DMEformer: A newly designed dynamic model ensemble transformer for crude oil futures prediction.

Crude oil futures prediction plays an important role in ensuring sustainable energy development. However, the performance of existing models is not satisfactory, which limits its further application. The poor performance mainly results from the lack of data mining of economic models and the poor stability of most data analysis models. To solve the above problems, this paper proposes a new dynamic model ensemble transformer (DMEformer). The model uses three different Transformer variants as base models. It not only ensures the difference of base models but also makes the prediction results of base models not to appear disparity. In addition, NSGA-II is adopted to ensemble the results of base models, which considers both the modeling stability and accuracy in the optimization. Finally, the proposed model adopts a dynamic ensemble scheme, which could timely adjust the weight vector according to the fluctuation of energy futures. It further improves the reliability of the model. Comparative experiments from the perspective of single models and ensemble models are also designed. The following conclusions can be drawn from the experimental results: (1) The proposed dynamic ensemble method can improve the performance of the base model and traditional static ensemble method by 16% and 5% respectively. (2) DMEformer can achieve better performance than 20 other models, and its accuracy and MAPE values were 72.5% and 2.8043%, respectively. (3) The proposed model can accurately predict crude oil futures, which provides effective support for energy regulation and sustainable development.

Neural network aided flexible joint optimization with design of experiment method for nuclear power plant inspection robot.

Introduction: The flexible joint is a crucial component for the inspection robot to flexible interaction with nuclear power facilities. This paper proposed a neural network aided flexible joint structure optimization method with the Design of Experiment (DOE) method for the nuclear power plant inspection robot. Methods: With this method, the joint's dual-spiral flexible coupler was optimized regarding the minimum mean square error of the stiffness. The optimal flexible coupler was demonstrated and tested. The neural network method can be used for the modeling of the parameterized flexible coupler with regard to the geometrical parameters as well as the load on the base of the DOE result. Results: With the aid of the neural network model of the stiffness, the dual-spiral flexible coupler structure can be fully optimized to a target stiffness, 450 Nm/rad in this case, and a given error level, 0.3% in the current case, with regard to the different loads. The optimal coupler is fabricated with wire electrical discharge machining (EDM) and tested. Discussion: The experimental results demonstrate that the load and angular displacement keep a good linear relationship in the given load range and this optimization method can be used as an effective method and tool in the joint design process.

Realistic Actor-Critic: A framework for balance between value overestimation and underestimation.

Introduction: The value approximation bias is known to lead to suboptimal policies or catastrophic overestimation bias accumulation that prevent the agent from making the right decisions between exploration and exploitation. Algorithms have been proposed to mitigate the above contradiction. However, we still lack an understanding of how the value bias impact performance and a method for efficient exploration while keeping stable updates. This study aims to clarify the effect of the value bias and improve the reinforcement learning algorithms to enhance sample efficiency. Methods: This study designs a simple episodic tabular MDP to research value underestimation and overestimation in actor-critic methods. This study proposes a unified framework called Realistic Actor-Critic (RAC), which employs Universal Value Function Approximators (UVFA) to simultaneously learn policies with different value confidence-bound with the same neural network, each with a different under overestimation trade-off. Results: This study highlights that agents could over-explore low-value states due to inflexible under-overestimation trade-off in the fixed hyperparameters setting, which is a particular form of the exploration-exploitation dilemma. And RAC performs directed exploration without over-exploration using the upper bounds while still avoiding overestimation using the lower bounds. Through carefully designed experiments, this study empirically verifies that RAC achieves 10x sample efficiency and 25% performance improvement compared to Soft Actor-Critic in the most challenging Humanoid environment. All the source codes are available at https://github.com/ihuhuhu/RAC. Discussion: This research not only provides valuable insights for research on the exploration-exploitation trade-off by studying the frequency of policies access to low-value states under different value confidence-bounds guidance, but also proposes a new unified framework that can be combined with current actor-critic methods to improve sample efficiency in the continuous control domain.