WSEAS Transactions on Power Systems
Print ISSN: 1790-5060, E-ISSN: 2224-350X
Volume 13, 2018
Average Displacement-Support Vector Machines via Lyapunov Exponents and Wave Approximate Function in Power Load Forecasting Model
Authors: , , , , ,
Abstract: With the development and improvement of the electricity market, the role of power load forecasting in competitive market is being recognized. It is very important to ensure the security and stability of electric network by accurate load forecasting. Meanwhile, the reform of the electricity market leads to a result that it is far more difficult for the match between power supply-side and demand-side, which brings new challenges to the power load forecasting industry. According to the chaotic and non-linear characters of power load data, and to prove the effectiveness and accuracy, the model of AD-SVM (average displacement-support vector machines) based on Lyapunov exponents and wave approximation function optimization was established. Firstly, the optimal time delay and embedding dimension sequence of time series are determined by average displacement method. Secondly the method of maximum Lyapunov exponents is devoted to selecting the appropriate embedding dimension and time delay, which proves the effectiveness of the accuracy invariants of phase-space; and then considering the volatility autocorrelation characteristics, we determine learning parameters based on the exploiting fluctuation approximation function and embedding dimension, which can greatly assist us with the prediction of future market; meanwhile, article applies KNN method to estimate noise standard deviation so as to determine the parameter selection. Finally, support vector machine is utilized for establishing the model of power load forecasting. In addition, for existing generalized autoregressive heteroscedasticity of phenomenon of LSSVM prediction error, In this paper, the prediction error is modified by GARCH model on the basis of the original and the prediction error of the auto regressive process is eliminated. Therefore, it can further enhance prediction accuracy. In order to prove the rationality of chosen dimension and the effectiveness of the model, any other random dimensions were selected to compare with the calculated dimension while BP algorithm was used to compare with the result of SVM. The results show that the model is effective and more accurate in the forecasting of short-term power load.
Search Articles
Keywords: Power load forecasting, Support vector machines, Lyapunov exponents, average displacement, Wave approximate function, embedding dimension, GARCH error correction
Pages: 258-271
WSEAS Transactions on Power Systems, ISSN / E-ISSN: 1790-5060 / 2224-350X, Volume 13, 2018, Art. #26