A combined model based on CEEMDAN and modified flower pollination algorithm for wind speed forecasting. (15th March 2017)

• Record Type: Journal Article
• Authors: Zhang, Wenyu; Qu, Zongxi; Zhang, Kequan; Mao, Wenqian; Ma, Yining; Fan, Xu
• Is Part Of: Energy conversion and management. Volume 136(2017)
• Page Start: 439
• Page End: 451
• Publication Date: 2017-03-15
• Subjects: complete ensemble empirical mode decomposition adaptive noise CEEMDAN -- no negative constraint theory NNCT -- flower pollination algorithm with chaotic local search CLSFPA -- China National Renewable Energy Center CNREC -- auto regressive moving average ARMA -- autoregressive integrated moving average ARIMA -- artificial neural network ANN -- radial basis function RBF -- support vector machine SVM -- back propagation BP -- wavelet neural network WNN -- particle swarm optimization PSO -- chaotic particle swarm optimization CPSO -- genetic algorithm GA -- differential evolution DE -- genetic simulated annealing algorithm GSA -- feed-forward neural network FNN -- Elman neural network ENN -- general regression neural network GRNN -- intrinsic mode function IMF -- square sum of the error SSE -- average absolute forecast error of n times forecast results MAE -- root mean-square forecast error RMSE -- average of absolute error MAPE -- Diebolde-Mariano (DM) test DM test; Wind speed prediction -- Combined model -- Weight coefficient optimization -- Complete ensemble empirical mode decomposition adaptive noise; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2017.01.022
• Languages: English
• ISSNs: 0196-8904

Abstract

Graphical abstract: Highlights: A CEEMDAN-CLSFPA combined model is proposed for short-term wind speed forecasting. The CEEMDAN technique is used to decompose the original wind speed series. A modified optimization algorithm-CLSFPA is proposed to optimize the weights of the combined model. The no negative constraint theory is applied to the combined model. Robustness of the proposed model is validated by data sampled from four different wind farms. Abstract: Wind energy, which is stochastic and intermittent by nature, has a significant influence on power system operation, power grid security and market economics. Precise and reliable wind speed prediction is vital for wind farm planning and operational planning for power grids. To improve wind speed forecasting accuracy, a large number of forecasting approaches have been proposed; however, these models typically do not account for the importance of data preprocessing and are limited by the use of individual models. In this paper, a novel combined model – combining complete ensemble empirical mode decomposition adaptive noise (CEEMDAN), flower pollination algorithm with chaotic local search (CLSFPA), five neural networks and no negative constraint theory (NNCT) – is proposed for short-term wind speed forecasting. First, a recent CEEMDAN is employed to divide the original wind speed data into a finite set of IMF components, and then a combined model, based on NNCT, is proposed for forecasting each decomposition signal. To improve the forecasting capacity of the combined model, a modified flower pollination algorithm (FPA) with chaotic local search (CLS) is proposed and employed to determine the optimal weight coefficients of the combined model, and the final prediction values were obtained by reconstructing the refined series. To evaluate the forecasting ability of the proposed combined model, 15-min wind speed data from four wind farms in the eastern coastal areas of China are used. The experimental results of this study show that: (a) the proposed CEEMDAN-combined model can take advantages of individual models and has the best performance among single models and the benchmark model; (b) the proposed CLSFPA is superior to FPA according to test functions and is effectively applied in optimizing the combined model; (c) the proposed algorithms are effective in high-precision wind speed predictions.