Associative reasoning-based interpretable continuous decision making in industrial production process

Abstract

Adaptive decision making in continuous production process is challenging for the uncertain characteristics of dynamics system state, decision-making time, as well as the correlation between the previous decision and the subsequent decision. In this paper, a data-driven continuous decision-making model with self-adaption and interpretability is proposed. For the uncertainty of industrial data, a new form of information granule with variable parameters is designed to describe the knowledge extracted from data. For the continuity of the decision-making process, a data-driven interpretable associative reasoning model is established by considering the similarity of knowledge and the attention mechanism. Aiming at the dynamic feature in the decision-making process, a data-driven associative learning method is proposed to obtain the adaptivity of the proposed model, which generates and updates routes according to the co-activation and simultaneous activities of the information granules. To reason decision results, the proposed model searches and records the reasoning routes according to the reasoning results and the objective at the decision points. Noisy chaotic time series data and a number of classification datasets are employed for experimental analyses, and the results indicate that the proposed method exhibits high reasoning accuracy, anti-noise ability, and interpretability. In addition, the experimental results based on industrial data indicate that the decision-making model is capable of giving a reasonable, effective and interpretable decision scheme and suitable for establishing a credible decision-making system.