With the recent increase in the use of fossil fuels to cope with the explosive demand for energy, diverse global problems, such as greenhouse gas and energy resource depletion, have attracted much attention.
Many efforts have been made to reduce greenhouse gas emissions.
Previous Research
Transition to a smart city is one of the representative efforts of the local government to reduce greenhouse gas emissions.
Smart cities can reduce greenhouse gas emissions by reducing traffic congestion and energy consumption via big data analysis and then introducing alternatives, such as electric vehicles and renewable energy.
The transition to a smart city has been accelerated and refined by various artificial intelligence (AI) technologies, which have already enhanced the performance of various ICT areas, such as communications, applications, content, and digital commerce.
The infrastructure for electricity forms the basis for maintaining the fundamental survival of members of society; hence, it is imperative to supply electricity stably and efficiently in an environmentally friendly manner.
Improving the productivity and quality of electricity is one of the most common issues requiring resolution.
Many studies have been conducted to solve this issue by applying AI technologies to energy data for diverse goals, such as analysis, management, and forecasting.
Load forecasting is a technique that predicts how much electricity will be used in the future, which enables electricity suppliers to produce electricity without waste.
Load forecasting can be classified into four categories based on the time resolution: very-short-term load forecasting, short-term load forecasting, mid-term load forecasting (MTLF), and long-term load forecasting.
The load forecasting type is chosen based on the requirements of the application.
Proposed Model
In this paper, we propose a novel transfer learning-based monthly load forecasting model for cities or districts using other domain data that have a high correlation coefficient with the target data.
We use the transfer learning technique to perform accurate hierarchical monthly electric load forecasting for metropolitan cities using public datasets.
By calculating the Pearson correlation coefficients (PCCs), we selected relevant domains that can improve the effectiveness of transfer learning.
We demonstrate that our proposed model can exhibit higher performance than popular statistical and ensemble methods.
Significance
Our proposed model can contribute to the accurate forecasting of mid-term electricity energy consumption in smart cities, which can be used to determine the system capacity, system operation and maintenance costs, and future grid expansion plans.
Proposed Model
The target domain is represented by district name and electricity category; other combinations of districts and categories become the source domain.
Monthly electric load forecasting model is constructed using similar electric load data from 124 source domains.
Each source data and target data is divided into a training and a test set.
To find similar load data from the source domains, the PCCs between the training sets of the source data and the target data are calculated.
Using the selected source data, a DNN-based forecasting model is constructed, and then fine-tuned using the training set of the target data.
The artificial neural network consists of three layers: an input layer, one or more hidden layers, and an output layer. Each layer consists of several nodes called perceptrons.
To determine the hyper-parameters of the DNN model, several comparative experiments are performed.
The number of hidden layers is set to five, and the activation function in a multilayer perceptron is SELU.
When limited data are available from the target domain for training, adaptation techniques, such as transfer learning, can be employed to improve the performance of a forecasting model.
To find similar source domain data, correlation analysis methods, such as the PCC analysis, can be used to determine the similarity between two domains.
Experiment
PCC HeatmapMAPE BoxplotAverage MAPE
The model's effectiveness was evaluated through extensive experiments using monthly electric load data from 25 districts in Seoul from January 2005 to December 2018.
The data was divided into a training set (January 2005 to December 2016) and a testing set (January 2017 to December 2018).
The experiments considered every combination of district and category as a target domain and the other data as source domains.
The PCC (Pearson correlation coefficient) values were calculated for each target domain and source domain combination.
The DNN models were constructed by selecting the top 10, 20, and 30 most similar domains in terms of PCC.
MAPE and NRMSE were used to evaluate the performance of the models, and the transfer learning-based MTLF models showed better performance than other machine-learning-based models.
댓글