Elman short-term wind power prediction based on the optimized seagull algorithm

Accurate prediction of wind farm power output can relieve the pressure of grid frequency regulation and peak regulation and improve grid stability. With the goal of improving power prediction accuracy and reducing overall prediction error, this paper proposes an Elman short-term wind power prediction model on the basis of an optimized seagull algorithm. Firstly, the Elman network is used as the base prediction model, and the seagull algorithm is applied to seek the best values for its weights. Secondly, the chaotic circle mapping with better initial characteristics is improved to equalize its sequence distribution for optimizing the population initialization. Then, to address the lack of local search capability, an optimized iterative approach using the sine cosine operator is used to achieve a balance between local exploitation ability and global search capability. Finally, after simulation and analysis of the actual data set, it is verified that the model has a better prediction effect.


Introduction
China is rich in wind energy resources.The wind power installed capacity has a large number of upgrades every year; the random nature of wind can cause fluctuations in electricity outputs, which brings a critical challenge to grid safety and stable operation.Effective grid stability can be improved by precise and reliable wind power prediction technology [1].
Recently, the integration of optimization algorithms with deep learning models has achieved better prediction results.Chen et al. [2] proposed an algorithm for a multi-objective artificial bee colony and used it to optimize the key parameters of a wavelet neural network; after constituting a prediction model for probabilistic interval prediction of power, it has good prediction results.Zhu [3] introduced the whale optimization algorithm (WOA) to improve the weight thresholds for the Elman model.The study constructed a forecasting model and then used it to forecast the high volatility and strength nonlinearity of wind power time series.It validated model forecasting effects under different forecasting time scales.Alqaness [4] used the seagull algorithm to optimize a dendritic neural regression model on wind power prediction, which has been validated on four wind power datasets for the accuracy of the prediction model.Wang and Liang [5] used the seagull algorithm to optimize BP network weight thresholds for short-term PV output power prediction, which can reduce each evaluation index to effectively improve the prediction accuracy.The above forecasting method of optimizing the base model by optimization algorithm can effectively improve the forecasting accuracy, but there is also the drawback of falling into local optimum.Therefore, optimization methods to further improve the predictive effect of the model need to be sought.

Elman short-term power prediction model based on optimized seagull algorithm
Elman neural network enhances its ability to handle time-series data with good nonlinear prediction capability by using a unique takeup layer [6].For the purpose of promoting the learning capability of the Elman neural network so as to prevent it from falling into partial optimality, its weight search is performed by the seagull optimization algorithm (SOA) [7].To address the problems of low fault tolerance of population update and lack of local search capability in the optimization search process, corresponding measures are used to improve the final prediction effect.

Population initialization for improved circle chaos mapping
For the problem of low fault tolerance of the algorithm population distribution, the chaotic circle mapping with relatively uniform initial sequence distribution is selected for improvement to make the population distribution more uniform.Its mapping function formula is as follows [8].
The improved circle mapping formula is as follows: where mod is the residual function; n is the number of iterations, given as 2000 in the paper.Distribution histograms of circle chaotic maps before and after the improvement are shown in Figure 1, and the improved sequence is more uniformly distributed.Using the improved circle mapping to optimize the population initialization process has enhanced the population diversity in distribution and broadened the search range of the gull species in the space.
(a) Before improvement (b) After improvement Figure 1.Distribution histogram of chaotic circle maps

Improvement of the iterative approach based on the sine and cosine operator
In the standard sine cosine algorithm [9], the position is updated by iterating using the following formula: sin( ) 0.5 cos( ) , 0.5 where Xi t indicates the present solution; Pi t denotes the current optimal individual's position at the i-th dimension when the iteration reaches the t-th time; r1, r2, r3, r4 are four control parameters, where r1=a×(1-t/T), and in the formula, t denotes iteration number at present, T denotes the maximal number of iterations, and a is set to 2 as a constant.
To address the problem of lack of local search capability during the iteration of the seagull optimization algorithm, sine and cosine operators are used to modify its iteration.The modified iteration formula can be denoted as:

Implementation of predictive models
In summary, the specific flow of the Elman short-term power prediction model based on an optimized seagull algorithm can be seen in Figure 2.And for the purpose of having a better quantification of the prediction performance of prediction models, three evaluation indexes, MAE, RMSE, and MAPE, are used to comprehensively evaluate and analyze the prediction results [10].

Analysis of specific experimental algorithms
The actual measurement data of the Gansu Tianshui wind farm is used for simulation verification.The future 30 time points are predicted by modeling.The prediction curves are shown in Figures 3 and 4, and the prediction result indicators are shown in Table 1  From the graphs, it can be concluded that the proposed model decreases in all evaluation indexes, which indicates that the model has a good predictive effect on the whole and can effectively reduce the prediction error and decrease the prediction dispersion.The precision of the model prediction is verified on the prediction results of partial power sequences.The model is verified to have good generalization capability by predicting the full field power.(In the table and figure, the Elman power prediction model under the optimized seagull algorithm is defined as ISOA-Elman for the sake of displaying conciseness).

Conclusion
Accurate prediction of wind farm output is significant to the stable and safe operation of the electricity system.In terms of improving the accuracy of the prediction results, this paper introduces the seagull algorithm based on the Elman neural network.In the course of the forecast, for the problem of low fault tolerance of population location, the improved Circle chaos mapping is proposed to optimize the population initialization process; to address the lack of local search capability, an optimized iterative approach using the sine cosine operator is used to achieve a balance between local exploitation capability and global search ability.The prediction model proposed in the article has a good prediction effect and can reduce the prediction error to a certain extent.It also has good applicability through example verification.In subsequent work, research can be conducted on data preprocessing to consider the impact of data on prediction performance and improve prediction accuracy by improving the data quality of the input model.

Figure 2 .
Figure 2. Wind power prediction model by Elman with optimized seagull algorithm

Figure 3 .Figure 4 .
Figure 3.Comparison curve of prediction results when part wind turbine sequences are included

Table 1 .
. Comparison of performance metrics under different models