Short-term prediction of wind power based on grey wolf optimization Elman neural network algorithm

The intermittence and fugitiveness of wind power is one of the significance ekements affecting the quality of wind power. Aiming at the random characteristics of wind power, the output power of short-term wind electricity is forecasted by collecting historical wind speed, power and other parameters of wind farm. A model of Elman network optimized by grey wolf arithmetic (GWO) is proposed to realize short period accurate forecast for wind power. The pattern is verified and analyzed on matlab. Compared with other prediction models, the prediction accuracy is higher, which shows the effectiveness and advancement of the model.


Introduction
Globally, wind power, as a non-polluting renewable energy, is experiencing rapid development [1] .Reasonable and accurate forecast can offer support for power system, which is helpful to make full use of wind resources, so as to improve the economic operation of power system.
Reference [2] optimized the empirical mode decomposition model by firefly algorithm to predict wind power output; in Reference [3], the SSA is used to optimize the BP neural network, which solves the problem that the BP network is easy to be trapped in local optimum.Reference [4] used GA to optimize Elman network to anticipate wind power output, but this measure has the problems of slow iteration speed and low prediction accuracy.

Wind power data processing
The output power of the fan can be expressed as: In the formula:   is rated wind speed;   is rated power;   is the wind energy utilisation factor; ρ is atmospheric density; A is the effective area of blade sweep; v is the measured wind velocity.
It goes to show from equation (1) that the output power is mainly related to the wind velocity, air density and the effective area of blade sweep.Among them, wind speed is one of the main influencing factors.When the wind velocity increases to twice the original, the output power of the fan may increase to eight times before the wind speed changes.Therefore, this paper analyzes the wind speed as the main influencing factor.
In order to reduce the absolute value difference between the data, the data is normalized, and the data value needs to be uniformly scaled to the range of [-1,1], and then input into the network model.The formula of data normalization is as follows: where x is the sequence to be normalized ;  ，  are the minimum and maximum.

Grey wolf optimization algorithm
Grey wolf optimization algorithm is an intelligence optimization arithmetic.The algorithm has the features of strong global search ability, good convergence properties, parameters less and easy fulfilment.
(1) Search and encircle prey.The mathematical formula is described as (3): ( 1) ( ) In the formula: T is the iterations;   () is the position vector of prey; X(T) is the placement of T generation grey wolf; X(T+1) represents the position vector of the T+1 generation grey wolf.
( 1) (3) Attack target.In the process of wolf hunting, the location information can be used to validate the best position for prey, and then gradually narrow the scope of the prey.Among them, the indicators mainly used to evaluate the training model effect and prediction performance include: MAPE; MSE; RMSE.The smaller the error value, the higher the model prediciton accurancy [5] .
Mean absolute percentage error: ( ) Root mean square error: In the formula:   is power prediction value;   is the real value; i is the actual number.

Example analysis
In this paper, the simulation is carried out on the MATLAB2018b platform.A total of 1800 sets of data points of numerical weather forecast and real operational data provided by a wind farm in Ningxia in 2017 are selected as samples.The Elman neural network with a structure of 3-7(7)-1 is created.The test results are shown in Figure 2 and Figure3: In this paper, four models are used for analysis and comparison, namely BP, Elman, GWO-BP and GWO-Elman.The same sample data set is input during the training process, and the prediction results and performance evaluation are contrasted.The training results of are shown in table 1 below.It goes to show that the three types of eeor values of the GWO-Elman model in the four prediction models are the smallest compared with the other three.The prediction accuracy is 16.876 % higher than that of the BP and 7.701 % higher than that of the Elman.

Conclusion
The short-term prediction of wind power can provide certain reference value for power grid dispatching and energy management, and help to realize the stable operation and reliable power supply.Accurate wind power forecasting is the key factor to ensure the normal operation of wind power generation system.The GWO algorithm is used to optimize the Elman neural network.The GWO-Elman is proposed to solve the issues of Elman neural network, such as accessible to be trapped in local optimum, slow rate of convergence and susceptibility to the beginning value.Through simulation analysis and comparison with other models, it is shown that the grey wolf optimization algorithm plays a major role in adjusting the Elman neural network, effectively in improving the stability and precision of the network.

4 ComparisonFigure 2 .Figure 3 .
Figure 2. Comparison of forecast and raw values.Distribution of forecast error

Table 1 .
Comparison of four models.