Total Sky Images Research Articles

The ultra-short-term photovoltaic power prediction based on multi-exposure high-resolution total sky images using deep learning

With the increasing installed capacity of renewable energy sources such as photovoltaic (PV) generation in the power system, ultra-short-term PV power prediction is more important for the stability. However, the prediction accuracy in the traditional methods cannot be guaranteed. Based on multi-exposure high-resolution total sky images (TSIs), this paper presents an advanced ultra-short-term PV prediction method using deep learning. To fully utilize the high-resolution images, an overlapping sliding window cutting and concatenating strategy are described to capture the global and local features of an image. The multi-exposure images are fused to provide more details about edge information and brightness. To better extract the image features and sequential features for PV prediction, a convolutional long–short-term memory model (CLSTM) with a multi-head self-attention mechanism is presented. The experiments use real-world datasets in the Zero Carbon Emission laboratory at Zhejiang University. The simulation results show that the proposed model can utilize TSIs to achieve the desired accuracy consistently. Under different weather conditions, the prediction accuracy of this model is improved by 49.1%–66% compared with that of other models.

Solar Radiation Prediction Based on Convolution Neural Network and Long Short-Term Memory

Photovoltaic power generation is highly valued and has developed rapidly throughout the world. However, the fluctuation of solar irradiance affects the stability of the photovoltaic power system and endangers the safety of the power grid. Therefore, ultra-short-term solar irradiance predictions are widely used to provide decision support for power dispatching systems. Although a great deal of research has been done, there is still room for improvement regarding the prediction accuracy of solar irradiance including global horizontal irradiance, direct normal irradiance and diffuse irradiance. This study took the direct normal irradiance (DNI) as prediction target and proposed a Siamese convolutional neural network-long short-term memory (SCNN-LSTM) model to predict the inter-hour DNI by combining the time-dependent spatial features of total sky images and historical meteorological observations. First, the features of total sky images were automatically extracted using a Siamese CNN to describe the cloud information. Next, the image features and meteorological observations were fused and then predicted the DNI in 10-min ahead using an LSTM. To verify the validity of the proposed SCNN-LSTM model, several experiments were carried out using two-year historical observation data provided by the National Renewable Energy Laboratory (NREL). The results show that the proposed method achieved nRMSE of 23.47% and forecast skill of 24.51% for the whole year of 2014, and it also did better than some published methods especially under clear sky and rainy days.

Using a Multi-view Convolutional Neural Network to monitor solar irradiance

In the last years, there is an increasing interest for enhanced method for assessing and monitoring the level of the global horizontal irradiance (GHI) in photovoltaic (PV) systems, fostered by the massive deployment of this energy. Thermopile or photodiode pyranometers provide point measurements, which may not be adequate in cases when areal information is important (as for PV network or large PV plants monitoring). The use of All Sky Imagers paired convolutional neural networks, a powerful technique for estimation, has been proposed as a plausible alternative. In this work, a convolutional neural network architecture is presented to estimate solar irradiance from sets of ground-level Total Sky Images. This neural network is capable of combining images from three cameras. Results show that this approach is more accurate than using only images from a single camera. It has also been shown to improve the performance of two other approaches: a cloud fraction model and a feature extraction model.

The Ultra-Short-Term Forecasting of Global Horizonal Irradiance Based on Total Sky Images

Solar photovoltaics (PV) has advanced at an unprecedented rate and the global cumulative installed PV capacity is growing exponentially. However, the ability to forecast PV power remains a key technical challenge due to the variability and uncertainty of solar irradiance resulting from the changes of clouds. Ground-based remote sensing with high temporal and spatial resolution may have potential for solar irradiation forecasting, especially under cloudy conditions. To this end, we established two ultra-short-term forecasting models of global horizonal irradiance (GHI) using Ternary Linear Regression (TLR) and Back Propagation Neural Network (BPN), respectively, based on the observation of a ground-based sky imager (TSI-880, Total Sky Imager) and a radiometer at a PV plant in Dunhuang, China. Sky images taken every 1 min (minute) were processed to determine the distribution of clouds with different optical depths (thick, thin) for generating a two-dimensional cloud map. To obtain the forecasted cloud map, the Particle Image Velocity (PIV) method was applied to the two consecutive images and the cloud map was advected to the future. Further, different types of cloud fraction combined with clear sky index derived from the GHI of clear sky conditions were used as the inputs of the two forecasting models. Limited validation on 4 partly cloudy days showed that the average relative root mean square error (rRMSE) of the 4 days ranged from 5% to 36% based on the TLR model and ranged from 12% to 32% based on the BPN model. The forecasting performance of the BPN model was better than the TLR model and the forecasting errors increased with the increase in lead time.

Clouds over Hyytiälä, Finland: an algorithm to classify clouds based on solar radiation and cloud base height measurements

Abstract. We developed a simple algorithm to classify clouds based on global radiation and cloud base height measured by pyranometer and ceilometer, respectively. We separated clouds into seven different classes (stratus, stratocumulus, cumulus, nimbostratus, altocumulus + altostratus, cirrus + cirrocumulus + cirrostratus and clear sky + cirrus). We also included classes for cumulus and cirrus clouds causing global radiation enhancement, and we classified multilayered clouds, when captured by the ceilometer, based on their height and characteristics (transmittance, patchiness and uniformity). The overall performance of the algorithm was nearly 70 % when compared with classification by an observer using total-sky images. The performance was best for clouds having well-distinguishable effects on solar radiation: nimbostratus clouds were classified correctly in 100 % of the cases. The worst performance corresponds to cirriform clouds (50 %). Although the overall performance of the algorithm was good, it is likely to miss the occurrences of high and multilayered clouds. This is due to the technical limits of the instrumentation: the vertical detection range of the ceilometer and occultation of the laser pulse by the lowest cloud layer. We examined the use of clearness index, which is defined as a ratio between measured global radiation and modeled radiation at the top of the atmosphere, as an indicator of clear-sky conditions. Our results show that cumulus, altocumulus, altostratus and cirriform clouds can be present when the index indicates clear-sky conditions. Those conditions have previously been associated with enhanced aerosol formation under clear skies. This is an important finding especially in the case of low clouds coupled to the surface, which can influence aerosol population via aerosol–cloud interactions. Overall, caution is required when the clearness index is used in the analysis of processes affected by partitioning of radiation by clouds.

Battery smoothing control for photovoltaic system using short-term forecast with total sky images

Smoothing power fluctuation from variable renewable energy (VRE) sources is one of the critical issues to increase their interconnection capacity to the power grids. Especially, in relatively small balancing areas, such as small islands and remote districts with a small tie-line capacity, the frequency control becomes difficult, and grid operators require countermeasures for VRE sources to reduce their power fluctuation amplitude. In this paper, we deal with a photovoltaic system with a battery energy storage and a camera to meet a grid code limiting its maximum power fluctuation. We propose a battery smoothing control using short-term forecast of photovoltaic system (PV) power output by deep learning techniques with historical PV power output data and total sky images taken by the camera. A numerical simulation is conducted for an actual system to verify that the proposed method contributes to the reduction of the battery capacity necessary for meeting the grid code.

An improved cloud recognition and classification method for photovoltaic power prediction based on total‐sky‐images

The rapid variation of irradiance due to cloud movements and occlusions is the main factor causing the fluctuation of photovoltaic (PV) power generation. Accurate recognition and classification of cloud is the prerequisite of improving the prediction accuracy of irradiance. Total-Sky-Images (TSIs) taken by ground-based cameras are a good solution to analyse the distribution of cloud in real-time and is suitable for ultra-short-term PV power prediction. Images are often processed by threshold-based segmentation algorithm, which is based on the brightness or greyscale value of pixels. However, the sunlight can change the brightness or greyscale distribution of those pixels which represent cloud and cause misclassification. This paper proposes a novel cloud classification method based on greyscale compensation values (GCVs) and Otsu algorithm to solve this problem and divides the total sky into three parts: clear sky, thin cloud and thick cloud. GCVs are greyscale values which extract from clear-sky TSIs. Experimental results show that this method can effectively improve the accuracy of cloud classification.

Analyzing of Cloud Macroscopic Characteristics in the Shigatse Area of the Tibetan Plateau Using the Total-Sky Images

AbstractThe macroscopic characteristics of clouds in the Tibetan Plateau are crucial to understanding the local climatic conditions and their impact on the global climate and water vapor cycle. In this study, the variations of cloud cover and cloud types are analyzed by using total-sky images of two consecutive years in Shigatse, Tibetan Plateau. The results show that the cloud cover in Shigatse presents a distinct seasonal difference that is characterized by low cloud cover in autumn and winter and high cloud cover in summer and spring. July is the month with the largest cloud coverage, and its average cloud cover exceeds 75%. The probability of clouds in the sky is the lowest in November, with an average cloud cover of less than 20%. The diurnal variations of cloud cover in different months also have considerable differences. Specifically, cloud cover is higher in the afternoon than that in the morning in most months, whereas the cloud cover throughout the day varies little from July to September. The dominant cloud types in different months are also not the same. The proportion of clear sky is large in autumn and winter. Stratiform cloud occupies the highest percentage in March, April, July, and August. The probability of emergence of cirrus is highest in May and June. The Shigatse region has clear rainy and dry seasons, and correlation analysis between precipitation and clouds shows that the largest cumulative precipitation, the highest cloud cover, and the highest proportion of stratiform clouds occur simultaneously in July.

An RGB channel operation for removal of the difference of atmospheric scattering and its application on total sky cloud detection

Abstract. The inhomogeneous sky background presents a great challenge for accurate cloud recognition from the total-sky images. A channel operation was introduced in this study to produce a new composite channel in which the difference of atmospheric scattering has been removed and a homogeneous sky background can be obtained. Following this, a new cloud detection algorithm was proposed that combined the merits of the differencing and threshold methods, named differencing and threshold combination algorithm (DTCA). Firstly, the channel operation was applied to transform 3-D RGB image to the new channel, then the circumsolar saturated pixels and its circularity were used to judge whether the sun is visible or not in the image. When the sun is obscured, a single threshold can be used to identify cloud pixels. If the sun is visible in the image, the true clear-sky background differencing algorithm is adopted to detect clouds. The qualitative assessment for eight different total-sky images shows the DTCA algorithm obtained satisfactory cloud identification effectiveness for thin clouds and in the circumsolar and near-horizon regions. Quantitative evaluation also shows that the DTCA algorithm achieved the highest cloud recognition precision for five different types of clouds and performed well under both visible sun and blocked sun conditions.

A total sky cloud detection method using real clear sky background

Abstract. The brightness distribution of sky background is usually non-uniform, which creates many problems for traditional cloud detection methods, including the failure of thin cloud detection in total sky images and significantly reducing retrieval accuracy in the circumsolar and near-horizon regions. This paper describes the development of a new cloud detection algorithm, named "clear sky background differencing (CSBD)", which is accomplished by differencing the original image and the corresponding clear sky background image using the images' green channel. First, a library of clear sky background images with a variety of solar elevation angles needs to be developed. The image rotation and image brightness adjustment algorithms are applied to ensure the two images being differenced have the same solar position and similar brightness distribution. Sensitivity tests show that the cloud detection results are satisfactory when the two images have the same solar positions. Several experimental cases show that the CSBD algorithm obtains good cloud recognition results visually, especially for thin clouds.

An automated cloud detection method based on the green channel of total-sky visible images

Abstract. Obtaining an accurate cloud-cover state is a challenging task. In the past, traditional two-dimensional red-to-blue band methods have been widely used for cloud detection in total-sky images. By analyzing the imaging principle of cameras, the green channel has been selected to replace the 2-D red-to-blue band for detecting cloud pixels from partly cloudy total-sky images in this study. The brightness distribution in a total-sky image is usually nonuniform, because of forward scattering and Mie scattering of aerosols, which results in increased detection errors in the circumsolar and near-horizon regions. This paper proposes an automatic cloud detection algorithm, "green channel background subtraction adaptive threshold" (GBSAT), which incorporates channel selection, background simulation, computation of solar mask and cloud mask, subtraction, an adaptive threshold, and binarization. Five experimental cases show that the GBSAT algorithm produces more accurate retrieval results for all these test total-sky images.

Factors affecting UV irradiance at selected wavelengths at Hoher Sonnblick

The effects of clouds and surface albedo on spectral UV irradiance are investigated at the mountain observatory Hoher Sonnblick (47.05°N, 12.95°E, 3106 m a.s.l.) in the Central Austrian Alps. Spectral UV radiation in the presence of clouds is measured with a Bentham spectroradiometer along with detailed cloud observations. Ratios of measured to modeled clear sky UV intensities (cloud modification factors) are evaluated for over 9000 spectral UV measurements to quantitatively provide the dependency of cloud modification factors on total cloud amount, cloud type, wavelength, and effective surface albedo. Effective albedo can be altered through changes in snowline height and clouds below the station. Average albedo is increased by 0.28 ± 0.15 due to cloud cover ≥ 4/8 below the observatory, accounting for cloud modification factors in the range of 1.02–1.14 in 75% of the investigated cases. Compared to snow-free conditions, an irradiance enhancement of 1.22 at 305 nm arises from a mean albedo of 0.68 and clear sky conditions, corresponding to a snowline height of 1000 m. Enhancements through augmented albedo can be considerably stronger under the presence of cloudiness. Enhancements through clouds are most commonly found with clouds below the station, accounting for about 40% of all cases of enhancements, often occurring in the presence of cumulus clouds (Cu, Sc, Ac) with an unobstructed solar disk. Analysis of total sky images reveals that enhancements can also be observed when the solar disk is obstructed by clouds or the sky is overcast.