Photovoltaic Fire Risk Multi-level Early Warning System

As an important part of new energy technology,the scale of photovoltaic power station has developed rapidly in recent years.But at the same time,photovoltaic fires have also become one of the important factors restricting the further development of photovoltaic technology.The subjecet of paper is multi-level fire warning technology in photovoltaic power station.And we design and realize the warning system through researching the fire detection technology.The research includes many types of photovoltaic fire detection technology,which is of great significance to the development of photovoltaic power station safe operation and maintenance technology.Aiming at the fire warning of photovoltaic,the paper completes a multi-level fire warning technology and realize the installation and testing in experimental photovoltaic power station.The multi-level fire warning system detection of fire hazards on the three scales including environment,component and power station.The main work of the paper includes the following parts:(1)Summarize the sources of fire insurance for photovoltaic power station through investigation.Develop detection techniques for the photovoltaic fire insurance which is high frequency and widespread.On the basis,design a multi-level fire warning scheme with comprehensive technical testing standards.In order to verify the robustness of the multi-level fire warning scheme,the author built an experimental photovoltaic power station platform,and equipped the platform with multi-level fire warning monitoring sensors such as weather station,camera and DC arc sensor.At the same time,complete the selection and packaging of the warning terminal according to the warning algorithm.(2)Design an environmental fire risk classification model based on the AHP hierarchy analysis algorithm and fuzzy comprehensive evaluation.The fire risk factor is selected through the environmental data of the photovoltaic power station,and the mathematical model of the fire risk factor is build by referring to the forest fire weather classification standard.Expert scoring is used to construct a judgment matrix and calculate factor weights.The calculation results of model weights meet the consistency test standards.After that,the fuzzy matrix is constructed to realize the fire risk classification.Finally,the classification model was tested according to the forest fire classification label of the National Meteorological Center and the meteorological data of the CIMISS platform.(3)Selection and calibration of photovoltaic DC arc detection sensors.Through investigation and analysis of the danger of photovoltaic arcs,select sensors for photovoltaic DC arcs and make threshold adjustments.The arc generator is made to conduct simulation experiments,and the threshold adjustment of the DC arc sensor are completed through the experiments of series arc and false detection arc based on the experimental photovoltaic power station platform.After the threshold of the sensor is adjusted,it can work normally in the experimental photovoltaic power station and the false detection rate is less than 8%.(4)Research on the detection algorithm of flame,smoke and combustibles with the YOLOV4-Tiny network model.Author built the data sets including flame,smoke and combustibles image and completed data annotations.Design experiments to verify the performance of the YOLOV4-Tiny network,YOLOV4 network and Fast-RCNN network on t HSI data set,and analyze the performance of the YOLOV4-Tiny network through horizontal comparison.The network model is carried in the processor of the fire warning terminal Jetson TX2 to achieve high-precision identification tasks.(5)Smoke detection technology based on distant view video.Firstly,adaptive median filtering is used to preprocess the video stream,and then the foreground image segmentation is completed by improving the Vibe algorithm.In terms of target feature extraction,the color space of the smoke image is converted to the HSI space to extract the saturation and brightness features,and the gray-level co-occurrence matrix is used to extract the texture features.The foreground area is detected by feature fusion and SVM classifier.The algorithm is tested on the video concentration with 92% recognition accuracy,and the running speed on Jetson TX2 processor GPU is not less than 12 FPS.(6)The software and hardware realization of multi-level fire warning terminal.The hardware part achieves IP67 protection level requirements through aviation plug conversion and chassis design to ensure the stable operation of terminal equipment at the power station site.The software part is based on PHP’s web development technology to realize the visualization of the monitoring terminal and the management of HSItorical data.All modules of the software platform are running in experimental photovoltaic power stations,and algorithm parameters can be adjusted through t HSI platform to improve algorithm performance.