米浩然,申森林,卫紫任,等.改进麻雀搜索算法优化的短期电力负荷CNN-BiLSTM融合预测模型[J].河南理工大学学报(自然科学版)，doi:10.16186/j.cnki.1673-9787. （ 2025120085  ）..

MI H R, SHEN S L, WEI Z R, et al．An improved SSA-optimized CNN-BiLSTM integrated model for short-term electricity load forecasting[J]. Journal of Henan Polytechnic University( Natural Science) ， doi: 10.16186/j.cnki.1673-9787. （ 2025120085  ）..

改进麻雀搜索算法优化的短期电力负荷CNN-BiLSTM融合预测模型(网络首发)

米浩然，申森林，卫紫任，李巍，李亚檑

国网河南省电力公司焦作供电公司，河南 焦作 454000

摘要: 目的 为提高电力负荷预测精度，降低预测难度，进行改进麻雀搜索算法优化的短期电力负荷CNN-BiLSTM融合预测模型研究。方法以河南省某地区110 kV变电站2022年的电力负荷数据为基础，提出一种基于改进麻雀搜索算法优化的卷积神经网络-双向长短期记忆网络（ISSA-CNN-BiLSTM）融合预测模型。首先，以日期信息、气象数据以及相似日负荷数据等构建特征集作为模型的输入变量，并对数据的异常值、缺失值和数据不规范等问题进行预处理。其次，采用CNN模型有效提取特征图中连续/非连续数据之间的潜在联系并构造时序序列的特征向量，再基于BiLSTM模型对其进行训练及预测。同时利用正余弦策略和柯西变异策略对传统的SSA进行强化，实现对模型超参数的自动寻优，得到预测结果。结果 仿真结果表明：相比较LSTM、PSO-LSTM、CNN-LSTM和Transformer-BiLSTM预测模型，ISSA-CNN-BiLSTM模型的预测误差MAPE降低为1.570 3%，平均绝对误差为0.427 6，均方根误差为0.645 3，预测精度得到明显提高。结论 ISSA-CNN-BiLSTM融合预测模型能够克服负荷预测随机因素的影响，具有更强的负荷时间相关性捕捉能力和泛化能力，可以为实际电力系统的高效运行和优化调度提供新的思路和方法。

关键词:短期负荷预测；CNN-BiLSTM融合模型；改进麻雀搜索算法；预测精度

doi: 10.16186/j.cnki.1673-9787. （ 2025120085   ）

基金项目:国网河南省电力公司科技项目(5217C0250001)

收稿日期：2025-12-26

修回日期：2026-04-28

网络首发日期：2026-06-05

An improved SSA-optimized CNN-BiLSTM integrated model for short-term electricity load forecasting (Online)

Mi Haoran, Shen Senlin, Wei Ziren, Li Wei, Li Yalei

Henan Jiaozuo Power Supply Company, Jiaozuo 454000,Henan,China；

Abstract: Objectives Accurate power load forecasting is one of the effective means to ensure stable operation of power grid and rational resource allocation. However, due to the high penetration of renewable energy, as well as the influence of meteorological conditions and diversified load structures, the uncertainty and difficulty of power load forecasting have further increased. Methods Based on the power load data of a 110kV substation in a certain region of Henan Province in 2022, an ISSA-CNN-BiLSTM integrated forecasting model optimized by an improved sparrow search algorithm (ISSA) was investigated. Firstly, by considering date information, meteorological data, and similar daily load data, a feature set was constructed as input variables for the model and the outliers, missing values and non-standard data formats were preprocessed. Secondly, the CNN model was employed to effectively extract the potential connections between sequential/non-sequential data in the feature map and construct the feature vectors of the time series. Then, BiLSTM model was used for training and prediction and the traditional SSA was enhanced by incorporating sine-cosine and Cauchy mutation strategies to achieve automatic optimization of model hyperparameters. Results Compared to the LSTM, PSO-LSTM, CNN-LSTM and Transformer-BiLSTM forecasting models, the ISSA-CNN-BiLSTM model exhibited the lower forecasting error, with the mean absolute percentage error (MAPE) as low as 1.570 3%, mean absolute error (MAE) of 0.427 6 and root mean square error (RMSE) of 0.645 3, which indicated a significant improvement in forecasting accuracy. Conclusions The ISSA-CNN-BiLSTM integrated forecasting model effectively mitigates the impact of stochastic factors in load forecasting. It exhibits enhanced capability in capturing temporal load correlations and possesses stronger generalization performance. This model can provide novel insights and methodologies for the efficient operation and optimal scheduling of actual power systems.

Key words: load forecasting; CNN-BiLSTM fusion model; ISSA; prediction accuracy