基于LightGBM及LSTM融合的科技园区短期负荷预测

Short Term Load Forecasting of Science and Technology Park Based on Integration of LightGBM and LSTM

在售电侧逐渐市场化的环境下,园区的电力负荷预测对电力市场及园区自身经济运行具有重要意义。为了提高科技园区负荷预测的精度和可靠性,提出采用K折交叉验证的LightGBM算法与LSTM算法的多特征融合算法。首先利用K折交叉验证的LightGBM算法训练第一层特征的预测模型,将其预测结果作为下一层LSTM模型训练的附加特征值。最后将两层模型预测的负荷值根据预测误差加权平均成最终的负荷预测值,结合数据算例表明,采用K折交叉验证的LightGBM算法提高模型的泛化能力,模型融合比单一的模型预测更具优势,能在一定程度上提高预测精度,减小预测误差。

In the environment of gradual marketization of the power sales side,the power load forecasting of the park is of great significance to the power market and the economic operation of the park itself.In order to improve the prediction accuracy and reliability of load forecasting in science and technology parks,a multifeature fusion algorithm of LightGBM and LSTM algorithm with K-fold cross validation is proposed in this paper.Firstly,the LightGBM algorithm with K-fold cross validation is used to train the prediction model of the first layer features,and the prediction results are used as the additional eigenvalues of the next layer LSTM model training.Finally,the load value predicted by the two-layer model is weighted and averaged into the final load prediction value according to the prediction error.Combined with the data example,the LightGBM algorithm using K-fold cross validation improves the generalization ability of the model,and the model fusion has more advantages than the single model prediction,which can improve the prediction accuracy and reduce the prediction error to a certain extent.