提出了 一种基于遗传算法优化的多头自注意力-长短期记忆(Genetic algorithm multi-head self-attention long short-term memory,GA-MSA-LSTM)神经网络的预测模型,预测空气中的苯污染物的浓度.将MSA加入LSTM神经网络中,使用遗传算法确定最优的"头"个数、时间步长和隐藏层神经元个数,以期提高苯浓度预测的精确度,优化输出结果.数据实验结果显示,GA-MSA-LSTM模型的预测值较LSTM模型和MSA-LSTM模型,能够更好地反映真实值的变化趋势,在均方根误差(Root mean square error,RMSE)、平均 绝对误差(Mean absolute error,MAE)和平均 绝对百分比(Mean absolute percentage error,MAPE)3个评价指标上也体现了其优越性,充分说明了该模型的有效性和可行性.研究表明,该模型具有普遍适用性,也可以应用于不同类型的时间序列数据分析.
A research for predicting benzene concentration based on GA-MSA-LSTM model
A prediction model based on genetic algorithm optimized multi-head self-attention long short-term memory(GA-MSA-LSTM)neural network has been proposed to predict the concentration of benzene pollutants in the air.The MSA model is added to LSTM neural network,which uses a genetic algorithm to determine the optimized number"head",the time steps and hidden layer neurons to improve the accuracy of benzene concentration prediction and optimize the output results.The data experimental results demonstrate that the predicted values of GA-MSA-LSTM model can better reflect the change tend of the true values than the LSTM model and the MSA-LSTM model.It also shows its superiority in the three evaluation indicators of root mean square error(RMSE),mean absolute error(MAE)and mean absolute percentage error(MAPE),which fully demonstrates the effectiveness and feasibility of the model.The research shows that this model has universal applicability and can be widely applied to various types of time series data analysis.
NETL
NSTL
万方数据
