【摘要】 为解决现有光伏功率预测方法存在效率低和非线性预测精度不高的问题，提出一种混合光伏功率预测模型。首先通过支持向量机（SVM）提取模块降低输入数据维度；然后利用平衡迭代规约和聚类（BIRCH）模块挖掘数据中的信息，划分特征库；最后根据光伏功率的波动特性，建立其对应的双向长短期记忆网络（BILSTM）预测模型。将提出的混合模型应用于欧洲中期天气预报中心（ECMWF）提供的真实数据集上进行预测，通过与8种主流的机器学习算法相比，该模型在测试数据集上的平均绝对误差（MAE）和均方误差（MSE）分别降低了4.3%～59.75%和35.65%～78.29%。此外，混合模型还具有良好的可解释性，使其在电力行业有广泛的应用前景。更多还原

【Abstract】 To tackle the challenges of inefficiency and inaccuracy in nonlinear photovoltaic power forecasting,a novel hybrid photovoltaic power forecasting model is introduced. Firstly,the input data dimensionality is reduced through the support vector machine(SVM)extraction module;Then, the balanced iterative reducing and clustering using hierarchies(BIRCH)clustering module is used to mine the information from the data and segment the feature library. Finally,a bi-directional long short-term memory network(bilstm)forecasting model is established according to the fluctuation characteristics of photovoltaic power output. When tested on realworld datasets from European Centre for Medium-Range Weather Forecasts(ECMWF),the proposed hybrid model significantly outperforms eight mainstream machine learning algorithms,mean absolute error(MAE)and mean squared error(MSE)are reduced by4.3% to 59.75% and 35.65% to 78.29% respectively. The model’s strong interpretability further underscores its potential for the widespread application in power industry.更多还原