NWP辅助复合神经网络预测误差修正的风储系统日前上报策略  

DAY-AHEAD REPORTING STRATEGY OF WIND STORAGE SYSTEM BASED ON NWP ASSISTED COMPOSITE NEURAL NETWORK PREDICTION ERROR CORRECTION

作　　者：李翠萍[1] 张冰[1] 李军徽 朱辉 朱星旭[1] 何俐 Li Cuiping;Zhang Bing;Li Junhui;Zhu Hui;Zhu Xingxu;He Li(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology,Ministry of Education(Northeast Electric Power University),Jilin 132012,China;China Three Gorges Renewables(Group)Co.,Ltd.,Beijing 100053,China)

机构地区：[1]现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学),吉林132012 [2]中国三峡新能源(集团)股份有限公司,北京100053

出　　处：《太阳能学报》2024年第10期86-96,共11页Acta Energiae Solaris Sinica

基　　金：吉林省自然科学基金联合基金(YDZJ202101ZYTS152)。

摘　　要：新能源电站出力存在强波动性导致巨额偏差考核支出,因此基于数值天气预报(NWP)和复合深度学习算法,提出一种计及误差预测修正的风储系统日前上报策略。首先通过改进的组合数据预处理算法对数据进行清洗以降低后续预测难度,建立基于分段式收敛粒子群算法(PCPSO)参数寻优的长短期记忆网络(LSTM)对分量分别进行预测,重构预测结果获取原预测曲线。其次考虑预测误差及NWP信息导入多输入反向传播神经网络(MIBP)获取误差预测曲线,使用非参数核密度函数修订该预测误差曲线后,以储能跟踪误差最小和储能全局调控能力最高为目的模拟储能运行获取最佳储能动作曲线,且叠加原预测曲线和最佳储能动作曲线获取最终日前上报曲线。最后通过仿真分析验证了上报策略的正确性与可行性。The output of new energy power stations has strong fluctuations that lead to huge deviation assessment expenditures.Therefore,based on numerical weather prediction(NWP)and composite deep learning algorithms,a day-ahead reporting strategy for wind-storage combined system that takes into account error prediction corrections is proposed.Firstly,the improved combined data preprocessing algorithm is used to clean the data to reduce the difficulty of subsequent predictions,and a long short-term memory network prediction network(LSTM)based on piecewise convergent particle swarm optimization(PCPSO)parameter optimization is established to predict the components respectively.Reconstruct the forecast results to obtain the original forecast curve.Secondly,the prediction error curve is obtained by considering the prediction error and the NWP information import multi-input backpropagation neural network(MIBP).After revising the prediction error curve using the non-parametric kernel density function,the optimal energy storage action curve is obtained by simulating the energy storage operation for the purpose of minimizing the energy storage tracking error and maximizing the global control ability of energy storage.Superimpose the original prediction curve and the optimal energy storage action curve to obtain the final day-ahead reporting curve.Finally,the correctness and feasibility of the reporting strategy are verified by simulation analysis.

关 键 词：风电 深度神经网络 粒子群优化 储能 日前上报策略 预测误差特征 NWP信息 

分 类 号：TM614[电气工程—电力系统及自动化]