Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hybrid Transformer Network for Different Horizons-based Enriched Wind Speed Forecasting

M. Madhiarasan, Partha Pratim Roy

TL;DR

Wind speed forecasting across multiple horizons is critical for reliable modern power systems but is challenged by nonstationarity and volatility. The authors propose an astute hybrid framework (ICEEMDAN-TNF-MLPN-RECS) that decomposes wind speed into IMFs with ICEEMDAN, forecasts each IMF with a Transformer Network, fuses the results to form a primary forecast, and applies a Multilayer Perceptron to predict residual errors for correction. Case studies on the Kethanur wind farm at hub heights of 65 m and 80 m show exceptionally low error indices, with MAEs in the range of $10^{-7}$ and strong horizon-specific performance, especially for very short-term forecasts. The approach demonstrates a scalable, horizon-aware forecasting paradigm that can reduce grid operator burden and enhance wind integration, with potential extensions to multivariate and real-time applications.

Abstract

Highly accurate different horizon-based wind speed forecasting facilitates a better modern power system. This paper proposed a novel astute hybrid wind speed forecasting model and applied it to different horizons. The proposed hybrid forecasting model decomposes the original wind speed data into IMFs (Intrinsic Mode Function) using Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). We fed the obtained subseries from ICEEMDAN to the transformer network. Each transformer network computes the forecast subseries and then passes to the fusion phase. Get the primary wind speed forecasting from the fusion of individual transformer network forecast subseries. Estimate the residual error values and predict errors using a multilayer perceptron neural network. The forecast error is added to the primary forecast wind speed to leverage the high accuracy of wind speed forecasting. Comparative analysis with real-time Kethanur, India wind farm dataset results reveals the proposed ICEEMDAN-TNF-MLPN-RECS hybrid model's superior performance with MAE=1.7096*10^-07, MAPE=2.8416*10^-06, MRE=2.8416*10^-08, MSE=5.0206*10^-14, and RMSE=2.2407*10^-07 for case study 1 and MAE=6.1565*10^-07, MAPE=9.5005*10^-06, MRE=9.5005*10^-08, MSE=8.9289*10^-13, and RMSE=9.4493*10^-07 for case study 2 enriched wind speed forecasting than state-of-the-art methods and reduces the burden on the power system engineer.

Hybrid Transformer Network for Different Horizons-based Enriched Wind Speed Forecasting

TL;DR

Wind speed forecasting across multiple horizons is critical for reliable modern power systems but is challenged by nonstationarity and volatility. The authors propose an astute hybrid framework (ICEEMDAN-TNF-MLPN-RECS) that decomposes wind speed into IMFs with ICEEMDAN, forecasts each IMF with a Transformer Network, fuses the results to form a primary forecast, and applies a Multilayer Perceptron to predict residual errors for correction. Case studies on the Kethanur wind farm at hub heights of 65 m and 80 m show exceptionally low error indices, with MAEs in the range of and strong horizon-specific performance, especially for very short-term forecasts. The approach demonstrates a scalable, horizon-aware forecasting paradigm that can reduce grid operator burden and enhance wind integration, with potential extensions to multivariate and real-time applications.

Abstract

Highly accurate different horizon-based wind speed forecasting facilitates a better modern power system. This paper proposed a novel astute hybrid wind speed forecasting model and applied it to different horizons. The proposed hybrid forecasting model decomposes the original wind speed data into IMFs (Intrinsic Mode Function) using Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). We fed the obtained subseries from ICEEMDAN to the transformer network. Each transformer network computes the forecast subseries and then passes to the fusion phase. Get the primary wind speed forecasting from the fusion of individual transformer network forecast subseries. Estimate the residual error values and predict errors using a multilayer perceptron neural network. The forecast error is added to the primary forecast wind speed to leverage the high accuracy of wind speed forecasting. Comparative analysis with real-time Kethanur, India wind farm dataset results reveals the proposed ICEEMDAN-TNF-MLPN-RECS hybrid model's superior performance with MAE=1.7096*10^-07, MAPE=2.8416*10^-06, MRE=2.8416*10^-08, MSE=5.0206*10^-14, and RMSE=2.2407*10^-07 for case study 1 and MAE=6.1565*10^-07, MAPE=9.5005*10^-06, MRE=9.5005*10^-08, MSE=8.9289*10^-13, and RMSE=9.4493*10^-07 for case study 2 enriched wind speed forecasting than state-of-the-art methods and reduces the burden on the power system engineer.
Paper Structure (20 sections, 37 equations, 8 figures, 6 tables)

This paper contains 20 sections, 37 equations, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. PROBLEM DESCRIPTION AND RELATED WORK CONCERN TO WIND SPEED FORECASTING
  3. TECHNICAL CONCEPT AND FRAMEWORK OF THE PROPOSED HYBRID WIND SPEED FORECASTING MODEL DESIGN
  4. Data Source
  5. Performance Error Index
  6. Proposed Astute Hybrid Approach Procedural Steps
  7. Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise
  8. Transformer Network
  9. Multilayer Perceptron Neural Networks
  10. Residual Error Correction Strategy
  11. PROPOSED FORECASTING MODEL EXPERIMENTAL RESULT ANALYSIS AND DISCUSSION
  12. Case Study 1: Proposed ICEEMDAN-TNF-MLPN-RECS Model Experimental Result
  13. Case Study 2: Proposed ICEEMDAN-TNF-MLPN-RECS Model Experimental Result
  14. COMPARATIVE PERFORMANCE ANALYSIS
  15. Performance Analysis for Case Study 1
  16. ...and 5 more sections

Figures (8)

  • Figure 1: The framework of the proposed hybrid wind speed forecasting model. The acquired wind farm data is split into two sets, one is used for the training phase, and the other is used for the testing phase. Then, decompose the wind speed series into subseries (IMFs and residuals) using ICEEMDAN. Each subseries is fed as input to the transformer network (TN) and performs the forecasting of the subseries. The fusion of all forecast subseries is done, reconstructed, and obtained the primary forecast wind speed. For each data point based residual error is computed that is fed as input to the MLPN network to perform the future residual error values forecasting because it can learn any complex problem. Finally, the forecast wind speed values are corrected by adding the forecast residual error values to the primary forecast wind speed. Thereafter, the performance is verified using performance error index computation. If acceptable, record the value, else do so until reaching the stopping criteria (max iterations).
  • Figure 2: Wind farm dataset collected location. The star in red color shows the wind farm location, Kethanur (latitude of 10.9153 N and longitude of 77.2657 N).
  • Figure 3: A detailed framework of the transformer network. Encoder and decoder connected through attention. The encoder doesn't care about the sequence order, but position embedding takes care of the sequence order. Encoder layer self-attention sub-layer, fully connected feed-forward sub-layer. Normalization is between each sub-layer. The output of the encoder is a dimension vector. A decoder comprises an input layer, masked self-attention, feed-forward sub-layer, and output layer.
  • Figure 4: Case Study 1 and 2: ICEEMDAN based on decomposed subseries. ICEEMDAN decomposed the collected case study 1 and 2 whole wind speed series into ten intrinsic mode functions (IMFs) and one residual subseries.
  • Figure 5: Case Study 1 and 2: Logarithmic scale comparison of forecast wind speed and ground truth wind speed. The proposed hybrid approach-based forecast wind speed results are precise and similar to the ground truth wind speed values.
  • ...and 3 more figures