Spatiotemporal Sequence-to-Sequence Clustering for Electric Load Forecasting

Abstract

Massive electrical load exhibits many patterns making it difficult for forecast algorithms to generalise well. Most learning algorithms produce a better forecast for dominant patterns in the case of weekday consumption and otherwise for less dominant patterns in weekend and holiday consumption. In view of this, there is the need to cluster the load patterns, so learning algorithms can focus on the patterns independently to produce forecasts with better accuracy for all cases. However, clustering time-series data breaks the time-series dependency, making model training difficult. This paper presents a novel sequence-to-sequence cluster framework to reform time-series dependency after clustering; this enables independent clusters to be modelled using Convolutional Neural Network-Gated Recurrent Unit, which learns spatiotemporal features for future forecasts. A real-world dataset by the Korea Power Exchange composed of nationwide consumption is used for case studies and experiments. Experimental results verify that the proposed study effectively improves the accuracy of electric load forecasting by about 50%, with a WAPE of 0.67%. The proposed method also speeds up the training process of the forecast algorithm by about 35%, given that only a subset of the dataset is trained due to clustering. Korea Water Resources Corporation has implemented the proposed method for load forecasting and system marginal price estimation.INDEX TERMS Convolutional neural network-gated recurrent unit (CNN-GRU), feature engineering, kmeans clustering, LightGBM classifier, sequence-to-sequence forecast, short-term load forecast (STLF).

Hourly Building Energy Consumption Prediction Using a Training Sample Selection Method Based on Key Feature Search

“…This supports the significance of occupancy behavior statistics noted in the literature [54] even more. The KFSS method, developed using occupancy behavior and outdoor meteorological information as important elements, improves the model's adaptability when compared to methods described in the literature [28,29]. The study of feature scenarios can give modelers guidance for improving models in actual modeling projects in comparison to the feature selection and feature extraction techniques described in the literature [40,55].…”

“…Using the k-means clustering method, for example, operational data from the literature [26,27] was used to classify energy consumption patterns into three categories. Acquah et al [28] produced strong predictive accuracy by classifying energy consumption patterns into three groups using K-means clustering with dynamic time wrapping. Chen et al [29] used fuzzy C-means (FCM) clustering to identify the usage patterns of energy consumption.…”

K-Means and Alternative Clustering Methods in Modern Power Systems

“…The K-means clustering algorithm is utilized to categorize users based on weather conditions or consumption levels, aiming to identify similarities among different load patterns [223], [224], [225], [226], [227]. Most studies employ this clustering algorithm to cluster users based on their load profiles.…”

“…The goal is to construct separate models for each cluster, thereby enhancing the accuracy of their predictive models. In [6], [226], [228], [229], [230], and [231], the authors applied the same methodology of clustering similar energy consumption patterns to develop a prediction model for residential load consumption. Additionally, temperature data was utilized to cluster users in order to build load forecasting models [224].…”