A Machine Learning Approach to Low-Cost Photovoltaic Power Prediction Based on Publicly Available Weather Reports

Application of Temporal Fusion Transformer for Day-Ahead PV Power Forecasting

The energy generated by a solar photovoltaic (PV) system depends on uncontrollable factors, including weather conditions and solar irradiation, which leads to uncertainty in the power output. Forecast PV power generation is vital to improve grid stability and balance the energy supply and demand. This study aims to predict hourly day-ahead PV power generation by applying Temporal Fusion Transformer (TFT), a new attention-based architecture that incorporates an interpretable explanation of temporal dynamics and high-performance forecasting over multiple horizons. The proposed forecasting model has been trained and tested using data from six different facilities located in Germany and Australia. The results have been compared with other algorithms like Auto Regressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost), using statistical error indicators. The use of TFT has been shown to be more accurate than the rest of the algorithms to forecast PV generation in the aforementioned facilities.

Solar Irradiance Forecast Based on Cloud Movement Prediction

Solar energy production based on a photovoltaic system is closely related to solar irradiance. Therefore, the planning of production is based on the prediction of solar irradiance. The optimal use of different energy storage systems requires an accurate prediction of solar irradiation with at least an hourly time horizon. In this work, a solar irradiance prediction method is developed based on the prediction of solar shading by clouds. The method is based on determining the current cloud position and estimating the velocity from a sequence of multiple images taken with a 180-degree wide-angle camera with a resolution of 5 s. The cloud positions for the next hour interval are calculated from the estimated current cloud position and velocity. Based on the cloud position, the percentage of solar overshadowing by clouds is determined, i.e., the solar overshadowing curve for the next hour interval is calculated. The solar irradiance is determined by normalizing the percentage of the solar unshadowing curve to the mean value of the irradiance predicted by the hydrometeorological institute for that hourly interval. Image processing for cloud detection and localization is performed using a computer vision library and the Java programming language. The algorithm developed in this work leads to improved accuracy and resolution of irradiance prediction for the next hour interval. The predicted irradiance curve can be used as a predicted reference for solar energy production in energy storage system optimization.

Day-Ahead Electric Load Forecast for a Ghanaian Health Facility Using Different Algorithms

Ghana suffers from frequent power outages, which can be compensated by off-grid energy solutions. Photovoltaic-hybrid systems become more and more important for rural electrification due to their potential to offer a clean and cost-effective energy supply. However, uncertainties related to the prediction of electrical loads and solar irradiance result in inefficient system control and can lead to an unstable electricity supply, which is vital for the high reliability required for applications within the health sector. Model predictive control (MPC) algorithms present a viable option to tackle those uncertainties compared to rule-based methods, but strongly rely on the quality of the forecasts. This study tests and evaluates (a) a seasonal autoregressive integrated moving average (SARIMA) algorithm, (b) an incremental linear regression (ILR) algorithm, (c) a long short-term memory (LSTM) model, and (d) a customized statistical approach for electrical load forecasting on real load data of a Ghanaian health facility, considering initially limited knowledge of load and pattern changes through the implementation of incremental learning. The correlation of the electrical load with exogenous variables was determined to map out possible enhancements within the algorithms. Results show that all algorithms show high accuracies with a median normalized root mean square error (nRMSE) <0.1 and differing robustness towards load-shifting events, gradients, and noise. While the SARIMA algorithm and the linear regression model show extreme error outliers of nRMSE >1, methods via the LSTM model and the customized statistical approaches perform better with a median nRMSE of 0.061 and stable error distribution with a maximum nRMSE of <0.255. The conclusion of this study is a favoring towards the LSTM model and the statistical approach, with regard to MPC applications within photovoltaic-hybrid system solutions in the Ghanaian health sector.

LSTM Networks for Overcoming the Challenges Associated with Photovoltaic Module Maintenance in Smart Cities

Predictive maintenance is a field of research that has emerged from the need to improve the systems in place. This research focuses on controlling the degradation of photovoltaic (PV) modules in outdoor solar panels, which are exposed to a variety of climatic loads. Improved reliability, operation, and performance can be achieved through monitoring. In this study, a system capable of predicting the output power of a solar module was implemented. It monitors different parameters and uses automatic learning techniques for prediction. Its use improved reliability, operation, and performance. On the other hand, automatic learning algorithms were evaluated with different metrics in order to optimize and find the best configuration that provides an optimal solution to the problem. With the aim of increasing the share of renewable energy penetration, an architectural proposal based on Edge Computing was included to implement the proposed model into a system. The proposed model is designated for outdoor predictions and offers many advantages, such as monitoring of individual panels, optimization of system response, and speed of communication with the Cloud. The final objective of the work was to contribute to the smart Energy system concept, providing solutions for planning the entire energy system together with the identification of suitable energy infrastructure designs and operational strategies.

Multi-Step Solar Irradiance Forecasting and Domain Adaptation of Deep Neural Networks

The problem of forecasting hourly solar irradiance over a multi-step horizon is dealt with by using three kinds of predictor structures. Two approaches are introduced: Multi-Model (MM) and Multi-Output (MO). Model parameters are identified for two kinds of neural networks, namely the traditional feed-forward (FF) and a class of recurrent networks, those with long short-term memory (LSTM) hidden neurons, which is relatively new for solar radiation forecasting. The performances of the considered approaches are rigorously assessed by appropriate indices and compared with standard benchmarks: the clear sky irradiance and two persistent predictors. Experimental results on a relatively long time series of global solar irradiance show that all the networks architectures perform in a similar way, guaranteeing a slower decrease of forecasting ability on horizons up to several hours, in comparison to the benchmark predictors. The domain adaptation of the neural predictors is investigated evaluating their accuracy on other irradiance time series, with different geographical conditions. The performances of FF and LSTM models are still good and similar between them, suggesting the possibility of adopting a unique predictor at the regional level. Some conceptual and computational differences between the network architectures are also discussed.