Conceived by Canadian researchers, the novel methodology combines picture processing and deep studying strategies to detect areas of photo voltaic panels coated in snow. It is reported to carry out higher than three different fashionable picture processing-based segmentation strategies.
Researchers from the University of Waterloo in Canada have developed a brand new automated mannequin for real-time snow cowl detection of PV panels and power loss estimation. The mannequin relies on picture processing and deep studying, utilizing a convolutional neural community (CNN) and a timelapse set of pictures to determine snow-covered areas versus clear ones.
“In chilly climates, the place snow cowl can final for days or even weeks and lead to as much as 34% power loss, well timed detection is crucial for maximizing power output in PV programs to attenuate these losses,” the researchers stated. “The capacity to autonomously detect snow in PV programs is crucial not just for maximizing power output but in addition for sustaining the integrity of the photo voltaic panels and making certain a steady provide of electrical energy.”
The proposed methodology is designed in 5 steps. First, a dataset of 250 pictures displaying PV panels in opposition to completely different backgrounds was obtained. Then, the pictures had been processed utilizing completely different picture transformation strategies to extend the dataset to 800 pictures. In this step, the analysis workforce manually created a binary black-and-white masks for every picture, the place the white pixels characterize the PV panels and the black pixels characterize the background.
At the following stage, all pairs of pictures and masks are fed to the U-Net CNN, a convolutional neural community particularly designed for picture segmentation duties. The mannequin is requested to view the dataset and determine the PV panels. Afterwards, it compares the outcomes with the ready masks, and if there are errors, the pictures are propagated again to the system.
In the fourth step, the mannequin is verified in opposition to the bottom fact. In the final step, the power loss as a consequence of snow accumulation is estimated based mostly on the variations within the protection of white pixels within the present picture and the designated non-snow picture. A mathematical mannequin from earlier literature was used to calculate the power loss.
“The presence of snow within the background could cause confusion as as to if it’s within the panel or not, including a layer of complexity to the correct estimation of snow protection,” the workforce defined. “To handle these challenges, a reference picture of the PV scene earlier than snowfall is used. This reference picture must be from the identical distance and angle because the take a look at picture and be obtained with out snow on the PV panels. Uniform digicam settings in each reference and take a look at pictures are vital to attenuate camera-induced variations.”
The novel methodology was examined on six new pictures that the mannequin was not educated on. The system’s snow detection and power loss are in contrast with the skills of fashionable segmentation strategies based mostly on picture processing, akin to Canny edge detection, Ok-Means segmentation, and picture thresholding. .
The outcomes present that the brand new methodology achieved a imply error of two.88% and a Dice coefficient rating of 0.91. The Dice rating is calculated because the ratio of predicted and floor fact areas. In compression, Ok-means clustering achieved a mean Dice rating of 0.47, Canny edge detection bought 0.60, and the thresholding methodology confirmed a rating of 0.63.
“The power loss was additionally estimated from the pictures with a imply power lack of 1.5 kWh / m2 / month from six completely different PV arrays,” defined the scientists. “The mannequin reliably predicts power loss as a consequence of snow accumulation with a imply error of 0.05 kWh/m2/month. The highest power loss is 0.23 kWh from a big PV array system consisting of within the 117 m2 space.The evaluation of the impact of the length of the snow protection on the power loss exhibits a saving potential of 0.13 kWh/m2 utilizing the well timed clearing of the snow protection.
The methodology is offered in “Using deep studying in direction of real-time snow cowl detection and power loss estimation for photo voltaic modules,” revealed in Applied Energy.
This content material is protected by copyright and is probably not reused. If you wish to cooperate with us and wish to reuse a few of our content material, please contact: [email protected].
Popular content material
