Parameter Evaluation Method for Power Demand Forecasting Methodology of a Clinic
- 17 September 2024
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industry Applications
- Vol. 61 (1) , 930-939
Abstract
In Japan, efforts to achieve a carbon-neutral society by 2050 are underway. As part of these efforts, hospitals and clinics are particularly focusing on energy management to reduce carbon dioxide emissions with peak-cut operations. Especially, peak-cut operations necessitate accurate power demand forecasting, often facilitated by neural networks (NNs). The multilayer perceptron (MLP) and the long short-term memory (LSTM) networks represent two prominent NNs employed for this purpose. Moreover, the mean absolute error (MAE), root mean square error (RMSE), and the ratio of RMSE to MAE are used as evaluation indicators. This study employs the MLP and LSTM for the power demand forecasting of a clinic considering factors such as data partitioning, structure of the NN (the number of layers and nodes), and data period. For MLP-based forecasting models, random partitioning is preferred to improve forecasting accuracy. Additionally, for four year or longer data periods, the learning outcomes are unaffected by the structure of the NN (the number of nodes and layers). Conversely, in LSTM-based forecasting models, the choice between random and block partitioning depends on the data period. Single- or double-layer configurations show better MAE and RMSE values than three- or six-layer configurations. Employing the ratio of RMSE to MAE as an evaluation indicator, the error distribution of the NN can be estimated. The characteristics of the forecasting model can be described from the error distribution. A value of 1.414 for the ratio of RMSE to MAE indicates a Laplace distribution, a value of 1.253 indicates a normal distribution. Therefore, a good forecasting model is characterized by low RMSE to MAE values and RMSE to MAE ratios between 1.253 and 1.414.Keywords
Funding Information
- JSPS KAKENHI (JP21K04559)
This publication has 17 references indexed in Scilit:
- Solar Power Forecasting Using Deep Learning TechniquesIEEE Access, 2022
- Pandemic-Aware Day-Ahead Demand Forecasting Using Ensemble LearningIEEE Access, 2022
- Load Forecasting Under Concept Drift: Online Ensemble Learning With Recurrent Neural Network and ARIMAIEEE Access, 2021
- Short-Term Energy Forecasting Framework Using an Ensemble Deep Learning ApproachIEEE Access, 2021
- A Simplified LSTM Neural Networks for One Day-Ahead Solar Power ForecastingIEEE Access, 2021
- Multi-model fusion short-term load forecasting based on random forest feature selection and hybrid neural networkIEEE Access, 2021
- Short-Term Load Forecasting Based on Integration of SVR and StackingIEEE Access, 2020
- LSTM Forecasts for Smart Home Electricity Usage2020
- A Hybrid Deep Learning Model with Evolutionary Algorithm for Short-Term Load Forecasting2019
- Learning to forget: continual prediction with LSTM1999