Classification Of Wind Speed Forecasting Methods

Currently, many different methods are used to predict wind speed. It is necessary to consider the most relevant methods used to predict wind speed, their features and scope of application, in order to identify the classes of methods applicable for solving the problem. Many papers have been published on both an integrated comprehensive review of modern approaches to predicting wind speed and for describing particular modified solutions. However, in any statement of the statement, all the works devoted to the study of methods for predicting wind speed consider these or other methods depending on the range of the performed forecast. However, referring to many studies in this field, wind forecasting can be divided based on the forecast horizon into four temporal range:

• Ultra-short-term forecast: From a few minutes to one hour ahead.
• Short-term forecast: From one hour to several hours ahead.
• Medium term forecast: From several hours to one week ahead.
• Long-term forecast: From one week to one year or more ahead.

Also, all methods can be divided to predict the wind speed according to the type of model used by this method to predict. As a result of the analysis of articles and literature devoted to predicting wind speeds, many of the most modern and most widely used techniques can be distinguished.

Categories of methods depending on the prediction model: Reference methods; Physical methods and methods for numerical weather prediction; Statistical methods; Artificial intelligence methods and Hybrid methods.

1 - Reference methods

Application of reference methods. There are several very simple ways to implement prediction methods, which at the same time can be very effective. Comparison with these methods allows us to assess the feasibility of implementing and applying more complex methods. In some cases, a reference method may be the best way to solve a string prediction problem. However, in most cases, the methods of this group are used to test more complex methods. This is done to make sure that the developed method gives a much lower error when predicting.

A. Average methods

In this method, the following value prediction is considered equal to the average calculation of all previous accumulated values. If the accumulated values for X1, …………, XT, then the forecast can be written as Method of constancy or Persistence Method. Persistence method uses a simple assumption that all predicted values of the wind speed or wind power are assumed to be equal to the last observed value. This means that If the measured wind speed and wind power at t are v(t) and P(t), then the forecasting wind speed and wind power at t+Δt can be formulated as the following term: v(t + Δt) = v(t) (1) P(t + Δt) = P(t) (2)

This approach is often used in comparison to assess the effectiveness of other methods, including methods of numerical prediction. The constancy methods often give better results than numerical prediction methods with ultra-short-term forecast range. The disadvantage of this method is the fact that the accuracy of the model decreases rapidly with an increase in the forecast period. Also, this approach will not be able to predict sharp gusts and weakening of the wind, which is extremely important for the intelligent control system of autonomous power complexes.

B. Seasonal method of constancy

This method is a variation of the constancy method for data that have a clear seasonal. Here, the predicted value is assumed to be the last observed value of the same period for which the forecast is made (for example, the same month of the previous year or the same hour of the previous day's day).

C. Offset method

The offset method is another formula for the method of constancy. This method allows the predicted value to increase or decrease over time, where the magnitude of change, called offset, is equal to the average observed change in the accumulated data. T + h time prediction values can be given as where T is the number of measurements. This expression is equivalent to drawing a line between the points of the first and last observation and extrapolating her into the future.

2 - Physical Methods

The value of wind speed is a very important parameter in wind power systems. However, the value of wind speed always depends on atmospheric weather conditions, therefore, in numerical prediction, the initial step in predicting wind speed is predicting future values of such weather variables as: temperature, relative humidity, light intensity, atmospheric pressure. All this is used in the Numerical Weather Prediction Model (NWP). This method is based on kinetic equations that take into account different meteorological data and convert wind speed to wind energy by solving a complex mathematical model. Methods using physical equations and NWP methods are usually applied to supercomputers because they require a very large amount of computing. It also leaves such methods unsuitable for use in embedded systems of autonomous power complexes. Current commercial energy forecasting methods are used for weather forecasting using NWP as input data. Physical systems, using input data from NWP, improve the output data (wind speed forecasts) to situational conditions in ways based on the physics of the lower atmospheric layer. The forecast of the most effective models for this category vary from 48 to 172 hours.

3 - Statistical methods

Data on changes in wind characteristics over the course of time transmitted by sensors from anemometer (wind speed, wind direction, turbulence intensity, air temperature) are ordered numerical values in a timely manner. Thus, data on wind characteristics values can be aggregated over a period of time, as time series. Thus, all known statistical methods can be used for time series analysis and forecasting for wind speed data analysis and forecasting. Statistical methods aim to determine the relationship between the speed of the wind in the time series. The methods of this class are simple, have small computational complexity and provide timely forecast. In many applications of these methods, the difference between the predicted and actual wind speeds in the recent past is used to modify the parameters of the models. The disadvantage of this method is that the error of prediction increases as the time of prediction increases that is, statistical time series and methods of neural networks are mainly intended for short-term forecasts. Statistical models for predicting wind speed have received considerable attention in recent years. The most commonly models used by researches include: Autoregressive Model (AR), Autoregressive Moving Average Model (ARMA), and Autoregressive Integrated Moving Average (ARIMA).

A new statistical method based on the AR model and analysis of independent components was presented, based on the obtained results, the proposed method obviously gives greater accuracy compared with direct predictions. Wind speed prediction for one hour using the Bayesian approach to distinguish wind resources Results showed the development and simulation of the AR model using the Bayesian approach be a useful tool in wind speed and wind forecasting. Auto regressive integrated moving average (ARIMA) model is one of the most common and repeated models used in random time series. The California ISO prototype prediction algorithm for short-term wind generation adopted a modified ARIMA model to calculate a predicted growth / decline factor 2. 5 hours ahead. Four ARMA-based methods were proposed to predict a range of wind velocity and direction. Such proposals have carried out good results, but each of them has its advantages and disadvantages, which should be evaluated for each situation.