Demand Response Method Considering Multiple Types of Flexible Loads in Industrial Parks

Jia Cui, Mingze Gao, Xiaoming Zhou, Yang Li, Wei Liu, Jiazheng Tian, Ximing Zhang

Abstract

With the rapid development of the energy internet, the proportion of flexible loads in smart grid is getting much higher than before. It is highly important to model flexible loads based on demand response. Therefore, a new demand response method considering multiple flexible loads is proposed in this paper to character the integrated demand response (IDR) resources. Firstly, a physical process analytical deduction (PPAD) model is proposed to improve the classification of flexible loads in industrial parks. Scenario generation, data point augmentation, and smooth curves under various operating conditions are considered to enhance the applicability of the model. Secondly, in view of the strong volatility and poor modeling effect of Wasserstein-generative adversarial networks (WGAN), an improved WGAN-gradient penalty (IWGAN-GP) model is developed to get a faster convergence speed than traditional WGAN and generate a higher quality samples. Finally, the PPAD and IWGAN-GP models are jointly implemented to reveal the degree of correlation between flexible loads. Numerical examples have been performed with the results proving that the proposed method is significantly better than the existing technologies in reducing load modeling deviation and improving the responsiveness of park loads.

Introduction

As smart grids integrate increasing amounts of renewable energy, the challenge of "peak shaving"—balancing supply and demand during high-usage periods—becomes more acute. Industrial parks are massive energy consumers with significant "flexible loads" (equipment that can shift or reduce power usage), yet modeling these loads accurately is notoriously difficult.

Current methods often fall into two traps: they are either too narrow, focusing only on specific residential loads like electric vehicles, or they rely on purely data-driven models that struggle with the high volatility and nonlinearities of industrial processes. This research proposes a hybrid framework that bridges this gap by combining physical constraints with advanced generative AI to create a "global" model for industrial machinery.

Methodology

Physical Process Analytical Deduction (PPAD)

Instead of treating the industrial park as a "black box," the PPAD model builds representations based on the actual physics of three primary load types:

  1. Rotating Loads: Modeled after steel rolling processes using trapezoidal and gate-shaped functions to represent power cycles.
  2. Heating Loads: Modeled after electric arc furnaces, accounting for "band-like" power fluctuations and harmonic noise inherent in smelting.
  3. Storage Loads: Utilizes a Thevenin dynamic model for lithium-ion batteries to track state-of-charge (SOC) and internal resistance variations.
Thevenin dynamic model used for characterizing lithium-ion battery storage loads within the PPAD framework.

Improved Wasserstein GAN with Gradient Penalty (IWGAN-GP)

To augment and correct industrial datasets, the authors employ Generative Adversarial Networks (GANs). Standard GANs often suffer from training instability and mode collapse. The proposed IWGAN-GP model incorporates a "Gradient Penalty" to enforce Lipschitz continuity, resulting in faster convergence and higher-quality synthetic samples that better reflect real-world volatility.

Architecture of the IWGAN-GP model showing the interaction between the generator and discriminator networks for sample generation.

By feeding high-entropy feature information from the physical PPAD models into the GAN, the AI is constrained by the laws of physics, preventing the generation of unrealistic load profiles. This synergy allows the model to handle data gaps while maintaining physical accuracy.

Experimental Results

The model was validated using real-world flexible load data from Anshan City, Liaoning Province, China (2019). The original data, collected at 1-hour intervals, was augmented to 15-minute intervals using the IWGAN-GP method to better capture transient fluctuations.

Modeling Accuracy

The proposed method was compared against standard benchmarks, including Fourier transforms and Back Propagation Neural Networks (BP-NET). For monthly load modeling, the IWGAN-GP achieved a Root Mean Square Deviation (RMSD) of 1.813 MW, significantly outperforming Fourier (4.667 MW) and BP-NET (7.320 MW).

Responsiveness Improvement

In simulation scenarios covering night, all-day, and daytime operations, the coordinated "H-R-S" (Heating-Rotating-Storage) configuration consistently showed the lowest unresponsiveness volume. The relative error of unresponsiveness was reduced by over 25% compared to existing technologies, demonstrating that coordinating different types of loads is far more effective than managing them in isolation.

Conclusion

This paper demonstrates that a hybrid approach combining physical analytical deduction with generative AI significantly improves the accuracy of demand response modeling in industrial parks. By moving beyond linear indicators and incorporating physics-informed AI, grid operators can better manage the nonlinear impacts of industrial machinery. Future work will focus on optimizing the computational overhead of training these models and deeper integration of industrial production throughput trade-offs into the demand response framework.

Summary updated

Abstract

With the rapid development of the energy internet, the proportion of flexible loads in smart grid is getting much higher than before. It is highly important to model flexible loads based on demand response. Therefore, a new demand response method considering multiple flexible loads is proposed in this paper to character the integrated demand response (IDR) resources. Firstly, a physical process analytical deduction (PPAD) model is proposed to improve the classification of flexible loads in industrial parks. Scenario generation, data point augmentation, and smooth curves under various operating conditions are considered to enhance the applicability of the model. Secondly, in view of the strong volatility and poor modeling effect of Wasserstein-generative adversarial networks (WGAN), an improved WGAN-gradient penalty (IWGAN-GP) model is developed to get a faster convergence speed than traditional WGAN and generate a higher quality samples. Finally, the PPAD and IWGAN-GP models are jointly implemented to reveal the degree of correlation between flexible loads. Meanwhile, an intelligent offline database is built to deal with the impact of nonlinear factors in different response scenarios. Numerical examples have been performed with the results proving that the proposed method is significantly better than the existing technologies in reducing load modeling deviation and improving the responsiveness of park loads.
BibTeX
@misc{cui2022demand,
  author = {Jia Cui and Mingze Gao and Xiaoming Zhou and Yang Li and Wei Liu and Jiazheng Tian and Ximing Zhang},
  title = {{Demand Response Method Considering Multiple Types of Flexible Loads in Industrial Parks}},
  year = {2022},
  eprint = {2205.11743},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.11743},
}
Figure