Scholarlink

<p>To solve the problems of incomplete feature extraction and long-term dependence of traditional methods on load time series, a ICEEMDAN-FE-CNN-BiLSTM short-term load forecasting model (SHSSA-ICEEMDAN-FE-CNN-BiLSTM) based on multi-strategy improved sparrow algorithm (SHSSA) was proposed. In order to solve the problems of slow convergence speed and low population diversity of the basic sparrow algorithm, the best point set strategy, golden Levy flight strategy, <i>t</i>-distributed disturbance strategy and dynamic scout allocation strategy are proposed to improve it. Firstly, the original load is decomposed using ICEEMDAN-FE, and then the historical load, meteorology, day type information, and electricity price integrated feature vector are constructed by CNN as the input data of BiLSTM, and the multi-strategy sparrow optimization algorithm is utilized to search for optimal combinations of hyperparameters in the BiLSTM to improve the prediction accuracy of the model. Then considering the problem of LSTM model in short-term forecasting of load only based on the current single moment, it is proposed to utilize BiLSTM with added sliding time window for the extraction and learning of load forward and backward bidirectional time series features, and the progress of predicting the next moment of load through the loads of multiple historical moments is achieved. The SHSSA-ICEEMDAN-FE-CNN-BiLSTM prediction model is verified to be more accurate and effective in load prediction by comparing with other prediction model test results. Through modal decomposition and hybrid neural network, the comprehensive analysis and accurate prediction of multi-feature influencing factors of power load are innovatively realized.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-term load forecasting of ICEEMDAN-FE and CNN-BiLSTM based on multi-strategy improved sparrow algorithm optimization

  • 期刊论文

  • 发表日期：

    2025年6月12日

查看全文
  • Renzhe Tang,
  • Taoyan Zhao,
  • Jiangtao Cao

摘要

To solve the problems of incomplete feature extraction and long-term dependence of traditional methods on load time series, a ICEEMDAN-FE-CNN-BiLSTM short-term load forecasting model (SHSSA-ICEEMDAN-FE-CNN-BiLSTM) based on multi-strategy improved sparrow algorithm (SHSSA) was proposed. In order to solve the problems of slow convergence speed and low population diversity of the basic sparrow algorithm, the best point set strategy, golden Levy flight strategy, t-distributed disturbance strategy and dynamic scout allocation strategy are proposed to improve it. Firstly, the original load is decomposed using ICEEMDAN-FE, and then the historical load, meteorology, day type information, and electricity price integrated feature vector are constructed by CNN as the input data of BiLSTM, and the multi-strategy sparrow optimization algorithm is utilized to search for optimal combinations of hyperparameters in the BiLSTM to improve the prediction accuracy of the model. Then considering the problem of LSTM model in short-term forecasting of load only based on the current single moment, it is proposed to utilize BiLSTM with added sliding time window for the extraction and learning of load forward and backward bidirectional time series features, and the progress of predicting the next moment of load through the loads of multiple historical moments is achieved. The SHSSA-ICEEMDAN-FE-CNN-BiLSTM prediction model is verified to be more accurate and effective in load prediction by comparing with other prediction model test results. Through modal decomposition and hybrid neural network, the comprehensive analysis and accurate prediction of multi-feature influencing factors of power load are innovatively realized.