【摘要】 电力负荷数据的多样性与复杂性，会导致负荷预测过程中出现超参数难以确定、拟合效果较差和预测精度不高等问题。针对以上问题，提出一种基于樽海鞘群算法的融入注意力机制的双向长短期记忆神经网络模型——SSA-AM-BiLSTM模型。该模型使用BiLSTM学习特征的内部变化规律，引入注意力机制为特征进行权重分配，并且利用樽海鞘群算法优化网络超参数。基于具体数据集进行的负荷预测仿真实验表明，相较于GRU、LSTM、AM-BiLSTM和PSO-AM-BiLSTM模型，所提出的SSA-AM-BiLSTM模型的MAPE分别减少了2.15%、1.93%、1.42%和0.45%，并且优化了拟合效果，显著提高了预测精度。更多还原

【Abstract】 Due to the diversity and complexity of power load data,there will be problems in the process of load forecasting,such as difficulty in determining hyper-parameters,poor fitting effect and low forecasting accuracy. To solve the above problems,a Bi-directional Long Short-Term Memory model incorporating attention mechanism based on Salp Swarm Algorithm(SSA-AM-BiLSTM model) is proposed.The model uses BiLSTM to learn the internal change rule of features,introduces the attention mechanism to distribute the weight of features,and uses Salp Swarm Algorithm to optimize the network hyper-parameters. The load forecasting simulation experiment based on specific data sets shows that,compared with GRU,LSTM,AM-BiLSTM and PSO-AM-BiLSTM models,the MAPE of the SSA-AM-BiLSTM model proposed in this paper is reduced by 2.15%,1.93%,1.42% and 0.45% respectively,and the fitting effect is optimized,which significantly improves the forecasting accuracy.更多还原