Modelling stochastic behaviour in simulation digital twins through neural nets

Reed, Sean; L�fstrand, Magnus; Andrews, John

Abstract

Discrete event simulation (DES) is a widely used technique for modelling systems where state changes occur at discrete points in time. Stochastic behaviour is represented through random variables associated with probability distributions, from which random variates are sampled during simulation to determine outcomes. However, the distribution of outcomes for an event in the real system being modelled often depend on characteristics of the current system state. This paper proposes the use of artificial neural networks (ANN) to determine the changing conditional distributions of random variables during a simulation. The ANN are pre-trained to predict distributions by learning from example pairs of input feature vectors and random variable outcomes. It enables complex, non-linear dependencies between these features and random variable outcome distributions to be accurately modelled, including distributions that are multi-modal. A major area of application is the development of digital twin models that closely mimic the complex stochastic behaviour of the connected physical twin by learning from the data the real system generates. The benefits of the approach introduced in the paper are demonstrated through a realistic DES model of load-haul-dump vehicle operations in a production area of a sublevel caving mine. Keywords: discrete event simulation; mixture density network; digital twin; artificial neural network; industry 4.0 1. Introduction Discrete event simulation (DES) is a popular modelling approach due to its ability to represent complex, dynamic systems with stochastic behaviour whilst requiring fewer simplifying assumptions compared to analytical models (Banks, 1998). Stochastic behaviour is incorporated in DES by using random variables to represent event outcomes, with outcome probabilities represented by a probability distribution function. Random variables in a DES

Citation

Reed, S., Löfstrand, M., & Andrews, J. (2022). Modelling stochastic behaviour in simulation digital twins through neural nets. Journal of Simulation, 16(5), 512-525. https://doi.org/10.1080/17477778.2021.1874844

• Journal Article Type: Article
• Acceptance Date: Jan 12, 2021
• Online Publication Date: Jan 26, 2021
• Publication Date: 2022
• Deposit Date: Jan 19, 2021
• Publicly Available Date: Jan 27, 2022
• Journal: Journal of Simulation
• Print ISSN: 1747-7778
• Electronic ISSN: 1747-7786
• Publisher: Taylor and Francis
• Peer Reviewed: Peer Reviewed
• Volume: 16
• Issue: 5
• Pages: 512-525
• DOI: https://doi.org/10.1080/17477778.2021.1874844
• Keywords: Modelling and Simulation; Software
• Public URL: https://nottingham-repository.worktribe.com/output/5246256
• Publisher URL: https://www.tandfonline.com/doi/full/10.1080/17477778.2021.1874844