Short-term commercial load forecasting based on peak-valley features with the TSA-ELM model

Authors	Mengran Zhou, Ziwei Zhu, Feng Hu, Kai Bian, Wenhao Lai, Tianyu Hu
Title	Short-term commercial load forecasting based on peak‐valley features with the TSA‐ELM model
Publication	Energy Science & Engineering
Volume	10
Issue	8
Pages	2622-2636
Year	2022
DOI	https://doi.org/10.1002/ese3.1203

Introduction

Background

• The demand for electricity in buildings is increasing year after year.
• Buildings account for 41.1% of primary energy and 74% of electricity.
• The commercial portion of electricity sales in the United States is 1.28 trillion kWh in 2020, accounting for 34.8% of total electricity sales.
• Inaccurate electricity load forecasts can cause deviations in load planning and layout, resulting in significant economic losses and energy waste.
• Building energy consumption forecasting is an important reference in developing various strategies to improve building energy performance.

Previous Research

• The widespread use of smart devices and BEMS (Building Energy Management System) has enabled the collection of higher resolution and more accurate electrical load and meteorological data.
• Building-level energy consumption forecasts can be broadly classified into three categories based on time horizon, namely STLF (Short-Term Load Forecasting), MTLF (Mid-Term Load Forecasting), and LTLF (Long-Term Load Forecasting).
• STLF (from a few minutes to a week in advance) has a direct impact on building operation and scheduling.
• Intelligent algorithms based on ML (Machine Learning) theories, particularly ANN (Artificial Neural Network), are used to deal with STLF problems.
• However, many ANN-based gradient-based methods, such as backpropagation or other variants, have some limitations in the field of electric load forecasting.

Proposed Model

• The purpose of this paper is to propose a simple and effective method for electricity demand forecasting.
• ELM (Extreme Learning Machine) possesses the characteristics of fast speed and strong learning performance and has been widely used in river flow and load forecasting.
• The input weights and hidden layer neuron thresholds of ELM are random, thus generating a series of nonoptimal parameters, which weakens the prediction performance of ELM.
• This problem can be alleviated by optimizing the algorithm to find the optimal model parameters.
• The TSA (Tunicate Swarm Algorithm) is a new intelligent algorithm and it has better performance than algorithms such as PSO (Particle Swarm Optimization), GWO (Gray Wolf Optimization), MVO (Multi-Verse Optimization), and EPO (Emperor Penguin Optimization).
• Therefore, this paper uses TSA to optimize the initial weights and thresholds of the ELM.
• Historical peak and valley loads are taken into consideration as input features to explore the impact of hysteresis peak and valley loads on load forecasting.
• The proposed method is called PV (Peak & Valley)-TSA-ELM, which is a short-term commercial electric load forecasting model based on the TSA-ELM algorithm with peak-valley features.

Significance

• The proposed model is expected to improve the accuracy of short-term load forecasting, which has a direct impact on building operation and scheduling, thereby reducing economic losses and energy waste.
• The use of the TSA to optimize the initial weights and thresholds of the ELM can enhance the prediction performance of the ELM.
• The consideration of historical peak and valley loads as input features is a novel approach to load forecasting.
• The PV-TSA-ELM model combines peak and valley features with the TSA-ELM algorithm, which is expected to improve the accuracy of short-term load forecasting.

Proposed Model

• The proposed model consists of a TSA-ELM algorithm for commercial STLF
• The TSA simulates the natural foraging process of tunicates through jet propulsion and swarm behaviors
• The ELM is an efficient single-layer feedforward neural network with input, hidden, and output layers
• The input weights and threshold of the hidden layer neurons are randomly initialized, and the corresponding output weights are calculated by generalized inverse matrix theory
• The MSE (Mean Squared Error) of the ELM is used as the fitness function in the optimization process
• The framework of the prediction model consists of two major steps: prediction of peaks and valleys, and prediction of short-term load
• The short-term load prediction uses MIC (Maximum Information Coefficient) feature selection to choose variables with MIC ≥ 0.8 as input features

Experiment

• The TSA-ELM algorithm outperforms other models in peak prediction across all three data sets, with the lowest RMSE (Root Mean Squared Error) and MAPE (Mean Absolute Percentage Error) values.
• The ELM and SVR (Support Vector Regression) algorithms optimized by TSA rank first and second, respectively, in valley prediction evaluation indicators.
• The optimized model has smaller prediction errors than the corresponding unoptimized algorithm.
• The prediction curves of the PV-TSA-ELM model for the last week of the three different data sets are very close to the real load data curve.
• The proposed method is compared to the corresponding SVR model and demonstrated to have merits.