Model-Free Renewable Scenario Generation Using Generative Adversarial Networks

Abstract

Scenario generation is an important step in the operation and planning of power systems with high renewable penetrations. In this work, we proposed a data-driven approach for scenario generation using generative adversarial networks, which is based on two interconnected deep neural networks. Compared with existing methods based on probabilistic models that are often hard to scale or sample from, our method is datadriven, and captures renewable energy production patterns in both temporal and spatial dimensions for a large number of correlated resources. For validation, we use wind and solar timesseries data from NREL integration data sets. We demonstrate that the proposed method is able to generate realistic wind and photovoltaic power profiles with full diversity of behaviors. We also illustrate how to generate scenarios based on different conditions of interest by using labeled data during training. For example, scenarios can be conditioned on weather events (e.g. high wind day, intense ramp events or large forecasts errors) or time of the year (e,g. solar generation for a day in July). Because of the feedforward nature of the neural networks, scenarios can be generated extremely efficiently without sophisticated sampling techniques.

Synthetic Dynamic PMU Data Generation: A Generative Adversarial Network Approach

“…The synthetic PMU data should be meaningful in such a way that the data generated at all the buses and the linkages should satisfy the spatial and temporal correlation property of the real data. This is significantly different from the standard application of GAN for image generation [10] or even wind and solar data generation in power system [7] which can be more appropriately be modelled as statistical phenomenons. However, in the PMU data generation, the spatial correlation is through the Kirchhoff's voltage and current laws which have to be satisfied at each time deterministically.…”

“…However, there are only few efforts applying GAN to power systems. One representative is the renewable scenario generation [7], which generates realistic-looking wind and photovoltaic power profiles with rich diversities under various scenarios of interests. The first effort applying GAN to PMU data generation is presented in [8], which learns the dynamics represented by the training dataset sampled from a single PMU and then produces a single synthetic PMU data stream.…”

“…As an alternative, there has been recent effort of introducing machine learning concepts such as Generative Adversarial Network (GAN) for synthetic data generation. GAN can progressively learn the underlying characteristics of the limitedvolume training data set to generate any large amount of credible synthetic data with realistic characteristics and rich diversity [7]- [9]. GAN-based data generation approaches are simulation-free and also free of several common issues like inaccurate generator and load modeling, and low simulation efficiency.…”

“…GAN-based data generation approaches are simulation-free and also free of several common issues like inaccurate generator and load modeling, and low simulation efficiency. In recent years, several data-driven GAN-based applications have been proposed, such as renewable energy data generation [7], single PMU data generation [8] and dynamic security assessment [9]. However, these approaches are not capable of creating PMU data streams from multiple locations in a network that possess meaningful electromechanical dynamics and satisfy Kirchhoff's laws.…”

A comparison of scenario generation methods for the participation of electric vehicles in electricity markets

“…Lee and Baldick 23 use the generalised dynamic factor model to generate load and wind power scenarios in such a way that the correlation structure between load and wind is preserved. Chen et al 24 and Vagropoulos et al 25 use neural networks in order to be able to capture both the linear and non-linear the dependencies in data. Chen et al 24 focus on generative models, where scenarios are generated based on unsupervised (machine) learning from historical data.…”