太陽能發電警示系統誤報問題改善之研究

A Study on Improving False Alarm Problem of Solar Photovoltaic Power

洪正鴻(Cheng -Hung Hung)

指導教授 ： 姜琇森

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2022年)

摘要

在太陽能發電過程中，太陽的角度和溫度的變化會影響直流模塊系列的轉換效率。在某些情況下，太陽能最高功率追蹤(Maximum power point tracking，MPPT)串列異常發電轉換效率不好，可能是因為樹枝的陰影或偏離的陽光角度，而非真正的故障。而基於發電轉換效率異常的故障警示系統並無法區別這樣的差異，導致警示頻率過高，其中又夾雜著誤判，造成監控人員無法有效控管情況(警示過多，不知太陽能版是否真的故障)，也間接造成維修人員的交通成本往上攀升的問題。
本研究收集2021年8月13日07時00分起至2022年4月13日17時30分止的太陽能發電資料，包含太陽日照度、模組溫度、MPPT轉換功率等欄位，共有252,540筆。這些經過雲端警示系統判斷為異常數據後，共有28,550筆。透過專家與監控人員檢視，將有遮陰的數據進行人工標籤化處理後，得到數據為2,837筆。為解決上述問題，本研究運用機器學習方法(決策樹 CART、隨機森林 Random Forest、支持向量分類 Support Vector Machine、MLP多層感知器分類器 Multilayer Perceptron)發展太陽能發電警示誤判檢測模型，以區別真實故障及遮陰影響的狀況。實驗結果發現以決策樹CART為基礎的太陽能發電警示誤判檢測模型，可以有效分辦真實故障及遮陰影響的狀況，可幫助監控人員瞭解是否為真正故障情形，避免警示頻率過高，使人員疏於警示，也降低第一線的維修人員因為誤判造成耗費過多人力成本與出勤費用。

關鍵字

太陽能設備 ； 設備誤報檢測 ； MPPT功率追蹤 ； 機器學習 ； 決策樹

並列摘要

In the process of solar power generation, changes in the angle and temperature of the sun will affect the conversion efficiency of power generation. In some cases, the solar maximum power point tracking (MPPT) tandem abnormality leads to poor power conversion efficiency. The reason may be the shadow of the branches or the deviated sunlight angle rather than a real fault. However, the fault warning system based on abnormal power conversion efficiency cannot distinguish such differences, resulting in an excessively high warning frequency mixed with misjudgment. As a result, the monitoring personnel cannot effectively control the situation (there are too many warnings, I do not know whether the solar panel is malfunctioning). It also indirectly causes the transportation cost of maintenance personnel to rise.
This study collects solar power generation data from 07:00 on August 13, 2021 to 17:30 on April 13, 2022, including variables such as solar insolation, module temperature, and MPPT conversion power, with a total of 252,540 records. There are a total of 28,550 cases of abnormal data judged by the cloud fault warning system. These abnormal data were inspected by experts and monitoring personnel, and after manual labeling of the data, 2,837 misjudgments records were obtained. To solve the above problems, this study uses machine learning methods (CART Decision Tree, Random Forest, Support Vector Machine, Multilayer Perceptron) to develop a solar power equipment false alarm detection model to distinguish between real faults and shading effects. The experimental results show that the solar power equipment false alarm detection model based on the decision tree can effectively separate the real faults and shading effects. It can help monitoring personnel understand whether it is a real fault condition and avoid warnings that are too frequent and cause personnel to ignore warnings. In addition, the results can also reduce the high labor cost and attendance cost caused by the misjudgment of front-line maintenance personnel.

並列關鍵字

Solar Power equipment ； Machine Learning ； Equipment False Alarm Detection ； Maximum Power Point Tracking ； Decision Tree

參考文獻

參考文獻(中文)

經濟部能源局(2021 a), 108/109年全國電力資源供需報告_經濟部能源局，https://ws.ndc.gov.tw/001/administrator/10/relfile/5653/30018/adb5a614-0e39-46b1-9a22-d49075499ac2.pdf, 2022讀取

參考文獻(英文)

Al-Alawi, S., Al-Hinai, H. (1998). An ANN-based approach for predicting global radiation in locations with no direct measurement instrumentation. Renewable Energy, 14(1-4), 199-204.

Benkercha, R., Moulahoum, S. (2018). Fault detection and diagnosis based on C4. 5 decision tree algorithm for grid connected PV system. Solar Energy, 173, 610-634.