Daily Average Wind Power Interval Forecasts Based on an Optimal Adaptive-Network-Based Fuzzy Inference System and Singular Spectrum Analysis

Zhongrong Zhang, Yiliao Song, Feng Liu, Jinpeng LiuSustainability 2016 · DOI

TL;DR

This paper introduces IFASF, a novel hybrid model that forecasts the range (upper and lower bounds) of daily average wind power. By combining Singular Spectrum Analysis (SSA) for noise reduction and an optimized Adaptive-Network-Based Fuzzy Inference System (ANFIS) with a Firefly Algorithm for parameter tuning, IFASF provides more reliable uncertainty information than traditional point forecasts, outperforming eight benchmark models across multiple wind farms.

What Problem Does This Paper Address?

The increasing integration of wind energy into power grids is hampered by its inherent variability and unpredictability. While accurate point forecasts (predicting a single value) are important, grid operators and independent power producers (IPPs) also need to understand the uncertainty associated with these forecasts. This uncertainty is crucial for operational planning, managing reserve capacity, and participating in electricity markets. Existing research has often focused on deterministic forecasts, leaving a gap in providing robust interval forecasts that capture the range of possible wind power generation. This paper aims to fill this gap by developing a method that directly estimates the upper and lower bounds of daily average wind power, thereby avoiding the accumulation of errors from deterministic forecasting steps.

Methods

The IFASF model is a hybrid approach built upon several key techniques:

  • Singular Spectrum Analysis (SSA): Used as a pre-processing step to denoise the wind power time series. SSA decomposes the time series into its constituent components, allowing for the separation of signal from noise, which is crucial for improving the accuracy of subsequent forecasting.
  • Adaptive-Network-Based Fuzzy Inference System (ANFIS): A neuro-fuzzy system that combines the learning capabilities of neural networks with the rule-based reasoning of fuzzy logic. ANFIS is used here to model the complex, non-linear relationships in wind power data.
  • Subtractive Clustering: Employed to automatically determine the optimal number of rules and membership functions for the ANFIS, reducing the need for expert knowledge.
  • Firefly Algorithm (FA): A metaheuristic optimization algorithm inspired by the flashing behavior of fireflies. In this paper, FA is used to optimize the neighborhood radius parameter of the subtractive clustering algorithm, which in turn influences the ANFIS's performance. This optimization is key to achieving an "optimal" ANFIS.
  • Interval Forecast (IF) Method: A novel approach that directly predicts the upper and lower bounds of wind power without relying on an intermediate deterministic forecast. This method is designed to avoid error accumulation.

The IFASF model integrates these components by first using SSA to denoise the data, then employing subtractive clustering (optimized by FA) to set up the ANFIS, and finally training two ANFIS models (one for the upper bound, one for the lower bound) using the interval forecast method.

Datasets

The study utilized wind power data from three different wind farms in Gansu Province, China. For each farm, 25 weeks of data (175 days) were collected and used for training and validation. The paper does not introduce new datasets but uses real-world wind power data to evaluate its proposed model against existing benchmarks.

Key Insights and Ideas

  • Direct Interval Forecasting is Key: The paper emphasizes that directly forecasting the upper and lower bounds of wind power, rather than first predicting a point value and then estimating uncertainty, significantly reduces accumulated errors and improves accuracy.
  • SSA as a Powerful Pre-processor: The inclusion of SSA as a denoising step demonstrably improves the performance of the ANFIS model, highlighting its effectiveness in handling the noisy nature of wind power data.
  • Firefly Algorithm for Optimal ANFIS: Using the Firefly Algorithm to tune the subtractive clustering's neighborhood radius is a novel way to optimize the ANFIS, leading to better interval forecasts.
  • Uncertainty is Crucial for Operations: The research underscores the practical importance of interval forecasts for grid operators and IPPs, providing essential information for planning and risk management beyond simple point predictions.
  • 70% Interval is More Meaningful: The study consistently found that 70% interval forecasts were more informative and achievable than 80% or 90% intervals, likely due to the inherent volatility of wind power.

Key Results

The IFASF model consistently outperformed eight benchmark models (including ARIMA, BPNN, ELM, ANFIS, and their SSA-enhanced versions) across three wind farms and for 70%, 80%, and 90% interval forecasts.

  • Improved Coverage: IFASF achieved higher Interval Forecast Coverage Probability (IFCP) compared to other models, meaning its predicted intervals more frequently contained the actual wind power values. For instance, in Experiment I, IFASF achieved an average IFCP of 76% for the 70% interval, significantly higher than ANFIS-SSA (72%) and ANFIS (65.14%).
  • Reduced Interval Width: The model also demonstrated good performance in terms of Interval Forecast Normalized Average Width (IFNAW) and Coverage Width-based Criteria (CWC), indicating a balance between accurate coverage and reasonably narrow prediction intervals.
  • Robustness: The model showed good adaptability across different wind farms and time horizons (daily average and six-hour average), suggesting its general applicability.

Limitations and Future Work

  • 90% Interval Forecasts are Challenging: The paper consistently notes that 90% interval forecasts are difficult to achieve with high coverage probability due to the inherent volatility of wind power. This suggests that for practical applications, confidence levels smaller than 10% might not be advisable.
  • Focus on Daily Average: While the paper addresses the need for daily average wind power forecasts, future work could explore more granular forecasting (e.g., hourly or sub-hourly) for finer operational control.
  • Data Limitations: The study primarily used wind power data. Incorporating other meteorological variables (like wind speed, direction, temperature, etc.) alongside the proposed hybrid model could potentially further enhance forecasting accuracy.
  • Computational Cost: While not explicitly detailed, complex hybrid models like IFASF can sometimes have higher computational demands, which might be a consideration for real-time applications.

The authors suggest that their work provides a strong foundation for future research in interval forecasting of wind power, offering a valuable reference for statistical and AI-based interval forecasting methods.

Summary updated

Abstract

Wind energy has increasingly played a vital role in mitigating conventional resource shortages. Nevertheless, the stochastic nature of wind poses a great challenge when attempting to find an accurate forecasting model for wind power. Therefore, precise wind power forecasts are of primary importance to solve operational, planning and economic problems in the growing wind power scenario. Previous research has focused efforts on the deterministic forecast of wind power values, but less attention has been paid to providing information about wind energy. Based on an optimal Adaptive-Network-Based Fuzzy Inference System (ANFIS) and Singular Spectrum Analysis (SSA), this paper develops a hybrid uncertainty forecasting model, IFASF (Interval Forecast-ANFIS-SSA-Firefly Alogorithm), to obtain the upper and lower bounds of daily average wind power, which is beneficial for the practical operation of both the grid company and independent power producers. To strengthen the practical ability of this developed model, this paper presents a comparison between IFASF and other benchmarks, which provides a general reference for this aspect for statistical or artificially intelligent interval forecast methods. The comparison results show that the developed model outperforms eight benchmarks and has a satisfactory forecasting effectiveness in three different wind farms with two time horizons.
BibTeX
@inproceedings{zhang2016daily,
  author = {Zhongrong Zhang and Yiliao Song and Feng Liu and Jinpeng Liu},
  title = {{Daily Average Wind Power Interval Forecasts Based on an Optimal Adaptive-Network-Based Fuzzy Inference System and Singular Spectrum Analysis}},
  booktitle = {Sustainability},
  year = {2016},
  doi = {10.3390/su8020125},
}
Figure