Abstract:Abstract Wind power generation holds immense importance in addressing the issue of global energy shortage, while precise wind power forecasting proves essential for effective management and dependable operation of wind power networks. This study introduces a hybrid deep learning model, encompassing complete ensemble empirical mode decomposition (CEEMD), sample entropy (SE), extreme learning machine (ELM) and time convolutional network (TCN), for accurately predicting short-term wind power output. First, CEEMD decomposed the original wind power into multiple submodes, which effectively reduced the series volatility. Then, the SE of intrinsic mode function sequence is calculated, and the subsequences with similar complexity are superimposed to reduce the calculation cost, improve the simulation accuracy and reduce the noise of the original wind power sequence. Secondly, the ELM model is established for each submode, and the prediction error of BiLSTM is predicted again using TCN to improve the efficiency and prediction performance of the hybrid model. Finally, the outcomes of each individual submode are amalgamated to yield the ultimate prediction outcome. To showcase the efficacy and dominance of the error compensation technique, several comparison models were established in the experiment. The results demonstrated that the suggested hybrid model exhibits superior predictive accuracy in the domain of wind power prediction. Compared with the comparison model, the improvement in MAPE and RMSE was 60.50 and 77.74%, respectively.

What problem does this paper attempt to address?

This paper attempts to solve the problems of accuracy and efficiency in wind power generation prediction. Specifically, the research aims to improve the accuracy of short - term wind power output prediction by introducing a hybrid deep - learning model (including Complete Ensemble Empirical Mode Decomposition (CEEMD), Sample Entropy (SE), Extreme Learning Machine (ELM) and Temporal Convolutional Network (TCN)). The paper points out that due to its randomness, intermittency and volatility, wind power generation brings great challenges to large - scale wind power grid - connection, so accurate wind power generation prediction is crucial for the effective management and reliable operation of the power grid. ### Main problems 1. **Volatility and non - linearity of wind power generation**: Wind power generation data has obvious non - linearity and non - stationarity, which makes it very difficult to predict directly. 2. **Computational complexity**: Directly using a single model for prediction requires a large amount of computational resources and is time - consuming. 3. **Prediction accuracy**: Existing prediction methods still have room for improvement in prediction accuracy, especially when dealing with the complex characteristics of wind power generation. ### Solutions 1. **CEEMD decomposition**: First, use the CEEMD method to decompose the original wind power generation time series into multiple sub - sequences, each sub - sequence contains different frequency components, thereby reducing the volatility of the sequence. 2. **Sample Entropy (SE)**: Calculate the sample entropy of each sub - sequence, stack sub - sequences with similar complexity to reduce computational cost, improve simulation accuracy, and reduce the noise of the original wind power generation sequence. 3. **ELM model**: Use the ELM model to predict each sub - sequence and evaluate the deviation between the predicted value and the actual observed value. 4. **TCN model**: Use the TCN model to predict the error sequence of the ELM model and integrate the error compensation result into the final prediction result, thereby improving the overall prediction performance. ### Experimental results Through comparative experiments with a variety of existing models (such as CEEMD - ELM - TCN, CEEMD - SE - ELM, ELM - TCN, ELM and TCN), the results show that the proposed CEEMD - SE - ELM - TCN hybrid model has a significant improvement in prediction accuracy. Specific indicators include Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), among which MAPE and RMSE are improved by 60.50% and 77.74% respectively. ### Conclusion This research proposes an effective wind power generation prediction model. By combining multiple deep - learning techniques, it significantly improves the accuracy and efficiency of short - term wind power generation prediction and provides strong support for large - scale grid - connection of wind power generation.

Wind power error compensation prediction model based on CEEMD-SE-ELM-TCN

Wind Power Short-Term Forecasting Hybrid Model Based on CEEMD-SE Method

Short-Term Wind Power Prediction Based on CEEMDAN-SE and Bidirectional LSTM Neural Network with Markov Chain

A multi-stage predicting methodology based on data decomposition and error correction for ultra-short-term wind energy prediction

A Multiscale Hybrid Wind Power Prediction Model Based on Least Squares Support Vector Regression–Regularized Extreme Learning Machine–Multi-Head Attention–Bidirectional Gated Recurrent Unit and Data Decomposition

Wind Speed Short-Term Prediction Based on Empirical Wavelet Transform, Recurrent Neural Network and Error Correction

Short-term wind power prediction based on ICEEMDAN decomposition and BiTCN–BiGRU-multi-head self-attention model

A dual-scale deep learning model based on ELM-BiLSTM and improved reptile search algorithm for wind power prediction

A short-term wind power prediction model based on CEEMD and WOA-KELM

Short-term wind power forecasting using the hybrid model of improved variational mode decomposition and Correntropy Long Short -term memory neural network

Short-term wind power prediction based on modal reconstruction and CNN-BiLSTM

Ultra-Short-Term Wind Power Prediction Based on eEEMD-LSTM

Multistep short-term wind power forecasting model based on secondary decomposition, the kernel principal component analysis, an enhanced arithmetic optimization algorithm, and error correction

Multi-step Wind Speed Prediction Based on an Improved Multi-Objective Seagull Optimization Algorithm and a Multi-Kernel Extreme Learning Machine.

A novel model based on CEEMDAN, IWOA, and LSTM for ultra-short-term wind power forecasting

Forecasting model for short-term wind speed using robust local mean decomposition, deep neural networks, intelligent algorithm, and error correction

Research on a Deep Ensemble Learning Model for the Ultra-Short-Term Probabilistic Prediction of Wind Power

Wind Power Prediction Based on EMD-KPCA-BiLSTM-ATT Model

Hybrid Model with Secondary Decomposition, Randomforest Algorithm, Clustering Analysis and Long Short Memory Network Principal Computing for Short-Term Wind Power Forecasting on Multiple Scales

Application of hybrid model based on CEEMDAN, SVD, PSO to wind energy prediction

An Integrated CEEMDAN to Optimize Deep Long Short-Term Memory Model for Wind Speed Forecasting