基于LSTM神经网络的数据驱动空间负荷预测方法  

Data driven spatial load forecasting method based on LSTM neural network

作　　者：李晶晶 张永敏 田桂林 崔胜胜 严洁 LI Jingjing;ZHANG Yongmin;TIAN Guilin;CUI Shengsheng;YAN Jie(State Grid Qinghai Electric Power Company Marketing Service Center,Xining 810000,China)

机构地区：[1]国网青海省电力公司营销服务中心,青海西宁810000

出　　处：《电子设计工程》2022年第22期154-157,164,共5页Electronic Design Engineering

摘　　要：目前的数据驱动空间负荷预测方法数据迭代训练损失值较大、预测速度较慢,难以应对呈几何指数增长的数据量。针对以上问题,以LSTM神经网络为基础提出了一种新的数据驱动空间负荷预测方法,分析神经网络内部的时序,避免数据消沉现象,确定训练数据空间的相关性。根据不同的神经元建立预测模型,通过数据预处理降低采集数据的维度大小,确保数据完整性。同时提供数据控制基础,控制模型输入输出量,统一格式标准,保证模型训练次序,结合LSTM神经网络结构,选择预测方法,完成数据驱动空间负荷预测。实验结果表明,所提方法能够有效减少数据迭代训练损失值,提高预测速度。Current data⁃driven spatial load forecasting methods have large data iteration training losses and slower forecasting speeds,making it difficult to cope with the exponentially increasing amount of data.In response to the above problems,a new data⁃driven spatial load forecasting method based on LSTM neural network is proposed,which analyzes the internal time sequence of the neural network,avoids the phenomenon of data depression,determines the relevance of the training data space,and establishes predictions based on different neurons during model training,data preprocessing is used to reduce the dimensionality of collected data to ensure data integrity,while providing a basis for data control,controlling model input and output,unifying format standards,ensuring model training order,combining LSTM neural network structure,and selecting predictions method to complete data-driven space load forecasting.The experimental results show that the proposed method can effectively reduce the data iterative training loss and improve the prediction speed.

关 键 词：LSTM神经网络 数据分析 驱动预测 负荷研究 空间负荷 数据预测 

分 类 号：TN301[电子电信—物理电子学]