注意力机制的长短时记忆神经网络航线订座需求预测

针对航线订座需求预测中存在的预测结果不稳定，偏差较大的问题，提出了一种基于注意力机制 的长短时记忆神经网络(Long Short-term Memory Neural Network，LSTM)航线订座需求预测模型。首先， 对采集得到的航线订座需求数据进行数据清洗与指标计算处理，接着，对处理后的指标数据基于注意力机 制做权重分配，然后进行 LSTM 航线订座需求预测模型的建立，从而得到航线订座需求的最终预测结果 值。将训练优化得到的模型应用于国内某航司的航线订座需求预测中，计算出预测结果。实验结果表明， 基于注意力机制的 LSTM 航线订座需求预测模型预测精度较高，以厦门-上海为例，预测结果在与真实值 的对比下，平均绝对误差(Mean Absolute Error，MAE)为 13.1，均方根误差(Root Mean Square Error，RMSE) 为 17.2，相比较于移动平均法，指数平滑法，循环神经网络（Recurrent Neural Network,，RNN），CNN-LSTM 混合模型有较好的预测效果。

Prediction of airline reservation demand based on attention mechanism with long short-term memory neural network

Aiming Aiming at the problem of unstable and large deviations in the forecast of route reservation demand, a Long Short-term Memory Neural Network (LSTM) route reservation demand prediction based on the attention mechanism is proposed. model. First, perform data cleaning and index calculation processing on the collected route reservation demand data, and then assign weights to the processed index data based on the attention mechanism, and then build the LSTM route reservation demand forecast model to obtain the route The final predicted value of the reservation demand. The model obtained by training optimization is applied to a domestic airline company''s route reservation demand forecast, and the forecast result is calculated. The experimental results show that the LSTM route reservation demand prediction model based on the attention mechanism has high prediction accuracy. Taking Xiamen-Shanghai as an example, the prediction result is compared with the real value, and the mean absolute error (MAE) is 13.1 , Root Mean Square Error (RMSE) is 17.2, which has better prediction effect than moving average method, exponential smoothing method, Recurrent Neural Network (RNN), CNN-LSTM hybrid model.