framework offers a high-quality and robust solution that significantly reduces the reliance on manual effort traditionally required in scheduling tasks. In the numerical experiment, the stability of the proposed agent during training is first demonstrated. Then we compare this trained agent on 192 instances with several existing approaches, namely a DRL-based approach, a metaheuristic algorithm, and a dispatching rule. The extensive experimental results demonstrate the scalability of our approach and its effectiveness across a variety of scheduling scenarios. Conclusively, our approach can thus solve the scheduling problems with high efficiency and flexibility, paving the way for application of DRL in solving complex and dynamic scheduling problems.

Symbolic Regression holds great potential for uncovering underlying mathematical and physical relationships from observed data. However, the vast combinatorial space of possible expressions poses significant challenges for previous online search methods and pre-trained transformer models, which mostly do not consider the integration of domain experts’ prior knowledge. To address these challenges, we propose the Symbolic Q-network, an advanced interactive framework for large-scale symbolic regression. Unlike previous transformer-based SR approaches, Symbolic Q-network leverages reinforcement learning without relying on a transformer-based decoder. Furthermore, we propose a co-design mechanism, where the Symbolic Q-network facilitates effective interaction with domain experts at any stage of the equation discovery process. Our extensive experiments demonstrate Sym-Q performs comparably to existing pretrained models across multiple benchmarks. Furthermore, our experiments on real-world cases demonstrate that the interactive co-design mechanism significantly enhances Symbolic Q-network’s performance, achieving greater performance gains than standard autoregressive models.