新能源电网中考虑特征选择的Bi-LSTM网络短期负荷预测

Short-term Load Forecasting in Renewable Energy Grid Based on Bi-directional Long Short-term Memory Network Considering Feature Selection

新能源电网中负荷对各特征因素更为敏感,当面对海量特征数据时,短期负荷预测方法面临着新的挑战。针对含有高维特征数据的新能源电网,提出一种考虑特征选择的双向长短期记忆(Bi-LSTM)网络短期负荷预测方法。先将样本数据按密度进行聚类后映射到权重诱导空间中,通过定义一种数据结构,以间隔之和最大为目标函数。为实现解空间的稀疏性,将正则项添加到目标函数中,并采用梯度下降法求解特征权值。经过预试验确定特征选择阈值等超参数,从而选出所需的特征因素。最后,使用Bi-LSTM网络基于选择后的数据进行负荷预测。以中国某地区新能源电网为例,验证了该方法的有效性,结果表明其与传统方法相比,具有更好的准确性和适用性。

Loads in the renewable energy grid is more sensitive to various feature factors,and there is a new challenge for shortterm load forecasting method when facing massive feature data.Aiming at the renewable energy grid with high-dimensional feature data,a short-term load forecasting method based on the bi-directional long short-term memory(Bi-LSTM)network considering feature selection is proposed.The sample data are firstly clustered and mapped into the weight-induced space according to density.By defining a structure of interval measurement data,the maximum interval sum is used as the objective function.In order to achieve the sparsity of the solution space,a regular term is added into the objective function,and the feature weights are solved by the gradient descent algorithm.Hyper-parameters such as the feature selection threshold are determined by pre-tests,and then the required feature factors are selected.Finally,the Bi-LSTM network is used to carry out load forecasting based on the selected data.Taking the renewable energy grid in certain area of China as an example,the effectiveness of the proposed method is verified.And compared with the traditional methods,the proposed method is more accurate and applicable.