基于高维数据和深度学习的短期电力负荷预测

Short-term Power Load Forecasting Based on High-dimensional Data and Deep Learning

精确高效的短期电力负荷预测在现代化电网建设和资源分配上具有重要的作用和研究意义。随着水电、风电等新能源逐渐接入,电网中数据维度飞速增长,面对具有海量特征的高维数据,短期电力负荷预测技术面临着重大的挑战。传统的短期电力负荷预测技术仅关注了某些特定的特征,因此很难对电力负荷数据在时间维度进行建模;基于循环神经网路的方法虽然能够关注数据中的历史信息,但是因为数据维度过高,导致重要信息丢失、模型无法收敛。为了解决这一问题,提出了一种基于注意力机制的长短期记忆模型(Attention-LSTM),算法使用卷积神经网络从电网高维数据中提取特征,然后使用LSTM对特征进行时序建模;同时引入Attention机制来对历史信息进行加权,减少重要历史信息的丢失。算法在中国东南某地区电力负荷数据上进行测试,预测精度明显优于目前常用的方法。

Accurate and efficient short-term power load forecasting has important role and research significance in modern power grid construction and resource allocation.With the gradual access of new energy sources such as hydropower and wind power,the data dimension in the power grid has grown rapidly.In the face of massive high-dimensional data,short-term power load forecasting technology is facing major challenges.The traditional short-term power load forecasting technology only focuses on certain specific characteristics,so it is difficult to model the power load data in the time dimension;although the method based on the recurrent neural network can pay attention to the historical information in the data,but because of the data dimension If it is too high,important information will be lost and the model will not converge.In order to solve this problem,an attention-based long-short-term memory model(Attention-LSTM)is proposed.The algorithm employ a deep convolutional neural network to extract features from the high-dimensional data of the power grid,and then uses LSTM to model the features in time series;Attention mechanism is also introduced to weight historical information to reduce the loss of important historical information.The algorithm is tested on power load data in a region in southeast China,and the prediction accuracy is significantly better than the currently used methods.