| The thesis addresses the need for an intelligent prediction system to carry out the task of time series prediction with causal information, where the traditional methods have proven to be inadequate. The research focuses on modeling a prediction scheme that represents and manages uncertainties that may arise in the prediction scheme. The proposed intelligent prediction architecture is composed of four modules: prediction with confidence distribution, a reinforcement learning algorithm, the Fuzzy Analytic Hierarchy Process (FAHP), and a performance assessment module. Prediction with a confidence distribution is derived to represent uncertainties and multiple trends with different confidence levels. A reinforcement learning algorithm is devised to reduce uncertainties in the prediction. The FAHP provides a means to adjust effects of causal information, which may abruptly change the prediction trend. Performance assessment metrics are employed to assess the performance of the prediction algorithms with multiple criteria. This architecture will provide meaningful prediction results and serve as a better prediction system for a variety of applications in engineering, business, etc. |
