Advanced local prediction and its applications in power and energy systems

Zhu, Lei Advanced local prediction and its applications in power and energy systems. PhD thesis, University of Liverpool.

Text ZhuLei_Sep2014_2007714.pdf - Unspecified Available under License Creative Commons Attribution. Download (2MB)

Abstract

Due to the global energy crisis and environmental concerns, the development of sustainable energy is considered by more and more countries. In order to make this target, energy demand management is significantly necessary in which forecasting the energy demand is the starting point. The accurate prediction of energy demand could help the energy sectors to make these operation decisions and policy properly. A novel approach, which is the support vector regression based local predictor with false neighbor filtered (FNF-SVRLP), is proposed. This method is an amelioration of the support vector regression based local predictor (SVRLP). SVRLP is a powerful prediction method which employs phase reconstruction algorithms, such as the correlation dimension and mutual information methods used in time series analysis for data preprocessing. Compared with the global prediction method, in a local prediction method, each predicting point has its own model constructed based on its nearest neighbors (NNs) reconstructed from the time series, and the fitness of NNs would mainly affect the model performance. However, it has been found that NNs may contain a class of false neighbors (FNs) which would decrease the fitting accuracy dramatically and lead to a poorer forecasting performance. Therefore, a new false neighbor filter is proposed to remove those false neighbors and keep the optimal nearest neighbors. Then, the FNF-SVRLP is proposed. Wind power is one of the most popular renewable energy. The increasing penetration of wind power into the electric power grid accompanied with a series of challenges. Due to the uncertain and variable nature of wind resources, the output power of wind farms is hard to control, which could lead to the instability of the power grid operation and the unreliability of electricity supplies. In order to slove this problem, the FNF-SVRLP based short-term wind power perdition model is presented. Through the comparison with the SVRLP based short-term wind power perdition and ARMA based short-term wind power perdition, it is found that the FNF-SVRLP based short-term wind power perdition model is much more accurate than the others. Due to the fact that natural gas is cleanest burning of all fossil fuel, it can be considered as an important adjunct to renewable energy sources such as wind or solar, as well as a bridge to the new energy economy. Different from the wind power, the customer consumption behavior could effect the natural gas demand. Therefore, the customer behavior based ``Advanced Model" with FNF-SVRLP is presented to undertake the natural gas prediction. The proposed FNF-SVRLP natural gas model is compared with the SVRLP and autoregressive moving average (ARMA) to show its superiority. In addition, a web sever based online natural gas demand perdition system has been set up to help the National Grid to obtain the accurate daily natural gas demand perdition easily and timely. It is found that the most kinds of energy demand data are non-stationary, the internal regularity between predicting point and its nearest-neighbors are much more complex than the stationary dataset. In order to help the local predictor to capture the internal regularity between predicting point and its nearest-neighbors more accurately, the morphological filter is proposed. the morphological filter is applied to decompose the non-stationary dataset into several subsequences, ranked form the low frequency subsequence to the high frequency subsequence. Through this way, the local predictor could capture the non-stationary dataset more accurate, and improve the final performance of prediction. The morphological filter is applied to decompose the non-stationary into several subsequences, ranked form the low frequency subsequence to the high frequency subsequence. Through this way, the local predictor could capture the non-stationary dataset more accurate, and improve the final performance of prediction. Moveover, an novel calculation method of structure element (SE) is introduced. Different form the conventional SE, this novel approach can optimize the scale and shape of SE to match the original signal. After that, a novel algorithm, which is mathematical morphology based local prediction with support vector regression (SVRLP-MM) is proposed. The real-world wind speed data has been used to evaluate the performance of SVRLP-MM. The final results presented demonstrate that SVRLP-MM based wind speed prediction model can achieve a higher prediction accuracy than the SVRLP based model and ARMA model based model by using the same real-world wind speed data.

Item Type:	Thesis (PhD)
Additional Information:	Date: 2014-09 (completed)
Subjects:	?? Q1 ?? ?? TK ??
Depositing User:	Symplectic Admin
Date Deposited:	11 Sep 2015 11:04
Last Modified:	17 Dec 2022 01:28
DOI:	10.17638/02007714
URI:	https://livrepository.liverpool.ac.uk/id/eprint/2007714