200914846 九、發明說明： 【發明所屬之技術領域】 本發明是有關於一種遞升式小波轉換，特別是以遞 升式小波轉換應用於電力品質事故之監測與辨識之方 法。 【先前技術】 隨著傳統產業朝向高科技及技術密集方向之發展， 製造設備採用高精密儀器為數可觀，惜因儀器設備對訊 號品質之變化曰趨靈敏，局部的異常訊號問題常易引發 大規模的設備誤動作或故障，致使產業競爭力蒙受衝 擊，抽失難以估計，故近來各類產業從業人員對於提升 訊號品質之要求，莫不深切期盼。著眼於此，對於異常 干擾訊號分析之能力實應予以強化，以俾提供遏止訊號 污柒之參考，並加強訊號品質之提升，進而增加儀器分 析精確度暨滿足客戶端之處理要求。然在訊號分析的過 程中，訊號的背景雜訊(N〇ise)造成測試結果誤判的機率 大幅增加。有鑑於此，為協助提升訊號分析準確率，發 展有效滤除雜訊之演算技巧確有其需要性與返切性。 過去諸多文獻業已提出若干訊號品質處理方案。其 中’由於類神經網路(Neural network)能夠利用學習範例 予以建構複雜的非線性系統模型，故此演算法常被廣泛 應用於訊號處理之範疇，以協助提供良好的訊號檢測模 式。惟該方法僅能運用至處理特定訊號，對於尚未訓練 的資料型態，其鑑別結果恐有誤判之虞。另對於訊號之 分析與監測而言，亦有文獻應用傅立葉轉換（F〇urier 200914846 transform)進行處理，惟當訊號屬非平穩狀態 (N〇n_Stationarystate)時，傅立葉轉換仍無法有效呈現訊 號相關特徵’應用^值未螓完善。此外’有馨於訊號常 具時變性’除須掌握其時間訊息外，尚須獲知其頻率訊 息，俾利協助執行相關應用。儘管短時傅立葉轉換 (Short-term F0urier transf〇rm)能夠以時頻域的方式表達 訊號特徵，惟待選定分析視SA小後，所求得之時-頻域 解析度仍十分有限’故尚存改善增強之處。近來，小波 轉換（Wavelet transform)技術被視為訊號處理最佳利器 之一。根據訊號之高低頻成分，小波函數分析視窗可作 適當的Μ縮或展開，進而在低頻部分能夠獲得高的頻率 解析度’、另在高頻部分則可獲得高的時間解析度，其時 位能力’更有助於精確觀測職暫態行為及 =握動U性，就訊號處理架構的完備性而言，相對於 ，、他方法實可提供較為理想之協助。 =非線性元件的廣泛使用，大量雜訊侵入待測訊 號，致使小波轉換在訊號處理上失去原有的準確性。所 非線性兀件上的雜訊而能重見原來的訊號， 來對訊號進行監測及分析亦是重要。 【發明内容】 =發明的目的在於提出一種小波轉換應用於電 一方法，對於電壓訊號此類的電子訊號遂以 六、异法之小波轉換對此訊號進行分解、上提 式、及正規化運算，絲 之後再以一逆遞升式演算法重建 原始的電子矾號，此舉得 羋j侍1乂佳的電子訊號品質訊號。 200914846 、電壓突降、電壓突升、 雜訊後藉由上述遞升式小 等步驟’而應用於雜訊干 於多個實施例中，以正常電壓 電壓中斷及瞬時脈衝分別加入 波轉換進行運算及其訊號重建 擾之濾除與偵測。 雷上述之目的’本發明提供—種小波轉換應用於 =子《之_方法，此_方法包含步驟⑷將一第一 電：訊號轉換為—具標么（per她)基準值之電子訊號， 之《步驟(b)使用一二階式解析小波換之 將^票么基準值之電子訊號解析為—高頻訊號及一= —’再來’步驟⑷藉由—臨界值來決定是否除去高頻 訊，，其中當尚頻訊號小於或等於此臨界值時，則將高 頻訊號設為零’而當高頻訊號大於此臨界值時，則保持 高頻訊號，並執行步驟（d)。在步驟中⑷中使用—逆二階 式解析小波轉換之運算步驟而對具標么基準值之電 進重建’以產生-第二電子訊號。最後，步驟⑷於檢 測，一電子訊號過程中，提供一判斷運算式，用以判斷 於第一電子訊號是否遭雜訊污染。 【實施方式】 β以下詳細地討論目前較佳的實施例。然而應被理解 的是，本發明提供許多可適用的發明觀念，而這些觀今 能被體現於很寬廣多樣的特定具體背景中。所討論的二 定具體的實施例僅是說明使用本發明的特定結構，而且 不會限制本發明的範圍。 關於小波轉換在信號雜訊消除的研究方面，目前最 主要的方法為Donoho所提出的小波係數臨界法。此方 200914846 法疋依據一般雜訊經由小波轉換之後，代表雜訊成分的 小波係數通常具有較細微的尺度，也就是說較小的小波 係數是由雜訊所產生。因此只要將這些較小的小波係數 遽除’即可達到過濾雜訊的目的。而設定一臨界值 (Threshold)亦是符合小波係數臨界法，如果小波係數低 於此臨界值，則將這些小波係數設為零。這些被變為零 的小波係數大部份是相對應於雜訊，而輸入信號的的係 數值通常高於此臨界值。臨界值的選取主要分為硬式臨 界值和軟式臨界值兩種，可依信號特性及精確度的要求 加以選擇適當的篩選方式。 故，本實施方式’以設定一臨界值，並以軟式臨界 值作為選取的一種方式。以下為本實例之驗證說明，如 第1圖所示，此圖為本發明之小波轉換應用於電子訊號 之偵測方法之流程圖，圖中，此方法包含下列步驟： 步驟11 :針對電壓、電流、功率、阻抗、頻率或 轉數等相關量之一電子訊號為簡化計算（simplify the computation)、易查出結果（Easier t〇 check the result)等 實鉍步驟遂以轉換一標么（Per-unit)系統，換言之，將電 ，、電流、功率、阻抗、頻率或轉數等相關量，個別選 定其基準值，並將各量的數值除以所對應的基準值產 ，各量的標么值。而在本實施例中，為區別多種電子訊 號之訊號名稱，特以將作為電壓信號之一第一電子訊號 轉換為一具標么基準值之電子訊號。 ^ v驟12 ·使用一二階式解析小波轉換之運算步驟 而將具標么基準值之電子訊號解析為一高頻訊號及一低 頻讯唬。於步驟b)不同於習知所採用的小波轉換之摺積 200914846 運算，於此本實施例中，小波轉換係採用二階遞升式演 算法之小波轉換，請參閱第2圖所示，此圖是遞升式演 算法之小波轉換之架構圖。搭配此圖，係說明此運算步 驟流程。 首先，將具標么基準值之電子訊號x[n]進行訊號分 解（splitting)’以產生一第一奇數序列之訊號χ〇[η]及一 第一偶數序列之訊號xe[n]; 接者，將第一奇數序列之訊號xo[n]及第一偶數序列 之訊號xe[n]分別代入一第一運算式及一第二運算式，以 產生一第一高頻成份訊號dj[n]&一第一低頻成份訊號 Sj[n]; 隶後’使用一第一正規化（normalization)參數KH及 一第二正規化參數KL對分別第一高頻成份訊號dj[n]& 第一低頻成伤§fL號Sj [η]進行相乘，以產生一高頻訊號 d[n]及一低頻訊號S[n],如式（1)及式（2)所示： d[n]=KH dj[n]...式（1) S[n] =KL Sj[n]”·式（2) 請注意，前述第一運算式及第二運算式為以下方程 式所表示’第一運算式為： dj[n]=dj-l[n]-Pj(Sj-l[n])，（j =1，2，.·.，η,且 η 是正整數） 其中藉由預測函數Pj預測dj[n]與dj-1 [η]間的差值 Pj(Sj-l[n])，再將dj-l[n]減去pj(Sj-l[n])而求得高頻訊號 dj[n] ’而S0[n]係為第一偶數序列之訊號xe[n] 〇 第二運算式為： 以[11]=8』-1[11]+11』((^[11])’（卜1，2，...,11，且11是正整數） 其中藉由更新函數Uj與dj[n]關係以補償Sj-l[n]， 200914846 以求得低頻訊號Sj[n]， d 0[n]為第一奇數序列之訊號 xo[n] ° 而在於步驟b)中第一運算式及第二運算式之運算過 程稱之上提式運算（lifting)。 步驟13 : 判斷高頻訊號是否小於或等於臨界值 λ，若是，則執行步驟14,若否，則執行步驟15。其中， 臨界值；I為： λ = σ」2\η(Ν) , σ = 其中，σ為雜訊之標準差，Ν為資料點個數， MAD(d[n])則代表d[n]之平均絕對離差（Mean absolute deviation)。 步驟14 : 將高頻訊號設為零，並執行步驟16。 步驟15 :保持高頻訊號，並執行步驟16。 步驟16 :使用一逆二階式解析小波轉換之運算步驟 而對具標么基準值之電子訊進行重建，以產生一第二電 子訊號，請參閱第3圖所示，此圖是遞升式演算法之小 波反轉換之架構圖。搭配此圖，係說明此運算步驟流程。 首先，使用前述第一正規化參數KH及第二正規化 參數K L而對高頻訊號d[n]及低頻訊號S[n]進行相除， 以產生一第二高頻成份訊號dj[n]及一第二低頻成份訊 號Sj [η]，如式（3)及式（4)所示： dj[n]=d[n]/KH...式（3)BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a step-up wavelet transform, and more particularly to a method for monitoring and identifying power quality accidents using a step-up wavelet transform. [Prior Art] With the development of traditional industries towards high-tech and technology-intensive directions, the use of high-precision instruments for manufacturing equipment is considerable. Unfortunately, due to the changes in signal quality of instruments and equipment, local abnormal signal problems often lead to large-scale problems. The malfunction or malfunction of the equipment has caused the industrial competitiveness to be affected, and the loss is difficult to estimate. Therefore, various industrial practitioners have recently been eagerly awaiting the improvement of signal quality. With this in mind, the ability to analyze anomalous interference signals should be enhanced to provide a reference for suppressing signal contamination and to enhance signal quality, thereby increasing instrumentation accuracy and meeting client processing requirements. However, during the signal analysis process, the background noise of the signal (N〇ise) caused a sharp increase in the probability of misjudging the test result. In view of this, in order to help improve the accuracy of signal analysis, the development of effective algorithms for filtering noise has its necessity and reciprocity. In the past, many literatures have proposed several signal quality processing solutions. Among them, because the neural network can use the learning paradigm to construct complex nonlinear system models, the algorithm is often widely used in the field of signal processing to help provide a good signal detection mode. However, this method can only be applied to the processing of specific signals. For the untrained data types, the identification result may be misjudged. In addition, for the analysis and monitoring of signals, the literature also applies Fourier transform (F〇urier 200914846 transform) for processing. However, when the signal is non-stationary state (N〇n_Stationarystate), the Fourier transform still cannot effectively display the signal-related features' The application value is not perfect. In addition, when there is a time when the signal is often changed, it is necessary to know the frequency information to assist in the implementation of the application. Although the Short-term F0urier transf〇rm can express the signal features in the time-frequency domain, the time-frequency domain resolution obtained after the selected analysis is small is still very limited. Improve the enhancements. Recently, Wavelet transform technology has been regarded as one of the best tools for signal processing. According to the high and low frequency components of the signal, the wavelet function analysis window can be appropriately collapsed or expanded, so that high frequency resolution can be obtained in the low frequency part, and high time resolution can be obtained in the high frequency part. The ability 'is more conducive to accurate observation of the transient behavior and = grip U, in terms of the completeness of the signal processing architecture, in contrast, his method can provide more ideal assistance. = Wide use of nonlinear components, a large amount of noise invades the signal to be tested, causing the wavelet conversion to lose its original accuracy in signal processing. The noise on the nonlinear components can revisit the original signal, and it is also important to monitor and analyze the signal. SUMMARY OF THE INVENTION The purpose of the invention is to provide a method for applying wavelet transform to an electric one. For an electronic signal such as a voltage signal, the signal is decomposed, uplifted, and normalized by a wavelet transform of six different methods. After the wire, the original electronic nickname is reconstructed by a reverse-step-up algorithm, which is a good signal quality signal. 200914846, voltage dip, voltage surge, and noise are applied to the noise in a plurality of steps by the steps of the above-mentioned step-up method, and the wave voltage conversion is performed by the normal voltage voltage interruption and the instantaneous pulse respectively. Its signal reconstruction is filtered and detected. The purpose of the above is 'the present invention provides a wavelet transform applied to the = sub-method, the method includes the step (4) converting a first electric: signal into an electronic signal having a reference value of (per). "Step (b) uses a second-order analytical wavelet to convert the electronic signal of the reference value into a high-frequency signal and a = 're-take' step (4) to determine whether to remove the high by the critical value Frequency, wherein when the frequency signal is less than or equal to the threshold, the high frequency signal is set to zero' and when the high frequency signal is greater than the threshold, the high frequency signal is maintained, and step (d) is performed. In step (4), the inverse second-order analytical wavelet transform operation step is used to reconstruct the electrical input of the reference value to generate a second electronic signal. Finally, in step (4), during the detection of an electronic signal, a judgment expression is provided for determining whether the first electronic signal is contaminated by noise. [Embodiment] β The presently preferred embodiment will be discussed in detail below. It should be understood, however, that the present invention provides many applicable inventive concepts which can be embodied in a wide variety of specific specific contexts. The specific embodiments discussed are merely illustrative of specific constructions of the invention and are not intended to limit the scope of the invention. Regarding the research of wavelet transform in signal noise cancellation, the most important method at present is the wavelet coefficient critical method proposed by Donoho. This method 200914846 After the general noise is converted by wavelet, the wavelet coefficients representing the noise components usually have a finer scale, that is, the smaller wavelet coefficients are generated by the noise. Therefore, as long as these smaller wavelet coefficients are removed, the purpose of filtering noise can be achieved. Setting a threshold (Threshold) is also in accordance with the wavelet coefficient critical method. If the wavelet coefficient is lower than the threshold, the wavelet coefficients are set to zero. Most of these wavelet coefficients that are zeroed are corresponding to noise, and the value of the input signal is usually higher than this threshold. The selection of the critical value is mainly divided into two types: hard critical value and soft critical value. The appropriate screening method can be selected according to the requirements of signal characteristics and accuracy. Therefore, the present embodiment 'sets a critical value and uses a soft critical value as a means of selection. The following is a verification description of the present example. As shown in FIG. 1, this figure is a flowchart of a method for detecting wavelet signal applied to an electronic signal according to the present invention. In the figure, the method includes the following steps: Step 11: For voltage, One of the related quantities of current, power, impedance, frequency, or number of revolutions is an actual step such as simplifying the computation, Easier t〇check the result, etc. -unit) system, in other words, the relevant values of electricity, current, power, impedance, frequency or number of revolutions are individually selected, and the value of each quantity is divided by the corresponding reference value. What value. In this embodiment, in order to distinguish the signal names of the plurality of electronic signals, the first electronic signal, which is one of the voltage signals, is converted into an electronic signal having a reference value. ^ vStep 12: The electronic signal with the reference value is parsed into a high frequency signal and a low frequency signal using a second-order analytical wavelet transform operation step. In step b), unlike the wavelet transform 200914846 which is conventionally used, in the present embodiment, the wavelet transform adopts a wavelet transform of a second-order step-up algorithm, as shown in FIG. 2, which is The architecture diagram of the wavelet transform of the step-up algorithm. With this figure, the flow of this operation step is explained. First, the electronic signal x[n] with the reference value is subjected to signal splitting to generate a first odd sequence signal χ〇[η] and a first even sequence signal xe[n]; The signal xo[n] of the first odd sequence and the signal xe[n] of the first even sequence are respectively substituted into a first arithmetic expression and a second operational expression to generate a first high frequency component signal dj[n ]& a first low frequency component signal Sj[n]; followed by a first normalization parameter KH and a second normalization parameter KL pair first high frequency component signal dj[n]& The first low frequency is sfL Sj [η] is multiplied to generate a high frequency signal d[n] and a low frequency signal S[n], as shown in equations (1) and (2): d[ n]=KH dj[n]...Formula (1) S[n] =KL Sj[n]"·(2) Note that the first arithmetic expression and the second arithmetic expression are represented by the following equations' The first expression is: dj[n]=dj-l[n]-Pj(Sj-l[n]), (j =1,2,.., η, and η is a positive integer) The function Pj predicts the difference Pj(Sj-l[n]) between dj[n] and dj-1 [η], and then subtracts dj-l[n] from pj(Sj-l[n]) The frequency signal dj[n] ' and S0[n] is the signal of the first even sequence xe[n] 〇 The second expression is: [11]=8』-1[11]+11』((^[ 11]) '(Bu 1, 2, ..., 11, and 11 is a positive integer) where the Sj-l[n], 200914846 is compensated by updating the function Uj and dj[n] to obtain the low frequency signal Sj[ n], d 0[n] is the signal xo[n] ° of the first odd sequence and the operation of the first and second expressions in step b) is called lifting. Step 13 : determining whether the high frequency signal is less than or equal to the threshold value λ, if yes, executing step 14, if not, executing step 15. wherein, the critical value; I is: λ = σ"2\η(Ν), σ = σ is the standard deviation of the noise, Ν is the number of data points, and MAD(d[n]) represents the mean absolute deviation of d[n]. Step 14: Set the high frequency signal to zero and perform step 16. Step 15: Maintain the high frequency signal and perform step 16. Step 16: reconstructing the electronic signal with the reference value using an inverse second-order analytical wavelet transform operation step to generate a second electronic signal, as shown in FIG. 3, which is a step-up algorithm. The architecture of the wavelet inverse conversion. With this figure, the flow of this operation step is explained. First, the high frequency signal d[n] and the low frequency signal S[n] are divided by using the first normalization parameter KH and the second normalization parameter KL to generate a second high frequency component signal dj[n] And a second low frequency component signal Sj [η], as shown in equations (3) and (4): dj[n]=d[n]/KH... equation (3)
Sj[n]=S[n]/KL...式（4) 接者，將第二高頻成份訊號dj[n]及第二低頻成份訊 200914846 號Sj[n]分別代入一第三運算式及一第四運算式以產生 一第二奇數序列之訊號xo[n]及一第二偶數序列之訊號 xe[n]； 最後，將第二奇數序列之訊號x0[n]及第二偶數序列 之訊號xe[n]進行合成以產生一第二電子訊號χ[η]。 請注意’前述第三運算式及第四運算式為以下方程 式所表示，第三運算式為： dj[n]- dj-l[n]+卩购^㈤），（』=12，几且η是正整數）。 第四運算式為： S刺=叫叫叫咖])’（j =12，η，且^是正整數）。 ^步驟17··於檢測第一電子訊號過程中，提供一判斷 運算式，以判斷於步驟d)中所重建的第一電子訊號是否 遭雜訊污染。 此一判斷運算式係為作為重建訊號之第二電子訊號 與輸入訊號之第一電子訊號之誤差平方值。此時，若第 一電子訊號遭受外來雜訊污染，此誤差平方值將遠大於 零。 、 經由上述技術内文可知，本專利所提演算法應用於 探討數種訊號，包含正常電壓波形、電壓突降（Sa幻、 壓突升（Swell)、電壓中斷（〇utage)及瞬時脈衝 dmpulse)。於模擬測試過程令，並將上述5種訊號加入 雜訊污染’爾後再續評估本文方法是否可祕雜訊，進 而重建原來的訊號波形。其中訊號頻率為6〇 Hz，取 頻率則為15.36他。另文中亦利用訊雜比 (Signal-t0-n〇ise ratio, SNR)，即待測訊號平均能量對雜 200914846 訊平均能量的比值’予以進行濾波效能評估，如不所示.Sj[n]=S[n]/KL... (4), the second high frequency component signal dj[n] and the second low frequency component signal 200914846 Sj[n] are respectively substituted into a third operation And a fourth arithmetic expression to generate a second odd sequence signal xo[n] and a second even sequence signal xe[n]; finally, the second odd sequence signal x0[n] and the second even number The sequence signal xe[n] is synthesized to generate a second electronic signal χ[η]. Please note that the third arithmetic expression and the fourth operational expression are represented by the following equations: the third operational expression is: dj[n]- dj-l[n]+卩购^(五)), (』=12, several η is a positive integer). The fourth expression is: S thorn = called called coffee])' (j = 12, η, and ^ is a positive integer). ^Step 17·· During the detection of the first electronic signal, a judgment expression is provided to determine whether the first electronic signal reconstructed in step d) is contaminated by noise. The judgment expression is the squared error of the first electronic signal as the reconstructed signal and the first electronic signal of the input signal. At this time, if the first electronic signal is contaminated by external noise, the squared error will be much larger than zero. According to the above technical knowledge, the algorithm proposed in this patent is applied to explore several kinds of signals, including normal voltage waveform, voltage dip (Sa illusion, Swell, voltage interrupt (〇utage) and transient pulse dmpulse ). In the simulation test procedure, and adding the above five kinds of signals to the noise pollution, we will continue to evaluate whether the method can be used to reconstruct the original signal waveform. The signal frequency is 6 〇 Hz and the frequency is 15.36. In addition, the signal-to-noise ratio (SNR), which is the ratio of the average energy of the signal to be measured to the average energy of the 200914846 signal, is also used for filtering performance evaluation, if not shown.
SNR = 10I〇gf2^]dB §fl雜比係為待測第一電子訊號（vs)之平均能量對雜 訊信號（ve)之平均能量的比值’予以進作濾波效能呼 估’且第一電子訊號（vs)之均能量為νβ(%])2，及雜訊 信號（ve)之平均能量為”括㈣，而坰代表第一電子訊 號、e[n]代表雜訊信號，及n為正整數。 測試1:正常電壓波形加入雜訊 本測试係將正常電壓波形加入雜訊後，再予以模擬 評估所提方法之可行性。第4A圖所示為輸入訊號。透 過新型小波轉換之解析與還原後，第4B圖顯示其訊號 重建結果。由圖中可知本文方法確具有濾除雜訊之效 果。其中’訊號未濾波前SNR值為3〇·3 dB，然於濾波 後，SNR值則提高至63.74 dB，換言之濾波後訊號品質 已愈趨完美。另第4C圖所示為重建訊號與輸入訊號之 誤差平方值，此結果使得雜訊污染更易於在時域被觀測 到。 測試2 :電壓突降加入雜訊 電壓突降實屬現今電力訊號品質問題中最常發生現 象之一，通常肇因於大型非線性負載之啟動、故障電流 的干擾或供電網路短暫解聯等狀況。根據IEEE Std. 1159-1995之規範，電壓值下降至系統額定電壓之1〇0/〇 至90%間，且至少持續0.5個週期甚或至i分鐘，均可 視為電壓突降。 12 200914846 於本測試中，電壓遞減至系統額定電壓之6〇%，且 持續3個週波。第5A圖所示為電壓突降加入雜訊之波SNR = 10I〇gf2^]dB §fl is the ratio of the average energy of the first electronic signal (vs) to the average energy of the noise signal (ve), which is the filter performance estimate and the first The average energy of the electronic signal (vs) is νβ(%))2, and the average energy of the noise signal (ve) is "four (4), and 坰 represents the first electronic signal, e[n] represents the noise signal, and n It is a positive integer Test 1: Normal voltage waveform is added to the noise This test adds the normal voltage waveform to the noise and then simulates the feasibility of the proposed method. Figure 4A shows the input signal. After analysis and restoration, Figure 4B shows the signal reconstruction result. It can be seen from the figure that the method has the effect of filtering out noise. The 'SNR before the signal is unfiltered is 3〇·3 dB, but after filtering, The SNR value is increased to 63.74 dB, in other words, the filtered signal quality is more perfect. The other 4C shows the squared error of the reconstructed signal and the input signal, which makes the noise pollution easier to observe in the time domain. Test 2: Voltage dip is added to the noise voltage drop One of the most common phenomena in today's power signal quality problems is usually caused by the start of large non-linear loads, the interference of fault currents or the transient disconnection of the power supply network. According to the IEEE Std. 1159-1995 specification, the voltage value It can be regarded as a voltage dip when it falls to 1〇0/〇 to 90% of the rated voltage of the system and lasts for at least 0.5 cycles or even i minutes. 12 200914846 In this test, the voltage is decremented to 6系统 of the rated voltage of the system. %, and lasts for 3 cycles. Figure 5A shows the voltage drop added to the noise wave
形，其SNR值為27.9 dB。透過本文方法之協助，第5B 圖顯示其濾波結果，此圖說明所提方法確具還原電壓突 降波形之能力，且其SNR值並提高至60.89 dB。此時， 計算重建訊號與輸入訊號之誤差平方值即可得第5匸圖 之監測波形，由此圖可知雜訊的出現與結束確均符合測 試2之模擬狀況。 測試3 :電壓突升加入雜訊 與電壓突降相較下’當電壓上升至高於系統額定電 壓10%以上’且此情況持續〇·5個週期或至丄分鐘之間， 即稱為電壓突升。當鄰近大型負翁除時，輸電系統之 ，壓調整機制反應過慢’即會產生電壓突升狀況。若儀 器設備本身無相關保護措施’則將因電壓瞬間過 致設備受損或誤動作。 第6A圖所不為測試波形，此波形係由電歷突升加 入雜訊所構成，且SNR值為34.03 dB。其中，電壓遞增 ，為系統額定㈣之50%，並持續3個週波。經本文^ =運异後’可得如第6B圖之重建訊號，此訊號 :為犯’測試結果均能證實文中方法具備滤除雜 訊之此力。另第6C圖則顯示雜訊監測波形，該結果對 於觀察雜訊變化幅度與出現起料間應可提供較為完善 之協助。 測試4 :電壓令斷加入雜訊 13 200914846 電壓中斷之際，電壓有效值將低於〇1 p.u.並會維持 0.5個週期以上之久。發生電壓中斷原因繁眾，常見的輸 ，線路故障造成地H網路供電巾止，或是發電機跳脫使 得區域供電需求不足，進而導致電壓逐漸下降終至斷電。 第7A目|波形為電壓中斷加入雜訊所合成而得。 其中，第4個至第6個週期之電壓值均為〇pu。現由所 提方法予以驗證該波形，可得如第7B圖之還原結果， 其中SNR值由最初26.23 dB經濾'波後增加至6〇 8 dB。 而雜訊隨時間變動之結果則顯示於第7c圖。換言之， 透過上述測試結果的呈現，可說明本文所提方法確有助 於雜訊的濾除與檢測。 測試5 :瞬時脈衝加入雜訊 、瞬時脈衝係屬-種在數毫微秒到數毫秒間變化數伏 特或至數千伏特的脈波。脈衝的起因通常肇生於雷擊、 2系統斷路跳脫及重新閉合，或系統補償電容器切換 ^情況。對使用者而言，瞬時脈衝的發生是毫無徵兆， =僅會造成資料傳輸錯誤，_過大的脈波亦可能使 件系統瞬間過載而解聯。 第8A圖為瞬時脈衝之波形，其中並加人雜訊予以 擾’且此波形之SNR值為33 〇4 dB。第8B圖則為其 ;慮波結果’此日寺SNR值已增加為64.95 dB，換言之本文 ^法確能有效還原瞬時脈衝之波形。另第8C圖則顯示 :成監測波形’由圖中可觀察出誤差平均值大於零之處 即為雜訊發生虛。 14 200914846 由上述内文可知，本實施例所採用的新型小波轉換 亦具有下述特點：適用於任一種小波轉換技術。其中僅 須調整預測函數pj、更新函數Uj、正規化參數ΚΗ與 以加減計算取代摺積運算，大幅降低轉換過程之 算量。所有運算均在空間域（Spatial domain)執行，可大 幅減少記憶體的使用。而若第一電子訊號為正弦波時， 口正弦波轉換後之高頻成分均為零，故重建後之訊號仍 為純正弦波，對訊號品質並不造成影響，所以可將將第 一電子訊號之高頻訊號設為零。 雖然本發明已以較佳實施例揭露如上，然其並非用 以限定本發明，任何熟習此技藝者，在不脫離本發明之 精神和範圍内，當可作各種之更動與潤飾，因此本發明 之保護範圍當視後附之申請專利範圍所界定者為準。 【圖式簡單說明】 為讓本發明之上述和其他目的、特徵、優點與實施 例能更明顯易懂，所附圖式之詳細說明如下： 第1圖係繪示本發明之小波轉換應用於電子訊號之债 測方法之流程圖； 第2圖係繪示本發明之小波轉換架構圖； 第3圖係繪示本發明之小波反轉換架構圖； 第4A圖係繪示正常電壓波形加入雜訊之波形圖； 第4B圖係繪示根據第4A圖之重建訊號之波形圖； 第5 A圖係繪示電壓突降加入雜訊加入雜訊之波形圖； 15 200914846 第5B圖係綠示根據第5 a圖之重建訊號之波形圖； 第5C圖係繪示計算重建訊號與輸入訊號之誤差平方值 瓜測結果之波形圖； ^6Α圖係續'示電壓突升加入雜訊之波形圖； 第6B圖係綠示第6 a圖之重建訊號之波形圖； • 第6C圖係繪示計算重建訊號與輸入訊號之誤差平方值 監測結果之波形圖； 第7 A圖係％示電壓中斷加入雜訊之波形圖； 第7B圖係繪示第7A圖之重建訊號之波形圖； 第7C圖係繪示計算重建訊號與輸入訊號之誤差平方值 ^ 監測結果之波形圖； 第8 A圖係繪示瞬時脈衝加入雜訊之波形圖； 第8B圖係繪示第8A圖之重建訊號之波形圖；以及 第8C圖係繪示計算重建訊號與輸入訊號之誤差平方值 監測結果之波形圖。 $元件符號說明】 11〜17 :步驟流程。 16Shape with an SNR of 27.9 dB. With the help of this method, Figure 5B shows the filtering results. This figure shows that the proposed method has the ability to restore the voltage burst waveform and its SNR value is increased to 60.89 dB. At this point, the squared error of the reconstructed signal and the input signal is calculated to obtain the monitoring waveform of the fifth graph. The graph shows that the appearance and end of the noise are consistent with the simulation of the test 2. Test 3: Voltage surge is added to the noise and the voltage dip is lower than 'when the voltage rises above 10% of the rated voltage of the system' and this condition lasts for 5 cycles or minutes, which is called voltage burst. Rise. When the proximity of a large negative Weng, the transmission system, the pressure adjustment mechanism reacts too slowly, that is, a voltage surge occurs. If the instrument itself has no relevant protective measures, the device will be damaged or malfunction due to voltage transients. Figure 6A is not a test waveform. This waveform consists of an electrical calendar spike plus noise, and the SNR value is 34.03 dB. Among them, the voltage is increasing, which is 50% of the system rated (four), and lasts for 3 cycles. After the article ^=Transportation, you can get the reconstruction signal as shown in Figure 6B. This signal: the test results can confirm that the method has the power to filter out the noise. The other 6C shows the noise monitoring waveform. This result can provide better assistance for observing the amplitude of the noise change and the occurrence of the material. Test 4: Voltage interrupts add noise 13 200914846 When the voltage is interrupted, the rms voltage will be lower than 〇1 p.u. and will last for more than 0.5 cycles. There are many reasons for the voltage interruption, the common transmission, the line fault caused by the H network power supply towel, or the generator tripping makes the regional power supply demand insufficient, which leads to the voltage gradually drop to the end of power failure. The 7A mesh|waveform is synthesized by adding a noise to the voltage interrupt. Among them, the voltage values of the fourth to sixth periods are all 〇pu. The waveform is now verified by the proposed method, and the reduction result as shown in Fig. 7B can be obtained, wherein the SNR value is increased from the initial 26.23 dB after filtering to 6 〇 8 dB. The result of the noise changes over time is shown in Figure 7c. In other words, through the above test results, it can be shown that the method proposed in this paper can help the filtering and detection of noise. Test 5: Transient pulses are added to the noise, and the instantaneous pulse is a pulse that varies by several volts or thousands of volts between a few nanoseconds and a few milliseconds. The cause of the pulse is usually caused by lightning strikes, 2 system trips and reclosing, or system compensation capacitor switching. For the user, there is no sign of the occurrence of the transient pulse, = only the data transmission error, _ excessive pulse wave can also make the system instantaneously overload and untie. Figure 8A shows the waveform of the instantaneous pulse, which is disturbed by adding noise and the SNR value of this waveform is 33 〇 4 dB. Figure 8B is for it; the wave result is increased to 64.95 dB on this day. In other words, the method can effectively restore the waveform of the instantaneous pulse. In addition, the 8C chart shows: the monitoring waveform ‘ from the figure can be observed that the error average is greater than zero, that is, the noise is imaginary. 14 200914846 It can be seen from the above text that the novel wavelet transform used in this embodiment also has the following features: It is applicable to any wavelet transform technology. Among them, only the prediction function pj, the update function Uj, the normalization parameter ΚΗ and the addition and subtraction calculation are used instead of the convolution operation, which greatly reduces the calculation process of the conversion process. All operations are performed in the spatial domain, which greatly reduces the use of memory. If the first electronic signal is a sine wave, the high frequency component after the sine wave conversion is zero, so the reconstructed signal is still a pure sine wave, which does not affect the signal quality, so the first electron can be The high frequency signal of the signal is set to zero. While the present invention has been described above by way of a preferred embodiment, it is not intended to limit the invention, and the present invention may be modified and modified without departing from the spirit and scope of the invention. The scope of protection is subject to the definition of the scope of the patent application. BRIEF DESCRIPTION OF THE DRAWINGS In order to make the above and other objects, features, advantages and embodiments of the present invention more obvious, the detailed description of the drawings is as follows: Figure 1 shows the application of the wavelet transform of the present invention. FIG. 2 is a flow chart of the wavelet transform architecture of the present invention; FIG. 3 is a diagram showing the wavelet inverse conversion architecture of the present invention; FIG. 4A is a diagram showing the normal voltage waveform added to the hybrid Waveform of the signal; Figure 4B shows the waveform of the reconstructed signal according to Figure 4A; Figure 5A shows the waveform of the voltage drop added to the noise to add noise; 15 200914846 Figure 5B shows the green signal According to the waveform diagram of the reconstructed signal of Figure 5a; Figure 5C shows the waveform of the squared value of the error of the reconstructed signal and the input signal; ^6Α图Continued to show the waveform of the voltage surge added to the noise Fig. 6B is a waveform diagram of the reconstructed signal of Fig. 6a; Fig. 6C is a waveform diagram of the squared value monitoring result of calculating the reconstructed signal and the input signal; Interrupt the waveform of adding noise; 7th Figure B shows the waveform of the reconstructed signal in Figure 7A; Figure 7C shows the waveform of the squared value of the error between the reconstructed signal and the input signal; the waveform of the monitoring result; Figure 8A shows the transient pulse added to the noise The waveform diagram of FIG. 8B is a waveform diagram of the reconstruction signal of FIG. 8A; and FIG. 8C is a waveform diagram for monitoring the error squared value of the reconstruction signal and the input signal. $Component Symbol Description] 11~17: Step flow. 16