Understanding Digital Twins
Factors for "Consistency of IT Systems"
- Confidentiality, Integrity, Availability, Privacy
- Resilience, Distributed system, Legacy support, Scalability
Requirements for the Reliability and Configuration of Digital Twin Platforms
- Network quality, Data acquisition system, Data synchronisation
- Big-data’s 5V : volume, velocity, veracity (진실성, 정확성), variability, variety
- Software flexibility, Precise vs Accurate
.jpg)
Statistical methods
The term 'twin'
- doesn't necessarily mean it's 100% identical to the physical world; rather, it's an expression that describes how changes in the physical world are linked to cyberspace in (near) real-time, and conversely, how decisions made in cyberspace are reflected back into the physical world.
Ensemble techniques
- divided into stacking, bagging, random forest, and boosting.
- Boosting: Models are trained sequentially, focusing on incorrect predictions. Superior to other techniques.
Reinforcement learning
- value optimisation, policy optimisation, imitation
- Value optimisation
- Takes actions that maximise the sum of cumulative rewards from a specific state until the final point in time. Q-learning-based derivatives (DQN, QR-DQN)
- Policy optimisation
- Optimises the policy that defines which action to take in a specific state. Policy-gradient, Actor-critic family (A2C, A3C, PPO, DDPG)