Dynamic memory matrix construction and fan fault early warning method based on characteristic parameter sorting

Abstract

The invention provides a dynamic memory matrix construction and fan fault early warning method based on characteristic parameter sorting, which comprises the following steps: 1) acquiring historical operation data of a power station blower, selecting part of data according to characteristic parameters to form a historical data set T, and acquiring test set data; 2) selecting a plurality of vectors from the historical data set T for each observation vector in the test set data to form a dynamic memory matrix D, and establishing an MSET model; 3) performing standardization processing on the dynamic memory matrix D, and performing same processing on the current observation vector; and 4) calculating an estimated value of the test set by using an MSET model, introducing a similarity function to carry out fault early warning, and giving an alarm when the similarity function is lower than an early warning threshold value. The method has the characteristics of high unit calculation precision, good adaptability, comprehensive monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of blower failure warning, and more specifically, to a method for constructing a dynamic memory matrix and a blower failure warning based on feature parameter sorting. Background Art

[0002] The operating condition of the blower during the operation of a thermal power plant unit will directly affect the combustion state of the furnace and the safety of the unit operation. Once a failure occurs, the unit will be shut down, causing economic losses to the power plant and even equipment accidents.

[0003] Therefore, it is necessary to monitor the operation status of the fan in real time. When its operation data is abnormal, it can timely and effectively detect the abnormality of the fan and guide the staff to perform equipment maintenance to reduce or avoid the losses caused by the failure of the fan. The existing fan fault warning technology has limited ability to identify fault information. It often selects part of the monitoring data to establish a warning model. In the implementation process, some fault information will inevitably be ignored, resulting in untimely fault warning and insufficient fault warning ability. Summary of the invention

[0004] In order to overcome the problems existing in the above-mentioned prior art, the present invention proposes a dynamic memory matrix construction and fan fault early warning method based on characteristic parameter sorting.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for constructing a dynamic memory matrix and warning of fan faults based on characteristic parameter sorting includes the following steps:

[0007] S1: Obtain historical operation data of the power plant blower, select part of the data according to the characteristic parameters to form a historical data set T, and obtain test set data to construct the historical data set T, including the following steps:

[0008] Step 1: Sort the historical data according to the current size, and divide the data into m groups according to the step size (Imax-Imin) / m;

[0009] Step 2: Sort each group of samples according to the characteristic parameters of the fan lubricating oil pressure, fan lubricating oil temperature, fan blade opening, fan outlet air pressure, and fan inlet air temperature to form data sets a, b, c, d, and e;

[0010] Step 3: Select m groups of samples from data sets a, b, c, d, and e respectively, and add them to the historical data set T;

[0011] Step 4: Remove duplicates from the historical data set T.

[0012] S2: For each observation vector in the test set data, select several vectors from the historical data set T to form a dynamic memory matrix D, and establish an MSET model, including the following steps

[0013] Step 1: According to the current value x of the new observation vector X, select the data with current magnitude within the range of x±5A in T to form a matrix N;

[0014] Step 2: According to the characteristic parameters X, namely, the lubricating oil pressure of the fan, the lubricating oil temperature of the fan, the opening of the fan blades, the outlet air pressure of the fan, and the inlet air temperature of the fan, the data within a certain range are selected in the matrix N to form matrices N1, N2, N3, N4, and N5;

[0015] Step 3: Calculate the number of repetitions of each vector in matrix N in matrices N1, N2, N3, N4, and N5;

[0016] Step 4: Prioritize vectors with more repetition times to form the dynamic memory matrix D.

[0017] S3: normalize the dynamic memory matrix D and perform the same process on the current observation vector;

[0018] S4: Use the MSET model to calculate the estimated value of the test set, introduce a similarity function for fault warning, and issue an alarm when the similarity function is lower than the warning threshold.

[0019] Preferably, the historical operating data of the power station fan includes fan bearing temperatures 1 to 9, fan motor front bearing temperature, fan motor rear bearing temperature, fan X-direction bearing vibration, fan Y-direction bearing vibration, fan lubricating oil pressure, fan lubricating oil temperature, fan current, fan rotor blade opening, fan outlet air pressure, and fan inlet air temperature.

[0020] Preferably, the expression of the dynamic memory matrix D is:

[0021]

[0022] D n*m It is a matrix with n rows and m columns, representing the monitoring status of n monitoring variables at m moments.

[0023] Preferably, an MSET model is established, through a dynamic memory matrix D n*m The prediction result of the observation vector is obtained, and the calculation formula is as follows:

[0024]

[0025] Where: X est —Estimation vector; X obs — observation vector; —The spatial distance between two vectors.

[0026] Preferably, the dynamic memory matrix is ​​standardized, and the formula is as follows:

[0027]

[0028] Where x is the variable, μ is the mean of x, and σ is the standard deviation of x.

[0029] Preferably, a similarity function is introduced as an evaluation index for fault warning, and the similarity function calculation formula is as follows:

[0030]

[0031] Where: X est —Estimation vector; X obs —Observation vector; t k is the kth moment.

[0032] Preferably, the fault warning threshold is calculated by a similarity function of historical normal operation data, and the calculation formula is as follows:

[0033]

[0034] Where: S w —Fault warning threshold; —Minimum value of the similarity function of historical normal operation data; k—threshold coefficient, and k>1. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A system flow chart of a method for constructing a dynamic memory matrix and warning of fan faults based on feature parameter sorting according to an embodiment of the present invention is shown.

[0036] Figure 2 A flowchart of constructing a historical data set of a dynamic memory matrix based on feature parameter sorting and a fan fault early warning method according to an embodiment of the present invention is shown.

[0037] Figure 3 A dynamic memory matrix construction flow chart of a dynamic memory matrix construction and fan fault warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0039] Figure 4 The prediction result of the X-direction bearing vibration of the blower according to a dynamic memory matrix construction and blower fault early warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0040] Figure 5The prediction result of the Y-axis bearing vibration of the blower according to a dynamic memory matrix construction and blower fault warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0041] Figure 6 The prediction result of the lubricating oil pressure of the blower of a dynamic memory matrix construction and blower fault warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0042] Figure 7 The prediction result of the lubricating oil temperature of the blower of a method for constructing a dynamic memory matrix based on characteristic parameter sorting and a blower fault warning according to an embodiment of the present invention is shown.

[0043] Figure 8 The prediction result of the air temperature at the inlet of the blower of a method for constructing a dynamic memory matrix based on characteristic parameter sorting and warning of blower failure according to an embodiment of the present invention is shown.

[0044] Fig. 9 The prediction result of the fan blade opening of a dynamic memory matrix construction and fan fault warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0045] Fig.10 The prediction result of the blower current of a dynamic memory matrix construction and blower fault early warning method based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0046] Fig.11 The prediction result of the outlet wind pressure of the blower of a method for constructing a dynamic memory matrix and warning of blower failure based on characteristic parameter sorting according to an embodiment of the present invention is shown.

[0047] Fig.12 The test set similarity function of a dynamic memory matrix construction based on feature parameter sorting and a wind turbine fault early warning method according to an embodiment of the present invention is shown. Fig.13 The average relative error of the MSET model constructed by two methods, namely, a dynamic memory matrix construction based on characteristic parameter sorting and a fan fault early warning method according to an embodiment of the present invention, is shown. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with the accompanying drawings and embodiments, so that the objects, advantages and technical solutions of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0049] like Figure 1The flowchart of a method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting is shown, and the method comprises the following steps:

[0050] 1. Obtain the historical operation data of the power plant fan, select part of the data according to the characteristic parameters to form the historical data set T, and obtain the test set data;

[0051] The historical operation data of the power station fan are obtained, including the fan bearing temperature 1-9, the fan motor front bearing temperature, the fan motor rear bearing temperature, the fan X-axis bearing vibration, the fan Y-axis bearing vibration, the fan lubricating oil pressure, the fan lubricating oil temperature, the fan current, the fan blade opening, the fan outlet pressure, and the fan inlet temperature. Select some data according to the characteristic parameters to form the historical data set T. The specific data selection process is as follows: Figure 2 As shown, the use of this method to select historical data sets reduces the historical operating data space and improves the model calculation speed. At the same time, the screened data can cover different operating states of the blower, ensuring the accuracy of the model.

[0052] 2. For each observation vector in the test set data, select several vectors from the historical data set T to form a dynamic memory matrix D, and establish the MSET model;

[0053] The specific selection process of the dynamic memory matrix D is as follows Figure 3 As shown in the figure, for each observation vector, data is selected within a certain range according to its characteristic parameters to form a dynamic memory matrix, which can ensure that the selected data is close to the running state of the current observation vector and improve the calculation accuracy of the model. The expression of the dynamic memory matrix D is:

[0054]

[0055] D n*m It is a matrix with n rows and m columns, representing the monitoring status of n monitoring variables at m moments.

[0056] 3. Standardize the dynamic memory matrix D and perform the same process on the current observation vector;

[0057] Since the present invention selects different data for each observation vector to form a dynamic memory matrix, the memory matrix corresponding to each observation vector is different. Therefore, it is necessary to first standardize the dynamic memory matrix and then perform the same processing on the observation vector. The standardization processing formula is as follows:

[0058]

[0059] Where x is the variable, μ is the mean of x, and σ is the standard deviation of x.

[0060] S4: Use the MSET model to calculate the estimated value of the test set, introduce a similarity function for fault warning, and issue an alarm when the similarity function is lower than the warning threshold.

[0061] Through the dynamic memory matrix D n*m The prediction result of the observation vector is obtained, and the calculation formula is as follows:

[0062]

[0063] The similarity function is introduced as the evaluation index of fault warning. The Euclidean distance is the most commonly used method to measure the spatial distance between two vectors. The larger the Euclidean distance, the lower the similarity between the two vectors and the less similar the operating states are. The sliding window method is used to calculate the average similarity. The sliding window size is 20. The similarity function calculation formula is as follows:

[0064]

[0065] Where: X est —Estimation vector; X obs —Observation vector; t k is the kth moment.

[0066] When the unit is operating normally, the fault warning threshold is calculated by the similarity function of the historical normal operating data, and the calculation formula is as follows:

[0067]

[0068] Where: S w —Fault warning threshold; —Minimum value of the similarity function of historical normal operation data; k—threshold coefficient, and k>1.

[0069] The example obtains 20273 sets of historical operation data of the blower, using Figure 2 The method screens the historical operation data and obtains 13051 sets of data to form the historical data set T, which reduces the historical operation data by about 35%. 400 sets of operation data before the failure occurred were selected to form the test set data, and the MSET model fault warning test was carried out. Figure 3 The method selects 150 groups of historical operation data for each group of observation vectors to form a dynamic memory matrix for model calculation.

[0070] The traditional method selects the memory matrix according to the Euclidean distance between vectors, that is, by calculating the Euclidean distance between the observation vector and all historical data sets, several historical data with the closest distance to the observation vector space are selected to form the memory matrix. This method can screen out the vector with the closest distance to the observation vector space from a large amount of historical data for model calculation, which can ensure the accuracy of the model to a certain extent. However, the Euclidean distance reflects the overall spatial distance between the two vectors, which only reflects the similarity of the two vectors in real numerical values, and cannot clearly reflect the operating status between the two vectors.

[0071] The first 100 groups of test set data were selected, and the model was calculated using the method of the present invention and the traditional method respectively. The size of the dynamic memory matrix constructed by the two methods was 150 groups. Through comparative calculation, it can be seen that the time consumed by the present invention to calculate the 100 groups of test set data is 19.8s, and the calculation time of the traditional method is 27.4s. Fig.13 The calculation results of the two methods show that the calculation accuracy of the temperature parameter calculated by the traditional method is higher, but the calculation accuracy of other characteristic parameters is lower than that of the method described in the present invention. Because the temperature parameters in the blower data are relatively uniform and there is no large step, the memory matrix selected by the traditional method by calculating the Euclidean distance often selects data that is very close to the temperature data of the observation vector, but other parameters, such as pressure, current, bearing vibration, etc., are far away from the observation vector space, resulting in a large deviation in the model calculation.

[0072] from Figures 4 to 11 It can be seen that before the fault develops, the prediction results of the MSET model for the X-axis bearing vibration of the fan, the Y-axis bearing vibration of the fan, the lubricating oil pressure of the fan, the lubricating oil temperature of the fan, the inlet air temperature of the fan, the fan blade opening, the fan current, and the outlet air pressure of the fan are very close to the actual measured values. Similarly, Fig.12 The similarity of sample points in the first half of the period is relatively high, and the minimum average similarity during this period is 0.7130.

[0073] Depend on Figure 4 , 5 , 6, 7, and 8 show that some time before the fault occurred, the predicted results of the fan X-axis bearing vibration, fan Y-axis bearing vibration, fan lubricating oil pressure, fan lubricating oil temperature, and fan inlet air temperature deviated from the actual measured values ​​to varying degrees, indicating that the operating state of the fan at this time deviated from the normal operating state, even in Fig. 9 , 10 The predicted results of the fan blade opening, fan current, and fan outlet pressure in 11 are close to the actual measured values, but other parameters can already reflect that the fan is in an abnormal operating state. Fig.12 It can be seen that the average similarity is constantly decreasing before the failure occurs.

[0074] Pick Fig.12 The lowest value of the data sample points in the middle and front part is 0.7130, which is the lowest average similarity of the normal operation data. Take k as 1.2 to calculate the fault warning threshold:

[0075]

[0076] Depend on Fig.12 It can be seen that at the 347th sample point, the average similarity is less than the fault warning threshold, and an alarm signal is issued, indicating that the present invention can detect abnormal operation of the blower in advance and remind the staff to perform equipment maintenance to avoid causing greater economic losses.

[0077] Beneficial Effects

[0078] The present invention uses the parameters of blower bearing temperature 1-9, blower motor front bearing temperature, blower motor rear bearing temperature, blower X-direction bearing vibration, blower Y-direction bearing vibration, blower lubricating oil pressure, blower lubricating oil temperature, blower current, blower rotor blade opening, blower outlet air pressure, and blower inlet air temperature to establish a fault warning model. The monitoring parameters are comprehensive and specific, and can effectively and accurately monitor the operating status of the blower, thereby improving the reliability of the blower fault warning.

[0079] The present invention utilizes the operating characteristic parameters of the blower, including the blower current, the blower lubricating oil pressure, the blower lubricating oil temperature, the blower blade opening, the blower outlet air pressure, and the blower inlet air temperature to screen out data that can effectively characterize the operating status of the blower from a large amount of historical data. This can greatly reduce the size of the historical operation database and shorten the time for constructing a dynamic memory matrix. The screened historical operation data set can represent different operating states of the blower, thereby ensuring the calculation accuracy of the model.

[0080] The present invention constructs a dynamic memory matrix for the observation vector at each moment, which can avoid the inclusion of data of other operating states in a single historical memory matrix, affecting the calculation accuracy of the MSET model; the dynamic memory matrix is ​​constructed using the operating characteristic parameters of the blower to ensure that the screened historical data is similar to the operating state of the current observation vector, effectively improving the calculation accuracy and fault warning capability of the MSET model.

Claims

1. A method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting, characterized in that: The steps include: S1: Obtain historical operation data of the power plant blower, select part of the data according to the characteristic parameters to form a historical data set T, and obtain test set data to construct the historical data set T, including the following steps: Step 1: Sort the historical data according to the current size, and divide the data into m groups according to the step size (Imax-Imin) / m; Step 2: Sort each group of samples according to the characteristic parameters of the fan lubricating oil pressure, fan lubricating oil temperature, fan blade opening, fan outlet air pressure, and fan inlet air temperature to form data sets a, b, c, d, and e; Step 3: Select m groups of samples from data sets a, b, c, d, and e respectively, and add them to the historical data set T; Step 4: Remove duplicates from the historical data set T; S2: For each observation vector in the test set data, select several vectors from the historical data set T to form a dynamic memory matrix D, and establish an MSET model, including the following steps: Step 1: According to the current value x of the new observation vector X, select the data with current magnitude within the range of x±5A in T to form a matrix N; Step 2: According to the characteristic parameters X, namely, the lubricating oil pressure of the fan, the lubricating oil temperature of the fan, the opening of the fan blades, the outlet air pressure of the fan, and the inlet air temperature of the fan, the data within a certain range are selected in the matrix N to form matrices N1, N2, N3, N4, and N5; Step 3: Calculate the number of repetitions of each vector in matrix N in matrices N1, N2, N3, N4, and N5; Step 4: Prioritize the vectors with more repetition times to form the dynamic memory matrix D; S3: normalize the dynamic memory matrix D and perform the same process on the current observation vector; S4: Use the MSET model to calculate the estimated value of the test set, introduce the similarity function for fault warning, and use the sliding window method to calculate the average similarity. When the similarity function is lower than the warning threshold, an alarm is issued.

2. A method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1, characterized in that: The historical operation data of the power station blower obtained in step S1 include blower bearing temperatures 1 to 9, blower motor front bearing temperature, blower motor rear bearing temperature, blower X-direction bearing vibration, blower Y-direction bearing vibration, blower lubricating oil pressure, blower lubricating oil temperature, blower current, blower rotor blade opening, blower outlet air pressure, and blower inlet air temperature.

3. The method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1 is characterized in that: The expression of the dynamic memory matrix D in step S2 is: D n*m It is a matrix with n rows and m columns, representing the monitoring status of n monitoring variables at m moments.

4. The method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1 is characterized in that: In step S2, the MSET model is established, and the dynamic memory matrix D n*m The prediction result of the observation vector is obtained, and the calculation formula is as follows: Where: X est —Estimation vector; X obs — observation vector; —The spatial distance between two vectors.

5. The method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1 is characterized in that: In step S3, the dynamic memory matrix is ​​normalized, and the formula is as follows: Where x is the variable, μ is the mean of x, and σ is the standard deviation of x.

6. The method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1 is characterized in that: In step S4, a similarity function is introduced as an evaluation index for fault warning, and the average similarity is calculated using a sliding window method. The similarity function calculation formula is as follows: Where: X est —Estimation vector; X obs —Observation vector; t k is the kth moment.

7. The method for constructing a dynamic memory matrix and warning of fan failure based on characteristic parameter sorting according to claim 1 is characterized in that: In step S4, the fault warning threshold is calculated by the similarity function of the historical normal operation data, and the calculation formula is as follows: Where: S w —Fault warning threshold; —Minimum value of the similarity function of historical normal operation data; k—threshold coefficient, and k>1.

Images