SUMMARY

For an operational system where the optimization is a combinatorial optimization problem, the optimization performance largely determines the solution quality. Moreover, there exists a trade-off between the computing time of the decision-making process and the optimization performance, which is particularly obvious in a system that conducts real-time operations. To obtain better solutions to the decision-making problem in a shorter time, many optimization algorithms are proposed to improve the searching efficiency in the solution space. However, information extraction from the environment is also essential for problem-solving. The environment information not only includes the optimization model inputs, but also contains details of the current situation that may provide different problem formulation and optimization strategies. Due to the time constraint and the computation time of visual processing algorithms, most conventional operational systems collect environment data from sensor platforms but do not analyze image data, which contains situational information that can assist with the decision-making process. To address this issue, this thesis proposes CNN-enabled visual data analytics and intelligent reasoning for real-time optimization, and a closed-loop optimization structure with discrete event simulation to fit the use of situational information in the optimization model. In the proposed operational system, CNNs are used to extract context information from image data, like the type and number of objects. Then reasoning techniques and methodologies are applied to deduct knowledge about the current situation to adjust optimization strategies and parameter settings. Discrete event simulation is conducted to test the optimization performance, and adjustments can be made on how situational information fits in the optimization process. To validate the feasibility and effectiveness, an application to occupancy-aware elevator dispatching optimization is presented.