Abstract:Short-term load forecasting (STLF) is crucial for the daily operation of power grids. However, the non-linearity, non-stationarity, and randomness characterizing electricity demand time series renders STLF a challenging task. Various forecasting approaches have been proposed for improving STLF, including neural network (NN) models which are trained using data from multiple electricity demand series that may not necessary include the target series. In the present study, we investigate the performance of this special case of STLF, called transfer learning (TL), by considering a set of 27 time series that represent the national day-ahead electricity demand of indicative European countries. We employ a popular and easy-to-implement NN model and perform a clustering analysis to identify similar patterns among the series and assist TL. In this context, two different TL approaches, with and without the clustering step, are compiled and compared against each other as well as a typical NN training setup. Our results demonstrate that TL can outperform the conventional approach, especially when clustering techniques are considered.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to explore the application of **Transfer Learning (TL)** in **Short - Term Load Forecasting (STLF)**, especially for the daily - ahead electricity demand time series in European countries. Specifically, the main objectives of the research are: 1. **Improve the accuracy of STLF**: By using transfer learning techniques, apply the knowledge obtained from one or more related tasks to new, unseen tasks, thereby improving the performance of the prediction model. 2. **Cross - country - level application**: Study the similarity of electricity demand patterns among different countries and share these patterns through transfer learning to achieve more accurate predictions. This helps to better regulate the balance between electricity production and consumption, reduce operating costs, and enhance the security and robustness of the system. 3. **Reduce training time and data requirements**: Through transfer learning, fine - tuning can be carried out on the basis of existing models, thereby significantly reducing the time and amount of data required for retraining the model, especially in the case of limited data. 4. **Verify the effectiveness of different TL methods**: Compare two different transfer learning methods - TL methods with and without a clustering step, and compare them with the traditional NN training method to evaluate their performance differences. #### Research background Short - term load forecasting is crucial for the daily operation of power systems because it involves decisions in power generation, transmission, and distribution. However, electricity demand time series have characteristics such as non - linearity, non - stationarity, and randomness, making STLF a challenging task. Although various prediction methods have been proposed, how to effectively use transfer learning to improve prediction accuracy remains a research - worthy problem. #### Main contributions 1. **Expand the scale of international transfer learning**: This research applies transfer learning at the pan - European level (27 EU countries), which is an important expansion of existing research and shows the possibility of sharing knowledge among different countries in a global prediction model. 2. **Select MLP as the base model**: Compared with complex deep - learning architectures, the research selects the relatively simple Multi - Layer Perceptron (MLP) as the base model because MLP performs well in the field of time - series prediction and is fast in calculation, easy to implement and adjust. 3. **Introduce cluster analysis**: In order to identify countries with similar electricity demand patterns, the research introduces a time - series clustering method. This helps to share knowledge more effectively during the transfer learning process, especially among countries with similar geographical, demographic, and economic conditions. #### Methodology The research adopts an Automated Machine Learning Operations (MLOps) pipeline, including the following main steps: - **Data pre - processing**: Clean duplicates, outliers, convert time zones, and fill in missing values. - **Load feature analysis**: Calculate the average load curve and analyze daily, weekly, and annual load patterns. - **Cluster analysis**: Use the hierarchical clustering method to group countries based on the similarity of their load curves. - **Model selection and training**: Select MLP as the prediction model and apply transfer learning techniques for training and optimization. - **Experimental design and evaluation**: Design three experimental schemes to evaluate the performance of the traditional NN training method, the TL method without clustering, and the TL method with clustering respectively. Through these methods, the research aims to verify the potential advantages of transfer learning in short - term load forecasting and provide valuable references for power system operators.

Transfer learning for day-ahead load forecasting: a case study on European national electricity demand time series

Enhanced Sequence-to-Sequence Deep Transfer Learning for Day-Ahead Electricity Load Forecasting

A Hybrid Short-Term Load Forecasting Method Based on Improved Ensemble Empirical Mode Decomposition and Back Propagation Neural Network

A comparative assessment of deep learning models for day-ahead load forecasting: Investigating key accuracy drivers

Fuzzy Clustering-Based Deep Learning for Short-Term Load Forecasting in Power Grid Systems Using Time-Varying and Time-Invariant Features

Hierarchical transfer learning with applications for electricity load forecasting

Combining fuzzy clustering and improved long short-term memory neural networks for short-term load forecasting

Transfer learning strategies for solar power forecasting under data scarcity

Transfer Learning for Electricity Price Forecasting

Efficient Residential Electric Load Forecasting via Transfer Learning and Graph Neural Networks

In Search of Deep Learning Architectures for Load Forecasting: A Comparative Analysis and the Impact of the Covid-19 Pandemic on Model Performance

Transfer Learning Featured Short-Term Combining Forecasting Model for Residential Loads with Small Sample Sets

Research on short-term load forecasting based on clustering and deep learning

Local Short Term Electricity Load Forecasting: Automatic Approaches

Transfer Learning in Transformer-Based Demand Forecasting For Home Energy Management System

A Multi-Step Time-Series Clustering-Based Seq2Seq LSTM Learning for a Single Household Electricity Load Forecasting

Hierarchical Clustering and Deep Learning for Short-Term Load Forecasting with Influenced Factors

Day-Ahead Electricity Load Forecasting with Multivariate Time Series

ES-dRNN: A Hybrid Exponential Smoothing and Dilated Recurrent Neural Network Model for Short-Term Load Forecasting

An Ensemble Framework for Short-Term Load Forecasting Based on TimesNet and TCN

Deep Learning for Time Series Forecasting: The Electric Load Case