一种基于序列分解和神经网络的供电成本预测方法

A Method for Power Supply Cost Prediction Based on Time-Series Decomposition and Neural Network

摘要: 随着国内各企业生产环境的变化，精益财务管理，智慧财务运营越来越重要，寻找一种更有效的成本计算方法成为各企业突破困境的方向，这对于企业发展规划、战略部署有着非常重要的作用。本文提出一种新颖的Prophet与LSTNet的组合模型，应用于电网成本分摊，该模型首先使用序列分解模型将数据解构，生成平稳光滑的子序列，以提升后续神经网络模型的训练效果，其次在LSTNet神经网络中加入注意力机制，学习序列的长短周期模式，充分发挥神经网络模型的非线性优势。实验结果表明，本文模型能够良好地预测具有长短周期性的非平稳成本序列。

Abstract: With the changes in the production environment of domestic enterprises, lean financial management and smart financial operation are becoming more and more important. Finding a more effective cost calculation method has become the direction for enterprises to break through the dilemma, which plays a very important role in enterprise development planning and strategic deployment. In this paper, a novel combination model of prophet and LSTNet is proposed for power grid cost allocation. Firstly, the model uses the sequence decomposition model to deconstruct the data and generate smooth sub-sequences to improve the training effect of the subsequent neural network model. And then, the attention mechanism is added to the LSTNet neural network to learn the long and short period pattern of the sequence, giving full play to the nonlinear advantage of neural network model. The experimental results show that the proposed model can well predict the non-stationary cost series with long and short periodicity.