Hybrid Forecasting Approach Based on GRNN Neural Network and SVR Machine for Electricity Demand Forecasting

Weide Li, Xuan Yang, Hao Li, Lili SuEnergies 2017 · DOI

TL;DR

This paper introduces a novel hybrid model, EEMD-SCGRNN-PSVR, for electricity demand forecasting. By decomposing the demand into waveform and trend components and applying specialized techniques to each, the model achieves significantly improved accuracy and robustness compared to existing methods.

What Problem Does This Paper Address?

Accurate electricity demand forecasting is crucial for efficient electricity market operation and power system management. However, electricity demand is inherently non-linear and influenced by numerous unpredictable factors like weather, holidays, and price fluctuations, making traditional forecasting methods struggle to achieve high accuracy. Existing models often fail to adequately address noise, seasonality, and complex underlying patterns in the data.

Methods

The proposed EEMD-SCGRNN-PSVR model is a sophisticated hybrid approach that tackles the forecasting problem in three stages:

  • Stage 1: Decomposition and Noise Reduction:
    • Ensemble Empirical Mode Decomposition (EEMD): This technique is used to decompose the original electricity demand signal into several Intrinsic Mode Functions (IMFs) and a residual. EEMD is an advancement over EMD, designed to mitigate mode mixing issues by adding white noise.
  • Partial Auto Correlation Function (PACF): PACF is employed to identify and filter out noise from the decomposed IMFs. The authors assume that the first IMF often contains noise, and the remaining IMFs represent the "waveform" component, while the residual represents the "trend" component.
  • Stage 2: Single Forecasting:
    • Waveform Component Forecasting:
      • Seasonal Adjustment (S): The waveform component is further processed to remove seasonal influences.
  • General Regression Neural Network (GRNN) with Cross-Validation (CV): GRNN, a non-linear regression model, is used to forecast the seasonally adjusted waveform component. Cross-validation is applied to optimize GRNN's parameters, specifically the smoothing factor (sigma), to achieve the best prediction accuracy. The seasonal index is then re-added to obtain the final waveform forecast.
  • Trend Component Forecasting: * Support Vector Regression Machine (SVR) optimized by Particle Swarm Optimization (PSO): SVR is used to model the trend component. The Particle Swarm Optimization (PSO) algorithm is employed to find the optimal parameters (like 'c' and 'sigma') for the SVR model, enhancing its predictive power for the trend.
  • Stage 3: Ensemble Forecasting:
    • The final electricity demand forecast is obtained by simply summing the predicted waveform component and the predicted trend component. This ensemble approach leverages the strengths of both individual forecasting modules.

Datasets

The study utilizes two real-world electricity demand datasets from Australia:

  • New South Wales (NSW): Data collected from April 1, 2015, to June 23, 2015.
  • Victorian State (VIC): Data collected from August 1, 2015, to October 23, 2015.

Both datasets consist of half-hourly electricity demand values. For each dataset, 11 weeks of data were used for training, and the subsequent week was used for testing. The paper does not introduce new benchmark datasets but uses these established sources for evaluation.

Key Insights and Ideas

  • Decomposition is Key: The core idea is that by separating the complex electricity demand signal into distinct waveform and trend components, each can be modeled more effectively using specialized techniques. This avoids the "curse of dimensionality" and noise interference that can plague direct forecasting of the raw signal.
  • Synergy of Techniques: The paper highlights the power of combining diverse techniques: EEMD for robust decomposition, PACF for noise identification, GRNN for non-linear waveform modeling, and SVR with PSO for trend forecasting. Each component addresses a specific challenge in the data.
  • Seasonal Adjustment Matters: Explicitly addressing seasonal patterns in the waveform component significantly improves the GRNN's ability to capture cyclical behavior, leading to better predictions.
  • Ensemble for Robustness: Combining the forecasts of the waveform and trend components through simple summation (ensemble) proves more accurate and robust than any single component's forecast or simpler hybrid models.

Key Results

The EEMD-SCGRNN-PSVR model demonstrated superior performance compared to three other models:

  • CGRNN: A baseline model using GRNN with cross-validation on the original data.
  • EMD-CGRNN-PSVR: Uses EMD for decomposition but lacks seasonal adjustment.
  • EEMD-CGRNN-PSVR: Uses EEMD but lacks seasonal adjustment.

Quantitative results (using RMSE, MAE, and MAPE) showed that EEMD-SCGRNN-PSVR achieved:

  • In NSW: Reduced RMSE by 29.2%, MAE by 27.8%, and MAPE by 30.5% compared to CGRNN.
  • In VIC: Reduced RMSE by 22.5%, MAE by 23.9%, and MAPE by 26.5% compared to CGRNN.

The model also exhibited better robustness, with lower MAPE fluctuations and bounds in simulations, indicating more stable performance across different scenarios.

Limitations and Future Work

  • Computational Complexity: The authors acknowledge that the EEMD-SCGRNN-PSVR model is more complex and computationally intensive than simpler models, requiring more time and memory. However, they suggest that advancements in computing power will mitigate this issue.
  • Generalizability: While the model performed well on two Australian datasets, further validation on diverse datasets from different regions and with varying characteristics would be beneficial.
  • Future Work: The authors suggest that this hybrid approach could be extended to forecast other related variables such as electricity price, wind speed, and tourism demand. They also imply that further research into optimizing the ensemble combination method could yield even better results.

Summary updated

Abstract

Accurate electric power demand forecasting plays a key role in electricity markets and power systems. The electric power demand is usually a non-linear problem due to various unknown reasons, which make it difficult to get accurate prediction by traditional methods. The purpose of this paper is to propose a novel hybrid forecasting method for managing and scheduling the electricity power. EEMD-SCGRNN-PSVR, the proposed new method, combines ensemble empirical mode decomposition (EEMD), seasonal adjustment (S), cross validation (C), general regression neural network (GRNN) and support vector regression machine optimized by the particle swarm optimization algorithm (PSVR). The main idea of EEMD-SCGRNN-PSVR is respectively to forecast waveform and trend component that hidden in demand series to substitute directly forecasting original electric demand. EEMD-SCGRNN-PSVR is used to predict the one week ahead half-hour’s electricity demand in two data sets (New South Wales (NSW) and Victorian State (VIC) in Australia). Experimental results show that the new hybrid model outperforms the other three models in terms of forecasting accuracy and model robustness.
BibTeX
@inproceedings{li2017hybrid,
  author = {Weide Li and Xuan Yang and Hao Li and Lili Su},
  title = {{Hybrid Forecasting Approach Based on GRNN Neural Network and SVR Machine for Electricity Demand Forecasting}},
  booktitle = {Energies},
  year = {2017},
  doi = {10.3390/en10010044},
}
Figure