Abstract:In this paper, we consider the problem of developing predictive models with limited data for energy assets such as electricity loads, PV power generations, etc. We specifically investigate the cases where the amount of historical data is not sufficient to effectively train the prediction model. We first develop an energy predictive model based on convolutional neural network (CNN) which is well suited to capture the interaday, daily, and weekly cyclostationary patterns, trends and seasonalities in energy assets time series. A transfer learning strategy is then proposed to address the challenge of limited training data. We demonstrate our approach on a usecase of daily electricity demand forecasting. we show practicing the transfer learning strategy on the CNN model results in significant improvement to existing forecasting methods.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to develop effective prediction models for energy assets (such as electric power load, photovoltaic power generation, etc.) in the case of limited historical data. Specifically, when the historical data is insufficient, traditional prediction models cannot be effectively trained, resulting in poor prediction performance. To this end, the author proposes a method based on convolutional neural network (CNN) and transfer learning to meet this challenge. ### Main problems 1. **Insufficient data**: In many application scenarios, especially in the case of new customers or new facilities, there is a lack of sufficient historical data to train accurate prediction models. 2. **Limitations of existing methods**: Existing prediction models usually require a large amount of historical data for training, and such data may be difficult to obtain in practical applications. ### Solutions To overcome the above problems, the author proposes the following solutions: 1. **Use convolutional neural network (CNN)**: Design a CNN model suitable for capturing daily, daily and weekly periodicities to process energy time - series data. 2. **Transfer learning strategy**: Through transfer learning, use public data sets to pre - train the model, and then fine - tune it with a small amount of data of target customers to improve the prediction performance. ### Method overview 1. **Data collection and pre - processing**: Obtain energy data of different types of customers from public data sets and normalize them. 2. **Pre - training**: Use public data sets to pre - train the CNN model and learn general features. 3. **Fine - tuning**: On the basis of pre - training, use the limited historical data of target customers to fine - tune the model, keep the learned general features, and at the same time adapt to the data characteristics of specific customers. 4. **Evaluation and verification**: Through experimental verification, compare the prediction performance of different models, including the SARIMA model, the CNN model without pre - training, and the situation of only using the pre - trained model. ### Experimental results The experimental results show that the proposed transfer learning strategy significantly improves the prediction performance, especially in the application of commercial and industrial (C&I) customers. Compared with other models, this method shows a lower mean absolute error (MAE) in all tested customers, which proves its effectiveness and superiority. ### Formula representation - **Mean absolute error (MAE)**: \[ \text{MAE}=\frac{1}{N}\sum_{i = 1}^{N}|x_i-\hat{x}_i| \] where \(x_i\) is the actual value, \(\hat{x}_i\) is the predicted value, and \(N\) is the number of samples. - **Model loss function**: \[ \text{Loss}=\sum_{i = 1}^{n}\sum_{t = 0}^{T}(P_i(t)-\hat{P}_i(t))^2 \] where \(P_i(t)\) is the actual power, \(\hat{P}_i(t)\) is the predicted power, \(n\) is the number of samples, and \(T\) is the time step. Through this method, the author has successfully solved the problem of developing efficient energy prediction models in the case of limited data, providing new ideas and technical means for short - term load prediction in energy systems.

Energy Predictive Models with Limited Data using Transfer Learning

Transfer Learning in Deep Learning Models for Building Load Forecasting: Case of Limited Data

A Review of Transfer Learning Approaches for Load, Solar and Wind Power Predictions

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

Time Series Prediction Method of Industrial Process With Limited Data Based on Transfer Learning

A multi-energy load forecasting method based on parallel architecture CNN-GRU and transfer learning for data deficient integrated energy systems

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

Powering Electricity Forecasting with Transfer Learning

Wind and Solar Power Generation Forecasting Based on Hybrid CNN-ABiLSTM, CNN-Transformer-MLP Models

Deep learning and transfer learning models of energy consumption forecasting for a building with poor information data

Energy forecasting model based on CNN-LSTM-AE for many time series with unequal lengths

Transfer Learning for Electricity Price Forecasting

Short-Term Load Forecasting for a Single Household Based on Convolution Neural Networks Using Data Augmentation

A hybrid transfer learning model for short-term electric load forecasting

An Integrated Transfer Learning Method for Power Generation Prediction of Run-off Small Hydropower in Data-scarce Areas

Predicting short-term energy usage in a smart home using hybrid deep learning models

A Day-Ahead Photovoltaic Power Prediction via Transfer Learning and Deep Neural Networks

Hierarchical transfer learning with applications for electricity load forecasting

Hybrid deep learning models with data fusion approach for electricity load forecasting

Data-driven short term load forecasting with deep neural networks: Unlocking insights for sustainable energy management

Transfer Learning on Transformers for Building Energy Consumption Forecasting -- A Comparative Study