CEEMDAN组合DISPSO-LSTM的短期电力负荷预测

Short-Term Power Load Forecasting Based on CEEMDAN and DISPSO-LSTM

针对短期电力负荷预测精度低与准确性差的问题,设计了CEEMDAN-DISPSO-LSTM混合预测模型。运用自适应噪声的完全集成经验模式分解算法以获取平稳负荷序列,采用改进的动态个体-群体粒子群算法求解长短期神经网络参数的最优值,利用最优参数值构建的LSTM网络进行负荷预测。仿真结果表明:与其他模型相比,该混合预测模型取得了较低的电力负荷预测误差,其误差评价指标RMSE、MAE和MAPE分别为43.71 MW,28.53 MW和0.81%,有效地提高了短期电力负荷的预测精度和准确性。

A CEEMDAN-DISPSO-LSTM hybrid forecasting model is designed to improve the precision and accuracy of short-term power load forecasting.The complete ensemble empirical mode decomposition with adaptive noise is used to obtain a stable load sequence,the improved dynamic individual-social particle swarm optimization algorithm is used to find the optimal value of the long-term and short-term neural network parameters,and then the LSTM network is built on the optimal parameter value for load forecasting.The simulation results show that compared with other models,the hybrid forecasting model works well with smaller power load forecasting errors.Its RMSE,MAE and MAPE are 43.71 MW,28.53 MW and 0.81%,respectively.It is concluded that the new model improves the precision and accuracy of power load forecasting.