201203137 六、發明說明： 【發明所屬之技術領域】 本發明是有關於一種資訊末端設備（remote terminal unit)且特別疋種有具有資料校正（data correction)功能 的資訊末端設備與資料校正方法。 • 【先前技術】 資吼末端設備目如已廣泛用於各遠端監控系統，以達 # 成自動化監控之目的。目前的技術都是將資訊末端設備的 資料直接取回遠端電腦主機進行運算，而未賦予資訊末端 設備過濾可能由現場引入雜訊之能力。如此，一旦混入正 常訊號之雜訊過高時，將造成擷取之資料產生誤差，而導 致監控系統主機可能會接收到錯誤的資料。 舉例來說，公共事業用自動化監控系統，例如：電力、 電L水力、運輸監控系統等，多數會採用集中監控方式 建構遠端監控系統。資訊末端設備一般多會配置於現場， 以負責擷取現場資料。由於資訊末端設備是配置於現場， 因此較容易受到現場各受監控端設備所引入之雜訊干擾， * ❿導致監控f料充滿了許多的不確定性，使得監控系統主 機可能會將來自於資訊末端設備的錯誤資料顯示於螢幕， 進而導致操作人員誤判資料與作出不正確的決策判斷。 欠資訊末端設備可以透過有線或無線傳輸方式將所擷取 的資料傳送給位於遠端的監控系統主機，若資訊末端設備 所傳輸的監控 > 料為錯誤資料，則也會導致遠端的監控系 201203137 統主機收到不正確的資訊。 目前雖有學術論文提出以模糊演算法來建立校正函數 的模型，做為校正誤差之作法，但採用的方式是二離線方 式做模糊演算法之計算’所以並無法達成即時校正之目的。 【發明内容】 本發明提供一種資料校正方法，用於—資 備。首先，提供一平滑參數。接著，將量測的多筆監控^ 料所構成的一輸入向量當作一訓練樣本饋入—廣義 網路，以H此確定該廣義類神賴路的多個權錢。然 根據該平滑參數與該些權重值產生—校正糊輸出值、< 盥種資訊末端設備，包括—信號收發模組 :/“置處理模組。信號收發模組接收量測的一 監控資料…： 神經網路來校正該 4以產生一校正預測估測值，其中哕 滑參數是透_號處‘執= 、’”丁、上所述’本發明所提出的資訊 的資料’在未送至遠端主機前，便先將收到 料做信號前置處理運算，以解決誤差影響控資 控系統的運作更為安全可#。更進_ 、，、化監 置處理是以廣義睛經網路作為 l上述^虎前 並搭配粒子群演算法找出廣義類神經網路方法^ 來做資料前處理。 ，㈣巾最佳平滑參數 201203137 【實施方式】 請參照第1圖，第1圖是本發明之實施例所提供的自 動化監控系統的系統方塊圖。自動化監控系統10包括受監 控端設備12、信號轉換器14、資訊末端設備16與監控系 統主機18。受監控端設備12可以是電力、電信、水力、運 輸監控系統等的受監控端設備，資訊末端設備16置放於受 . 監控端設備12的附近，用以擷取來自於受監控端設備12 的信號。在受監控端設備與資訊末端設備丨6之間具有信號 φ 轉換器14，信號轉換器14可以將受監控端設備π的信號 轉換乘資訊末端設備16可以接受與處理的信號。舉例來 发，文監控端設備12可以是現場端之受監控電氣設備，然 而，^號轉換器14可以是電力轉換器。現場端之受監控電 氣。又備所產生的類比訊號可以透過信號轉換器Μ變成標準 電流信號，並且標準電流信號可以被傳送至資訊末端設備 16。 鲁 資訊末端設備16包括信號收發模組166與信號前置處 /模、、且168 ’彳5號收發模組166接收信號轉換器丨4所轉換 • 2之信號的量測資料’信號前置處理單元168用以將量測 貝料進行校正處理，並產生校正預測輸出值。接著，信號 2板組166，校正預測輸出值傳送至監控系統主機18， =控系統主機18分析接收到的校正賴輸出值，以檢視受 •控端設備12的狀況。 要說明的疋，#號前置處理模組16使用廣義類神經網 路作為資訊末端設備16的資料前處理之關鍵技術。廣義類 5 201203137 神經網路學習速度快，且僅需少量數據即可快速逼近目標 函數而達到收斂的效果。除此之外，信號前置處理模組16 還使用粒子群演算法來做平滑參數的最佳化，粒子群演算 法常被用來處理最佳化的問題，同時也具備了快速收斂， 只需要設定較少參數即可完成等優點。如此，透過粒子群 演算法將可以減少了以往使用廣義類神經網路調整參數的 問題，利用最佳化廣義類神經網路以達到快速，有效地解 決問題。據此，此實施例中的信號前置處理模組16可以透 過廣義類神經網來校正資訊末端設備16所接收到的錯誤監 控資料。 接著叫參照第2圖，第2圖是本發明之實施例所提 仏的自動化監控系統之各信號與資料的示意圖。在步驟 S22，里測信號會混入干擾雜訊而形成量測信號。在步驟 S24，里測信號被轉換成資訊賣端設備所能接受與處理的監 控-貝料，且此監控資料會被自動化監控系統2〇中的資訊末 端設備所接收。因為，監控資料混入了干擾雜訊，因此若 干擾雜efL過大，將影響監控資料的正確性，而導致監控系 統主機28會顯示錯誤的監控資料給操作人員。據此，在步 驟，資訊末端設備會對監控資料進行前置信號處理， 以校正錯誤的監控資料。上述前置信號處理是使用上述廣 義類神經網進行處理，以產生校正預測輸出值，並傳送給 監控系統主機28。 接著，請參照第3圖，第3圖是本發明之實施例所提 供資料杈正方法的流程圖。一開始，步驟S30與步驟S40 201203137 會同時被執行。在步驟S40，藉由各種演算法產生平滑參 數，並將平滑參數提供給廣異類神經網路使用，其中合適 的平滑參數會將雜訊數據做平滑處理，以做出較正確估 計。在步驟S30，將量測的監控資料構成的輸入向量當作 訓練樣本饋入廣義神經網路的輸入層，以藉此確定廣義類 神經網路的權重值，其中廣義類神經網路會將訓練樣本的 輸入特徵向量與輸出特徵向量分別存到輸入神經元與輸出 神經元間的連結。此處量測的監控資料例如是類比資料， • 但在其他實施例中，監控資料亦可以是數位資料。 接著，在步驟S32，根據平滑參數及輸入層至隱藏層 之權重值，計算隱藏層各神經元之輸出值。接著，在步驟 S34，根據隱藏層至輸出層的權重，計算輸出層神經元之輸 出值，並正規化輸出層神經元之輸出值，以產生了校正預 測輸出值。如此，將可以減少混入干擾雜訊時的誤差影響。 要說明的是，在步驟S40,產生平滑參數的其中一種 演算法可以是粒子群演算法。請參照第4圖，第4圖是本 ® 發明之實施例所提供的產生平滑參數之方法流程圖。首 先，在步驟S42，初始化學習常數、疊代次數、平滑參數 '粒子個數及其範圍。接著，在步驟S44，定義適應函數(fitness function)。之後，在步驟S46，根據定義函數，計算各平滑 參數粒子所對應的適應值。然後，在步驟S48，對平滑參 數所對應之適應值進行排列，並據此選出最優適應值所對 應的平滑參數。之後，在步驟S50，根據此最優適應值所 對應的平滑參數，更新平滑參數粒子的位置與速度。 201203137 接著，在步驟S52，判斷誤差值或疊代次數是否滿足 終止條件。若疊代次數已經達到預設的數目或平滑參數收 斂值已不再變動，則進行步驟S54,否則，則回到步驟S46。 在步驟S54，輸出最優適應值所對應的平滑參數，並將此 平滑參數饋入廣義類神經網路。 綜上所述，本發明所提出的資訊末端設備會將接收到 的資料，在未送至遠端主機前，便先將這些量測的監控資 料做信號前置處理運算，以解決誤差影響，並使自動化監 控系統的運作更為安全可靠。更進一步地說，上述信號前 置處理是以廣義類神經網路作為即時資料前處理的方法， 並搭配粒子群演算法找出廣義類神經網路中最佳平滑參數 來做資料前處理。' 【圖式簡單說明】 第1圖是本發明之實施例所提供的自動化監控系統的 系統方塊圖。 第2圖是本發明之實施例所提供的自動化監控系統之 各信號與資料的示意圖。 第3圖是本發明之實施例所提供資料校正方法的流程 圖。 第4圖是本發明之實施例所提供的產生平滑參數之方 法流程圖。 【主要元件符號說明】 201203137 10、20 :自動化監控系統 12 :受監控端設備 14 :信號轉換器 16 :資訊末端設備 166 :信號收發模組 168:信號前置處理模組 18 :監控系統主機 S22、S24、S26、S30、S32、S34、S40、S42、S44、S48 癱 S50、S52、S54 :步驟201203137 VI. Description of the Invention: [Technical Field] The present invention relates to a remote terminal unit and particularly to an information terminal device and a data correction method having a data correction function. • [Prior Art] The end equipment of the asset has been widely used in remote monitoring systems to achieve the goal of automatic monitoring. The current technology is to retrieve the information of the end device of the information directly back to the remote computer host for calculation, and does not give the information to the end device to filter the ability to introduce noise from the scene. As a result, if the noise mixed with the normal signal is too high, the data will be inaccurate, and the monitoring system host may receive the wrong data. For example, public utilities use automatic monitoring systems, such as power, electricity, water, transportation monitoring systems, etc., most of them will use centralized monitoring to build remote monitoring systems. Information terminal equipment is generally deployed on site to take charge of on-site data. Since the information terminal equipment is configured on the site, it is more susceptible to noise interference introduced by the monitored equipment on the site. * The monitoring equipment is full of many uncertainties, so that the monitoring system host may come from the information. The error data of the end device is displayed on the screen, which causes the operator to misjudge the data and make an incorrect decision. The information terminal device can transmit the captured data to the remote monitoring system host through wired or wireless transmission. If the monitoring device transmitted by the information terminal device is incorrect, it will also cause remote monitoring. The system 201203137 host received incorrect information. At present, although there are academic papers that propose a fuzzy algorithm to establish a model of the correction function, as a method of correcting the error, the method adopted is the calculation of the fuzzy algorithm in the second offline mode. Therefore, the purpose of immediate correction cannot be achieved. SUMMARY OF THE INVENTION The present invention provides a data correction method for use in preparing. First, a smoothing parameter is provided. Then, an input vector composed of the measured plurality of monitoring materials is fed as a training sample into the generalized network, thereby determining the plurality of weights of the generalized class. According to the smoothing parameter and the weight values, the calibration paste output value, < the information end device, including the signal transceiver module: / "the processing module. The signal transceiver module receives the measurement data. ...: the neural network to correct the 4 to generate a corrected prediction estimate, wherein the slip parameter is the _ number at the 'execution=, '', and the above-mentioned 'information of the information presented by the invention' is not Before being sent to the remote host, the signal processing is pre-processed to solve the error, and the operation of the control system is safer. Further processing _,,, and monitoring is based on the generalized eye network as the above-mentioned ^ tiger front and with the particle swarm algorithm to find the generalized neural network method ^ to do data pre-processing. (Four) towel optimal smoothing parameter 201203137 [Embodiment] Please refer to Fig. 1, which is a system block diagram of an automatic monitoring system according to an embodiment of the present invention. The automated monitoring system 10 includes a monitored terminal device 12, a signal converter 14, an information end device 16, and a monitoring system host 18. The monitored end device 12 may be a monitored end device of the power, telecommunication, hydraulic, transportation monitoring system, etc., and the information end device 16 is placed in the vicinity of the monitored end device 12 for extracting from the monitored end device 12 signal of. Between the monitored end device and the information end device 丨6, there is a signal φ converter 14, which can convert the signal of the monitored end device π by the signal that the information end device 16 can accept and process. For example, the monitoring device 12 can be a monitored electrical device at the field, however, the number 14 converter can be a power converter. The monitored electricity at the field. The analog signal generated by the standby device can be converted into a standard current signal through the signal converter, and the standard current signal can be transmitted to the information terminal device 16. The Lu information end device 16 includes a signal transceiving module 166 and a signal pre-mode/module, and the 168 '彳5 transceiver module 166 receives the signal of the signal converter 丨4 to convert the signal of the signal. The processing unit 168 is configured to perform a correction process on the measured bedding material and generate a corrected predicted output value. Next, the signal 2 board group 166, the corrected predicted output value is transmitted to the monitoring system host 18, and the control system host 18 analyzes the received corrected output value to view the condition of the controlled device 12. To illustrate, the ## pre-processing module 16 uses a generalized neural network as a key technique for data pre-processing of the information end device 16. Generalized Class 5 201203137 Neural network learning is fast, and only a small amount of data is needed to quickly approach the target function to achieve convergence. In addition, the signal pre-processing module 16 also uses the particle swarm algorithm to optimize the smoothing parameters. The particle swarm algorithm is often used to deal with the optimization problem, and also has a fast convergence, only Need to set fewer parameters to complete and so on. In this way, the particle swarm optimization algorithm can reduce the problem of adjusting parameters using the generalized neural network in the past, and optimize the generalized neural network to solve the problem quickly and effectively. Accordingly, the signal pre-processing module 16 in this embodiment can correct the error monitoring data received by the information terminal device 16 through the generalized neural network. Referring next to Fig. 2, Fig. 2 is a schematic diagram showing signals and data of an automated monitoring system as proposed in an embodiment of the present invention. In step S22, the measured signal is mixed with interference noise to form a measurement signal. In step S24, the measured signal is converted into a monitoring-being material that can be accepted and processed by the information selling device, and the monitoring data is received by the information terminal device in the automatic monitoring system. Because the monitoring data is mixed with interference noise, if the interference efL is too large, the correctness of the monitoring data will be affected, and the monitoring system host 28 will display the wrong monitoring data to the operator. Accordingly, in the step, the information terminal device performs pre-signal processing on the monitoring data to correct the erroneous monitoring data. The pre-signal processing described above is processed using the above-described broad-type neural network to generate a corrected predicted output value and transmitted to the monitoring system host 28. Next, please refer to Fig. 3, which is a flow chart showing a method of data correction according to an embodiment of the present invention. Initially, step S30 and step S40 201203137 will be executed simultaneously. In step S40, smoothing parameters are generated by various algorithms, and the smoothing parameters are provided to the broad heterogeneous neural network for use, wherein the appropriate smoothing parameters smooth the noise data to make a more accurate estimation. In step S30, the input vector formed by the measured monitoring data is fed as a training sample to the input layer of the generalized neural network, thereby determining the weight value of the generalized neural network, wherein the generalized neural network will train The input feature vector and the output feature vector of the sample are respectively stored in a connection between the input neuron and the output neuron. The monitoring data measured here is, for example, analog data, but in other embodiments, the monitoring data may also be digital data. Next, in step S32, the output values of the neurons of the hidden layer are calculated based on the smoothing parameters and the weight values of the input layer to the hidden layer. Next, in step S34, the output value of the output layer neuron is calculated based on the weight of the hidden layer to the output layer, and the output value of the output layer neuron is normalized to generate a corrected predicted output value. In this way, it is possible to reduce the influence of errors when mixing interference noise. It is to be noted that, in step S40, one of the algorithms for generating the smoothing parameter may be a particle swarm algorithm. Please refer to FIG. 4, which is a flow chart of a method for generating smoothing parameters provided by an embodiment of the present invention. First, in step S42, the learning constant, the number of iterations, the smoothing parameter 'number of particles and its range are initialized. Next, in step S44, a fitness function is defined. Thereafter, in step S46, the adaptive value corresponding to each smoothing parameter particle is calculated based on the definition function. Then, in step S48, the adaptation values corresponding to the smoothing parameters are arranged, and the smoothing parameters corresponding to the optimal fitness values are selected accordingly. Thereafter, in step S50, the position and velocity of the smoothing parameter particles are updated according to the smoothing parameter corresponding to the optimal fitness value. 201203137 Next, in step S52, it is judged whether or not the error value or the number of iterations satisfies the termination condition. If the number of iterations has reached the preset number or the smoothing parameter convergence value has not changed, step S54 is performed, otherwise, the process returns to step S46. In step S54, the smoothing parameter corresponding to the optimal fitness value is output, and the smoothing parameter is fed into the generalized neural network. In summary, the information terminal device proposed by the present invention will perform the pre-processing of the measured monitoring data before the remote data is sent to the remote host to solve the error impact. And make the operation of the automated monitoring system safer and more reliable. Furthermore, the above signal preprocessing is based on a generalized neural network as a method of real-time data pre-processing, and the particle swarm algorithm is used to find the optimal smoothing parameters in the generalized neural network for data pre-processing. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a system block diagram of an automated monitoring system provided by an embodiment of the present invention. Figure 2 is a schematic diagram of signals and data of an automated monitoring system provided by an embodiment of the present invention. Fig. 3 is a flow chart showing a method of data correction provided by an embodiment of the present invention. Figure 4 is a flow diagram of a method for generating smoothing parameters provided by an embodiment of the present invention. [Main component symbol description] 201203137 10, 20: Automation monitoring system 12: Monitored terminal device 14: Signal converter 16: Information terminal device 166: Signal transceiver module 168: Signal pre-processing module 18: Monitoring system host S22 , S24, S26, S30, S32, S34, S40, S42, S44, S48 瘫S50, S52, S54: steps