Use this URL to cite or link to this record in EThOS:
Title: Revealing certainties from uncertainties through data mining and data modelling
Author: Gu, Yuanlin
ISNI:       0000 0004 8504 8053
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
In model identification, the existence of uncertainty normally generates negative impact on the accuracy and performance of the identified model. This thesis focuses on the development of three novel methods to deal with model uncertainty, which are the robust model structure selection (RMSS) method, cloud-NARX model and machine learning enhanced nonlinear autoregressive moving average with exogenous inputs (MLENARMAX) model. First, the RMSS method is developed for model identification problems with small size data and multi-datasets. The proposed method can reduce the model structure uncertainty and therefore improve the model performances. The RMSS method is applied to two real data applications, which are the modelling of Kp index and modelling of cortical response. Second, the cloud-NARX model is proposed. The cloud-NARX model uses cloud model and cloud transformation to quantify the uncertainty throughout the structure detection, parameter estimation and model prediction. The cloud-NARX model is applied to predict AE index 1 hr ahead. The new predicted band can be generated to forecast system output with confidence interval. The cloud-NARX method provides a new way to evaluate the model based on uncertainty analysis and reveal the reliability of model, and visualize the bias of model prediction. Third, the MLE-NARMAX model is developed. The MLE-NARMAX model is established based on a NARMAX model structure, which is composed of the most important candidate features (variables). With an extra neural network sub-model, the MLE-NARMAX model is enhanced by the machine learning methods so that the model performance can be improved. The MLE-NARMAX model is applied to predict appliance energy use 10 minutes ahead and predict Dst index 3 hours ahead. The proposed model provides a new way for data modelling problems through machine learning approach with a simple/sparse, interpretable and transparent model structure.
Supervisor: Wei, Hualiang ; Balikhin, Michael ; Bigg, Grant Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available