A Stacked Autoencoder Application for Residential Load Curve Forecast and Peak Shaving

TL;DR

This paper introduces a stacked autoencoder-based forecasting method to enhance peak shaving in residential load management, effectively balancing generation and demand for improved energy efficiency.

Contribution

It presents a novel SAE-based load forecasting approach that improves peak shaving capabilities by considering both generation and demand-side factors.

Findings

Effective peak shaving demonstrated in case study

Improved load forecasting accuracy

Enhanced BESS utilization for peak management

Abstract

For the last ten years, utilities have observed on-going transitions on consumers' load curves. The previously flat load curves have more frequently turned into duck-shape. This is jointly caused by the increasing household loads as well as the popularity of rooftop solar photovoltaic. Such load curve transition challenges the operational flexibility of the existing systems and greatly increases the per-MWh energy costs. Peak shaving, in this context, becomes a critical task to demand-side management. Owing to the development of Battery Energy Storage Systems (BESS), numerous peak shaving strategies have been developed and implemented. In this paper, by applying a stacked autoencoder (SAE)-based residential peak load curve forecasting technology, we further lift the peaking shaving capabilities of BESSs to a new level. The proposed strategy takes into account the welfares of both generation-side and demand-side and reaches an optimal balance. A comprehensive case study using smart meter data demonstrates the effectiveness of the proposed method in peak shaving application.