Local Cold Load Pick-up Estimation Using Customer Energy Consumption Measurements

TL;DR

This work tackles local CLPU estimation after outages by forecasting the energy not served $\hat{E}_o$ at the customer level using timestamped measurements and an ARIMA-based approach with hourly resolution. It introduces a service-point architecture with dynamic model-order selection to predict CLPU peak $P_{CLPU}$, energy, and duration, and compares ARIMA against LSTM and HWES, finding ARIMA superior in both accuracy and computation speed. Validation uses 50 real households and GridLab-D simulations, showing good alignment between ARIMA forecasts and physical load models, and demonstrating the method’s utility for restoration planning. The approach enables targeted, rapid restoration decisions and potential deferral of grid upgrades, particularly in cold climates, by providing simple, local forecasts with low computational requirements.

Abstract

Thermostatically-controlled loads have a significant impact on electricity demand after service is restored following an outage, a phenomenon known as cold load pick-up (CLPU). Active management of CLPU is becoming an essential tool for distribution system operators who seek to defer network upgrades and speed up post-outage customer restoration. One key functionality needed for actively managing CLPU is its forecast at various scales. The widespread deployment of smart metering devices is also opening up new opportunities for data-driven load modeling and forecast. In this paper, we propose an approach for customer-side estimation of CLPU using time-stamped local load measurements. The proposed method uses Auto-Regressive Integrated Moving Average (ARIMA) modeling for short-term foregone energy consumption forecast during an outage. Forecasts are made on an hourly basis to estimate the energy to potentially recover after outages lasting up to several hours. Moreover, to account for changing customer behavior and weather, the model order is adjusted dynamically. Simulation results based on actual smart meter measurements are presented for 50 residential customers over the duration of one year. These results are validated using physical modeling of residential loads and are shown to match well the ARIMA-based forecasts. Additionally, accuracy and execution speed has been compared with other state-of-the-art approaches for time-series forecasting including Long Short Term Memory Network (LSTM) and Holt-Winters Exponential Smoothing (HWES). ARIMA-based forecast is found to offer superior performance both in terms of accuracy and computation speed.

Figures (13)

• Figure 1: Feeder CLPU.
• Figure 2: Single dwelling CLPU.
• Figure 3: CLPU estimation using the estimated value of energy not served during the outage.
• Figure 4: ARIMA model order selection algorithm.
• Figure 5: CLPU assessment algorithm overview.