Abstract
In this paper, we investigate two types of photovoltaic (PV) systems (on-grid and off-grid) of different sizes and propose a reliable PV forecasting method. The novelty of our research consists in a weather data-driven feature engineering considering the operation of the PV systems in similar conditions and merging the results of deterministic and stochastic models, namely Machine Learning algorithms (Random Forest—RF, eXtreme Gradient Boost—XGB) and Deep Learning algorithms (Deep Neural Networks—DNN, Gated Recurrent Unit—GRU) into a Hybrid Meta-learning Forecasting method. It combines the estimations of the above-mentioned algorithms with relevant features to predict the PV output using a Long Short-Term Memory model. To design the PV forecast for off-grid systems, that are equally important for prosumers, and approximate the potential power of these systems, the level of load and charging state of the batteries are considered. In this context, feature engineering using the weather and PV output data, including PV characteristics, is relevant to obtaining a performant and robust PV forecast for various use cases taking into account the size and connectivity of the PV systems. On average, the Mean Absolute Error and Mean Absolute Percentage Error have halved compared to values obtained with deterministic methods and are 25% lower than the stochastic models.
Similar content being viewed by others
Notes
Abbreviations
- \(t, h\) :
-
Time interval of the records and the corresponding hour
- \(\Delta t\) :
-
Interval between two time steps
- \(T\) :
-
Number of records corresponding to the training interval
- \(S,s\) :
-
Total number of sources for the weather web API, index of the API source
- \(n,i\) :
-
Number of inverters and corresponding index
- \({XW}^{t}\) :
-
Input set built from weather variables
- \({P}_{i}^{t}\), \(\widehat{{P}_{i}^{t}}\) :
-
Generated and predicted power of inverter i at time t (W)
- \({X}_{i}^{t}\) :
-
Input of the ML&DL models for inverter i (array)
- \({y}_{i}^{t}\) :
-
Target variable of the output power of the inverter i (W)
- \(k\) :
-
Index of the stochastic models (RF, XGB, DNN, GRU) used in predictions
- \(\widehat{{y}_{i,k}^{t}}\) :
-
Predicted values of the ML&DL models for the inverter i (W)
- \({{\text{Wind}}}_{s}^{t}\) :
-
Wind speed at time t provided by the API source s (m/s)
- \({{\text{Cloud}}}_{s}^{t}\) :
-
Cloud cover at time t provided by the API source s (%)
- \({{\text{Temp}}}_{s}^{t}\) :
-
Temperature at time t provided by the API source s (°C)
- \({{\text{Hum}}}_{s}^{t}\) :
-
Humidity at time t provided by the API source s (%)
- \({{\text{Pressure}}}_{s}^{t}\) :
-
Pressure at time t provided by the API source s (Pa)
- \({{\text{UV}}}_{s}^{t}\) :
-
UV index at time t provided by the API source s \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{Precip}}}_{s}^{t}\) :
-
Precipitation at time t provided by the API source s (mm/h)
- \({{\text{Sol}}}_{i}^{t}\) :
-
Effective solar irradiance at time t calculated based on pyranometer measurements for inverter i \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{POA}}}_{i}^{t}\) :
-
Plane of Array irradiance at time t measured by the pyranometer for inverter i \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{Sol}}}_{{\text{STC}}}\) :
-
Reference irradiance or solar irradiance under STC \(\left( {1000\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({\text{SF}}\) :
-
Soiling factor of the modules (approx. 0.97)
- \({n}_{m}, m\) :
-
Number of the installed modules and index of the module
- \({{\text{PD}}}_{m}^{t}, {{\text{PD}}}_{i}^{t}\) :
-
PV module, respectively, PV output of the inverter i calculated using deterministic model based on the measurements of pyranometer (W)
- \({{\text{Sol}}}_{i,s}^{t}\) :
-
Effective solar irradiance at time t determined for the source s of the weather API for the inverter i \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{Sol}}}_{i,0}^{t}\) :
-
Effective solar irradiance at time t determined for the clear sky model for the inverter i \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{P}_{{\text{STC}}}}_{m}\) :
-
Rated capacity of the module m under Standard Test Conditions (STC)
- \({{f}_{p}}_{m}\) :
-
Derating factor of the module m
- \({{\alpha }_{p}}_{m}\) :
-
Temperature coefficient power of the module m(%/°C)
- \({T}_{{\text{cell}} m}^{tf}\) :
-
Cell temperature of the module m (°C)
- \({{T}_{{\text{STC}}}}_{m}\) :
-
Cell temperature under STC (25 °C)
- \({{\eta }_{{\text{Pc}}}}_{m}\) :
-
Power conditioning efficiency of the module m
- \({A}_{m}\) :
-
Area of the module m (m2)
- \({{\eta }_{{\text{STC}}}}_{m}\) :
-
Module efficiency under STC (%)
- \({\eta }_{m}^{t}\) :
-
Module efficiency at time t (%)
- \({{\text{Type}}}_{i}\) :
-
Inverter type on-grid, off-grid, \({{\text{Type}}}_{i}\in \left\{{\text{onG}},\mathrm{ offG}\right\}\)
- \({c}_{j},{n}_{c}\) :
-
Convolution coefficients applied in Savitzky–Golay filter, total number of coefficients
- \({{\text{PLoad}}}_{i}^{t}\) :
-
Load power measured by the inverter i at time t (W)
- \({{\text{PBat}}}_{i}^{t}, {{\text{SOC}}}_{i}^{t}\) :
-
Battery power and SOC measured by the inverter i at time t (%)
- \({{\text{DNIextra}}}^{t}\) :
-
Direct normal extraterrestrial solar irradiance \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{DNI}}}_{s}^{t},{{\text{DHI}}}_{s}^{t},{{\text{GHI}}}_{s}^{t}\) :
-
Direct normal irradiance, diffuse horizontal irradiance and global horizontal irradiance determined for source s \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{AirT}}}_{s}^{t}\) :
-
Atmospheric transmittance at time t calculated based on \({{\text{CC}}}_{s}^{t}\) provided by the API source s (%)
- \({Z}^{t}\) :
-
Apparent zenith angle (°)
- \({{\text{AZ}}}^{t}\) :
-
Solar azimuth angle (°)
- \({\beta }_{i}\) :
-
Tilt angle of the PV arrays connected to inverter i (°)
- \({\Phi }_{i}\) :
-
Azimuth angle of the PV arrays connected to inverter i (°)
- \(\rho \) :
-
Albedo or ground reflectance
- \({{\text{Aoi}}}_{i}^{t}\) :
-
Angle of incidence on the PV arrays connected to inverter i at time t (°)
- \({{\text{POA}}}_{i,s}^{t}, {{\text{POAd}}}_{i,s}^{t},{{\text{POAsky}}}_{i,s}^{t}, {{\text{POAg}}}_{i,s}^{t}\) :
-
Plane of Array total (global) irradiance, direct (beam), sky diffuse and ground diffuse irradiance for source s \(\left( {\frac{{\text{W}}}{{{\text{m}}^{2} }}} \right)\)
- \({{\text{CSol}}}_{i,s}^{t}\) :
-
Weather data-driven variable of the cloud cover and effective solar irradiance
- \({{\text{hSol}}}_{i,s}^{t}\) :
-
Weather data-driven variable of the time interval and effective solar irradiance
- \({{\text{Pmin}}}_{i,s}^{t},{{\text{Pstd}}}_{i,s}^{t}, {{\text{Pmax}}}_{i,s}^{t}, {{\text{Pmean}}}_{i,s}^{t}\) :
-
The analytical values of the generated power for each value of \({{\text{hSol}}}_{i,s}^{t}\) of the inverter i
- \({\text{RMSE}}, {R}^{2}, {\text{MAPE}}, {\text{MAE}}\) :
-
Evaluation metrics of the forecast performance
- \(\widehat{{{\text{PD}}}_{m,s}^{t}}, \widehat{{{\text{PD}}}_{i,s}^{t}}\) :
-
PV module, respectively, PV output of the inverter i estimated for source s using deterministic model (W)
- \({XM}_{i}^{t}\) :
-
Input of the HMF model
- \(p\) :
-
Number of variables (features) of the input \({XM}_{i}^{t}\)
- \({\omega }_{i}\),\({a}^{t}\) :
-
Variables of the LSTM model
- \(d\) :
-
Daily time sequences used in LSTM (\(d=96\))
References
Cho D, Valenzuela J (2020) Optimization of residential off-grid PV-battery systems. Sol Energy. https://doi.org/10.1016/j.solener.2020.08.023
Bâra A, Oprea S-V, Oprea N (2023) How fast to avoid carbon emissions: a holistic view on the RES, storage and non-RES replacement in Romania. Int J Environ Res Public Health. https://doi.org/10.3390/ijerph20065115
Fernández-González R, Puime-Guillén F, Panait M (2022) Multilevel governance, PV solar energy, and entrepreneurship: the generation of green hydrogen as a fuel of renewable origin. Util Policy 79:101438. https://doi.org/10.1016/j.jup.2022.101438
Oprea SV, Bâra A (2020) Ultra-short-term forecasting for photovoltaic power plants and real-time key performance indicators analysis with big data solutions. Two case studies—PV Agigea and PV Giurgiu located in Romania. Comput Ind. https://doi.org/10.1016/j.compind.2020.103230
Shouman ER, El Shenawy ET, Khattab NM (2016) Market financial analysis and cost performance for photovoltaic technology through international and national perspective with case study for Egypt. Renew Sustain Energy Rev. https://doi.org/10.1016/j.rser.2015.12.074
Dongol D, Feldmann T, Schmidt M, Bollin E (2018) A model predictive control based peak shaving application of battery for a household with photovoltaic system in a rural distribution grid. Sustain Energy Grids Netw. https://doi.org/10.1016/j.segan.2018.05.001
Oprea SV, Bâra A (2021) Edge and fog computing using IoT for direct load optimization and control with flexibility services for citizen energy communities. Knowl Based Syst. https://doi.org/10.1016/j.knosys.2021.107293
Rodríguez-Gallegos CD, Vinayagam L, Gandhi O, Yagli GM, Alvarez-Alvarado MS, Srinivasan D et al (2021) Novel forecast-based dispatch strategy optimization for PV hybrid systems in real time. Energy. https://doi.org/10.1016/j.energy.2021.119918
Mandelli S, Brivio C, Colombo E, Merlo M (2016) Effect of load profile uncertainty on the optimum sizing of off-grid PV systems for rural electrification. Sustain Energy Technol Assess. https://doi.org/10.1016/j.seta.2016.09.010
Richter B, Golla A, Welle K, Staudt P, Weinhardt C (2021) Local energy markets—an IT-architecture design. Energy Inf. https://doi.org/10.1186/s42162-021-00164-6
Liu C, Chai KK, Zhang X, Chen Y (2021) Peer-to-peer electricity trading system: smart contracts based proof-of-benefit consensus protocol. Wirel Netw. https://doi.org/10.1007/s11276-019-01949-0
Oprea SV, Bâra A (2021) Devising a trading mechanism with a joint price adjustment for local electricity markets using blockchain. Insights for policy makers. Energy Policy. https://doi.org/10.1016/j.enpol.2021.112237
Kuznetsova E, Anjos MF (2021) Prosumers and energy pricing policies: When, where, and under which conditions will prosumers emerge? A case study for Ontario (Canada). Energy Policy. https://doi.org/10.1016/j.enpol.2020.111982
Gigoni L, Betti A, Crisostomi E, Franco A, Tucci M, Bizzarri F, Mucci D (2018) Day-ahead hourly forecasting of power generation from photovoltaic plants. IEEE Trans Sustain Energy. https://doi.org/10.1109/TSTE.2017.2762435
Ayop R, Tan CW, Syed Nasir SN, Daud MZ, Yiew LK, Nordin NM, Bukar AL (2022) The performances of partial shading adjuster for improving photovoltaic emulator. Int J Power Electron Drive Syst. https://doi.org/10.11591/ijpeds.v13.i1.pp528-536
Abdel-Nasser M, Mahmoud K (2019) Accurate photovoltaic power forecasting models using deep LSTM-RNN. Neural Comput Appl. https://doi.org/10.1007/s00521-017-3225-z
Agga A, Abbou A, Labbadi M, El Houm Y (2021) Short-term self consumption PV plant power production forecasts based on hybrid CNN-LSTM, ConvLSTM models. Renew Energy. https://doi.org/10.1016/j.renene.2021.05.095
Luo X, Zhang D, Zhu X (2021) Deep learning based forecasting of photovoltaic power generation by incorporating domain knowledge. Energy. https://doi.org/10.1016/j.energy.2021.120240
Androniceanu A, Georgescu I (2023) The impact of CO emissions and energy consumption on economic growth: a panel data analysis. Energies. https://doi.org/10.3390/en16031342
Han Y, Wang N, Ma M, Zhou H, Dai S, Zhu H (2019) A PV power interval forecasting based on seasonal model and nonparametric estimation algorithm. Sol Energy. https://doi.org/10.1016/j.solener.2019.04.025
El-Baz W, Tzscheutschler P, Wagner U (2018) Day-ahead probabilistic PV generation forecast for buildings energy management systems. Sol Energy. https://doi.org/10.1016/j.solener.2018.06.100
David M, Boland J, Cirocco L, Lauret P, Voyant C (2021) Value of deterministic day-ahead forecasts of PV generation in PV + storage operation for the Australian electricity market. Sol Energy. https://doi.org/10.1016/j.solener.2021.06.011
Nguyen TN, Müsgens F (2022) What drives the accuracy of PV output forecasts? Appl Energy 323:119603
VanDeventer W, Jamei E, Thirunavukkarasu GS, Seyedmahmoudian M, Soon TK, Horan B et al (2019) Short-term PV power forecasting using hybrid GASVM technique. Renew Energy. https://doi.org/10.1016/j.renene.2019.02.087
Kushwaha V, Pindoriya NM (2019) A SARIMA-RVFL hybrid model assisted by wavelet decomposition for very short-term solar PV power generation forecast. Renew Energy. https://doi.org/10.1016/j.renene.2019.03.020
Das S (2019) Short term forecasting of solar radiation and power output of 89.6kWp solar PV power plant. Mater Today Proc. https://doi.org/10.1016/j.matpr.2020.08.449
Pierro M, Gentili D, Liolli FR, Cornaro C, Moser D, Betti A et al (2022) Progress in regional PV power forecasting: a sensitivity analysis on the Italian case study. Renew Energy. https://doi.org/10.1016/j.renene.2022.03.041
Qin J, Jiang H, Lu N, Yao L, Zhou C (2022) Enhancing solar PV output forecast by integrating ground and satellite observations with deep learning. Renew Sustain Energy Rev 167:112680
Netsanet S, Zheng D, Zhang W, Teshager G (2022) Short-term PV power forecasting using variational mode decomposition integrated with Ant colony optimization and neural network. Energy Rep. https://doi.org/10.1016/j.egyr.2022.01.120
King DL, Boyson WE, Kratochvil JA (2004) Photovoltaic array performance model. Sandia Report No. 2004–3535. https://doi.org/10.2172/919131
Hossain M, Mekhilef S, Danesh M, Olatomiwa L, Shamshirband S (2017) Application of extreme learning machine for short term output power forecasting of three grid-connected PV systems. J Clean Prod. https://doi.org/10.1016/j.jclepro.2017.08.081
El-Rafey E, El-Sherbiny M (1988) Load/weather/insolation database for estimating photovoltaic array and system performance in Egypt. Sol Energy. https://doi.org/10.1016/0038-092X(88)90056-4
Olatomiwa L, Mekhilef S, Huda ASN, Ohunakin OS (2015) Economic evaluation of hybrid energy systems for rural electrification in six geo-political zones of Nigeria. Renew Energy. https://doi.org/10.1016/j.renene.2015.04.057
Savitzky A, Golay MJE (1964) Smoothing and differentiation of data by simplified least squares procedures. Anal Chem. https://doi.org/10.1021/ac60214a047
Buchman I (2001) Batteries in a portable world: a handbook on rechargeable batteries for non-engineers. Chemistry & ….
Pal AM, Das S (2015) Analytical model for determining the sun’s position at all time zones. Int J Energy Eng 5(3):58–65. https://doi.org/10.5923/j.ijee.20150503.03
Reno MJ, Hansen CW, Stein JS (2012) Global horizontal irradiance clear sky models: implementation and analysis. SANDIA REPORT SAND2012–2389 Unlimited Release Printed March 2012
Holmgren F, Hansen WC, Mikofski AM (2018) pvlib python: a python package for modeling solar energy systems. J Open Sour Softw. https://doi.org/10.21105/joss.00884
Campbell GS, Norman JM (1998) An introduction to environmental biophysics. Introd Environ Biophys. https://doi.org/10.1007/978-1-4612-1626-1
Hay JE, Davies JA (1980) Calculation of the solar radiation incident on an inclined surface. In: Proceedings first canadian solar radiation data workshop
Klucher TM (1979) Evaluation of models to predict insolation on tilted surfaces. Sol Energy. https://doi.org/10.1016/0038-092X(79)90110-5
Lave M, Hayes W, Pohl A, Hansen CW (2015) Evaluation of global horizontal irradiance to plane-of-array irradiance models at locations across the United States. IEEE J Photovolt. https://doi.org/10.1109/JPHOTOV.2015.2392938
Reindl DT, Beckman WA, Duffie JA (1990) Evaluation of hourly tilted surface radiation models. Sol Energy. https://doi.org/10.1016/0038-092X(90)90061-G
Oprea SV, Bâra A (2022) Feature engineering solution with structured query language analytic functions in detecting electricity frauds using machine learning. Sci Rep. https://doi.org/10.1038/s41598-022-07337-7
Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, et al (2017) LightGBM: a highly efficient gradient boosting decision tree. In: Advances in neural information processing systems
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553):436–444
Su Y, Kuo CCJ (2019) On extended long short-term memory and dependent bidirectional recurrent neural network. Neurocomputing. https://doi.org/10.1016/j.neucom.2019.04.044
Goodfellow I, Bengio Y, Courville A (2016) Deep learning an MIT Press book. In Nature
Acknowledgements
This work was supported by a grant of the Ministry of Research, Innovation and Digitization, CNCS-UEFISCDI, project number PN-III-P4-PCE-2021-0334, within PNCDI III.
Author information
Authors and Affiliations
Contributions
A.B. and S.V.O. wrote the main manuscript text and A.B. prepared all figures and the concept idea, including the revision of the manuscript. S.V.O. created all table and performed the checking of the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Oprea, SV., Bâra, A. On-grid and off-grid photovoltaic systems forecasting using a hybrid meta-learning method. Knowl Inf Syst 66, 2575–2606 (2024). https://doi.org/10.1007/s10115-023-02037-8
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10115-023-02037-8