TW201120310A - The state telemetry technology and fault diagnosing system in large-scale wind power farms - Google Patents

The state telemetry technology and fault diagnosing system in large-scale wind power farms Download PDF

Info

Publication number
TW201120310A
TW201120310A TW098142015A TW98142015A TW201120310A TW 201120310 A TW201120310 A TW 201120310A TW 098142015 A TW098142015 A TW 098142015A TW 98142015 A TW98142015 A TW 98142015A TW 201120310 A TW201120310 A TW 201120310A
Authority
TW
Taiwan
Prior art keywords
wind
generator
fault
neural network
output
Prior art date
Application number
TW098142015A
Other languages
Chinese (zh)
Inventor
meng-hui Wang
Guan-Jie Huang
Jen-Cheng Yang
Original Assignee
Nat Univ Chin Yi Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nat Univ Chin Yi Technology filed Critical Nat Univ Chin Yi Technology
Priority to TW098142015A priority Critical patent/TW201120310A/en
Publication of TW201120310A publication Critical patent/TW201120310A/en

Links

Classifications

    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02EREDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00Energy generation through renewable energy sources
    • Y02E10/70Wind energy
    • Y02E10/72Wind turbines with rotation axis in wind direction
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04SSYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention is the telemetry technology and fault diagnosing system for wind power farm. Firstly, it takes a set of fault signal from a simulation diagnosis system. All fault conditions are obtained from the installed sensors in the wind power system, and then they provide a diagnosis system based on the extension neural network (ENN) for fault diagnosis. It can quickly diagnose the fault status of wind power system to improve the maintained efficiency.

Description

201120310 六、發明說明: 【發明所屬之技術領域】 本發明係關於一種風力發電場狀態遙測技術及故障診 斷系統,尤指一種利用感測器檢測風力發電機特定部位的 特徵訊號’再利用事先模擬的故障狀況以自動判斷風力發 電機故障狀況之相關技術。 【先前技術】 _ 一十一世紀全球暖化現象日益嚴重,已引起世界各國 重視’降低二氧化碳的排放量及發展再生能源,已經是全 球人類必須面對的潮流。風能是一種清潔的能源,風力發 電在新能源行業中成長最快,根據財團法人國家實驗研究 院科技政策研究與資訊中心提供之統計數據,2〇〇6年全世 界發電容量高達74,2〇〇MW,足見風力發電越來越受到各 國重視’全球發展風力發電的成長率約達35〇/〇 ;在美國、 德國和丹麥等先進國家甚至高達5〇%以上。我國是一個天 鲁然能源短缺的國家,絕大多數皆需仰賴進口;另外隨著環 保意識的抬頭,民間對新電廠的設立,不論是火力發電廠 或核能電廠的興建皆遭遇極大的反對聲浪,隨著石化能源 饧格的攀升,再生能源的研究與開發是目前解決國内能源 問題的方法之一。 目别國内已有設置大型風力發電場,在秋冬季持續且 強勁的東北季風下運作,風力發電系統可以提供相當不錯 的電力,並相對降低傳統電廠的供電量,對國内能源供應 八有相田大的效益。因此我國目前不管是政府或民間對於 201120310 風力發電場均有大量投資,而在風力發電場中,風力發電 機為噪音主要來源,為避免干擾人群及考量風力條件,大 型風力發電場的位置大都選在偏遠地區或海上(離岸型), 且風力發電機間為避免風流干擾,風力機間的排列分佈必 須非常的分散。因此,風力發電場的設置須面臨如下之難 題: 1 ·風力發電場地處偏遠,故距離監控中心相當遠,災 害防治、設備支援和通訊都相當困難。 2·風力發電場風勢強勁(尤其是颱風)、環境惡劣對風 力機葉片及内部機構損害極大。 3_風力機之間的排列分佈必須分散,使監控系統設計 較為複雜。 由於風力發電機長期暴露在外,對於故障的發生率也 相對提高,例如2刪年1G月間,台電香山二號風力發電 機組疑似變壓器過熱,以致引燃齒輪箱内齒輪油引起大火 ’且因風力機高度過高造成救援困難,造成風力發電機的 損壞嚴重,此次大火可謂勞民傷財’損失慘重。因此,如 何預防風力發電機事故發生,確保風力發電場安全穩定的 運轉、提高其管理效率、降低營運及維修成本,需要一套 功能完善、性能穩定的運轉狀態監測和故障診斷系統,已 成為大型風力發電場研究相當重要的項目之一。 目前全球投入此再生能源之研究非常的多,但國内、 外在故障診斷和狀態偵測之相關研究並不多僅有少數成 果發表,在全世界大力推展再生能源之際,我國資訊相關 技術相當發達’有必要積極投人故障珍斷和狀態债測之相 201120310 之關鍵技術,以避免過 主及降低營運和維修成 機監控及診斷之相關研 關研究,來發展符合台灣運轉環境 度依賴進口物品’以期達到技術自 本。再者,近年來國内外風力發電 究如下:201120310 VI. Description of the Invention: [Technical Field] The present invention relates to a wind power field state telemetry technology and a fault diagnosis system, and more particularly to a feature signal for detecting a specific part of a wind power generator by using a sensor The fault condition is related to the technology for automatically determining the fault condition of the wind turbine. [Prior Art] _ The global warming phenomenon in the 11th century has become increasingly serious, and has attracted the attention of all countries in the world. 'Reducing carbon dioxide emissions and developing renewable energy is already a trend that must be faced by humans all over the world. Wind energy is a clean energy source. Wind power generation is the fastest growing in the new energy industry. According to statistics provided by the Science and Technology Policy Research and Information Center of the National Experimental Research Institute of the Foundation, the world's power generation capacity is as high as 74,2 in 2 years. 〇〇 MW, it is obvious that wind power is getting more and more attention from all countries. The global growth rate of wind power generation is about 35〇/〇; in advanced countries such as the United States, Germany and Denmark, it is even more than 5%. China is a country with a shortage of energy, and most of them need to rely on imports. In addition, with the rise of environmental awareness, the establishment of new power plants by the people, whether it is a thermal power plant or a nuclear power plant, has encountered great opposition. With the rising of petrochemical energy, the research and development of renewable energy is one of the methods to solve domestic energy problems. In view of the large wind farms in the country, which operate under the continuous and strong northeast monsoon in autumn and winter, the wind power system can provide quite good power and relatively reduce the power supply of traditional power plants. The benefits of Aida. Therefore, in China, whether it is the government or the private sector, there is a large investment in the 201120310 wind farm. In the wind farm, wind turbines are the main source of noise. In order to avoid interference and consider the wind conditions, the location of large wind farms is mostly selected. In remote areas or offshore (offshore), and to avoid wind disturbances between wind turbines, the distribution of wind turbines must be very dispersed. Therefore, the setting of wind farms must face the following difficulties: 1 • The wind power generation site is remote, so it is quite far from the monitoring center. Disaster prevention, equipment support and communication are quite difficult. 2. Wind farms are strong (especially typhoons), and the environment is extremely damaging to wind turbine blades and internal mechanisms. 3_The distribution of the wind turbines must be dispersed, making the design of the monitoring system more complicated. Due to the long-term exposure of wind turbines, the incidence of failures has also increased relatively. For example, during the 1G month of 1 year, Taipower No. 2 wind turbine generators suspected that the transformers were overheated, causing the gear oil inside the gearbox to cause a fire. The height is too high, resulting in difficulties in rescue, resulting in serious damage to wind turbines. The fire can be described as a huge loss of labor and wealth. Therefore, how to prevent wind turbine accidents, ensure the safe and stable operation of wind farms, improve their management efficiency, and reduce operating and maintenance costs requires a well-functioning and stable operation state monitoring and fault diagnosis system, which has become a large-scale Wind farm research is one of the most important projects. At present, there are many studies on the global investment in renewable energy, but there are not many researches on domestic and external fault diagnosis and state detection. Only when a lot of achievements are promoted in the world, China's information technology It is quite developed that it is necessary to actively invest in the key technologies of 201120310, which is the fault of the fault and state debt measurement, to avoid over-the-counter and reduce the related research and research on the operation and maintenance of machine monitoring and diagnosis, to develop the dependence on Taiwan's operating environment. Imported items 'in order to achieve technology from the present. Furthermore, wind power research at home and abroad in recent years is as follows:
1·風力機並聯供電线之研究:因為受限於 素,風力發電無法長時間提供穩定的電力,所以通常會 上辅助電源…市電、太陽能、蓄電池、或其他類型之 Μ能源’來穩定負載所需之功率’彡中需要考慮到前端 的交流/直流轉換器、後端提供市電的直流,交流轉換器 ,以及考慮最大功率追蹤等問題。 、π 2·風力發電機之狀態監控:目前國内、外相關文獻的 方法有: ⑴振動分析:將感測器安裝於特定位置,測量 動頻率; (2)油質分析:可分成分析油中雜質與油的成份兩部分 ,此檢測方法大多必須停機,所以應用於風力發電機狀態 Φ 檢測並不十分理想; ^ (3)聲音監測:此方法大多用將聲音感測器裝設於齒輪 相内,藉由齒輪運轉聲音來判斷機組狀態; (4)電機效應:包含電機機械放電測量、開關切換速度 測量、開關接點測量、變壓器油質分析等,其目的在於提 早發現故障、分析故障問題與建立歷史資料三大方向。 3.提焉風能效率之研究:相關文獻以風速與輸出功率 的關聯度,以現有之風場實際測量所在位置的風速、經度 緯度’輸出功率’應用人工智慧方法評估風場效益,藉 201120310 由數據庫的建立,做為新的風場選擇之評估依據,增加風 場建設的經濟效益,以及經由流體力學等數學模擬,探討 風力機風葉擷取最大風能的條件與葉片數量對系統效應之 影響。 4. 風力發電模型之建構及保護模擬:風力機並聯電力 系統時,必須考慮電壓、電流、頻率和同步問題,若系統 故障發生時,應快速切離電力系統。在文獻上探討的系統 故障有短路故障、頻率變動、電壓驟降等故障之模擬分析 ,目的是希望系統發生不同程度之擾動時,可預先得知擾 動對系統之影響,適時利用保護設備,使發電系統回復穩 定狀態。 ~ 5. 風力發電機的故障診斷,大致可分成: (1)訊號分析方法:在機構方面,根據機械故障診斷的 基本原理,透過安裝在風力機齒輪箱的感測器,所收集到 的震動信冑,透過頻譜分析、波型分析、倒頻譜分析等方 法,以掌握風力機運轉時異常或故障的初期徵兆,比起以 常規方法檢測或手摸和耳聽的檢測方法,可更早檢測到機 械結構的故障;在電力方面’以電力錶頭進行相關參數的 量測’如電壓、電流'相位、有效功率'無效功率等,利 用所擷取之參數進一步分析,以判斷故障發生的類型。 (2)人工智慧方法:當系統特徵參數過於複雜時,多數 研究者會採用人工智慧方法協助故障診斷’這些方法包括 專家系統、神經網路和模糊理論等方法,透過學習步驟, 進行故障診斷,可提高診斷的準確率。 由上述可知,大型風力發電機長期暴露在外,且因風 201120310 力機高度過高救援困難,如何預防風力發電機事故發生, 確保風力發電場安全穩定的運轉及降低維修成本,需要一 套功能完善、性能穩定的狀態監測和故障診斷系統,而既 有故障診斷系統在特徵參數過於複雜時,必須採用人工智 慧協助故障診斷,但人工智慧本身仍需要特殊的人為參數 ’可能相對延長診斷及學習的時間,故仍有進一步檢討及 改善的空間。 鲁 【發明内容】 因此,本發明主要目的在提供一種風力發電場狀態遙 測技術及故障診斷系統,其採用可拓類神經網路進行故障 診斷,不需要特殊的人為參數,即可大幅縮短診斷及學習 的時間。 為達成前述目的採取的主要技術手段係令前述系統包 括: ' 複數感測器,分設於風力發電機内各個指定的位置上 •進行訊號收集,以取得各指定檢測位置的特徵訊號; 一處理單元,係分別與各感測器連接,又處理單元内 建一可拓類神經網路及多種模擬故障狀況;其中,該可拓 類神經網路包括一輸入層及一輸出層,其中輸入層^將各 特徵訊號進行分類並建構成物元模型,再送入可拓類神經 網路,以運算出各類特徵訊號的可拓距離最小值進而由 輸出層輸出判斷的故障狀況; 由於本發明採用可拓類神經網路進行故障診斷,具 架構簡單、學習速度快、診斷準確率高等優點,同時對 201120310 具雜訊之輸入資料具有相當高的容錯能力。 【實施方式】 關於本發明一可行實施例的系統架構請參閱第一圖所 不,主要係令一故障診斷系統與一風力發電機(2〇)連 結,以便在風力發電機(20)工作時監測其運轉狀況,並在 異常狀況發生時迅速診斷出其故障原因,以利於及時維修 或爭取搶修時機;其中: # 該風力發電機(20)包括一葉片(21)、一齒輪箱(22)、一 發電機(23)及一變壓器(24);其中,葉片(21)具有一輪軸 (210),並透過其輪軸(21〇)與齒輪箱(22)連結,而帶動齒 輪箱(22)内的齒輪組,又齒輪箱(22)的輸出端係與發電機 (23)的轉軸(230)連結,由齒輪箱(22)輸出的動力帶動發電 機(23)運轉並發電,所產生的電力經過變壓器(24)變壓後 進行供電或併網。 又本發明主要係在風力發電機(2〇)各元件的特定位置 •或輸出端安裝感測器(25)〜(28),以取得各元件感測項目的 特徵訊號,其檢測的項目大致包括: 1.輪軸(210)、轉軸(230)的運轉狀況檢測:檢測輪軸 (210)、轉轴(230)故障的方法很多,最普遍的方法是使用 震動分析’由於設備的設計安裝誤差及故障出現時,會在 輪軸(210)、轉軸(230)處產生震動,故本發明選擇在其輪 轴(21 0)及轉軸(230)的轴承(211 )(231)處設感測器(25),用 以感測輪軸(210)、轉軸(230)的震動狀況,再將震動感測 訊號經過頻域轉換,以分析出輪軸(21〇)、轉軸(23〇)的狀 201120310 態訊息,再根據狀態訊息判斷故障的項目,而可能發生的 故障項目包含磨損、斷裂、潤滑不良等。 2·齒輪箱(22)運轉狀況檢測:由於風力發電機(2〇)長 時間的運轉’傳動機構扮演由風能轉為機械能的重要角色 ,如能減少齒輪之間的損失’即可提高風力機的輸出效率 ,其監測項目有潤滑油油位、油溫度訊號、齒輪振動訊號 等,故本發明在齒輪箱(22)内分設感測器(26),以分別檢 測其油位及油溫是否正常,以確認齒輪箱(22)的運轉狀況 〇 參 3·發電機(23)運轉狀況檢測:發電機(23)是將機械能 轉為電能之主要元件,需考慮到輸出電壓、電流、相位與 當時風速之相對關係’可估計發電機之輸出功率曲線與經 過長時間運轉後,效率滑落到一定標準以下時提供警報; 因此本發明在發電機(23)的輸出端設有一電流電壓轉換器 (31)’以檢測其輸出電壓、電流與相位,另配合一風速檢 測器(33)及一風向檢測器(34)檢測運轉當時的風速風向訊 φ 號’與電流電壓轉換器(31)檢測所得輸出功率訊號一起送 至故障診斷系統(1 〇)進行判讀。另一方面,發電機(23)的 機身上亦設有感測器(27),以感測發電機(23)的震動情況 〇 4.變壓器(24)狀況檢測:係在變壓器(24)上安裝感測 器(28)以檢測其溫度變化;除此以外,該變壓器(24)的輸 出端進一步設有一電流電壓轉換器(32),以測量其電壓、 電流輸出大小》 除前述狀況檢測外’另可進一步以其他感測器檢測葉 201120310 片(21)及發電機(23)之轉速、齒輪箱(22)内機油壓力等。 前述各感測器(25)~(28)安裝位置的選定,可利用紅外 線熱顯像儀觀測運轉中的風力發電機(2〇),以取得其軸承 (211 )(231 )、齒輪箱(22)、發電機(23)、變壓器(24)之熱影 像,做為溫度偵測器安裝位置之參考點。又前述感測器 (25)〜(28)的輸出訊號係分別透過一換能器(TrarjSC|ucer) (29)與故障診斷系統(1〇)連結。 該故障s爹斷系統(10)包括有一處理單元(11) '複數的 Φ 類比數位轉換器(1 2)(1 3)、一顯示器(14)及一資料匯流排 (15);其中:該處理單元(ή)係透過資料匯流排(15)分別與 類比數位轉換器(12)(1 3)、顯示器(14)及風力發電機(20)上 各個換能器(29)連接’並進一步透過類比數位轉換器 (12)(13)分別與兩電流電壓轉換器(31)(32)及風速檢測器 (33)、風向檢測器(34)連接,以取得發電機(23) '變壓器 (24)的輸出電壓、電流與風速、風向訊號;再透過各換能 器(29)與風力發電機(2〇)上的各組感測器(25)~(28)連接。 φ 而處理單元(1彳)係將來自風力發電機(20)的輸出功率 、溫度感測、震動感測、油位感測及風速、風向等訊號利 用MATLAB / SIMULINK作為軟體介面,而利用相關訊號 分析軟體與資料探勘軟體,觀察訊號時頻圖、頻譜圖、週 期訊號與非週期訊號’進行初步的訊號來源點分析,再用 小波轉換、傅利葉轉換或希爾伯特-黃轉換(Empirical Mode Decomposition,EMD) ’萃取訊號的特徵值,以減少 建立故障碼的數目’並作為診斷系統之故障分類碼的資料 庫;而經過萃取特徵值的訊號即視為一特徵訊號,則進一 201120310 步利用内建的可拓類神經網路進行錯誤診斷。 請參閱第一圖所示,其揭矛古_ ,其包括-輸入層⑷)與一輪出層叫=經網路之架構 輸出層(42)之間建構可拓類神經網路(43);主輸 入的特徵訊號分類並建構成—物” 1 物兀模型後,再送入可拓類 神經網路(43)中’其中,輪 輸入層(41)的節點數量是由 模型之特徵數量所決定,而輪屮思“〇、 向輸出層(42)的節點數量則是由 特徵訊號的類別數量決定,並分 刀別用以存放計算後之可拓 距離,最後在輸出層(41)由最小值的可拓距離決定特徵訊 號所屬的類別。 可拓類神經網路(43〗盥甘au θ'5)與其他類神經網路方法相同,包 括學習與辨識兩個步驟;其中: 可拓類神經網路的學習法分為非監督式的#習與監督 式學習,本實施例採用監督式學習法來調整權重,藉由不 斷地學習與訓練來進行調整與修正權重。而可拓類神經網 路之監督式學習的演算步驟係如下所述: ϋ Ί1.#立輸入與輸出之權重值1. Research on parallel power supply lines for wind turbines: Because wind power is limited by the fact that wind power cannot provide stable power for a long time, it usually has auxiliary power supply...mains, solar energy, battery, or other types of energy to stabilize the load. The required power '彡 needs to consider the front-end AC/DC converter, the back-end to provide mains DC, AC converter, and consider the maximum power tracking. , π 2 · State monitoring of wind turbines: At present, the domestic and foreign related literature methods are: (1) Vibration analysis: the sensor is installed at a specific location to measure the dynamic frequency; (2) Oil quality analysis: can be divided into analytical oil In the two parts of impurities and oil, most of the detection methods must be stopped, so the detection of wind turbine generator state Φ is not very satisfactory; ^ (3) Sound monitoring: This method mostly uses the sound sensor installed in the gear In the phase, the state of the unit is judged by the sound of the gear running; (4) Motor effect: including mechanical discharge measurement of the motor, switch switching speed measurement, switch contact measurement, transformer oil quality analysis, etc., the purpose is to detect faults early and analyze faults. Problems and the establishment of historical data in three major directions. 3. Research on wind energy efficiency: The related literature uses the correlation between wind speed and output power to estimate the wind field benefit by applying the artificial intelligence method to the wind speed and longitude latitude 'output power' of the existing wind field. From the establishment of the database, as the basis for the evaluation of new wind farm selection, increase the economic benefits of wind farm construction, and through mathematical simulations such as fluid mechanics, explore the conditions of wind turbine blades to extract maximum wind energy and the number of blades on the system effect. influences. 4. Wind power model construction and protection simulation: When the wind turbine is connected to the power system, the voltage, current, frequency and synchronization problems must be considered. If the system fails, the power system should be quickly cut off. The system faults discussed in the literature have short-circuit faults, frequency fluctuations, voltage dips and other faults. The purpose is to hope that when the system has different degrees of disturbance, the influence of the disturbance on the system can be known in advance, and the protection equipment can be used in time. The power generation system returns to a steady state. ~ 5. Wind turbine fault diagnosis can be roughly divided into: (1) Signal analysis method: In terms of mechanism, according to the basic principle of mechanical fault diagnosis, the vibration collected by the sensor installed in the wind turbine gearbox Letterhead, through spectrum analysis, waveform analysis, cepstrum analysis and other methods, to grasp the initial signs of abnormal or faulty operation of the wind turbine, can be detected earlier than the detection method by conventional method or hand touch and ear hearing To the failure of the mechanical structure; in the aspect of electric power, 'measure the relevant parameters with the power meter', such as voltage, current 'phase, effective power', invalid power, etc., and further analyze the parameters taken to determine the type of fault occurrence. . (2) Artificial intelligence method: When the system characteristic parameters are too complicated, most researchers will use artificial intelligence to assist in fault diagnosis. These methods include expert systems, neural networks and fuzzy theory, and through the learning steps, fault diagnosis, Improve the accuracy of the diagnosis. It can be seen from the above that large-scale wind turbines are exposed for a long time, and because of the difficulty in rescue of winds 201120310, how to prevent wind turbine accidents, ensure the safe and stable operation of wind farms and reduce maintenance costs, a set of functions is needed. Stable condition monitoring and fault diagnosis system, while the fault diagnosis system must use artificial intelligence to assist fault diagnosis when the characteristic parameters are too complicated, but artificial intelligence itself still needs special artificial parameters 'may be relatively prolonged diagnosis and learning Time, so there is still room for further review and improvement. [Invention] Therefore, the main object of the present invention is to provide a wind power field state telemetry technology and a fault diagnosis system, which uses an extension type neural network for fault diagnosis, and can greatly shorten the diagnosis without special human parameters. The time of study. The main technical means for achieving the above objectives are that the system includes: 'Multiple sensors, which are located at various designated positions in the wind turbine. • Signal collection to obtain characteristic signals of each specified detection position; The system is connected to each sensor, and the processing unit has an extension-type neural network and a plurality of simulated fault conditions. The extension-type neural network includes an input layer and an output layer, wherein the input layer is The feature signals are classified and constructed to form a matter-element model, and then sent to the extension-type neural network to calculate the minimum value of the extension distance of each type of characteristic signal and then the fault condition judged by the output layer output; The fault-based neural network for fault diagnosis has the advantages of simple structure, fast learning speed and high diagnostic accuracy. At the same time, it has a fairly high fault tolerance for the input data of 201120310 with noise. [Embodiment] Referring to the first figure, the system architecture of a feasible embodiment of the present invention is mainly for connecting a fault diagnosis system to a wind power generator (2〇) so that when the wind power generator (20) is working. Monitor the operation status and quickly diagnose the cause of the failure when the abnormal condition occurs, so as to facilitate timely maintenance or seek emergency repair time; wherein: # The wind turbine (20) includes a blade (21) and a gear box (22) a generator (23) and a transformer (24); wherein the blade (21) has an axle (210) and is coupled to the gearbox (22) through its axle (21〇) to drive the gearbox (22) The inner gear set, and the output end of the gear box (22) is coupled to the rotating shaft (230) of the generator (23), and the power outputted by the gear box (22) drives the generator (23) to operate and generate electricity. After the transformer is transformed by the transformer (24), the power is supplied or connected to the grid. Further, the present invention mainly installs sensors (25) to (28) at specific positions of the components of the wind turbine (2〇) or at the output end to obtain characteristic signals of sensing items of the respective components, and the items to be detected are roughly Including: 1. Operation detection of axle (210) and shaft (230): There are many ways to detect the failure of axle (210) and shaft (230). The most common method is to use vibration analysis. When the fault occurs, vibration will occur at the axle (210) and the rotating shaft (230). Therefore, the present invention selects a sensor at the bearing (211) (231) of the axle (21 0) and the rotating shaft (230) ( 25), to sense the vibration condition of the axle (210) and the rotating shaft (230), and then frequency-domain-convert the vibration sensing signal to analyze the state of the axle (21〇) and the shaft (23〇). Then, according to the status message, the faulty item is judged, and the faulty items that may occur include wear, breakage, poor lubrication, and the like. 2. Gearbox (22) operation detection: Due to the long-term operation of the wind turbine (2〇), the transmission mechanism plays an important role in converting wind energy into mechanical energy, such as reducing the loss between gears. The output efficiency of the wind turbine, the monitoring items include lubricating oil level, oil temperature signal, gear vibration signal, etc., so the present invention separates the sensor (26) in the gear box (22) to detect the oil level and Whether the oil temperature is normal, to confirm the operation status of the gear box (22) 〇 3 3 · generator (23) operating condition detection: the generator (23) is the main component to convert mechanical energy into electrical energy, taking into account the output voltage, The relative relationship between current, phase and current wind speed can be used to estimate the output power curve of the generator and provide an alarm when the efficiency falls below a certain standard after a long period of operation; therefore, the present invention provides a current at the output of the generator (23). The voltage converter (31)' detects the output voltage, current and phase, and cooperates with a wind speed detector (33) and a wind direction detector (34) to detect the wind speed and direction of the wind at the time of the operation. Feeding voltage converter (31) with the detection signal to output the resulting fault diagnosis system (1 billion) for interpretation. On the other hand, the generator (23) is also provided with a sensor (27) to sense the vibration of the generator (23). 4. Transformer (24) condition detection: on the transformer (24) The sensor (28) is installed to detect the temperature change; in addition, the output of the transformer (24) is further provided with a current-voltage converter (32) to measure the voltage and current output size. 'Other sensors can be used to detect the speed of the blade 201120310 (21) and generator (23), the oil pressure in the gearbox (22), and so on. The selection of the mounting positions of the sensors (25) to (28) can be performed by using an infrared thermal imager to observe the wind turbine (2〇) in operation to obtain its bearing (211) (231) and gear box ( 22), the thermal image of the generator (23) and the transformer (24), as the reference point for the installation position of the temperature detector. Further, the output signals of the sensors (25) to (28) are respectively connected to the fault diagnosis system (1〇) through a transducer (TrarjSC|ucer) (29). The fault s breaking system (10) includes a processing unit (11) 'complex Φ analog digital converter (1 2) (1 3), a display (14) and a data bus (15); wherein: The processing unit (ή) is connected to each transducer (29) on the analog digital converter (12) (13), the display (14) and the wind turbine (20) through the data bus (15) and further Connected to the two current-to-voltage converters (31) (32) and the wind speed detector (33) and the wind direction detector (34) through the analog-to-digital converter (12) (13) to obtain the generator (23) 'transformer ( 24) The output voltage, current and wind speed, wind direction signal; and then through each transducer (29) and the various sets of sensors (25) ~ (28) on the wind turbine (2 〇). φ and the processing unit (1彳) uses the MATLAB / SIMULINK as the software interface for the output power, temperature sensing, vibration sensing, oil level sensing, wind speed, wind direction and other signals from the wind turbine (20). Signal analysis software and data exploration software, observe signal time-frequency diagram, spectrogram, periodic signal and non-periodic signal' for preliminary signal source point analysis, and then use wavelet transform, Fourier transform or Hilbert-Huang transform (Empirical Mode) Decomposition, EMD) 'Extract the characteristic value of the signal to reduce the number of fault codes established' and use it as a database of fault classification codes for the diagnostic system; and the signal after extracting the feature value is regarded as a characteristic signal, then use the 201120310 step Built-in extension-type neural network for error diagnosis. Referring to the first figure, it is constructed that an extension type neural network (43) is constructed between the input layer (4) and the round output layer (42). The characteristic signal of the main input is classified and constructed into a “object”. After the object model is sent to the extensional neural network (43), the number of nodes in the input layer (41) is determined by the number of features of the model. And the number of nodes to the output layer (42) is determined by the number of categories of feature signals, and is used to store the calculated extension distance, and finally at the output layer (41) by the minimum. The extension distance of the value determines the category to which the feature signal belongs. The extension-like neural network (43〗 盥 Gan au θ'5) is the same as other neural network methods, including two steps of learning and identification; among them: the learning method of extension-type neural network is divided into unsupervised #习与监督式学习, this embodiment uses the supervised learning method to adjust the weight, and adjusts and corrects the weight by continuous learning and training. The calculation steps of the supervised learning of the extensional neural network are as follows: ϋ Ί 1.# The weight value of the input and output
Vk2 n. Vfj -….,/i (1) 步驟302 :讀取訓練樣本資料與特徵數允,如式 (2)所示: (2) 步驟303 :計算出每項特徵之權重中心值,以&表示 ,如下所示: ^ 卜幻’(3) (4) 201120310 % 而經,巧範圍可由學習資料所決定如下所示: wij=min\x^ (5) (6) 料( w^=^k·} 其中代表可拓類神經網路之輸入端學習資 步驟304 :利用#以下式(7)開始計算可拓距離(請配Vk2 n. Vfj -...., /i (1) Step 302: Read the training sample data and the feature number, as shown in equation (2): (2) Step 303: Calculate the weight center value of each feature, Expressed in & as follows: ^ 幻幻'(3) (4) 201120310 % By the course, the skill range can be determined by the learning materials as follows: wij=min\x^ (5) (6) Material (w ^=^k·} which represents the input of the extension-type neural network. Step 304: Start the calculation of the extension distance using #(7) below (please match
合參閱第四圖所示): 〜卜%1 Μ Κ-^/) 十1 • 2 (7) 步驟305找k,= ’如果k*= k,並跳到步 kem 〆 驟307 ;若資料類別不相等k*表k,則繼續步驟306之動 作0 步驟306 :調整k類別與k*類別之權重值。 1.更新權重上、下限值大小。 = ^·_〇ω + η(4 - 2kj_old ) (8) = ^j_oId +r}(Xy~ Zkj〇ld) (9) -'ν_〇ω) (10) (11) 2 ·更新權重中心值大小。 ” _ ^kj_new + ^j^new) Zkj、new 2 _ _ (WK J k j _new 2 (12) (13) 步驟307 :重複步驟303至步驟307之步驟,直到所 有學習資料皆讀取並完成學習分類完畢。 步驟308 :當所有資料之分類程序都已達到收斂狀態 12 201120310 若否則返回步驟303繼 或總誤差率到達到目標值則停止 續。 所述: 之辨識演算 即可進行辨識或 當可拓類神經網路完成學習程序後 分類,演算步驟包括: /驟401 . 5賣取可拓類神經網路的權重值矩陣; 步驟402:計算中間值大小; ’See also the fourth figure): ~b%1 Μ Κ-^/) Ten 1 • 2 (7) Step 305 find k, = 'if k*= k, and jump to step kem 307 307; If the categories are not equal k* table k, then the action of step 306 is continued. Step 306: Adjust the weight values of the k category and the k* category. 1. Update the weight upper and lower limit values. = ^·_〇ω + η(4 - 2kj_old ) (8) = ^j_oId +r}(Xy~ Zkj〇ld) (9) -'ν_〇ω) (10) (11) 2 · Update weight center The value size. _ ^kj_new + ^j^new) Zkj, new 2 _ _ (WK J kj _new 2 (12) (13) Step 307: Repeat steps 303 to 307 until all learning materials are read and completed. The classification is completed. Step 308: When all the classification procedures of the data have reached the convergence state 12 201120310 If otherwise, return to step 303 or the total error rate to reach the target value, then stop. The: Identification algorithm can be identified or After the learning neural network completes the learning process, the calculus steps include: /Step 401. 5 Selling the weight value matrix of the extension type neural network; Step 402: Calculating the intermediate value size;
步驟403 :讀取測試樣本; 步驟404 : 步驟405 : 屬類別; 5十算測試樣本與各類別之可拓距離,· 尋找最小可拓距離,藉以判斷測試樣本所 步驟406 .完成辨識所有樣本即停止運算,否則回 驟403讀取下一筆測試樣本。 如前揭所述,由於硬體安裝的誤差或葉片損壞等因素 均將使轉㈣轉時產生震動,若將震動訊號經過頻域轉換 可刀析出a備的狀態訊息;又由於風力發電機⑽)須長 時間運轉,其齒輪箱(22)容易内部齒輪之間的磨損,使得 風力發電機(20)的輸出效率降低,而齒輪箱(22)内的油位 及油溫是否正常也是判斷齒輪箱(2 2)正常運轉與否的關鍵 之一,因此本發明利用多組感測器(25)〜(28)在風力發電機 (20)的齒輪箱(22)取出潤滑油油位、油溫度、齒輪振動等 特徵Λ號’作為前述可拓類神經網路(43)的測試樣本。又 發電機(23)係將機械能轉為電能之主要元件,故須考慮到 輸出電壓、電流、相位與當時風速之相對關係,因而本發 13 201120310 明利用電流電壓轉換器(31)由發電機(23)輸出端取出電氣 k號作為診斷用的測試樣本。 再者,本實施例係先模擬出8種可能的故障(運轉)狀 況’分別如下列: (1)正常狀況。 (2) 葉片一片故障。 (3) 葉片兩片故障。 (4) 風力發電機欠相。Step 403: Read the test sample; Step 404: Step 405: Category; 5 Calculate the distance between the test sample and each category, · Find the minimum extension distance, and then judge the test sample. Step 406. Complete identification of all samples. Stop the operation, otherwise go back to step 403 to read the next test sample. As mentioned above, due to hardware installation error or blade damage, etc., the vibration will be generated when the turn (4) is turned. If the vibration signal is frequency-domain converted, the state information of the device can be analyzed by the knife; and the wind generator (10) ) It must be operated for a long time, and its gear box (22) is easy to wear between the internal gears, so that the output efficiency of the wind turbine (20) is lowered, and whether the oil level and oil temperature in the gear box (22) are normal is also judged. (2 2) One of the keys to normal operation, so the present invention uses a plurality of sets of sensors (25) to (28) to take out the lubricating oil level and oil temperature in the gear box (22) of the wind power generator (20). , the gear vibration and other characteristics nickname 'as the test sample of the aforementioned extension-like neural network (43). In addition, the generator (23) is the main component that converts mechanical energy into electrical energy, so the relative relationship between the output voltage, current, phase and the current wind speed must be taken into consideration. Therefore, the present invention 13 201120310 uses a current-to-voltage converter (31). The electrical k number is taken out at the output of the motor (23) as a test sample for diagnosis. Furthermore, this embodiment first simulates eight possible fault (operation) conditions as follows: (1) Normal condition. (2) One blade is faulty. (3) Two blades fail. (4) The wind turbine is out of phase.
(5) 齒輪箱油量不足。 (6) 齒輪箱溫度較高。 (7) 齒輪箱油溫較高。 (8) 風力發電機發生多重故障。 又處理單元(20)係利用10種特徵訊號包括:葉片轉; 、發電機轉速、發電機輸出電壓、發電機輸出電流、發$ 機:出功率、葉片軸承振幅值、齒輪箱軸承振幅值、齒李 知溫度、齒輪箱油溫和齒⑥箱機油墨力等送入可拓類神爱 2進行運算,藉以診斷風力發電機。除葉片轉速與發負 :轉逮外,其他特徵訊號均來自與故障診斷系統(1〇)連與 、各組感測器(25卜(28)及電流電麼轉換器(31 )(32)。 再者’故障診斷系統(1〇)診斷風力發電機(2〇)故障狀 :的方式係先利用可拓類神經網路判斷第"種特徵訊 ^ ’⑽其為何種轉速且轉速是否正常,㈣認前述狀態 匕:再由風力發電機(20)取得其他特徵訊號並送入可拓類 隍壯、“ μ 』㈣轉巾的風力發f機(20)其故 障狀況係屬前列的何種狀況。 14 201120310 另》月參閱第四圖所不,本發明主要係令風力發電場内 的每一風力發電機(20)分別連接一故障診斷系統⑽,為 方便進行遠端監控,各故障診斷系統(1〇)可分別連接一無 線網路傳輸單元(50),以便透過無線網路與一無線集線器 (60)連結,再由無線集線器_透過網際網路與遠端的控 制中。(70)連,结,以便故障診斷系統(1〇)將其診斷資料傳 送至遠端的控财心(7〇),以執行遠端集_遙控。 由上述可知,本發明如何在風力發電機的各個指定位 # 置上取得特徵訊號,並輸入至故障診斷系統的處理單元, 利用可拓類神經網路進行辨識,以自動出診斷風力發電機 的故障狀況,利用前述自動診斷技術具有下列優點: 1 利用故障診斷系統的預警功能,可預先瞭解風力發 機的運轉狀況,預先執行保養或維修,以降低重大事故 發生狀況。 2_透過風力發電機運轉狀態的監控,可有效減少停機 維修的時間,增加系統可靠度。 _ 3 可由故障診斷系統可得知風力發電機發生何種故障 ’並節省分析故障類型的步驟,使維修人員能快速準備備 料進行維修。 4·若進一步應用無線網路的技術,可由遠端集中監控 Μ力發電場所有風力發電機的運轉狀況。 5 ·持續的紀錄運轉狀況,建立未來故障預防的歷史資 '料庫’以作為維修及可靠度分析之重要依據。 15 201120310 【圖式簡單說明】 第一圖 第二圖 第三圖 線圓。 、今丹 , *公吗 〇 係本發明之可知 J拓類神經網路牟-立 係本發明可,、不思圖。 奴乃J拓類神經網路 J峪之可拓距離特性# 係本發明各故障診斷系統透過無線網路與每 第四圖 端控制中心連線的示意圖。 【主要元件符號說明】 (10)故障診斷系統 (12)(13)類比數位轉換 (14)顯示器 (20)風力發電機 (210)輪軸 (22)齒輪箱 (24)變壓器 (29)換能器 (33)風速檢測器 (41)輸入層 (43)可拓類神經網路 (60)無線集線器 (1 1)處理單元 器 (15)資料匯流排 (21)葉片 (211)(231)軸承 (23)發電機 (25)~(28)感測器 (31 )(32)電流電壓轉換器 (34)風向檢測器 (42)輸出層 (50)無線網路傳輸單元 (70)控制中心 16(5) The gearbox has insufficient oil. (6) The gearbox temperature is high. (7) The gearbox oil temperature is high. (8) Multiple failures in wind turbines. The processing unit (20) utilizes 10 kinds of characteristic signals including: blade rotation; generator speed, generator output voltage, generator output current, generator: output power, blade bearing amplitude value, gearbox bearing amplitude value, The tooth Lizhi temperature, the gearbox oil temperature and the tooth of the 6-box machine are sent to the extension class Shenai 2 for calculation to diagnose the wind turbine. In addition to the blade speed and the negative: the transfer, other characteristic signals are from the fault diagnosis system (1〇), each group of sensors (25 (28) and current converter (31) (32) Furthermore, the 'diagnostic system (1〇) diagnoses the fault of the wind turbine (2〇): the first method is to use the extensional neural network to judge the "characteristics of the characteristics" (10) Normally, (4) recognize the above state: the wind generator (20) obtains other characteristic signals and sends it to the extension type, and the "μ" (four) turntable wind power machine (20) is in the forefront of fault conditions. 14 201120310 Another month refers to the fourth figure. The present invention mainly relates to each wind turbine (20) in the wind farm being respectively connected to a fault diagnosis system (10), for convenient remote monitoring, each fault The diagnostic system (1) can be connected to a wireless network transmission unit (50) for connection to a wireless hub (60) via a wireless network, and then by the wireless hub _ through the Internet and remote control. 70) Connect, knot, so that the fault diagnosis system (1〇) will The diagnostic data is transmitted to the remote control center (7〇) to execute the remote set_remote. It can be seen from the above that the present invention obtains the characteristic signal on each designated position of the wind power generator and inputs it to the fault diagnosis. The processing unit of the system uses the extension type neural network for identification to automatically diagnose the fault condition of the wind turbine. The above automatic diagnosis technology has the following advantages: 1 Using the early warning function of the fault diagnosis system, the wind generator can be known in advance. The operation status, pre-executed maintenance or repair, to reduce the occurrence of major accidents. 2_ Through the monitoring of the operation status of the wind turbine, it can effectively reduce the downtime and increase the reliability of the system. _ 3 can be known by the fault diagnosis system What kind of failure occurs in the wind turbine' and saves the steps of analyzing the type of fault, so that the maintenance personnel can quickly prepare the materials for maintenance. 4. If the technology of the wireless network is further applied, the wind turbine can be centrally monitored by all the wind turbines. Operation status. 5 · Continuous record operation status, establish the history of future failure prevention The 'repository' is used as an important basis for maintenance and reliability analysis. 15 201120310 [Simple description of the diagram] The first diagram, the second diagram, the third diagram, the circle of the circle, the present, the public, the public, the knowledge of the invention J The invention is applicable to, and does not reflect on. The extension distance characteristic of the slave neural network J峪 is the fault diagnosis system of the present invention through the wireless network and every fourth figure. Schematic diagram of the connection of the terminal control center [Description of main component symbols] (10) Fault diagnosis system (12) (13) Analog digital conversion (14) Display (20) Wind turbine (210) Axle (22) Gear box (24 Transformer (29) transducer (33) wind speed detector (41) input layer (43) extension type neural network (60) wireless hub (1 1) processing unit (15) data bus (21) blade (211) (231) Bearing (23) Generator (25) ~ (28) Sensor (31) (32) Current-to-Voltage Converter (34) Wind Direction Detector (42) Output Layer (50) Wireless Network Transmission Unit (70) Control Center 16

Claims (1)

  1. 201120310 七、申請專利範圍: 1 ·—種風力發電場狀態遙測技術及故障診斷系統,包 括: 不旻数感測器,分設於風力發電機内各個指定的位置上 進行訊號收集,以取得各指定檢測位置的特徵訊號; 一處理單元,係分別各感測器連接,又處理單元内建 一可拓類神經網路及多種模擬故障狀況;其中,該可拓類 神經網路包括一輸入層及一輸出層,其中輸入層係將各特 徵訊號進行分類並建構成物元模型,再送入可拓類神經網 路’以運算出各類特徵訊號的可拓距離最小值,進而由輸 出層輸出判斷的故障狀況。 2.如申請專利範圍第1項所述之風力發電場狀態遙 測技術及故障診斷系統,該風力發電機包括一葉片、一齒 輪箱、一發電機及一變壓器;該葉片具有一輪軸,用以與 Ud輪箱連、’、。,而帶動齒輪箱内的齒輪組,又齒輪箱的輸出 端係與發電機的轉軸連結; 各感測器係分別設於前述葉片之輪轴袖承上、發電機 轉轴上及齒輪箱上’用以分別檢㈣片轴承振幅值、發電 機抽承振幅值、齒輪箱溫度及齒輪箱油溫等。 3專利&圍帛2項所述之風力發電場狀態遙 測技術及故障診斷L & & ,, _ A 類士MM你 診斷系統進一步包括複數的 類比數位轉換器、一鞀 理單元係透過資料匯::一貝料匯流排;…該處 及風力發電機上久 與類比數位轉換器、顯示器 轉換器分別與兩電攻蕾广^接,並進—步透過類比數位 "·<堅轉換器及一風速檢測器、一風向 17 201120310 檢測器連接,兩電流電壓轉換器係分別與發電機及變壓器 的輸出端連接,以取得發電機'變壓器的輸出電壓、電流 與風速'風向訊號作為特徵訊號;另透過各換能器與風力 發電機上的各組感測器連接β 4_如申請專利範圍第3項所述之風力發電場狀態遙 測技術及故障診斷系統,可拓類神經網路採取監督式學習 ’其演算步驟包括: 步驟301 ··建立輸入與輸出之權重值·· • 步驟302 :讀取訓練樣本資料與特徵數k ; 步驟303 :計算出每項特徵之權重中心值; 步驟304:利用可拓類神經網路之輸入端學習資料開 始計算可拓距離; 步驟305 .找k*,如果k*= k,並跳到步驟3〇7 ;若 k*孕k,則繼續下一步驟3〇6 ; 步驟306:調整k類別與k*類别之權重值;包括更新 權重上、下限值大小及權重中心值大小; ,步驟307:重複步驟3〇3至步驟3〇7之步驟,直到所 有學習資料皆讀取並完成學習分類完畢; 步驟308··當所有資料之分類程序都已達到收斂狀態 或總誤差率到達到目標值則停止,若否則返回步驟綱繼 續。 ,如申明專利知圍第4項所述之風力發電場狀態遙 測技術及故障珍斷系統,可拓類神經網路進行辨識驟 包括: 步驟401 :讀取可拓類神經網路的權重值矩陣’· 18 201120310 步驟402 ··計算中間值大小; 步驟4 0 3 :讀取測試樣本; 步驟404 ,計异測試樣本與各類別之可拓距離; 步驟405:尋找最小可抬距離,藉以判斷測試樣 屬類別; 步驟406:完成辨識所有樣本即停止運算, 驟403讀取下一筆測試樣本。 11少 6.如申請專利範圍第]至5項中任_項所述 發電場狀態遙測技術及故障診斷系統,該故障 力 —步設有—無㈣路傳輪單元,錢過無線^與^统進 的控制中心連結。 、遠端 八、圖式:(如次頁)201120310 VII. Scope of application for patents: 1 · Wind power field state telemetry technology and fault diagnosis system, including: 旻 感 感 , , , , , , , , , , , , , , 风力 风力 风力 风力 风力 风力 风力 风力 风力 风力 风力 风力 风力 风力Detecting the characteristic signal of the position; a processing unit is connected to each sensor, and the processing unit has an extension type neural network and a plurality of simulated fault conditions; wherein the extension type neural network includes an input layer and An output layer, wherein the input layer classifies each feature signal and constructs a matter-element model, and then sends it to the extension-type neural network to calculate the extension distance minimum of each feature signal, and then is judged by the output layer output. The fault condition. 2. The wind farm state telemetry technology and fault diagnosis system according to claim 1, wherein the wind power generator comprises a blade, a gear box, a generator and a transformer; the blade has an axle for Connected with the Ud wheel box, ',. And driving the gear set in the gear box, and the output end of the gear box is coupled with the rotating shaft of the generator; each of the sensors is respectively disposed on the axle sleeve of the blade, the generator shaft and the gear box 'Use to separately check the (four) bearing amplitude value, generator pumping amplitude value, gearbox temperature and gearbox oil temperature. 3 patent & cofferdam 2 wind farm state telemetry technology and fault diagnosis L &&, _ A class MM your diagnostic system further includes a plurality of analog digital converters, a processing unit through Data sink:: a billiard busbar; ... the wind turbine and the wind generator for a long time and analog digital converter, display converter and two electric attack buds are connected, and advance through the analog digital "·< The converter and a wind speed detector, a wind direction 17 201120310 detector connection, two current and voltage converters are respectively connected with the output of the generator and the transformer to obtain the generator 'transformer output voltage, current and wind speed' wind direction signal The characteristic signal is connected to each group of sensors on the wind power generator through each transducer. 4) The wind power field state telemetry technology and fault diagnosis system described in claim 3, the extension type neural network The road adopts supervised learning. The calculation steps include: Step 301 · Establish weight values for input and output. · Step 302: Read training sample data and feature number k; Step 303: Calculate the weight center value of each feature; Step 304: Start calculating the extension distance by using the input learning data of the extension type neural network; Step 305. Find k*, if k*=k, and jump to Step 3〇7; if k* pregnancy k, proceed to the next step 3〇6; Step 306: Adjust the weight values of the k category and the k* category; including the upper and lower limit values of the update weight and the weight center value; Step 307: Repeat steps 3〇3 to 3〇7 until all the learning materials are read and the learning classification is completed. Step 308··When all the data classification procedures have reached the convergence state or the total error rate arrives Stop at the target value, otherwise return to the step to continue. For example, the wind power field state telemetry technology and the fault jerk system described in claim 4, the extension type neural network for identifying the steps includes: Step 401: reading the weight value matrix of the extension type neural network '· 18 201120310 Step 402 · Calculate the intermediate value size; Step 4 0 3: Read the test sample; Step 404, Calculate the extension distance of the test sample and each category; Step 405: Find the minimum liftable distance, to judge the test The genus category; step 406: complete the identification of all samples to stop the operation, and 403 to read the next test sample. 11 less 6. If the power generation field state telemetry technology and fault diagnosis system described in any of the items in the patent scopes 1-5 to 5, the fault force-step is provided - no (four) road transmission unit, money over the wireless ^ and ^ Connected to the control center. , remote, eight, schema: (such as the next page)
TW098142015A 2009-12-09 2009-12-09 The state telemetry technology and fault diagnosing system in large-scale wind power farms TW201120310A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW098142015A TW201120310A (en) 2009-12-09 2009-12-09 The state telemetry technology and fault diagnosing system in large-scale wind power farms

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW098142015A TW201120310A (en) 2009-12-09 2009-12-09 The state telemetry technology and fault diagnosing system in large-scale wind power farms

Publications (1)

Publication Number Publication Date
TW201120310A true TW201120310A (en) 2011-06-16

Family

ID=45045090

Family Applications (1)

Application Number Title Priority Date Filing Date
TW098142015A TW201120310A (en) 2009-12-09 2009-12-09 The state telemetry technology and fault diagnosing system in large-scale wind power farms

Country Status (1)

Country Link
TW (1) TW201120310A (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103176128A (en) * 2013-03-28 2013-06-26 华南理工大学 Method and system for forcasting state of wind generating set and diagnosing intelligent fault
TWI481780B (en) * 2011-08-02 2015-04-21 Univ Nat Sun Yat Sen An adjustable-pitch control of wind power system and method thereof
TWI648466B (en) * 2016-03-04 2019-01-21 日商日立製作所股份有限公司 Control device for a plurality of wind power generation devices, control method for a wind power plant or a plurality of wind power generation devices
TWI657404B (en) * 2016-12-23 2019-04-21 財團法人船舶暨海洋產業研發中心 Offshore wind farm management system and method thereof
CN109973331A (en) * 2019-05-05 2019-07-05 内蒙古工业大学 A kind of fan blade of wind generating set fault diagnosis algorithm based on bp neural network
TWI728535B (en) * 2019-10-31 2021-05-21 國立勤益科技大學 Monitor system and method thereof
TWI731502B (en) * 2019-12-09 2021-06-21 國立臺灣科技大學 Intelligent detection method and system for power equipment failures
TWI732660B (en) * 2020-08-20 2021-07-01 國立勤益科技大學 Wind power generator fault diagnosis system and method

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI481780B (en) * 2011-08-02 2015-04-21 Univ Nat Sun Yat Sen An adjustable-pitch control of wind power system and method thereof
CN103176128A (en) * 2013-03-28 2013-06-26 华南理工大学 Method and system for forcasting state of wind generating set and diagnosing intelligent fault
TWI648466B (en) * 2016-03-04 2019-01-21 日商日立製作所股份有限公司 Control device for a plurality of wind power generation devices, control method for a wind power plant or a plurality of wind power generation devices
TWI657404B (en) * 2016-12-23 2019-04-21 財團法人船舶暨海洋產業研發中心 Offshore wind farm management system and method thereof
CN109973331A (en) * 2019-05-05 2019-07-05 内蒙古工业大学 A kind of fan blade of wind generating set fault diagnosis algorithm based on bp neural network
TWI728535B (en) * 2019-10-31 2021-05-21 國立勤益科技大學 Monitor system and method thereof
TWI731502B (en) * 2019-12-09 2021-06-21 國立臺灣科技大學 Intelligent detection method and system for power equipment failures
TWI732660B (en) * 2020-08-20 2021-07-01 國立勤益科技大學 Wind power generator fault diagnosis system and method

Similar Documents

Publication Publication Date Title
TW201120310A (en) The state telemetry technology and fault diagnosing system in large-scale wind power farms
CN202710100U (en) Wind generating set remote on-line health state monitoring and fault diagnosing system
CN202735788U (en) Wind generating set running state remote-monitoring and fault diagnosis system
CN102620807A (en) System and method for monitoring state of wind generator
CN103343728A (en) Wind generating set remote on-line multi-mode health state monitoring and fault diagnosis system
CN202041178U (en) On-line monitoring system for power transmission line sag
CN105787584A (en) Wind turbine malfunction early warning method based on cloud platform
CN101858778A (en) Vibration monitoring-based wind generator set automatic fault diagnosis method
CN102434388A (en) Health status online monitoring device of wind generating set and monitoring method of monitoring device
CN102156043A (en) Online state monitoring and fault diagnosis system of wind generator set
CN110469462B (en) Wind turbine generator system intelligent state monitoring system based on multiple templates
CN103835882A (en) Large wind generating set state monitoring and fault diagnosis system
CN105337575B (en) Photovoltaic plant status predication and method for diagnosing faults and system
CN101825893A (en) Centralized and remote control monitoring, and fault diagnosis system of wind turbine
CN103940611B (en) Rolling bearing self adaptation method for detecting abnormality under a kind of Wind turbines variable working condition
CN104564542B (en) A kind of fault diagnosis system and its diagnostic method based on big data technology
CN206056648U (en) A kind of Wind turbines dynamo bearing condition monitoring system
CN104767482A (en) Method for diagnosing aging and short circuit of photovoltaic module online
CN105205569A (en) Draught fan gear box state on-line evaluation model building method and on-line evaluation method
CN103759891A (en) Blade imbalance on-line fault diagnosis method of doubly-fed wind generator
CN205260225U (en) Long -range central monitoring system of health condition is synthesized to fan
CN202453182U (en) Fault diagnosis device of gearbox of wind generation set
Smarsly et al. Structural health monitoring of wind turbines observed by autonomous software components–2nd level monitoring
CN110905736A (en) Fan operation monitoring system
CN202994473U (en) Wind power generation unit vibration on-line monitoring and fault diagnosis system