基于IWOA-SA-Elman神经网络的短期风电功率预测

SHORT-TERM WIND POWER PREDICTION BASED ON IWOA-SA-ELMAN NEURAL NETWORK

由于风力发电的随机性和不确定性使其短期功率的预测工作十分困难,而神经网络模型依靠其强大的自学习能力在风电功率预测领域有着广泛的应用。但神经网络预测精度受初始权重影响较大,且易出现过拟合的问题。为此构建一种基于改进鲸鱼算法和模拟退火组合优化的Elman神经网络短期风电功率预测模型,模型首先利用改进鲸鱼算法结合模拟退火策略获得高质量神经网络初始权值,接着引入正则化损失函数防止其过拟合,最后以西班牙瓦伦西亚某风电场陆上短期风电功率为研究对象,将该算法与BP、LSTM、Elman、WOA-Elman、IWOA-Elman 5种神经网络算法进行算法性能测试对比,结果表明IWOA-SA-Elman神经网络模型预测误差最小,验证了该算法的合理性和有效性。

Due to the randomness and uncertainty of wind power generation,it is very difficult to predict its short-term power,and the neural network model has a wide range of applications in the field of wind power prediction relying on its powerful self-learning ability.However,the prediction accuracy of neural network is greatly affected by the initial weight,and prone to over-fitting problems.In this paper,an Elman neural network short-term wind power prediction model based on improved whale optimization algorithm(IWOA)and simulated annealing(SA)combined optimization is constructed.Firstly,the improved whale optimization algorithm combined with simulated annealing strategy is used to obtain the initial weights of high-quality neural network,and then the regularization loss function is introduced to prevent overfitting.Finally,the short-term wind power of a wind power plant in Valencia,Spain is taken as the research object.The algorithm is compared with back propagation(BP),long short-term memory(LSTM),Elman,WOA-Elman and IWOAElman neural network algorithms.The results show that the prediction error of IWOA-SA-Elman neural network model is the smallest,which verifies the rationality and effectiveness of the algorithm.