• Tidak ada hasil yang ditemukan

View of W ON WEATHER FORECASTING BASED S REVIEW ON WEATHER FORECASTING BASED SOLAR ENERGY PREDICTION TO GENERATE MAXIMUM POWER EFFICIENCY

N/A
N/A
Protected

Academic year: 2023

Membagikan "View of W ON WEATHER FORECASTING BASED S REVIEW ON WEATHER FORECASTING BASED SOLAR ENERGY PREDICTION TO GENERATE MAXIMUM POWER EFFICIENCY"

Copied!
7
0
0

Teks penuh

(1)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 12

REVIEW ON WEATHER FORECASTING BASED SOLAR ENERGY PREDICTION TO GENERATE MAXIMUM POWER EFFICIENCY

Jay Kumar Pandey

Research Scholar, Dr. A. P. J. Abdul Kalam University, Indore (M.P.) –INDIA Dr. Vikas Kumar Aharwal

Assistant Professor, Dr. A. P. J. Abdul Kalam University, Indore (M.P.) –INDIA

Abstract - During periods of bright daylight, photovoltaic solar panels offer a tempting fuel source. Energy conserved in plenty would be used later in the evening or on a shady day.

When evaluating whether to utilize stored forms of energy in the present or for the future, the ability to predict the environment on several timescales is crucial. Because of the solar cell's non-minor I-V quality, predicting stored energy in the short term is difficult. Long- term forecast energy is difficult due to the inconsistency of the climate concept. Using the results from a solar panel, we examine these constraints, as well as their relationship to climate perceptions. While manually constructing complex prediction models on a massive scale may be achievable, developing a million users across the grid is a difficult task. To solve the problem, we utilize a machine learning approach to examine the possibilities of autonomously building models for solar power generation that are site-specific and dependent on India's Metrological Department (IMD) weather forecasts. For creating prediction models, we study linear least squares and support vector machines as illustrations of regression methods. Multiple kernel functions are being used. We utilize historical IMD forecasts and sun peak intensity from a weather station deployment to validate each model's accuracy for nearly a year. While it may be feasible to physically construct complex large-scale solar farm prediction models, doing it for a large number of households throughout the lattice is a challenging task. According to our results, Support Vector Machines (SVM)-based prediction models designed to utilize seven different weather parameters are 28% more authentic than existing forecast-based models for our site.

Keywords: weather forecast; energy prediction; fuzzy decision trees; solar power System;

photovoltaic cell; transmission power planning; Climate;

1 INTRODUCTION

Solar power availability is not assured at any specific time and location; it is associated with the time of day, as well as current and previous climatic parameters. Because meteorological services surface reflectivity, we should be able to use weather forecasts around the clock to our advantage when arranging activities that require solar energy.

There are three intriguing questions to think about: (1) Can we anticipate the amount of energy we'll be able to capture for the future based on today's regular weather forecasts?

(2) How precise are our measured real insulation and other local climate parameters for us to make that assumption right now? (3) What’s the best way to incorporate these 2 sources for prediction? The solutions to these problems are important, not only for research but also for implementation reasons. [6]. A solar structure is now developed using photovoltaic cells made of silicon in sequence, which consists primarily of silicon PN-junctions with a greater surface area. Photons from received light transfer electrons from the valence to the conduction band. As a result, an external load can produce DC and, using the across the junction capacitance, there's also a band-gap voltage. The band-gap is effectively the cell's open-circuit voltage. Its short-out current isn't always solely determined by the optical power that has been incident. The interior impedance to which an outside load should be coordinated to achieve peak power transmission to the load is termed their ratio. One of the ultimate targets of the grid of future initiatives is to raise the usage of alternative sources of energy such as the sun and wind. The Renewable Standard for Portfolios, for example, aims for intermittent renewables to account for up to 25% of energy generation [10], with renewables accounting for 33% of total generation by 2021 [11]. It is difficult to integrate renewable energy into the system in a significant way because its system supply is unreliable and uncontrolled. The present-day electric grid allows families to use virtually any amount of electricity at any time, but it is not yet equipped to handle large amounts of unpredictable generation.

(2)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 13

Most states' present net metering laws reflect this, allowing the consumer to sell energy generated in house renewables back to the power grid while placing minimum limits on the overall number of involved customers as the main step of power given per client.

[12]. In Tamilnadu, for example, the total number of involved clients is limited to 1% of all subscribers. Utilities limit renewables' contribution because, unlike energy demand, renewable generation is unpredictable, so grid planning should be done in advance. It is difficult to work on the subject of instinctively creating models that reliably estimate renewable energy using India's Metrological Department (IMD) weather forecasts to promote the best arrangement and lower the hindrance to the negligible part of renewables in the force matrix.

We put some machine learning algorithms to the test to construct prediction models based on past IMD predicted data and connect them to data from solar photovoltaic cells.

After being based on prior forecasts and in formulated data, our prediction models utilize IMD Predict for a specific area to make assumptions for future generations across different time horizons. The collection of short and long processes of energy is created in a profession that needs sophisticated, dynamic, ongoing control dependent on genuine, necessary information and the ability to anticipate and get ready for future requirements. Examples are [16] and [17].

The major purpose of this work is to show how a fuzzy prediction method, specifically Fuzzy Decision Trees (FDTs), may increase energy assumption precision. For our early attempt at evaluating energy gain in real situations, we chose FDTs over other techniques such as neural networks, because FDTs provide human-like results with explicit rules that will allow us to predict the energy gain in real situations. To make the results better in the future, not only are important aspects identified automatically but so are their interactions.

FDTs now have the advantage of dealing with both symbolic (in this case, foggy, luminous, and rainstorm are examples of climate classes) and measurable data (such as temperature) at the same time.

2 SOLAR PANELS

A. Achieving the most power from a solar panel

Estimate the parameters that will allow the solar photovoltaic system to be directly connected to the consumption. Figure 1.1 represents the highest power point (MPP), which would be the direction on the I-V curve where the area shown on the curve is the highest.

For optimal Choose a solar panel that is well-designed for accuracy and ease of calibration to the planned load. This is difficult, however, because with luminosity, the I-V curve, and also the MPP, there are remaining changes. It also probably depends on the temperature of the panel, which is impacted by light. MPPTs (maximum power point trackers) are switching power converters and actively measuring power converters that can switch the actual consumption to sustain the MPPT operational point. Several techniques, specifically those based on fuzzy systems, have been given [20] and are still being found [18]. A simple comparison is included.

Figure 2.1: The ideal solar panel

VOC is the open-circuit voltage (when RL =) and Isc is the short circuit current (when RL =).

The I-V characteristic is shown by the black curve, the amount of power that is accessible to an outer load (the I-V) is represented by the grey curve, and the doted-blue Charge line represents the operational coordinate (V, I) highlighted by the red symbol at which the charge will deliver highest feasible power from the panel.

(3)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 14 3 SOLAR ENERGY IN REAL-LIFE WEATHER SITUATIONS

The solar collector should be tested with some specific test settings, which include a temperature of 25.5 °C and illumination of 1000 W/m2 (1.0 sun) with a volume of air of 1.6 (AM 1.5) filtered solar spectra, according to the National Renewable Energy Laboratory. On a clear day, the goal is to estimate the brightness and brightness of sunlight incident on a sun-facing 37.5° Celsius tilted surface at an angle of 40.81° Celsius above the skyline. In the continental United States, this design, with the panel pointed directly at the sun, approximates solar on the top geometrically near the warmer month’s equinoxes. At the earth's surface, however, insulation is rarely as high as the required 1000 W/m2.Furthermore, as previously stated, to examine the creation of electrical energy accurately with actual climate circumstances, genuine lighting and temperature swings must be observed. Remember that a panel that performs well in the NREL territory will almost certainly perform poorly in any legitimate setting to investigate solar electricity generation with achievable expectations under natural climate conditions; it is recommended to combine a feasible arrangement with an MPPT. However, there is a slew of commercial devices on the market, each of which uses a proprietary algorithm that isn't publicly published, and none of which is arguably the best or even the industry standard.

As a result, we structure our computation in such a manner that we may implement an ideal MPPT technique, i.e., we collect all the information possible first and then compute the actual optimal point later.

4 WEATHER PREDICTION

The solar collector-based observation systems are situated in a safe off-campus location with a wide central solid-angle view to the south. Another Windows PC on campus with dependable Internet access downloads current and forecast weather data from the India Meteorological Department regularly. The Website Weather RSS Feed [10] records the climate conditions every hour during the day because weather conditions fluctuate rapidly.

We recorded the climate conditions every hour because weather conditions change slowly.

To match the forecast to the underlying situation, we used the weather service's 48 standard categories. To eliminate prediction uncertainty, we've chosen to use the prediction just before bedtime in the morning. Other, more sophisticated approaches could take a forecast's testing or pattern into account.

a. Data aggregation

At first sight, the problem of integrating the data appears easy, but it is rather difficult.

Because it is more natural and predictable, we like to work in a one-day cycle. Energy forecasting across a short and long time horizon could be the focus of future research. As a result, to calculate the energy released by solar photovoltaic cells throughout a single day, we must begin with the power computation obtained previously in 11 minutes. Our first step consists of selecting from each of these series the highest power. In this manner, we simulate a good MPPT. As a result, to calculate the energy produced by the panel for a day, we must begin with the current measurement taken every 10–12 minutes. The first step is to select the maximum power from each of these series. As a result, we can replicate an ideal MPPT. Then, assuming nothing changes for the next 10 minutes, we integrate for the day to estimate the actual output produced. As shown in the table, the assumption is provided.

5 DATA ANALYSIS

Starting in January 2020, for 10 months, we collect climate forecast samples as well as measurable sun strength data. For the past two years, we have been gathering historical predicted data from the IMD at https://mausam.imd.gov.in/. In India, the IMD gives previous textual predictions for local city-side regions. Each forecast provides hourly projections for each metric for the next 1 hour to 6 days. Climate parameters include humidity with temperature, dew point, wind speed, foggy cover, precipitation approximation, and relative humidity. The percentage of cloud analysis in the atmosphere (from 0% to 100%) is referred to as "sky cover." The percentage of cloud coverage in the atmosphere (from 0% to 100%) is referred to as "sky cover."

We possessed data for 76 days in a row between the end of April and the beginning of July 2020. The average daily energy output was 343 watt-hours with a standard

(4)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 15

deviation of 179 watt-hours. According to Table 1, nearly half of the days are "fair," whereas the other half are "cloudy" or "rainy," as per Table 1. "Fair" days generate max energy to compare previous "partly cloudy" days, which one is best to "mostly cloudy," "cloudy," or

"shower" days, as one might anticipate. This consistency in diagrams and energy descriptions provides us with trust in different models, especially the majority aggregation procedure. The daily energy production has a lot of variability’s, but it's more or less constant for each category. We have confidence in our model because of the consistency in semantic and energy descriptions, particularly the majority aggregation approach. The daily energy generation fluctuates a lot, but it's rather consistent for each division.

The climate prediction accuracy for the analysis period was 61% when using the conventional set of rules. (Of the days' correct prediction at sunrise) Two or more variables contribute to this low percentage: There appears to be a risk aversion and a category mismatch in prediction. We can see a shift in the frequency of rainy days in Table 2(a), which depicts the number of projected climate conditions (predictions). In this we found 5 days with "showers," but 37 days with "showers" or "thunderstorms" predicted. This mismatch could be due to the weather forecaster's aversion to risk. The climate forecast will be a "rainy day" even if it pours for an hour in the middle of the day. On the other hand, our majority observation, which is appropriate on a sunny day, would result in a divided mismatch for energy projection. Likewise, we can see that the numeral and labeling of grouping vary between the 2 sets by comparing the labels on Tables 1 and 2a. In the present weather observations, the labels that are in the prediction do not appear. There aren't any "cloudy" predictions or "sunny" forecasts, for example. This suggests a deeper, structural issue: class borders are ill-defined. It will be called a mismatch if we expect

"mainly cloudy" and see "partly cloudy." By combining labels such as "cloudy" and "partly cloudy" into an "overcast" class, new weather classes could be developed. However, early research suggests that forecast accuracy does not improve at around this phase since the descriptions are similarly hazy or capricious.

6 PREDICTION MODELS

i. Linear Least Squares Regression

We start by using a linear least-squares regression technique to forecast solar strength.

Linear least-squares extrapolation is a basic and most commonly used method for evaluating the terms between a dependent or response variable, such as solar intensity, and a set of independent variables or predictors. The regression seeks to minimize the addition of squared differences between practical solar intensity and solar intensity projected using a linear approximation of anticipated weather indicators. In the prediction model below, the linear least-squares procedure is implemented on the 8 months of training data to produce a factor for each metric.

Solar Intensity = 1.18*Day + 77.9*Temp + 33.11*Dew Point +22.8*Wind Speed - 96.9*Sky Cover - 49.15*Precipitation -43.4*Humidity

ii. Support Vector Machines

After looking at a range of supervised learning techniques utilizing Support Vector Machines (SVM) [14], in classification and regression research, SVMs, which generate hyperplanes in a multidimensional space, have increasingly gained traction. The accuracy of SVM regression is determined by the kernel function and parameters chosen. We looked at three different SVM kernels in our research: a Radial Basis Function (RBF) kernel, a Linear Kernel, and a Polynomial Kernel to transfer data taken from input high-dimensional structure space. An SVM employs the kernel function. We chose SVMs over other learning algorithms because of their sparseness and ability to deal with non-linearity in data. We construct SVM regression with the linear kernel function on our training datasheet to use the LibSVM framework, which supports several SVM estimation methods [15].

iii. Eliminating Redundant Information

Many weather metrics have a high correlation with one another, as we saw in the previous section. As a result, our SVM regression models contain numerous redundant datasets, decreasing the accuracy of their projections. PCA is a commonly used technique for removing duplicated data from an initial dataset and therefore decreasing its dimensionality [16].To clean up redundant information from our statically collected data, we apply the

(5)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 16

principle component analysis algorithm. The PCA calculation changes a bunch of conceivably correlated input factors into a bunch of uncorrelated factors called head parts through asymmetrical change. The quantity of unique factors is not exactly equivalent to the number of essential parts. Under the constraint of being symmetrical to the primary segment, the greatest conceivable change is in the primary head segment, and the most extreme conceivable change is in the second head segment, and so on.

iv. Using Existing Models as a Reference

Finally, we match our regression-based prediction techniques to successfully developed models. First, we match our proposed model to that of a past-predicts-future (PPF) model, which predicts the subsequent day's solar insolation based on the previous day's solar insulation. While the models are extremely accurate when weather conditions find consistency, they are unable to predict abrupt weather changes. Second, we compare cloudy to cloudy, a simple model that just relies on the sky state to make the prediction [13]. The clouded model has been demonstrated to be more accurate than other PPF versions in the literature [17]. While the model can forecast weather changes, it does not take into account the impact of numerous meteorological metrics on solar intensity.

7 ENERGY AND CLIMATE PREDICTION

As per the record information description above, the task is to estimate how much energy we will produce that day before the sun rises. All approaches may be divided into two groups: direct methods, in which the efficiency value is determined using a "black-box"

structure (typically based on regression, like in Neuronal Networks [16] [17]), and indirect methods, in which a climate class is assumed first.

We will focus on the latter in this study. This method allows for the prediction of energy to be done using the energy of national weather forecasting techniques without any additional adjustments. The advantage of the broad formulation is that it expands the number of feasible algorithms. We've chosen to employ Fuzzy Decision Trees in this case since they're not only capable of dealing with large amounts of data, but they're also able to cope with ambiguous situations. However, we must also give a rationale for the forecast.

i. Fuzzy Decision Trees

FDTs are a subtype of the decision tree that is an improvising of traditional decision trees.

They were primarily employed in Machine Learning to cope with numerical and/or fuzzy values in training sets [22] [23] [24]. In addition, these trees introduced a softer classification of instances, resulting in a more fluid conclusion. As a result, degrees of decision and class membership are supplemented.

ii. Baseline prediction

To measure our strategy's performance, we'll need to establish a distance metric and a baseline. To evaluate how well we can assume the energy output of a solar cell, we find out the average of the absolute values of the divergence between the assumed and measured energy for the presented models.

Constant average assumption:

We assume that the normal energy for an ideal area can be accurately predicted but that it remains constant across time. To do so, we compute the average energy detected across the entire period after the fact. It's significant to mention that this is the maximum point that can be accomplished in real-world scenarios. A continuous approximation will extend the proposed energy.

For the future, energy would be the same as today’s: This is a normal method for minute series analysis, particularly in the prediction of climate forecasts.

• Pure weather assumption-based prediction: We propose that the expected energy class be based on the weather forecast. This is the logical way to approach the problem: "If today is likely to be bright, and we produce on average energy E on sunny days, then today we should notice energy E."

8 CONCLUSION

(6)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 17

Previous solar energy harvesting prediction models were mostly based on the recent past.

Unfortunately, these systems are not capable of anticipating changes in weather patterns.

Since they depend on the collections of numerous information sources from the nation over, the IMD's weather forecasts can provide adequate notice. When we assume that a solar panel generates approximately the same number of power, we see an insignificant increase of 152 W-hr between what is predicted and what is observed (constant prediction, baseline).

The disparity between measured and simulated rises when we use the naive model, which predicts that coming next day energy, is equal to today. This illustrates that energy varies quickly and that a constant assumption provides a basis that is difficult to transcend.

REFERENCES

1. Anuj Gupta, Kapil Gupta, Sumit Saroha, Solar Irradiation Forecasting Technologies: A Review, 2020: Vol 39 Iss 3-4 2020, Published 2021-07-09, https://doi.org/10.13052/spee1048-4236.391413, ISSN: 1546-0126 (Online Version)

2. Abdelhakim El hendouzi1 and Abdennaser Bourouhou2, Solar Photovoltaic Power Forecasting, Journal of Electrical and Computer Engineering / 2020, Volume 2020 |Article ID 8819925 | https://doi.org/10.1155/2020/8819925

3. K.U.JaseenaBinsu C.Kovoor, Deterministic weather forecasting models based on intelligent predictors: A survey, Journal of King Saud University - Computer and Information Sciences,Available online 24 September 2020, https://doi.org/10.1016/j.jksuci.2020.09.009

4. Min-Hee Chung, Estimating Solar Insolation and Power Generation of Photovoltaic Systems Using Previous Day Weather Data, Hindawi

5. Advances in Civil Engineering Volume 2020, Article ID 8701368, 13 pages,https://doi.org/10.1155/2020/8701368

6. Seul-Gi Kim, Jae-Yoon Jung, and Min-Kyu Sim *, A Two-Step Approach to Solar Power Generation Prediction Based on Weather Data Using Machine Learning, Sustainability 2019, 11, 1501; doi:10.3390/su11051501, 9 March 2019; Published: 12 March 2019

7. Abuela, M.; Chowdhury, B. Improving Combined Solar Power Forecasts Using Estimated Ramp Rates:Data- driven Post-processing Approach. IET Renew. Power Gener. 2018, 12, 1127–1135. [CrossRef]

8. Kim, S.; Jung, J.-Y.; Sim, M. Machine Learning Methods for Solar Power Generation Prediction based on Weather Forecast. In Proceedings of the 6th International Conference on Big Data Applications and Services (BigDAS2018), Zhengzhou, China, 19–22 August 2018.

9. Voyant, C.; Notton, G.; Kalogirou, S.; Nivet, M.L.; Paoli, C.; Motte, F.; Fouilloy, A. Machine learning methods for solar radiation forecasting: A review. Renew. Energy 2017, 1, 569–582. [CrossRef]

10. Abedinia, O.; Raisz, D.; Amjady, N. Effective prediction model for Hungarian small-scale solar power output.

IET Renew. Power Gener. 2017, 11, 1648–1658. [CrossRef]

11. Yahoo! Weather RSS Feed [Online]. Available: http://developer.yahoo.com/weather/ (accessed: 2010, Jan) 12. Dropbox Documentation [Online]. Available: https://www.dropbox.com/about (accessed: 2010, Jan)

13. Arduino Duemilanuove Datasheet [Online]. Available:

http://www.arduino.cc/en/Main/ArduinoBoardDuemilanove (accessed: 2010, Jan)

14. Peder Bacher, Henrik Madsen, Henrik Aalborg Nielson, “Online short-term solar power forecasting”, Informatics and Mathematical Modelling, Richard Pedersens Plads, Technical University of Denmark, Denmark, 22 May 2009.

15. Lin Phyo Naing Srinivasan, D., “Estimation of solar power generating capacity”, IEEE 11th International Conference on Probabilistic Methods Applied to Power Systems (PMAPS), 14-17 June 2010, Singapore 16. Hong-Tzer Yang, Jian-Tang Liao, Xiang-He Su, “A fuzzy-rule based power restoration approach for a

distribution system with renewable energies”, FUZZ-IEEE 2011: 2448-2453

17. Davide Caputo, Francesco Grimaccia, Marco Mussetta, Riccardo Enrico Zich, “Photovoltaic plants predictive model by means of ANN trained by a hybrid evolutionary algorithm”, IJCNN 2010: 1-6

18. Francesco Grimaccia, Marco Mussetta, Riccardo Enrico Zich, “Neuro-fuzzy predictive model for PV energy production based on weather forecast”, FUZZ-IEEE 2011: 2454-2457

19. Irwan Purnama, Yu-Kang Lo, Huang-Jen Chiu, “A fuzzy control maximum power point tracking photovoltaic system”, FUZZ-IEEE 2011: 2432-2439

20. Esram, T., Chapman, P.L., "Comparison of Photovoltaic Array Maximum Power Point Tracking Techniques", IEEE Transactions on Energy Conversion, Vol. 22 (2) pp. 439-449, 2007

21. Chung-Yuen Won, Duk-Heon Kim, Sei-Chan Kim, Won-Sam Kim and Hack-Sung Kim, "A new maximum power point tracker of photovoltaic arrays using fuzzy controller", 25th Annual IEEE Power Electronics Specialists Conf. (PESC'94), pp. 396-403, Taipei, Taiwan, Jun 1994.

22. Kyohei Kurohane, Tomonobu Senjyu, Atsushi Yona, Naomitsu Urasaki, Tomonori Goya, Toshihisa Funabashi: A Hybrid Smart AC/DC Power System. IEEE Trans. Smart Grid 1(2): 199-204 (2010)

23. Yuan, Y. & Shaw, M. Induction of Fuzzy Decision Trees Fuzzy Sets and systems, 1995, 69, 125-139.

24. Janikow, C. Z. Fuzzy Decision Trees: Issues and Methods IEEE Transactions on Systems, Man and Cybernetics, 1998, 28, 1-14.

25. Marsala, C. & Bouchon-Meunier, B. An Adaptable System to Construct Fuzzy Decision Trees Proc. of the NAFIPS'99, 1999, 223-227.

26. Quinlan, J. R. Induction of Decision Trees Machine Learning, 1986, 1, 86-106.

27. Breiman, L.; Friedman, J.; Olshen, R. & Stone, C. Classification And Regression Trees Chapman and Hall, 1984.

28. Marsala, C. & Bouchon-Meunier, B. Ranking Attributes to Build Fuzzy Decision Trees: a Comparative Study of Measures IEEE World Congress on Computational Intelligence, 2006, 1777-1783.

(7)

VOLUME: 08, Special Issue 11, (IC-MLSDG 2021) Paper id-IJIERM-VIII-XI, November 2021 18

29. Marsala, C. Gradual Fuzzy Decision Trees to Help Medical Diagnosis. IEEE World Congress on Computational Intelligence, 2012, Brisbane, Australia, June 2012 (to appear)

Referensi

Dokumen terkait

Of referring to the statistical calculation result, it is observed that there is significant difference between the mean scores of pretest result of the control class when

The Internet of Things (IoT)-based off-grid solar power plant monitoring system manages to monitor the input/output of solar modules and battery conditions in