The accurate forecast of the photovoltaic generation (PVG) process is essential to develop optimum installation sizing and pragmatic energy planning and management. This paper proposes a PVG forecast model for a PVG/Battery installation. The forecasting strategy is built on a Medium-Term Energy Forecasting (MTEF) approach refined dynamically every hour (Dynamic Medium-Term Energy Forecasting (DMTEF)) and adjusted by means of a Short-Term Energy Forecasting (STEF) strategy. The MTEF predicts the generated energy for a day ahead based on the PVG of the last 15 days. As for STEF, it is a combination between PVG Short-Term (ST) forecasting and DMTEF methods obtained by selecting the least inaccurate PVG estimation every 15 minutes. The algorithm results are validated by measures taken on a 3 KWp standalone PVG/Battery installation. The proposed approaches have been integrated into a management algorithm in order to make a pragmatic decision to ensure load supply considering relevant constraints and priorities and guarantee the battery safety. Simulation results show that STEF provides accurate results compared to measures in stable and perturbed days. The NMBE (Normalized Mean Bias Error) is equal to −0.58% in stable days and 26.10% in perturbed days.

Accurate PVG forecast is necessary in most photovoltaic systems especially for decision making of energy planning and management and for system sizing. In literature, several models were proposed in order to ensure PVG forecasting. They are classified into three major categories: statistical, physical and satellite-derived methods [

Physical methods use Numerical Weather Prediction (NWP) models to forecast meteorological variables such as radiation and temperature with high forecasting accuracy. Forecasted meteorological parameters are fitted as inputs to the photovoltaic generator model to provide PVG forecast. Physical approaches require more complex calculations and input data than statistical approaches [

In order to improve the PVG forecasting results obtained in previous work in [

Section two presents the PVG forecasting strategy and provides details about MTEF, DMTEF and STEF approaches. Obtained results are discussed in section three with reference to a representative case study. Finally, some notes are highlighted by way of conclusion to this work in the fourth and last section.

The PVG forecasting approach consists of three sorts of dynamic forecasting (

The MTEF approach consists of performing a dynamic estimation of

The

where

The obtained

where

The time horizon selection is performed in order to test the forecasting accuracy of ARMA model by using the MTEF approach. The test is performed based on two kinds of data base. A data base of 365 days of the year 2019 and a seasonal data base organized into three seasons: a cold season (November, December, January and February), a moderate season (March, April, May and October), and a hot season (June, July, August and September). Furthermore, MTEF strategy is executed on three different time horizons (5 days, 10 days and 15 days) in order to select the best time horizon with the highest forecasting accuracy.

In order to evaluate the PVG forecasting accuracy, daily predicted and measured data were analyzed by computing the NRMSE and the NMBE errors for each prediction type:

Time horizon | Date | NRMSE (%) | NMBE (%) |
---|---|---|---|

15 days | 01/01/2019 to 15/01/2019 | 1.57 | −0.98 |

10 days | 01/01/2019 to 10/01/2019 | 1.81 | −1.73 |

5 days | 01/01/2019 to 5/01/2019 | 3.94 | −7.02 |

Database | Saison | NRMSE (%) | NMBE (%) |
---|---|---|---|

15 days | Hot | 0.85 | −1 |

Moderate | 1.68 | −0.83 | |

Cold | 2.40 | −3.25 | |

10 days | Hot | 1.07 | −1.38 |

Moderate | 1.93 | −1.77 | |

Cold | 2.80 | −3.64 | |

5 days | Hot | 2.02 | −0.95 |

Moderate | 4.48 | −11.41 | |

Cold | 5.81 | −9.95 |

The MTEF reveals the best PVG forecasting accuracy using 15 days’ time horizon for the year 2019 data base with NRMSE = 1.57% and NMBE = −0.98%.

The MTEF reveals the best PVG forecasting accuracy over 15 days’ time horizon for the three seasons. The best season for the selected time horizon is the hot season with NRMSE = 0.85% and NMBE = −1%. The obtained results are explained by the high weather stability in the hot season compared to moderate and cold season.

The DMTEF approach consists of performing the phenomenon of dynamicity by effectuating an hourly adjustment of the

The STEF approach is a combination of PVG Short-Term (ST) forecasting and DMTEF approach. It consists of adjusting every 15 min the

where

Otherwise,

The MTEF, DMTEF and STEF approaches are to be integrated into a management algorithm of a standalone PVG/Battery installation (

In order to test the forecasting accuracy of ARMA model based on MTEF approach, a data base of 15 days for a stable and perturbed period is adopted considering the obtained results of time horizon selection presented in

Both figures reveal a good forecasting accuracy results of MTEF approach based on ARMA model in both periods.

Period | NRMSE (%) |
---|---|

Stable | 5.53 |

Perturbed | 11.36 |

Both figures show good results but not optimal. From sunrise to sunset, a significant dissimilarity is noticed between the distribution of the forecasted

Day | NMBE (%) |
---|---|

Stable | −4.48 |

Perturbed | 49.61 |

The DMTEF approach is used to perform dynamicity on the MTEF approach by effectuating an adjustment every hour of the estimated

The obtained results reveal an optimization of PVG estimation based on DMTEF. The proposed dynamicity offers a noticeable increasing of The PVG estimation accuracy between the distribution of the adjusted

Day | NMBE (%) |
---|---|

Stable | −0.91 |

Perturbed | 29.20 |

The STEF approach makes use of ST forecasting in order to forecast PVG every 15 min considering the previous measured 2 h based on the ARIMA model. Such approach performs an adjustment of the previous estimated

The obtained results reveal an optimization of PVG estimation based on ST forecasting. Such approach offers a noticeable increasing of The PVG estimation accuracy between the distribution of the estimated

The obtained results reveal an optimization of PVG estimation based on STEF. An increasing of the PVG estimation accuracy is clearly noticed between the distribution of the adjusted

Day | NMBE (%) |
---|---|

Stable | −0.58 |

Perturbed | 26.10 |

The standalone PVG/Battery installation system is used to test the impact of the accuracy of the proposed STEF approach on the loads operation in compared to what it was previously with MTEF approach. The battery bank is used as a precaution source in anticipation of worst cases (lack of load supply by PVG source observed after performing STEF approach). As for loads, the major adjustment will appear in the water pump profile considering the continuous supply of the cold room.

The obtained results of

A dynamic forecasting of Photovoltaic Energy Generation (PVG) was developed and experimentally tested in a management system of a standalone PVG/Battery installation. The methodology consists of forecasting PVG for three time period ahead: daily, hourly and for each 15 min. First, the MTEF approach forecasts the PVG using an ARMA model for a day ahead based on 15 days data base of measures. Then the DMTEF approach adjusts the hourly obtained PVG by MTEF approach considering the relative error. Finally, a STEF approach combines DMTEF approach and PVG Short-Term (ST) forecasting, based on ARIMA model, to adjust estimated PVG every 15 min. Obtained results show good concordance between the three kinds of forecast with measures. Besides, the load operation is improved and the battery safety is respected in stable and perturbed days. Future works aims to improve the accuracy of the proposed strategy in more than one day (stable and perturbed) and with different PVG installations (Grid connected and hybrid PVG systems).

Daily estimated Photovoltaic power Generation (W)

Hourly estimated Photovoltaic power Generation (W)

Estimated Photovoltaic power Generation every couple of minutes (W)

ARMA order parameters

Auto-regressive coefficient

Moving average coefficient

White noise

Time

Sunrise time

Sunset time

Backward shift operator

Backward difference

Polynomial of order p

Polynomial of order q

Photovoltaic Generation

Medium-Term Energy Forecasting

Dynamic Medium-Term Energy Forecasting

Short-Term Energy Forecasting

Normalized Mean Bias Error

Machine Learning

Support Vector Machine

Artificial Neural Network

Fuzzy Logic

Auto-Regressive Moving Average

Numerical Weather Prediction

Short-Term

Relative Error

Auto-Regressive Integrated Moving Average

Normalized Root Mean Square Error

State of Charge

This work has been developed thanks to the support of SIME Laboratory, Tunisia and the Department of Electric, Electronic and Computer Engineering, University of Catania, Italy.