Increasing the Accuracy of Hourly Multi-Output Solar Power Forecast with Physics-Informed Machine Learning

Abstract

Machine Learning (ML)-based methods have been identified as capable of providing up to one day ahead Photovoltaic (PV) power forecasts. In this research, we introduce a generic physical model of a PV system into ML predictors to forecast from one to three days ahead. The only requirement is a basic dataset including power, wind speed and air temperature measurements. Then, these are recombined into physics informed metrics able to capture the operational point of the PV. In this way, the models learn about the physical relationships of the different features, effectively easing training. In order to generalise the results, we also present a methodology evaluating this physics informed approach. We present a study-case of a PV system in Denmark to validate our claims by extensively evaluating five different ML methods: Random Forest, Support Vector Machine, Convolutional Neural Networks (CNN), Long-Short Term Memory (LSTM) and a hybrid CNN–LSTM. The results show consistently how the best predictors use the proposed physics-informed features disregarding the particular ML-method, and forecasting horizon. However, also, how there is a threshold regarding the number of previous samples to be included that appears as a convex function.

A Review on Machine Learning Applications for Solar Plants

“…The study [ 64 ] reveals that the output power with the insolation and the air temperature has a linear and nonlinear correlation, correspondingly. Recently, researchers have been more interested in the ML application to increase the accuracy of the forecasters [ 61 , 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 ].…”

“…Due to forecast power, in [ 69 , 70 ], researchers integrated a PV-performance model into ML methods such as RF, SVR, CNN, LSTM, and hybrid CNN-LSTM. The results indicated that the proposed ML models provide the best performance regardless of the model’s type and forecasting horizon ( Table 7 ).…”

SOLETE, a 15-month long holistic dataset including: Meteorology, co-located wind and solar PV power from Denmark with various resolutions

“…To build the dataset, these files were imported, only selecting the interesting metrics, and averaged to obtain 5 min and hourly resolution. Data source location • Institution: Risø DTU National Laboratory for Sustainable Energy • City: Roskilde • Country: Denmark • Latitude and longitude: 55.6867, 12.0985 Data accessibil- ity Repository name: DTU Data Data identification number: https://doi.org/10.11583/DTU.17040767 Direct URL to data: https://data.dtu.dk/articles/dataset/The_SOLETE_dataset/17040767 The SOLETE dataset/17,040,767 Corresponding Reference: [1] Repository name: DTU Data Software release number: https://doi.org/10.11583/DTU.17040626 Direct URL to release: https://github.com/DVPombo/SOLETE The SOLETE platform/17,040,626 Corresponding Reference: [2] Related research articles [3] D. V. Pombo, H. W. Bindner, S. V. Spataru, P. E. Sørensen, & P. Bacher, Increasing the Accuracy of Hourly Multi-Output Solar Power Forecast with Physics-Informed Machine Learning, Sensors 22 (3) (2022) 749. [4] D. V. Pombo, P. Bacher, C. Ziras, H. W. Bindner, S. V. Spataru, & P. E. Sørensen, Benchmarking Physics-Informed Machine Learning-based Short Term PV-Power Forecasting Tools, Under Review.…”

“… The dataset is complemented by a Git repository [2] including three Python scripts employing only Open Access libraries. The first script simply imports the data, while the second showcases the methodology discussed in [3] and [4] to build physics-informed ML-models for solar power forecasting. This resource is particularly useful for students and researchers first starting in the machine learning-based forecasting field.…”

“…MLForecasting.py This file is directly related to the dataset's twin publications [3 , 4] . It is a self contained example of how to build a physics informed solar power forecaster based on different off-the-shelf ML algorithms.…”