Generative Adversarial Networks for Stochastic Grid Planning Considering Individual Distributed Energy Resource

Abstract

To consider the future impact of distributed energy resources (DERs) on grid planning, certain transmission and distribution system operators assume future profiles of conventional loads (CLs) and DERs and calculate the net load by adding them together. As certain DERs such as energy storage (ES) are operated according to grid conditions and renewable energy generation, the profiles must be assumed considering the correlation between CLs and DERs for calculating the net load. In addition, a spatial correlation is vital to properly understanding grid constraints on lines and transformers by load flow analysis. In this study, we proposed a method using a generative adversarial network to generate various profiles that simulate three-dimensional correlations such as time, space, and type of DERs. The objective is to show that the proposed method can adequately quantify the uncertainty of the future load assumption for grid planning, which is becoming increasingly difficult with the expansion of DERs. It was shown that the proposed method can account for the peak and off-peak deviations of individual DERs and appropriately assume the probability distribution of future aggregated loads. In fact, we evaluated the validity of the probability distribution using a reliability diagram and found that the deviation from the perfect reliability line was reduced by 33% on average against the compared method. Furthermore, we performed load flow analysis using the load profiles generated by the proposed method and confirmed that the load flows on lines and transformers can also be assumed with appropriate probability distributions.

Keywords

• Distributed energy resources
• generative adversarial networks
• grid planning
• scenario generation