为充分利用某油田地区的太阳能与风力资源，验证该地区风光互补发电系统施行的可行性，提出了一种将粒子群算法（Particle Swarm Optimization，PSO）和长短期记忆神经网络(Long Short-Term Memory，LSTM)以及注意力机制（Attention Mechanism）相结合的风光互补系统发电功率预测方法。引入粒子群算法对模型的隐藏神经元个数和初始学习率进行优化，获取最优参数并结合LSTM捕捉历史序列的相关性。引入注意力机制，通过权重分配聚焦关键时间步，自动识别重要特征维度，提高预测可靠性。仿真结果表明，文章所提模型可以有效捕捉数据的变化趋势，模型的平均绝对值误差、均方误差、均方根误差、决定系数等评价指标均优于LSTM、LSTM-Attention模型，证明PSO-LSTM-Attention模型在发电功率预测性能上表现优异。经实验验证，该系统风光互补发电预测模型均方根误差RMSE值稳定在0.2357，平均绝对误差MAE值为0.1456，拟合优度R2可达0.9435，相比于LSTM、LSTM-Attention模型均有所提升，预测效果较好。

In order to make full use of the solar and wind resources in an oilfield area and verify the feasibility of wind-solar complementary power generation system in this area, the article proposes a method combining Particle Swarm Optimization and Long Short-Term Memory and Attention mechanism to predict the power generation of wind-solar complementary system. Firstly, the particle swarm algorithm is introduced to optimize the number of hidden neurons and the initial learning rate of the model, to obtain the optimal parameters and to capture the correlation of historical sequences by combining with LSTM. Finally, the attention mechanism is introduced to focus on the key time step through weight allocation, automatically identify the important feature dimensions, and improve the prediction reliability. Simulation results show that the model proposed in the article can effectively capture the data trends, and the average absolute value error, mean square error, root mean square error, and coefficient of determination of the model are better than LSTM and LSTM-Attention, which proves that the PSO-LSTM-Attention model performs well in generating power prediction performance. After experimental verification, RMSE value of the wind and solar hybrid power generation prediction model of this system is stabilized at 0.2357, MAE value is 0.1456, and R2 can be up to 0.9435, which is improved compared with LSTM and LSTM-Attention, and the prediction effect is better.

辽宁省教育厅基本科研项目（JYTMS20231450）；辽宁省教育厅基本科研项目（LJ212510148022）；辽宁石油化工大学科学研究基金（2023XJJL-009）。