Predictive model deep learning driven forecasting for photovoltaic power generation

Abstract

Abstract: The present invention is a predictive model deep learning driven forecasting for photovoltaic power generation a method claim utilizing a Long Short-Term Memory (LSTM) neural network enhanced by feature selection methods using Dragonfly (DF) and Firefly (FF) algorithms therein, a hybrid deep learning model (LSTM-DF-FF) is developed to predict system performance, and its accuracy is assessed using performance metrics. Under ideal conditions, the model achieved a relative error of only 3.5%. However, predictions were less accurate on cloudy and rainy days, with relative errors of 7.8% and 10.1%, respectively and the LSTM-DF-FF model showed a high correlation with actual performance, achieving a correlation coefficient of 0.997 in clear conditions and 0.991 during overcast weather.

Specification

Description:Title of the invention:
Predictive model deep learning driven forecasting for photovoltaic power generation


Field of the invention:
The present invention relates to the field of solar photovoltaic and particular relates to field of a predictive model deep learning driven forecasting for photovoltaic power generation.

Prior art to the invention:
1. A patent document with application number "US20230140233" titled "Renewable energy system and electrical grid" is described here, "A system, method, and solar photovoltaic (PV) network for solar PV variability reduction with reduced time delays and battery storage optimization are described. The system includes a Moving Regression (MR) filter; a State of Charge (SoC) feedback control; and a Battery Energy Storage System (BESS). The MR filter, SoC feedback control and BESS are configured to provide smoothing of solar PV variabilities. The MR filter is a non-parametric smoother that utilizes a machine learning concept of linear regression to smooth out solar PV variations at every time step."

wherein, the present invention is a predictive model deep learning driven forecasting for photovoltaic power generation.

Objects of the invention:
It is a primary object of the present invention is a predictive model deep learning driven forecasting for photovoltaic power generation.
Summary of the invention:
An aspect of the present invention of a predictive model deep learning driven forecasting for photovoltaic power generation.





Detailed description:
The following specification particularly describes the invention and the manner in which it is to be performed.

The present invention will be further described in detail below through specific embodiments.

An embodiment of the present invention is a Building Integrated Photovoltaic (BIPV) technology is a rapidly advancing field that leverages solar energy for power generation within modern buildings, transforming them from mere energy consumers to energy producers. BIPV systems' efficiency is influenced by location, seasonal variations, and environmental conditions. Accurate performance forecasting of BIPV systems is essential for optimizing energy predictions and consumption patterns. This study focuses on predicting the output power of BIPV systems using inputs like solar radiation, ambient temperature, and wind speed in hot and humid environments. A novel approach is introduced, utilizing a Long Short-Term Memory (LSTM) neural network enhanced by feature selection methods using Dragonfly (DF) and Firefly (FF) algorithms. The hybrid deep learning model (LSTM-DF-FF) was developed to predict system performance, and its accuracy was assessed using performance metrics. Under ideal conditions, the model achieved a relative error of only 3.5%. However, predictions were less accurate on cloudy and rainy days, with relative errors of 7.8% and 10.1%, respectively. The LSTM-DF-FF model showed a high correlation with actual performance, achieving a correlation coefficient of 0.997 in clear conditions and 0.991 during overcast weather.
, Claims:Claims:

I claim,
1. A predictive model deep learning driven forecasting for photovoltaic power generation a method claim utilizing a Long Short-Term Memory (LSTM) neural network enhanced by feature selection methods using Dragonfly (DF) and Firefly (FF) algorithms therein, a hybrid deep learning model (LSTM-DF-FF) is developed to predict system performance, and its accuracy is assessed using performance metrics, and
wherein, under ideal conditions, the model achieved a relative error of only 3.5%. However, predictions were less accurate on cloudy and rainy days, with relative errors of 7.8% and 10.1%, respectively and the LSTM-DF-FF model showed a high correlation with actual performance, achieving a correlation coefficient of 0.997 in clear conditions and 0.991 during overcast weather.

Documents