ArticlePDF Available

Forecasting degradation rates of different photovoltaic systems using Robust Principal Component Analysis and ARIMA

Wiley
IET Renewable Power Generation
Authors:

Abstract and Figures

Degradation rates based on forecasting of performance ratio, Rp, time series are computed and compared with actual degradation rates. A three year forecasting of monthly Rp, measured from PV connected systems of various technologies is performed using the seasonal ARIMA (SARIMA) time series model. The seasonal ARIMA model is estimated using monthly Rp measured over a 5 year period and based on this model forecasting is implemented for the subsequent three years. The degradation rate at the end of the forecasting period, eighth year, is computed using a robust principal component analysis (RCPA) based methodology. The degradation rates obtained for various (PV) systems are then compared to the ones obtained using the actual eight year data.
Content may be subject to copyright.
A preview of the PDF is not available
... Temporal aggregation into monthly intervals was applied to mitigate inconsistencies due to irregular publication patterns and clustering of publication dates, creating a uniform time series structure (Granger, 1990). Aggregating data monthly reduces noise and improves the stability of trend analysis, which is essential for accurate model estimation in time series analysis (Pieri et al., 2017). This is particularly relevant for minimizing inconsistencies in publication patterns and stabilizing trend analysis. ...
... This process is especially beneficial when the objective is to isolate a stable forecast that is not contaminated by the intervention effects (Pieri et al., 2017). PCA reduces dimensionality, ensuring robust and independent components, crucial for improving accuracy and stability in multivariate time series forecasts. ...
... Synthesizing the counterfactual scenario from uncorrelated principal components enabled a stable forecast for comparing actual post-intervention impacts. This method allowed Pieri et al., (2017) to capture natural degradation patterns despite external irregularities. The synthetic baseline approach helps construct a theoretical trajectory of citation impacts and analyze shifts in research dissemination that alter impact metrics like FWCI. ...
Article
This research aims to assess the causal effects of the global crisis, namely the COVID-19 pandemic, on the dissemination of research in open science by examining changes in research impact using field-normalized measures. This is done by conceptualizing the pandemic-related changes as a coercive force that accelerated the adoption of open science practices, thus improving their recognition in different scientific disciplines. Using a Causal Autoregressive Integrated Moving Average (C-ARIMA) model, the study contrasts counterfactual scenarios to evaluate the potential impact of research in the absence of the pandemic against what was empirically observed. The pandemic considerably increased the FWCI of open science research, reaching a peak in February 2022. The overall effect suggests continued recognition of open science after the pandemic. The average monthly increase in FWCI was 0.335 above the expected growth, with a relative effect ranging from 15 to 23%. The pandemic hastened the influence of open science, with fields like social science, computer science, medicine, and engineering experiencing significant unexpected changes, while business management and accounting, or arts and humanities do not noted a significant effect. Pandemic restrictions and an increased need for transparency were likely major catalysts to improve the visibility and impact of open science in academia.
... Zhou et al. [6] proposed another spatio-temporal probabilistic forecasting model utilizing monotone broad learning system and the Copula theory. Elena et al. [7] used an auto-regressive integrating time series model to forecast the power of photovoltaic systems connecting of various technologies. Zhao et al. [8] proposed an auto-regressive and moving average model for simulating the chronological characteristics of the solar radiation. ...
... For a given query vector, this attention mechanism calculates the attention between the query vector and key values, and applies it as weights to the objective values, resulting in the attention values. The calculation formula is expressed by Equations (5)- (7). ...
Article
Full-text available
Recent years, the tremendous number of distributed photovoltaic are integrated into low‐voltage distribution network, generating a significant amount of operational data. The centralized cloud data centre is unable to process the massive data precisely and promptly. Therefore, the operational status of distributed photovoltaic systems in low‐voltage distribution network becomes difficult to predict. However, edge computing in the distribution network enable local processing of data to improve the real‐time and reliability of the forecasting service. In this regard, this paper proposes a distributed photovoltaic short‐term power forecasting model based on lightweight AI algorithms. Firstly, based on the Pearson correlation coefficient method, an analysis is conducted on the historical operational data in the network to extract important meteorological features that are correlated with the photovoltaic power output. Secondly, a distributed photovoltaic power forecasting model for the distribution network is constructed based on the Xception and attention mechanism. Finally, the model is trained using pruning, which involves removing redundant parts of the model, resulting in a compact and efficient forecasting model. By conducting validation on real‐world datasets, the results demonstrate that the model presented in this article possesses a smaller size and higher forecasting accuracy compared to other state‐of‐the‐art forecasting models.
... Subsequently, using the classic PR, the effect of seasonality is considered in the present study. Classical Decomposition [6] separates the signal into three terms: Signal = Trend + Seasonality + Irregular. ...
... Jordan/Fourier [6] ( ) = 0 + 1 sin (( 6 ) • ( − )) + 1 sin (( 3 ) • ( − )) (7) ...
... Statistical methods such as ARIMA [9], physical methods, such as solar cell modelling [10,11], and satellite images [12] are also used to forecast the power generated by solar modules. However, machine learning and especially deep learning have shown impressive performance in forecasting as these techniques can deal with complex data [13]. ...
Article
Full-text available
Deep learning is efficiently used for photovoltaic power generation forecasting to handle the intermittent nature of solar energy. However, big data are required for training deep networks which are not available for newly installed plants. Therefore, in this study, a novel strategy is proposed to train a deep learning model using a transfer learning technique to cop up with the unavailability of enough training datasets. A new 400 kWp solar power plant installed in the Himalayan region is considered as a case study to evaluate the proposed model. The proposed approach utilizes solar radiation data to train a deep neural network and then fine-tune the model using the power generation data from the plant. The network architecture is optimized using grey wolf optimizer to find the best suitable model for the data. The evaluation results show that the same model can achieve higher performance in generation forecasting with percentage error improved by 2% and R-value increased by 7.7% after applying transfer learning. Moreover, SHapley Additive exPlanation and Partial Dependence Plots are used to interpret the model behavior and showed that the model is mostly dependent on the previous generation values (up to 4 days) followed by the temperature and solar radiation.
... This method analyzes historical power output data along with related influencing factors to identify their correlation patterns and forecast future output power accordingly. Frequently used statistical methods encompass the Auto-Regression and Moving Average (ARMA) model, along with its variations [9][10][11], and the Autoregressive Integrated Moving Average (ARIMA) model [12,13]. Machine learning approaches mainly include models like Support Vector Machines (SVMs) [14], Artificial Neural Networks (ANNs) [15], and Extreme Learning Machines (ELMs) [16]. ...
Article
Full-text available
The accurate probabilistic forecasting of ultra-short-term power generation from distributed photovoltaic (DPV) systems is of great significance for optimizing electricity markets and managing energy on the user side. Existing methods regarding cluster information sharing tend to easily trigger issues of data privacy leakage during information sharing, or they suffer from insufficient information sharing while protecting data privacy, leading to suboptimal forecasting performance. To address these issues, this paper proposes a privacy-preserving deep federated learning method for the probabilistic forecasting of ultra-short-term power generation from DPV systems. Firstly, a collaborative feature federated learning framework is established. For the central server, information sharing among clients is realized through the interaction of global models and features while avoiding the direct interaction of raw data to ensure the security of client data privacy. For local clients, a Transformer autoencoder is used as the forecasting model to extract local temporal features, which are combined with global features to form spatiotemporal correlation features, thereby deeply exploring the spatiotemporal correlations between different power stations and improving the accuracy of forecasting. Subsequently, a joint probability distribution model of forecasting values and errors is constructed, and the distribution patterns of errors are finely studied based on the dependencies between data to enhance the accuracy of probabilistic forecasting. Finally, the effectiveness of the proposed method was validated through real datasets.
... The r-squared determination coefficient was quite high, confirming the good choice of the model. Because they control how the moving operator q and the autoregressive operator p are applied, the ACF and PACF functions are crucial [83]. This approach controls errors due to seasonal variation and other flaws, making it very versatile [84]. ...
Article
Full-text available
Biogas is a renewable energy source (RES). The aim of this research was to analyze the perspectives of electricity production from biogas in the European Union. The main source of information was data from Eurostat. We analyzed electricity production from biogas in the European Union (EU). The scope of this research was data from 2012 to 2021. First, we presented biogas production by feedstock type across the world. Then, we presented changes in electricity production from biogas in the EU. We used different methods to evaluate the changes in biogas production. First, we used the ARiMA (Autoregressive Moving Average) model to evaluate the stationarity of the time series. Our electricity production from biogas data proved to be stationary. Second, we elaborated on the prognosis of future changes in electricity production from biogas. The largest producer of biogas is the EU, and it is produced from crops, animal manure, and municipal solid waste. Our research found that the largest production from biogas in 2021 took place in Germany, Italy, and France. These countries have the greatest potential for electricity production from biogas, and they have spent significant funds on facilities and technology. Such countries as Ireland, Greece, Spain, France, Croatia, Poland, Portugal, Romania, Finland, and Sweden increased their electricity production from biogas in 2021 compared to 2020. According to our prognosis, the global production of biogas will increase from 62.300 TWh to 64.000 TWh in 2019–2026 (2.7% increase). In 2022–2026, such countries as Estonia (60.4%), Latvia (29.6%), Croatia (27.6%), Slovenia (10.9%), and Poland (8.2%) will increase their electricity production from biogas the most. In 2022–2026, such countries as Italy (0.68%), Portugal (1.1%), Greece (1.5%), Slovakia (2.3%), and Germany (2.6%) will increase their electricity production from biogas the least. Only Romania (−17.6%), Finland (−11.5%), Lithuania (−9.1%), and Malta (−1.06%) will decrease their production of electricity from biogas in 2022–2026. Such countries as Bulgaria (2344%), Denmark (590.9%), Croatia (449%), and France (183%) increased biogas consumption in 2013–2022. A decrease in the inland consumption of biogas in 2013–2022 was observed in Spain, Cyprus, Latvia, Luxembourg, Austria, and Slovenia.
... A summary of current deep learning-based techniques for predicting wind and solar power was published by Wang et al [9]. Some scientists use factual techniques to predict Solar Power generation, such as Autoregressive Moving Normal (ARMA) [10], Autoregressive Incorporated Moving Normal (ARIMA) [11], and Autoregressive Moving Normal Model with Exogenous Data Sources (ARMAX) [12]. However, these models are insufficient to increase the nonlinear time series data of SPG's forecasting precision. ...
Article
Full-text available
Prediction of Solar power generation plays an important role to improve the efficiency ofeconomic dispatch function and reduce the dependence on fossil fuels and help in the energy managementsystem. For time series solar energy prediction multiple models were introduced but these model trains arebased on yearly historical data. A big data collection containing many missing values makes these modeltraining more complicated that’s why In this paper, an efficient energy prediction model is proposed for theprediction of time series solar energy based on short predicted weather training data. Two complimentarymodels are based on linear regression and a knowledge based neural network is exploited to predict futuresolar power, with offline training. The LR is structured under the direction of the proposed input methodparameter selection and used when training data is enough. KBNN is used for existing advantagespredictive models are also very important when training data is not enough. According to test findings usingreal data sets. An LR model can deal effectively with linear data, but a KBNN model can cope effectivelywith nonlinear behavior. Additionally, the results demonstrate the effectiveness of LR showing a correlationcoefficient (R2) is 98% with a root mean square error of 45 and KBNN shows a correlation coefficient (R2)is 99% with a root mean square error of 44 in providing a reliable version, The results additionally showthe functionality of LR and KBNN in imparting a dependable version, especially when the short trainingdataset is available.
... [4]. Various statistical forecasting models have been developed by the researcher such as autoregressive moving average (ARMA) [5], autoregressive integrated moving average (ARIMA) [6], autoregressive integrated moving average with exogenous inputs (ARMAX) [7], and statistical time series model [8] which supports linear data, but it does not give promising results for nonlinear data. Various methods have been presented in the literature to handle nonlinear data such as optimization algorithms, machine learning, fuzzy-based system, and hybridization of the algorithms. ...
Article
Forecasting Solar Power is an important aspect for power trading companies. It helps in energy bidding, planning, and control. The challenge in forecasting is to predict nonlinear data, which can be fulfilled by the computation technique and machine learning model. ML models have high accuracy for time-series forecasting, but their accuracy is poor for nonlinear forecasting. To enhance the ML model for nonlinear prediction, an optimization algorithm is used for training. This paper presents how the computation technique is incorporated into the machine learning model and compared it with the conventional model. CSA-ANN and SOA-ANN models are developed and forecast solar power for a-day-ahead, three-day-ahead, and a week-ahead solar power generation by considering time, irradiation, and temperature as input parameters for the model. The models are compared with ANN, DE-ANN, and PSO-ANN since these models are widely used. Upon comparison, ANN gives the best result for short-term prediction but is unable to predict midterm and long-term predictions, whereas this problem is overcome by SOA-ANN, which is done by changing its training algorithm, and its performance is measured via statistical parameters such as MAE, MSE, MAPE, and R2. The percentage improvement of SOA-ANN is obtained with these statistical parameters as 6.54%, 16.05%, 1.67%, and 3.61%. Hence, SOA-ANN gives best result as compared to other models.
Article
Full-text available
This paper provides a review of methodologies for measuring the degradation rate, RD, of photovoltaic (PV) technologies, as reported in the literature. As presented in this paper, each method yields different results with varying uncertainty depending on the measuring equipment, the data qualification and filtering criteria, the performance metric and the statistical method of estimation of the trend. This imposes the risk of overestimating or underestimating the true degradation rate and, subsequently, the effective lifetime of a PV module/array/system and proves the need for defining a standardized methodology. Through a literature search, four major statistical analysis methods were recognized for calculating degradation rates: (1) Linear Regression (LR), (2) Classical Seasonal Decomposition (CSD), (3) AutoRegressive Integrated Moving Average (ARIMA) and, (4) LOcally wEighted Scatterplot Smoothing (LOESS), with LR being the most common. These analyses were applied on the following performance metrics: (1) electrical parameters from IV curves recorded under outdoor or simulated indoor conditions and corrected to STC, (2) regression models such as the Photovoltaics for Utility Scale Applications (PVUSA) and Sandia models, (3) normalized ratings such as Performance Ratio, RP, and PMPP/GI and, (4) scaled ratings such as PMPP/Pmax, PAC/Pmax and kWh/kWp. The degradation rate results have shown that the IV method produced the lowest RD and LR produced results with large variation and the largest uncertainty. The ARIMA and LOESS methods, albeit less popular, produced results with low variation and uncertainty and with good agreement between them. Most importantly, this review showed that the RD is not only technology and site dependent, but also methodology dependent.
Article
Grid-connected photovoltaic (PV) systems have become a significant constituent of the power supply mix. A challenge faced by both users and suppliers of PV systems is that of defining and computing a reliable metric of annual degradation rate while in service. This paper defines a new measure to calculate the degradation rate of PV systems from the PV field measured performance ratio (PR). At first, the PR time series is processed by conventional principal component analysis, which yields seasonality as the dominant data feature. The environment, operating conditions, uncertainty, and hardware used for monitoring influence the outdoor measurements unpredictably. These influences are viewed as perturbations that render the dominant feature obtained by PCA unsuitable to be used in a degradation rate definition. Robust principal component analysis (RPCA) is proposed to alleviate these effects. The new measure is defined as the area enclosed by the time series of the corrected by the RPCA annual monthly PR values. The degradation rates obtained for different technologies are compared with those obtained in previous studies. The results have shown that the degradation rates estimated by RPCA were in good agreement with previous investigations and provided increased confidence due to mitigation of uncertainty.
Conference Paper
Long-term reliability is critical to the cost effectiveness and commercial success of photovoltaic (PV) products. Today most PV modules are warranted for 25 years, but there is no accepted test protocol to validate a 25-year lifetime. The qualification tests do an excellent job of identifying design, materials, and process flaws that are likely to lead to premature failure (infant mortality), but they are not designed to test for wear-out mechanisms that limit lifetime. This paper presents a method for evaluating the ability of a new PV module technology to survive long-term exposure to specific stresses. The authors propose the use of baseline technologies with proven long-term field performance as controls in the accelerated stress tests. The performance of new-technology modules can then be evaluated versus that of proven-technology modules. If the new-technology demonstrates equivalent or superior performance to the proven one, there is a high likelihood that they will survive versus the tested stress in the real world.
Article
We evaluate and compare several forecasting techniques using no exogenous inputs for predicting the solar power output of a 1 MWp, single-axis tracking, photovoltaic power plant operating in Merced, California. The production data used in this work corresponds to hourly averaged power collected from November 2009 to August 2011. Data prior to January 2011 is used to train the several forecasting models for the 1 and 2 h-ahead hourly averaged power output. The methods studied in this work are: Persistent model, Auto-Regressive Integrated Moving Average (ARIMA), k-Nearest-Neighbors (kNNs), Artificial Neural Networks (ANNs), and ANNs optimized by Genetic Algorithms (GAs/ANN). The accuracy of the models is determined by computing error statistics such as mean absolute error (MAE), mean bias error (MBE), and the coefficient of correlation (R2) for the differences between the forecasted values and the measured values for the period from January to August of 2011. This work also addresses the accuracy of the different methods as a function of the variability of the power output, which depends strongly on seasonal conditions. The findings show that the ANN-based forecasting models perform better than the other forecasting techniques, that substantial improvements can be achieved with a GA optimization of the ANN parameters, and that the accuracy of all models depends strongly on seasonal characteristics of solar variability.
Article
A brief discussion of Statistical Quality Control Charting procedures is first presented with special reference to the relevance of the objectives and assumptions. An approach to the design of discrete feedforward and feedback control schemes, which are of great importance for example, in the chemical industry, is then given. This approach to control employs discrete stochastic and dynamic models discussed in Part I of this paper (Box and Jenkins, 1968) and has a close link with the forecasting problems discussed there. The control algorithms obtained are ideally suited to discrete digital computer control. However, for common simple situations the algorithms may be represented by suitable charts or nomograms which may be employed to obtain improved manual control. The paper ends with a discussion of a problem typical of that arising in the parts manufacturing industry.
Article
For a large scale implementation of photovoltaics (PV) in the urban environment, building integration is a major issue. This includes installations on roof or facade surfaces with orientations that are not ideal for maximum energy production. To evaluate the performance of PV systems in urban settings and compare it with the building user’s electricity consumption, three-dimensional geometry modelling was combined with photovoltaic system simulations. As an example, the modern residential district of Scharnhauser Park (SHP) near Stuttgart/Germany was used to calculate the potential of photovoltaic energy and to evaluate the local own consumption of the energy produced. For most buildings of the district only annual electrical consumption data was available and only selected buildings have electronic metering equipment. The available roof area for one of these multi-family case study buildings was used for a detailed hourly simulation of the PV power production, which was then compared to the hourly measured electricity consumption. The results were extrapolated to all buildings of the analyzed area by normalizing them to the annual consumption data. The PV systems can produce 35% of the quarter’s total electricity consumption and half of this generated electricity is directly used within the buildings.
Article
Long-term reliability is critical to the cost effectiveness and commercial success of photovoltaic (PV) products. Today most PV modules are warranted for 25 years, but there is no accepted test protocol to validate a 25-year lifetime. The qualification tests do an excellent job of identifying design, materials, and process flaws that are likely to lead to premature failure (infant mortality), but they are not designed to test for wear-out mechanisms that limit lifetime. This paper presents a method for evaluating the ability of a new PV module technology to survive long-term exposure to specific stresses. The authors propose the use of baseline technologies with proven long-term field performance as controls in the accelerated stress tests. The performance of new-technology modules can then be evaluated versus that of proven-technology modules. If the new-technology demonstrates equivalent or superior performance to the proven one, there is a high likelihood that they will survive versus the tested stress in the real world.