摘要
本文通过建立基于麻雀搜索算法(SSA)优化的太阳能发电混合模型,对太阳能发电功率进行预测。首先,通过改进的完全经验模态分解(ICEEMDAN)技术对太阳能发电数据进行预处理,将原始数据分解为固有模态函数(IMFs)和残余部分,从而有效去除噪声和非平稳成分。然后,将预处理后的IMFs和残余数据输入到双向长短时记忆网络(BiLSTM)进行建模,进一步使用ODE-LSTM技术稳定地学习长时间依赖关系。为提升该预测模型的性能,采用SSA进行超参数优化。实验结果表明,通过SSA优化的ODE-BiLSTM模型的均方误差(MSE)为116.78、平均绝对误差(MAE)为206.38,判定系数(R²)为0.7527。在这些指标表现上均优于传统的ARIMA、单一LSTM及随机森林模型,展现出更高的预测精度和鲁棒性。本研究验证了混合模型和优化算法在太阳能发电预测领域的潜力,为智能电网中新能源的高效管理提供了有力支持。
In this paper, solar power generation power is predicted by establishing a hybrid model for solar power generation based on the optimisation of Sparrow Search Algorithm (SSA). In the first place, the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) technique is utilized for preprocessing solar power generation data, decomposing the original data into Intrinsic Mode Functions (IMFs) and residuals to effectively remove noise and non-stationary components. Subsequently, the preprocessed IMFs and residual data are input into a Bidirectional Long Short-Term Memory network (BiLSTM) for modeling, with further steadily learning for long-term dependencies achieved through ODE-LSTM technology. To enhance the performance of the predictive model, SSA is employed for hyper parameter optimization. Experimental results demonstrate that the SSA-optimized ODE-BiLSTM model outperforms traditional ARIMA, standalone LSTM, and Random Forest models in terms of Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R²) metrics, exhibiting superior predictive accuracy and stability. This study confirms the potential of hybrid models and optimization algorithms in the field of solar power generation forecasting, providing robust support for efficient management of renewable energy within smart grids.
出处
《统计学与应用》
2024年第3期904-913,共10页
Statistical and Application
