Emanuele Ogliari
Politecnico di Milano, Department of Energy, Department Member
This work proposes and evaluates a method for the nowcasting of solar irradiance variability in multiple time horizons, namely 5, 10, and 15 min ahead. The method is based on a Convolutional Neural Network structure that exploits infrared... more
This work proposes and evaluates a method for the nowcasting of solar irradiance variability in multiple time horizons, namely 5, 10, and 15 min ahead. The method is based on a Convolutional Neural Network structure that exploits infrared sky images acquired through an All-Sky Imager to estimate the range of possible values that the Clear-Sky Index will possibly assume over a selected forecast horizon. All data available, from the infrared images to the measurements of Global Horizontal Irradiance (necessary in order to compute Clear-Sky Index), are acquired at SolarTechLAB in Politecnico di Milano. The proposed method demonstrated a discrete performance level, with an accuracy peak for the 5 min time horizon, where about 65% of the available samples are attributed to the correct range of Clear-Sky Index values.
Research Interests:
The penetration of nonprogrammable renewable energy sources, namely wind and solar technology, has greatly increased in the last decades and nowadays the shift toward green energy sources represents a priority worldwide. The high... more
The penetration of nonprogrammable renewable energy sources, namely wind and solar technology, has greatly increased in the last decades and nowadays the shift toward green energy sources represents a priority worldwide. The high variability of the primary source challenges the grid operators in ensuring the stability and reliability of the electric grid. Machine learning algorithms, and in particular artificial neural networks, are one of the most reliable methods for photovoltaic (PV) energy production forecast. This article proposes a new ensemble method based on the probabilistic distribution of the trials, the probabilistic ensemble method (PEM). The proposed method has been tested on a three-years real case study, where the available days have been clustered according to the solar irradiation forecast. The days where the worst performance, in terms of nRMSE, was recorded mostly belonged to the totally cloudy days class, that has been therefore selected for the analysis. The PEM has been compared with the ensemble based on the mean value, achieving an improvement in the nRMSE metric up to 4.79% in 2017 in the totally cloudy days class.
This dataset includes PV power production measured on the SolarTech Lab, Politecnico di Milano, Italy. Data are freely available for scientific research purpose and further data validation.In particular, the dataset is composed of the... more
This dataset includes PV power production measured on the SolarTech Lab, Politecnico di Milano, Italy. Data are freely available for scientific research purpose and further data validation.In particular, the dataset is composed of the following variables and specifics, with a time resolution of 1 minute:Timestamp: column with time recordings; the data format is "dd-MM-yyyy hh:mm:ss", with the time always expressed in Central European Time (CET).Pm: power recordings from the PV module (W); module tilt: 30°.Tair: ambient temperature (°C) measured by the weather station described in SolarTech Lab website (http://www.solartech.polimi.it/instrumentation/).GHI: measured Global Horizontal Irradiance (W/m2).GPOA: global irradiance measured on the plane of array (30°).Ws: measured wind speed (m/s).Wd: measured wind direction (°), assuming 0° east, positive south.It is worth noticing that this dataset includes original measurements, i.e. these raw data can be used for any additional...
Research Interests: Computer Science and IEEE
Research Interests:
Research Interests:
Application of Machine Learning in forecasting renewable energy sources (RES) is increasing: in particular, several neural networks have been employed to perform the day-ahead photo-voltaic output power forecast. The aim of this paper is... more
Application of Machine Learning in forecasting renewable energy sources (RES) is increasing: in particular, several neural networks have been employed to perform the day-ahead photo-voltaic output power forecast. The aim of this paper is to consider different training approaches in order to improve the accuracy of the PV power prediction, with particular attention to day-ahead and intra-day forecasts. Additionally, novel error metrics, specifically proposed for the defined task, are compared with traditional ones, showing the best approach for the different considered cases. The results will be validated over a 1-year time range of experimentally measured data, for a PV module installed in the Solar Tech Lab in the department of Energy at Politecnico di Milano.
Research Interests:
Accurate forecast of aggregate end-users electric load profiles is becoming a hot topic in research for those main issues addressed in many fields such as the electricity services market, the load shedding and the virtual power plants.... more
Accurate forecast of aggregate end-users electric load profiles is becoming a hot topic in research for those main issues addressed in many fields such as the electricity services market, the load shedding and the virtual power plants. Hence, load forecast is an extremely important task which should be understood more in depth. In this research paper the dependency of the day-ahead load forecast accuracy on the basis of the data typology employed in the training of LSTM has been inspected. A real case study of an Italian industrial load with samples recorded every 15 minutes for the year 2017 and 2018 was studied. The effect in the load forecast accuracy of different type of samples, which have been classified into “holidays”, “load shedding” and “maintenance” in the training dataset has been investigated by calculating the most commonly used error metrics showing the importance of data employed in load forecast.
Abstract In the last decade, wind has experienced a strong expansion reaching 591 GW (2018) of installed capacity worldwide. The higher penetration of variable renewable energy sources (wind and solar) has led to a growing demand for... more
Abstract In the last decade, wind has experienced a strong expansion reaching 591 GW (2018) of installed capacity worldwide. The higher penetration of variable renewable energy sources (wind and solar) has led to a growing demand for reliable forecast methods, to properly integrate these sources in the electric grid, decreasing the cost of electricity production and power curtailments. The present work proposes diverse wind power predictive approaches based on a physical model, artificial neural networks and an hybridization of the two. The time series used is composed of two-years hourly measurements of a wind farm in Italy, consisting of 24 wind turbines with a nominal power of 0.66 MW. To ensure an adequate reliability and robustness of the results obtained from the performance evaluation, it was chosen to use eight different error metrics and to evaluate the accuracy considering two different predictive situations (yearly and daily), using the persistence model as benchmark. The evaluations of predictive performances, regarding both the analyses, confirmed the superiority of data-driven approaches in the daily wind power prediction. More in detail, the hybrid model managed to reduce the MAE, the NRMSE and the SS values, compared to persistence, by 50%, 47.82% and 47.68%, respectively.
Research Interests:
Research Interests:
Accurate photovoltaic (PV) prediction has a very positive effect on many problems that power grids can face when there is a high penetration of variable energy sources. This problem can be addressed with computational intelligence... more
Accurate photovoltaic (PV) prediction has a very positive effect on many problems that power grids can face when there is a high penetration of variable energy sources. This problem can be addressed with computational intelligence algorithms such as neural networks and Evolutionary Optimization. The purpose of this article is to analyze three different hybridizations between physical models and artificial neural networks: the first hybridization combines neural networks with the output of the five-parameter physical model of a photovoltaic module in which the parameters are obtained from a datasheet. In the second hybridization, the parameters are obtained from a matching procedure with historical data exploiting Social Network Optimization. Finally, the third hybridization is PHANN, in which clear sky irradiation is used as an input. These three hybrid methods are compared with two physical approaches and simple neural network-based forecasting. The results show that the hybridizat...
Research Interests:
The inherently non-dispatchable nature of renewable sources, such as solar photovoltaic, is regarded as one of the main challenges hindering their massive integration in existing electric grids. Accurate forecasting of the power output of... more
The inherently non-dispatchable nature of renewable sources, such as solar photovoltaic, is regarded as one of the main challenges hindering their massive integration in existing electric grids. Accurate forecasting of the power output of the solar plant might therefore play a key role towards this goal. In this paper, we compare several machine learning and deep learning algorithms for intra-hour forecasting of the output power of a 1 MW photovoltaic plant, using meteorological data acquired in the field. With the best performing algorithms, our data-driven workflow provided prediction performance that compares well with the present state of the art and could be applied in an industrial setting.
Research Interests:
We compare the 24-hour ahead forecasting performance of two methods commonly used for the prediction of the power output of photovoltaic systems. Both methods are based on Artificial Neural Networks (ANN), which have been trained on the... more
We compare the 24-hour ahead forecasting performance of two methods commonly used for the prediction of the power output of photovoltaic systems. Both methods are based on Artificial Neural Networks (ANN), which have been trained on the same dataset, thus enabling a much-needed homogeneous comparison currently lacking in the available literature. The dataset consists of an hourly series of simultaneous climatic and PV system parameters covering an entire year, and has been clustered to distinguish sunny from cloudy days and separately train the ANN. One forecasting method feeds only on the available dataset, while the other is a hybrid method as it relies upon the daily weather forecast. For sunny days, the first method shows a very good and stable prediction performance, with an almost constant Normalized Mean Absolute Error, NMAE%, in all cases (1% < NMAE% < 2%); the hybrid method shows an even better performance (NMAE% < 1%) for two of the days considered in this analysi...
Research Interests:
Photovoltaics, among renewable energy sources (RES), has become more popular [...]
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
In order to optimize the investment costs in electrical infrastructures, it is increasingly important to exploit power system components as close as possible to their thermal limits. The temperature of overhead transmission lines is one... more
In order to optimize the investment costs in electrical infrastructures, it is increasingly important to exploit power system components as close as possible to their thermal limits. The temperature of overhead transmission lines is one of the major aspects to take into account to ensure safe and reliable grid operation: its estimation allows Transmission System Operators to efficiently manage the grid and to maintain a constant power supply service, eventually implementing proper corrective actions. In the present paper, two physical approaches for overhead lines temperature estimation, namely IEEE and CIGRE models, are tested and compared on four datasets coming from different Italian overhead lines. In detail, they are involved in the five-minutes-ahead conductor temperature estimation. A successful application of the proposed models requires a proper assessment of the line angle, a crucial input parameter to get reliable results. However, it is often unavailable and difficult to be correctly estimated. A specific procedure, aimed at computing the optimal line angle that maximize the accuracy of line temperature estimation, is proposed in this work.