基于改进Prophet算法的碳排放量预测方法

Carbon Emission Prediction Method Based on Improved Prophet Algorithm

摘要: 为了响应“十四五”纲要提出的碳达峰碳中和要求，企业需要有效地控制自身碳排放量，如何基于历史碳排放时序变化特征进一步对未来碳排放趋势进行预估对于企业的双碳规划与减碳工作有着至关重要的作用。本文尝试使用基于改进Prophet的预测模型对时间序列的碳排放数据进行预测，在数据预处理后，先使用Prophet算法进行预训练，将所有时间段内的残差放入LSTM神经网络进行残差的预测，再将预测获得的残差与Prophet原预测值相加，得到改进后的预测值。通过该值与原预测值的对比，能够发现该改进模型相较于原本的Prophet和LSTM模型在预测准确度(基于平均绝对误差指标度量)上分别有12%和82%的提升。

Abstract: In order to respond to the requirements of the carbon peak and carbon neutrality in the “14th Five Year Plan”, Enterprises are asked for effectively controlling their own carbon emissions. How to ad-ditionally predict the carbon emission trend in the future, based on the historical carbon emission time sequential changing data plays a vital role in the dual carbon planning and carbon reduction work of enterprises. This article attempts to use an improved Prophet model to predict carbon emissions data in time series. After data preprocessing, the Prophet algorithm is first used for pre-training, then the residuals in all time periods are put into the LSTM neural network for residu-al prediction. After that, the predicted residuals are added to the original Prophet prediction values to obtain the improved prediction values. By comparing these values with the original prediction values, it can be found that the prediction accuracy (based on the Mean Absolute Percentage Error) of the improved model is improved by 12% and 82% respectively compared with the original Prophet and LSTM models.