High-Resolution Peak Demand Estimation Using Generalized Additive Models and Deep Neural Networks

Extracted main text — title through conclusion, appendix excluded. This is what our citation measures are computed over, published so the extraction can be checked by eye.

1,151,745 characters · 1 sections · 22 citation commands

Rendered from LaTeX for readability, not typeset faithfully. Citation keys are highlighted; maths is left as source; figures, tables and equation environments are summarised rather than reproduced; unrecognised commands are greyed out so nothing is silently dropped. Email addresses are removed.

This text was truncated for display. The citation measures were computed over the complete text.

\journal{Energy and AI}

frontmatter\ead{[email removed]} \address[a]{University of Duisburg-Essen, Germany} \begin{abstract} This paper covers predicting high-resolution electricity peak demand features given lower-resolution data. This is a relevant setup as it answers whether limited higher-resolution monitoring helps to estimate future high-resolution peak loads when the high-resolution data is no longer available. That question is particularly interesting for network operators considering replacing high-resolution monitoring predictive models due to economic considerations. We propose models to predict half-hourly minima and maxima of high-resolution (every minute) electricity load data while model inputs are of a lower resolution (30 minutes). We combine predictions of generalized additive models (GAM) and deep artificial neural networks (DNN), which are popular in load forecasting. We extensively analyze the prediction models, including the input parameters' importance, focusing on load, weather, and seasonal effects. The proposed method won a data competition organized by Western Power Distribution, a British distribution network operator. In addition, we provide a rigorous evaluation study that goes beyond the competition frame to analyze the models' robustness. The results show that the proposed methods are superior to the competition benchmark concerning the out-of-sample root mean squared error (RMSE). This holds regarding the competition month and the supplementary evaluation study, which covers an additional eleven months. Overall, our proposed model combination reduces the out-of-sample RMSE by 57.4% compared to the benchmark. \end{abstract} \begin{keyword} Electricity Peak Load \sep Generalized Additive Models \sep Artificial Neural Networks \sep Prediction \sep Combination \sep Weather Effects \sep Seasonality \JEL Q41 \sep C53 \end{keyword}

Introduction

The decentralization of the energy system renders accurate electricity demand forecasts more critical than ever. Spikes in demand can produce strain on the networks. These issues will likely increase due to the increased use of lower-carbon technologies such as heat pumps and electric vehicles. Monitoring is expensive in the long term since it requires initial investments and sustained maintenance. Network operators are, therefore, particularly interested in peak-demand estimates.

This paper presents a winning method for high-resolution peak-demand prediction using generalized additive models (GAM) and deep neural networks (DNN). We developed this method for solving a data competition organized by Western Power Distribution (WPD), a British network operator responsible for the Midlands, South West, and Wales. The goal was to predict high-resolution minimum and maximum peak load using only low-resolution data and weather information. This is a relevant setup as it answers whether limited higher-resolution monitoring helps to estimate future high-resolution peak loads when the high-resolution data is no longer available. This is particularly interesting for network operators since consistent high-resolution monitoring is very costly zheng2013smart, suanduleac2021high. {Therefore, replacing high-resolution monitoring with data science methods is an economically attractive scenario. If the results for a particular substation generalize to other substations, only low-resolution data would be necessary to obtain high-resolution features.}

Peak load estimates are relevant at various scales. While Nationwide forecasts are essential for climate change issues and resource planning mughees2021deep, lee2022national, network operators are usually more interested in regional peak load predictions, which are also the topic of this paper. Another relevant topic for network operators is peak load shaving uddin2018review, lissa2021deep. The latter deals with flattening the load curve. While this topic is also highly relevant for network operators, it is less related to the topic of this paper. In the following paragraphs, we provide an overview of research on peak load estimation.

There are various papers dealing with peak load prediction at different scales, e.g, forecasting short-term load of individual households sun2018probabilistic, chou2018forecasting, office buildings chen2017short, low-voltage feeders haben2019short, cities guo2018deep, and nations lee2022national. The results indicate that weather data like temperature, humidity, wind speeds, and air-pressure help explain peak loads xie2016relative, dehalwar2016electricity, cai2019day, muzaffar2019short. {The influence of windspeed on electricity loads is discussed in detail in acarouglu2021comprehensive. hong2015weather deals with the selection of suitable weather stations.} In contrast, haben2019short, who supplemented their model with deterministic factors like the day of the week, found little to no evidence for an impact of the temperature on electricity load.

The applied techniques for peak load forecasting are diverse. They include models like time-varying autoregressive (ARTV) vu2017short, recurrent and convolutional neural networks (RNN, CNN) cai2019day, long-short term memory (LSTM) networks muzaffar2019short, gradient boosting aguilar2021short, and multistep approaches zhang2017short, sheng2020short, fan2021forecasting. The techniques proposed in this paper have also been applied to load forecasting. GAMs are widely used in global pierrot2011short and local goude2013local, ziel2022smoothed short-term electricity load forecasting, and in forecasting high-resolution demand smart-meter data amato2021forecasting. The literature on artificial intelligence models based on neural networks in electricity load forecasting is much broader. A wide range of artificial neural networks, including multilayer perceptron networks (MLP), CNNs, and RNNs, are used in peak load tasre2011daily, pallonetto2022forecast and load hosein2017load, amarasinghe2017deep, khwaja2020joint, walser2021typical, memarzadeh2021short, shaqour2022electrical, bashir2022short, khan2022efficient forecasting.

Summarizing the research contribution of this manuscript, we

itemize• propose a winning method for predicting high-resolution power load with low-resolution data. • analyze the descriptive data characteristics of relevant input data (esp. load and weather inputs) in detail for suitable usage for predictive analytics. • design machine learning models, GAMs and DNNs, which combined yield the best performance among competitors. • provide a rigorous predicting study that ensures the robustness of the method. • study and interpret the explanatory power of relevant inputs and discuss hyperparameter tuning results.

The remainder of this paper is structured as follows. Section (ref) introduces the data and provides an extensive descriptive data analysis. Section (ref) describes the models that we developed. The models' performance is measured using an evaluation study. Section (ref) contains the study design as well as the results. Section (ref) concludes.

figure[figure omitted — 392,276 chars of source]