Research On Image Processing Based Photovoltaic Modules Hotspot Detection

Powered by the world’s growing energy crisis and the population caused by the firing of fossil fuel, the investment of developing the clean and renewable energy grows fast around the world. More than that, The Exploiting of environment friendly energy system is also one of our country’s sustainable development strategies. Along with the new energy becoming more and more important, the Photovoltaic System, which is one of the environment friendly and renewable energy collecting system, has developed rapidly. But in the PV system’s working process, the hotspot phenomena which would impact the output power of PV system and reduce the life of the battery needs to be overcome immediately.When we talk about hotspot phenomena which occurs in the photovoltaic system, it always describe a physics appearance that the battery cells which have hotspot act as loads in the circuit that consuming the power of the solar energy which is produced by the PV system they belong to. And the essence why these cells change themselves into loads is their alterations of characters that result in the discordance with the other cells. In general, the reason which could cause this malfunction in PV system is partial shading when it works. In order to solve such a problem which is caused by external condition, a real-time and practical hotspot detection and detection method which could extend the life span of solar batteries and reduce the costs of generating electric power has a significant meaning.In this paper, a hotspot detection method which is based on the utilization of infrared imaging technology in the PV system’s faults detection is proposed to solve this phenomenon which is caused by partial shadowing. In this solution method, the main content includes: the preprocessing of infrared images which are captured at the producing phase in the PV system; the coding method, which could reduce the data size for the next detection step, for the working state of solar battery cells; the detection phase that could alert the hotspot fault before it happens.In the preprocessing phase, the YCb Cr color space model is utilized, which could eliminate the illumination channel from the image. As a result, this step could reduce the size of image to facilitate the next step’s computation. Then, convolutional neural network which takes advantages of its universal approximation acts as a encoder that maps the processed infrared image to the matrix of battery cell’s working state. Such step successfully classifies the working state into 4 classes. This phase makes a further step to reduce the size of the information; At the last step which performs detection operation, an extensible spiking neural network which could dynamically increase the number of neurons according to the amount of knowledge is proposed to fulfill such task that takes advantage of its ability of computation with time series information. The proposed spiking neural network model could control the size of network and modify the content of knowledge the network learns.At last, this hotspot detection method has been proved the feasibility and effectiveness by an infrared simulation video which describes a phenomena that a solar panel works with partial shading. This infrared video is simulated by MATLAB. Furthermore, the realization of detection method is performed by the CUDA, the high performance computing platform, which could satisfy the requirement of real-time detection.