【摘要】 电力负荷的准确预测能有效保持电网运行的稳定性，提高经济效益和社会效益。为了提高负荷预测的精准度，首先利用麻雀搜索算法(SSA)和变分模态分解(VMD)对输入的原始负荷进行模态分解，降低了电力负荷数据随机性与非平稳性；然后利用双向长短期记忆-随机森林(BiLSTM-RF)组合模型对分解后的子模态进行特征提取和预测。对某地区公开数据的性能验证与模型对比分析结果表明，改进的BiLSTM-RF(+BiLSTM-RF)组合模型在决定系数(R~2)、均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分误差(MAPE)这四个预测精度指标方面分别达到了0.973、300.31、134.91、0.037。与传统的支持向量机(SVM)、长短期记忆(LSTM)网络、门控循环单元(GRU)等单一预测方法，以及未改进的BiLSTM-RF组合预测方法相比，+BiLSTM-RF组合模型有更好的预测表现。更多还原

【Abstract】 Accurate prediction of power load can effectively maintain the stability of power grid operation and improve economic and social benefits. To improve the accuracy of load forecasting, the input raw load is first modally decomposed using sparrow search algorithm(SSA) and variational mode decomposition(VMD), which reduces the stochasticity and non-stationarity of the power load data; and then the decomposed sub modalities are feature extracted and predicted using the bi-directional long short-term memory-random forest(BiLSTM-RF) combination model. The results of the performance validation and model comparison analysis on the public data of a region show that the improved BiLSTM-RF(+BiLSTM-RF) combined model achieves 0.973,300.31,134.91,0.037 in the four prediction accuracy indexes, namely, the coefficient of determination(R~2), the root mean square error(RMSE), the mean absolute error(MAE), and the mean absolute percentage error(MAPE). Compared with the traditional single prediction methods such as support vector machine(SVM), long short-term memory(LSTM) network, gated recurrent unit(GRU), and the unimproved BiLSTM-RF combination prediction method, the +BiLSTM-RF combination model has better prediction result.更多还原