基于深度学习的智能电网电力负荷精准预测方法研究

Research on accurate prediction method of electric load for smart grid based on deep-learning

针对因电力系统中短期电力负荷预测不准确,导致智能电网无法有效协调电能生产、运输、分配的问题,为降低因过载或低负荷所造成的资源浪费,减少不必要的二氧化碳排放,本文提出了一种新的深度学习方法来解决此类电网短时电力负荷可靠预测问题。该方法利用卷积神经网络建立能量预测计算模型,利用CNN自适应数据特征挖掘特性、量化电力不确定性,利用丢弃正则化对深度网络结构进行优化,采用深度森林对所提取的数据特征进行学习并建立预测模型,以实现对电力负荷的准确预测,解决电力随机波动造成的现有预测方法精度下降问题。经过基于实际负载数据验证,在电力负荷不确定波动情况下,该方法能准确预测电力负荷,且精度比目前较为流行的方法高,有望成为解决智能电网核心问题的重要技术支撑。

In order to solve the problem of inaccurate short-term power load prediction in power systems,which leads to the inability of smart grids to effectively coordinate the production,transportation and distribution of electrical energy,and to reduce the waste of resources caused by overload or low load and unnecessary CO_(2)emissions.This paper proposes a new deep learning method to solve the problem of reliable short-term power load prediction in such grids.This new method uses the convolution neural network(CNN)to establish the prediction calculation model,where the adaptive data mining ability of the CNN is utilized to quantify the uncertainty of electric load,the dropout regularization is adopted to optimize the structure of the deep network,and the random forest is employed to learn the extracted features from the CNN to establish the prediction model to realize precision prediction of electric load and solve the forecasting accuracy degradation caused by random load fluctuation.After verification based on the actual load data,the method can accurately predict the power load under the uncertain fluctuation of power load,and the accuracy is higher than the current popular methods,which is expected to be an important technical support to solve the core problems of smart grid.