基于长短时记忆网络-纵横交叉算法的含高比例新能源电力市场日前电价预测

Accurate day-ahead electricity price prediction can assist power market participants in making rational decisions. With a high proportion of new energy being integrated into the power system, the difficulty of day-ahead electricity price prediction is continuously increasing. In order to improve the prediction accuracy of day-ahead electricity price in a power market with a high proportion of new energy, a day-ahead electricity price prediction model based on singular spectrum analysis (SSA) and crisscross optimization (CSO) to optimize the long short - term memory network (LSTM) is proposed. Firstly, SSA is used to decompose the original data into a trend sequence, a periodic sequence and a residual sequence. Secondly, an LSTM multi - step prediction model is established for each sub - sequence. Aiming at the problem that the parameters of the fully connected output layer of LSTM are prone to fall into local optima, a strategy of secondary training of LSTM is proposed. After the LSTM model is trained well, the CSO algorithm is used to fine - tune the weight coefficients and biases between the fully connected layers. Finally, all the predicted sequences are superimposed to obtain the final electricity price prediction value. Modeling and prediction are carried out using the data of the Nordic Denmark DK1 power market, and the experimental results show that the proposed method can effectively improve the prediction accuracy of day - ahead electricity price.

精准的日前电价预测能够协助电力市场参与者做出合理的决策。随着高比例新能源接入电力系统,日前电价的预测难度不断加大。为了提升含高比例新能源电力市场日前电价的预测精度,提出了一种基于奇异谱分析(singular spectrum analysis,SSA)和纵横交叉算法(crisscross optimization,CSO)优化长短时记忆网络(long short-term memory,LSTM)的日前电价预测模型。首先,使用SSA将原始数据分解成趋势序列、周期序列和剩余序列;其次,对各子序列建立LSTM多步预测模型,针对LSTM的全连接输出层参数易陷入局部最优的问题,提出了二次训练LSTM策略,在训练好LSTM模型后,使用CSO算法对全连接层间的权系数与偏置进行微调;最后,将所有预测序列进行叠加即得最终的电价预测值。以北欧丹麦DK1电力市场数据进行了建模预测,实验结果表明所提方法能够有效提高日前电价的预测精度。