Complexity

Research Article

An Advanced Hybrid Forecasting System for Wind Speed Point Forecasting and Interval Forecasting

Table 1

Comparison on the advantages and disadvantages of existing wind speed forecasting methods.


Category	Advantage	Disadvantage	Method sample	Sample advantage	Sample disadvantage

NWP models	Suitable for long-term forecast	Hard to deal with short-term horizons and costs a lot of computing time and resources	Computational fluid dynamics (FLUENT, CFD)	High parallel efficiency	High requirement of dataset length, limited simulation range
			Mesoscale numerical model (MM5, WRF)	Accurate simulation and a large number of alternative parameterization schemes	Suitable for weather forecast, not specifically optimized for wind prediction
			Physical schema (predictor)	Can predict spatial properties	Expensive calculation cost

Statistical strategies	Wide application, less running time	Assumption difficult to achieve	Time series (ARIMA, ARCH, GARCH, GM (1, 1))	Simple model assumptions, good self-fitness	Poor extrapolation, limited prediction range
Statistical strategies	Wide application, less running time	Assumption difficult to achieve	Cluster	Strong fault tolerance, simple human-computer interaction	Low accuracy, lack of systematicness

Machine learning	Excellent robustness, little model risk, wide application	High complexity, high requirements of knowledge	Neural network (FNN, BPNN, GRNN, ELMAN)	Excellent fitting effect nonlinear property	The model is unstable, high degree of data dependence
			Artificial intelligence (PSO, GA, BA, DA)	Easy to be understood and combined with other methods	Easy to fall into local extremum
			Data denoising (EMD, EEMD, CEEMD, SSA)	Eliminate the negative impact of noise data	The mode mixing problem in EMD and the residual noise in EEMD
			Deep learning (CNN, RNN, LSTM)	Strong fault tolerance, simple human-computer interaction	Low accuracy when the amount of data is small, lack of systematicness

Review of the previous literature shows that there is no single forecasting model that can be considered the best and applied in all cases for wind speed forecasting, as there are considerable differences among the wind speed time series.