【摘要】 利用辽宁省气象局提供的地面观测降水资料，构建了具有多元时间特征的降水数据，采用变分模态分解方法(variational mode decomposition, VMD)组合遗传算法(genetic algorithm, GA)对双向长短时记忆神经网络(bidirectional long short-term memory, BiLSTM)进行优化，建立基于VMD-GA-BiLSTM的月降水量预测模型，并与BiLSTM、VMD-BiLSTM和GA-BiLSTM进行实验对比，应用均方根误差(root mean square error, RMSE)、平均绝对误差(mean absolute error, MAE)和R~2决定系数作为模型评价指标。实验结果表明：VMD-GA-BiLSTM模型的R~2决定系数达到0.98,RMSE和MAE表现更低，验证了VMD-GA-BiLSTM模型在时间序列预测方面的优势。更多还原

【Abstract】 Based on the ground observation precipitation data provided by Liaoning Meteorological Bureau, the precipitation data with multiple temporal characteristics were constructed, the VMD method and genetic algorithm(GA) were used to optimize the BiLSTM, and the monthly precipitation prediction model based on VMD-GA-BiLSTM was established, and the experimental comparison was carried out with BiLSTM, VMD-BiLSTM and GA-BiLSTM, and the determination coefficients of RMSE, MAE and R~2 were used as the model evaluation indexes. Experimental results showed that the R~2 determination coefficient of the VMD-GA-BiLSTM model reached 0.98, and the RMSE and MAE performance were lower, which verified the advantages of the VMD-GA-BiLSTM model in time series forecasting.更多还原