Quality Evaluation and Prediction of PV Modules by Utilizing LM-Based Back Propagation Neural Network

應用倒傳遞類神經網路於太陽能電池模組之品質評價研究

蔡明忠(Ming-Jong Tsai) ； 陳思羽(Szu-Yu Chen) ； 黃如斌(Ru-Bing Huang) ； 彭成瑜(Cheng-Yu Peng)

《中國機械工程學刊》 37卷5期 (2016/10) Pp. 479-486

摘要

This paper presents quality evaluation of PV modules using back propagation neural networks (BPNN) with one hidden layer and four neurons, four inputs and three outputs. The key input parameters were adopted on internationally verified standards for silicon modules (International Electro-technical Commission, IEC61215). Firstly, the correlation mechanism between the module quality and defects in solar cell modules is investigated. Then, the key factors affecting quality evaluations of photovoltaic modules for classifications were developed. An estimated model is then constructed using artificial neural networks technology and the weight values between defect factors were automatically allocated by giving several learning data. Four key input parameters were the attenuation ratio of maximum output power and the EL image quantitative indicators, insulation resistance and wet leakage resistance regularization indicators. The output qualities were assigned evaluation classifications of A, B, and C. By using 30 set of simulated Training data, a best neural networks model with 4 internal nodes is obtained. From validation evaluation results for another 15 set of test simulated data, the successful rate is about 93.3%.

關鍵字

PV modules ； Quality Evaluation ； Back Propagation Neural Networks (BPNN)

並列摘要

本文應用倒傳遞類神經網路於太陽能電池模組之品質評價研究。以IEC61215規範為標準，設計類神經網路的訓練資料，四個輸入、三個輸出、一層隱藏層，隱藏層有四個神經元。首先，探討太陽能電池模組常見的缺陷以及這些缺陷與品質之間的關係，接著，從以上缺陷中選出四個關鍵參數作為品質評價模型的輸入，本研究選用的關鍵參數為經過IEC61215規範試驗後太陽能電池模組的絕緣電阻、濕漏電阻，以及量測試驗前後太陽能電池模組的最大功率衰減比例、EL影像量化指標的衰減比例。最後，依據品質評價模型的輸出形式，將評價分為三級（A、B、C）。訓練資料有30組，測試資料有15組，其預測正確率為93.3%（14/15）。