基于LSTM网络与误差补偿的预测模型

Prediction Model Based on LSTM Network and Error Compensation

随着时代发展,空气质量逐渐受到人们的重视,所以对未来空气污染物的变化预测显得尤为重要。首先,针对PM2.5的非线性变化以及变化所具有的周期性,选取完整年度数据进行训练和预测,使用对非线性序列数据拟合效果较好的LSTM网络作为初步预测模型,选择合适的滑动窗口,使用训练数据,建立了LSTM网络预测模型。由于LSTM网络预测结果中存在相邻年份误差分布相似,但整年分布不均匀的现象,使用FCM对训练数据及误差进行模糊聚类。通过聚类中心,对当前预测数据进行分类,并利用聚类结果,得到当前预测数据的误差补偿值,对LSTM网络的当前预测结果进行误差补偿,得到最终预测结果。最后,通过合肥2017年至2021年的空气污染数据对该方法进行了验证,结果表明,所建模型的效果优于其他对比模型,具有一定的可行性与有效性。

With the development of the times, people pay more and more attention to air quality, so it is particularly important to predict the change of air pollution in the future. Firstly, aiming at the nonlinear change and periodicity of PM2.5,select the compete annual data for training and prediction. LSTM network, which has excellent fitting effect on the nonlinear sequence data, is used as the preliminary prediction model. The LSTM network prediction model is established by selecting the appropriate sliding window and using the training data. For the phenomenon of similar error distribution in adjacent years but uneven distribution in the whole year in the prediction results of LSTM network, the training data and errors are fuzzy clustered by FCM. Through the clustering results, the current prediction data are classified, and the error compensation value of the current prediction data is obtained by using the clustering center. Using the error compensation value to modify the prediction result of LSTM network, the final prediction result is obtained. Finally, the proposed method is verified by the air pollutant data set of Hefei from 2017 to 2021. The results show that the model is better than other comparison models, which is feasible and effective.