Considering Multi-scale Inputs and Optimizing the Short-term Load Prediction of CNN-BiGRU
Short-term load forecasting is an important prerequisite for market planning and can effectively ensure the safe and stable operation of the power system.In order to solve the problems of strong randomness and large volatility of power load,a short-term power load prediction method based on CEEMDAN-PE-SSA-CNN-BiGRU was proposed.Firstly,for the complex and changeable power load data,CEEMDAN(fully adaptive noise set empirical mode decomposition)was used as a subsequence,PE(permutation entropy)of the subsequences was calculated,and the subsequences with similar entropy values are reconstructed to obtain a new sequence,which reduces the influence of the non-stationary series of the original data on the prediction accuracy and optimizes the computational cost.Secondly,the characteristics of the recombinant sequence were analyzed,and the multi-scale input was used to extract the data features by using the CNN(convolutional neural network),which was input to BIGRU(bidirectional gated recurrent unit network)for training,and SSA(sparrow search algorithm)was used to optimize the hyperparameters.Finally,the normalized new series data was input into the prediction model to obtain the prediction sequence,and the final prediction data was obtained by summarizing all the prediction sequences.Using the model proposed,taking the Spanish electricity load as an example,compared with the single model and the combined model,the experimental results show that the model has a better prediction effect.
万方数据
维普
