An Integrated Energy System Operating Scenarios Generator Based on Generative Adversarial Network

Suyang Zhou, Zijian Hu, Zhi Li Zhong, Di He, Meng JiangSustainability 2019 · DOI

TL;DR

This paper introduces a Generative Adversarial Network (GAN) to create realistic operating scenarios for Integrated Energy Systems (IES), addressing the scarcity of diverse data needed for data-driven control strategies. The generated data significantly improves the performance of a Q-learning-based control agent for renewable energy and storage systems, demonstrating the GAN's effectiveness in enriching training datasets.

What Problem Does This Paper Address?

Integrated Energy Systems (IES) are crucial for balancing energy security and environmental concerns, but their complex nature and the interdependencies of various energy sources make them challenging to model. While data-driven approaches offer a promising alternative to traditional modeling, they are hampered by the lack of sufficient and diverse operational data. Specifically, existing datasets for IES often lack the variety of scenarios (e.g., normal operation, fault conditions) required to train robust control strategies. This paper aims to bridge this gap by generating synthetic, yet realistic, IES operational data to enhance data-driven methods.

Methods

The core of the paper is a Generative Adversarial Network (GAN) designed to generate IES operating scenarios. The GAN consists of two main components:

  • Generator (G): A deconvolutional neural network that takes random noise as input and outputs synthetic IES operational data (energy price, load, generator output). Its goal is to produce data that is indistinguishable from real data.
  • Discriminator (D): A convolutional neural network that tries to differentiate between real operational data and the data generated by G.

The generator and discriminator are trained adversarially. The generator learns to produce more realistic data to fool the discriminator, while the discriminator improves its ability to detect fake data. The paper also proposes a model-free control strategy using the Q-learning algorithm for a renewable energy and storage system. This algorithm learns optimal control actions by interacting with the environment and receiving rewards, and it requires a rich dataset for effective training. The generated data is used to augment the original dataset for training this Q-learning agent.

Datasets

  • Training and Evaluation Data: The authors collected real load, wind, solar, energy price, and weather data from France. This data serves as the "original data set" for training the GAN and evaluating the Q-learning agent.
  • Generated Data: The paper introduces a novel benchmark of synthetic IES operating scenarios generated by the proposed GAN. This generated data is used to expand the original dataset, creating an "extended data set" for training.

Key Insights and Ideas

  • GANs for IES Scenario Generation: The paper demonstrates that GANs are a viable and effective tool for generating realistic and diverse operational scenarios for complex Integrated Energy Systems, overcoming the limitations of scarce real-world data.
  • Data Augmentation for Reinforcement Learning: The generated data significantly improves the performance of a Q-learning agent. This highlights that data diversity, not just quantity, is crucial for training robust RL agents in dynamic systems like IES.
  • Capturing Temporal Dynamics: The GAN is shown to capture time-sequence characteristics, producing generated data that exhibits similar visual and statistical properties to the original data, including trends and autocorrelations.
  • Enriching Scenarios for Robust Control: By generating scenarios that might not be present in the original dataset, the GAN helps the Q-learning agent learn control strategies for a wider range of operating conditions, leading to better decision-making and system stability.

Key Results

  • Data Quality: Visualizations and quantitative metrics (CDF similarity of 83.43%, autocorrelation coefficient similarity of 88.72%) show that the generated data closely resembles the original data distribution and temporal characteristics.
  • Q-Learning Performance: Training a Q-learning agent with the GAN-augmented dataset resulted in a higher average hourly income (an increase of approximately 12,000€) compared to training with the original dataset alone.
  • System Stability: The agent trained on the expanded dataset also led to more stable energy storage fluctuations, reducing excessive power usage and associated penalty fees.

Limitations and Future Work

  • Limited Factors Considered: The current GAN model primarily focuses on generating operational data. Future work should incorporate more factors, such as IES network architecture and the transport characteristics of different energy types, to create a more comprehensive integrated energy system scenario generation method.
  • Integration with Classical Methods: The authors suggest that data-driven methods can complement, rather than replace, classical analysis or pretreatment techniques.
  • Fault Scenario Generation: The paper infers that the lack of specific fault scenarios in training data can hinder model accuracy. Future work could focus on generating generated fault samples to improve model generality and fault detection capabilities.

Summary updated

Abstract

The convergence of energy security and environmental protection has given birth to the development of integrated energy systems (IES). However, the different physical characteristics and complex coupling of different energy sources have deeply troubled researchers. With the rapid development of AI and big data, some attempts to apply data-driven methods to IES have been made. Data-driven technologies aim to abandon complex IES modeling, instead mining the mapping relationships between different parameters based on massive volumes of operating data. However, integrated energy system construction is still in the initial stage of development and operational data are difficult to obtain, or the operational scenarios contained in the data are not enough to support data-driven technologies. In this paper, we first propose an IES operating scenario generator, based on a Generative Adversarial Network (GAN), to produce high quality IES operational data, including energy price, load, and generator output. We estimate the quality of the generated data, in both visual and quantitative aspects. Secondly, we propose a control strategy based on the Q-learning algorithm for a renewable energy and storage system with high uncertainty. The agent can accurately map between the control strategy and the operating states. Furthermore, we use the original data set and the expanded data set to train an agent; the latter works better, confirming that the generated data complements the original data set and enriches the running scenarios.
BibTeX
@inproceedings{zhou2019integrated,
  author = {Suyang Zhou and Zijian Hu and Zhi Li Zhong and Di He and Meng Jiang},
  title = {{An Integrated Energy System Operating Scenarios Generator Based on Generative Adversarial Network}},
  booktitle = {Sustainability},
  year = {2019},
  doi = {10.3390/su11236699},
}
Figure