Gas Turbine Condition Monitoring Method Based on Multivariate State Autoregression and Kalman Filter

Abstract

The present invention discloses a gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering, which includes: collecting historical operating exhaust gas temperature data of all measuring points of the gas turbine exhaust gas temperature in the full load operation range, preprocessing the historical operating exhaust gas temperature data to obtain a preprocessing vector set, and obtaining a standard state vector matrix for the gas turbine exhaust gas temperature operation reference; obtaining the real-time state vector of the gas turbine exhaust gas temperature, calculating the estimated value of the real-time state vector according to the multivariate state autoregression estimation and the standard state vector matrix, and smoothing the residual to obtain a smoothed residual; performing Kalman filtering on the dispersion of all measuring points of the gas turbine exhaust gas temperature to obtain a dispersion estimated value; judging whether the smoothed residual exceeds the limit three times continuously through the threshold in the normal operation range, and at the same time, judging whether the dispersion estimated value exceeds the dispersion threshold, so as to realize the monitoring of the gas turbine exhaust gas temperature. The gas turbine condition monitoring method of the present invention improves the sensitivity of exhaust gas temperature monitoring and can give early warnings.

Description

Technical Field

[0001] The present invention relates to the field of automatic control technology, and in particular to a method for monitoring the state of a combustion engine based on multivariate state autoregression and Kalman filtering. Background Art

[0002] In order to improve the efficiency of heavy-duty gas turbines, the temperature before the turbine is getting higher and higher. The temperature before the turbine of the latest generation of H-class heavy-duty gas turbines reaches 1450°C. Such a high temperature means that the combustion chamber and its transition section are subject to greater thermal stress. These components will inevitably break and be damaged if they are in high-temperature operation for a long time. Therefore, it is particularly necessary to monitor the operating status of the gas turbine, especially the combustion chamber, and give early warnings in time.

[0003] At present, the monitoring system of gas turbines in power plants generally adopts the exhaust temperature dispersion index. The control system will promptly alarm when the exhaust temperature dispersion is large, reminding the operating personnel to take measures. When the exhaust temperature dispersion is too large, the control system will directly issue a shutoff command to prevent the accident from further expanding. In actual applications, the dispersion cannot give alarm information for slow-changing faults in a timely manner, and "after-the-fact" diagnosis will occur. When the combustion monitoring of the gas turbine issues a combustion fault alarm, the combustion system of the gas turbine has already been seriously damaged. Summary of the invention

[0004] Based on this, the present invention provides a method for monitoring the state of a gas engine based on multivariate state autoregression and Kalman filtering. Whether the exhaust temperature is abnormal is determined by a multivariate state autoregression model and a Kalman filtering model of the exhaust temperature mean, thereby determining the operating state of the gas engine.

[0005] To achieve the above object, the present invention adopts the following technical solution: a method for monitoring the state of a combustion engine based on multivariate state autoregression and Kalman filtering, which specifically comprises the following steps:

[0006] Step 1: Collect the historical exhaust temperature data of all the exhaust temperature measuring points of the gas turbine in the full load operation range; the historical exhaust temperature data is represented by the measuring point vector [x j ], where x j is the exhaust temperature measurement value of the jth measuring point;

[0007] Step 2: pre-process the collected historical exhaust temperature data, remove abnormal values and duplicate values, and obtain a pre-processing vector set;

[0008] Step 3: Determine the preprocessing vector set according to the operation state vector library to obtain a standard state vector matrix for the exhaust temperature operation reference of the gas turbine;

[0009] Step 4: Obtain the real-time state vector of the gas turbine exhaust temperature. Calculate the estimated value of the real-time state vector according to the multivariate state autoregressive estimation and the standard state vector matrix of the gas turbine exhaust temperature operation reference, and calculate the residual between the real-time state vector and the estimated value. Use the moving average method to smooth the residual between the real-time state vector and the estimated value to obtain the smoothed residual;

[0010] Step 5: Calculate the dispersion of all measurement points of the gas turbine exhaust temperature, and perform real-time processing using the Kalman filter to obtain the estimated value of the filtered dispersion;

[0011] Step 6: Determine whether the smoothed residual exceeds the limit through the threshold of the normal operation interval. At the same time, determine whether the estimated value of the dispersion continuously exceeds the dispersion threshold three times, so as to realize the monitoring of the gas turbine exhaust temperature.

[0012] Further, step 3 includes the following sub-steps:

[0013] Step 31: Find the minimum value and the maximum value of the exhaust temperature data in the preprocessing vector set, equally divide the entire change interval according to the minimum value and the maximum value, and sort the preprocessing vector set from small to large;

[0014] Step 32: Establish a state vector matrix with the preprocessing vector corresponding to the minimum value as the starting boundary and the preprocessing vector corresponding to the maximum value as the ending boundary;

[0015] Step 33: Select the corresponding preprocessing vector according to the divided distance in the sorted preprocessing vector set;

[0016] Step 34: Calculate the dissimilarity within the selected preprocessing vector. If the dissimilarity is qualified, add the selected preprocessing vector to the state vector state matrix; otherwise, randomly select within the neighborhood of the selected preprocessing vector, and repeat step 34;

[0017] Step 35: Repeat steps 33 - 34 until the divided distance reaches the ending boundary to obtain the standard state vector matrix of the gas turbine exhaust temperature operation reference.

[0018] Further, the judgment process for qualified dissimilarity is as follows: If the difference between the exhaust temperature data of more than 2 / 3 of the measurement points in the selected preprocessing vector exceeds 0.1 °C, the dissimilarity is qualified.

[0019] Further, the calculation process of the estimated value of the real-time state vector is as follows:

[0020] (a) The standard state vector matrix of the gas turbine exhaust temperature operation reference and the real-time state vector Obtain the distance operator through the Mahalanobis distance operator:

[0021]

[0022] (b) Combine the distance operator with the Mercer kernel theory to calculate the standard state vector matrix for the reference of the exhaust gas temperature operation of the gas turbine The weight coefficient w(k) of the k-th state vector in

[0023]

[0024] (c) Calculate the estimated value of the real-time state vector according to w(k)

[0025]

[0026] where S is a diagonal matrix composed of the standard deviations of the exhaust gas temperature data of the gas turbine exhaust gas temperature measurement points, h is the width coefficient of the Mercer kernel function, and n is the number of state vectors in the standard state vector matrix for the reference of the gas turbine exhaust gas temperature operation.

[0027] Furthermore, the calculation process of the dispersion S of all measurement points of the gas turbine exhaust gas temperature is as follows:

[0028]

[0029] where is the average exhaust gas temperature of all measurement points, is the compressor outlet temperature, the temperature unit is ℉, and (100) indicates that it is added only when the operating condition of the gas turbine changes.

[0030] Furthermore, the calculation process of the estimated value of the dispersion is as follows:

[0031]

[0032] where is the estimated value of the dispersion of the measurement points at the (k - 1)th moment, S is the dispersion of the measurement points, H is the identity matrix, K k is the first update matrix, is the P matrix at the previous moment, P K is the second update matrix, R is the noise matrix.

[0033] Furthermore, the process of obtaining the threshold value of the normal operation interval is as follows: Obtain the historical operation exhaust gas temperature data of the normal operation interval, combine the standard state vector matrix for the reference of the gas turbine exhaust gas temperature operation, apply the multivariate state autoregressive estimation to calculate the estimated value of the normal historical operation exhaust gas temperature data, and conduct statistical analysis on the residuals between the historical operation exhaust gas temperature data and the estimated value to obtain the residual distribution characteristics, and use the median of the residuals as the threshold value of the normal operation interval.

[0034] Further, the implementation process of step 6 is as follows:

[0035] (i) If the smoothed residual exceeds the limit three times continuously and the residual of the dispersion estimate value exceeds the threshold, it is determined that the measuring point corresponding to the exhaust gas temperature of the gas turbine is abnormal, and it is judged that there may be a fault in the relevant combustion chamber part;

[0036] (ii) If the number of times the smoothed residual exceeds the limit is less than three, but the residual of the dispersion estimate value exceeds the threshold, it is considered that there is a fault in the exhaust gas temperature measuring point and calibration needs to be checked;

[0037] (iii) If the smoothed residual does not exceed the limit three times continuously and the dispersion estimate residual does not exceed the dispersion threshold, it is considered that the current operating state of the gas turbine exhaust gas temperature belongs to the normal range.

[0038] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses the state estimation method combined with the method of dispersion Kalman to perform real-time monitoring of the gas turbine exhaust gas temperature. Moreover, the multivariate autoregressive state estimation method has high sensitivity to parameter changes, and can detect the slight deterioration of equipment through the change of the correlation relationship between parameters at the early stage of the development of slow-varying faults, so as to improve the sensitivity of exhaust gas temperature monitoring. At the same time, combined with the trend change of the Kalman filter of dispersion, the reliability of gas turbine exhaust gas temperature monitoring and diagnosis is improved. Description of the Drawings

[0039] Figure 1 It is a flow chart of the method for obtaining the reference standard state vector matrix of the gas turbine exhaust gas temperature measuring point in the present invention;

[0040] Figure 2 It is the residual curve of the multivariate state autoregressive estimation for 12 exhaust gas temperature measuring points of a certain gas turbine unit in the present invention;

[0041] Figure 3 It is the trend chart of the estimated value of the dispersion after Kalman filtering in the present invention. Detailed Embodiment

[0042] The technical solution of the present invention will be further explained below with reference to the drawings.

[0043] The present invention provides a gas turbine state monitoring method based on multivariate state autoregression and Kalman filtering, which specifically includes the following steps:

[0044] Step 1: Collect the historical operating exhaust gas temperature data of all measuring points of the gas turbine exhaust gas temperature in the full load operating range; the historical operating exhaust gas temperature data is composed of a set of measuring point vectors [x j , where x j is the exhaust gas temperature measurement value of the jth measuring point;

[0045] Step 2: Preprocess the collected historical exhaust gas temperature data during operation, eliminate outliers and duplicate values, and obtain a preprocessing vector set. In the present invention, an outlier refers to a value that exceeds the normal operating range, and the measuring point vector corresponding to the outlier is deleted; a duplicate value refers to that the difference in exhaust gas temperature measurement values of all measuring points in any two groups of measuring point vectors in the historical exhaust gas temperature data during operation is less than 0.1 °C, then one group of measuring point vectors is deleted.

[0046] Step 3: According to the method for determining the operating state vector library, obtain a standard state vector matrix for reference of the gas turbine exhaust gas temperature during operation from the preprocessing vector set, and reduce the number of state vectors in the standard state vector matrix through the method for determining the operating state vector library of equidistant partitioning and dissimilarity degree, while being able to meet the coverage range and diversity of the state vectors in the standard state vector matrix; as Figure 1 , specifically including the following sub-steps:

[0047] Step 31: Find the minimum value and the maximum value of the exhaust gas temperature data in the preprocessing vector set, and perform equidistant partitioning on the entire change interval according to the minimum value and the maximum value, and at the same time arrange the preprocessing vector set from small to large;

[0048] Step 32: Taking the preprocessing vector corresponding to the minimum value as the starting boundary and the preprocessing vector corresponding to the maximum value as the ending boundary, establish a state vector matrix;

[0049] Step 33: Select the corresponding preprocessing vector according to the divided distance in the sorted preprocessing vector set;

[0050] Step 34: Calculate the dissimilarity degree inside the selected preprocessing vector. If the dissimilarity degree is qualified, add the selected preprocessing vector to the state vector state matrix; otherwise, randomly select within the neighborhood of the selected preprocessing vector, and repeat Step 34;

[0051] Step 35: Repeat Steps 33 - 34 until the divided distance reaches the ending boundary to obtain a standard state vector matrix for reference of the gas turbine exhaust gas temperature during operation.

[0052] In the present invention, the judgment process for qualified dissimilarity degree is as follows: If the difference between any two measuring points of the exhaust gas temperature data of more than 2 / 3 of the measuring points in the selected preprocessing vector exceeds 0.1 °C, it is considered that the dissimilarity degree is qualified.

[0053] Step 4: Obtain the real-time state vector of the gas turbine exhaust temperature. Calculate the estimated value of the real-time state vector based on the multivariate state autoregressive estimation and the standard state vector matrix of the gas turbine exhaust temperature operation reference. There is a strong correlation relationship among the gas turbine exhaust temperature measurement points. The multivariate state autoregressive estimation method is suitable for modeling between strongly correlated parameters, and the exhaust temperature measurement point data is rich, which can provide good support for the standard state vector matrix of this estimation method. The accuracy and real-time performance of the estimation model are easy to control. By calculating the residual between the real-time state vector and the estimated value, use the moving average method to smooth the residual between the real-time state vector and the estimated value to obtain the smoothed residual; as Figure 2 , taking 12 measurement points as an example, obtain the residual curve between the real-time state vector and the estimated value through the above method, Figure 2 which shows the residual between the estimated values and the real-time measured values of 12 measurement points. It can be seen from this that the regression estimation effect of the multivariate state autoregressive estimation is good. Most of the deviation values are within 1°C, and the maximum deviation basically does not exceed 1.5°C. If converted to the relative error, it is basically controlled within 0.2%, fully meeting the requirements of engineering accuracy.

[0054] The calculation process of the estimated value of the real-time state vector in the present invention is as follows:

[0055] (a) Obtain the distance operator by using the Mahalanobis distance operator for the standard state vector matrix of the gas turbine exhaust temperature operation reference and the real-time state vector :

[0056]

[0057] (b) Combine the distance operator with the Mercer kernel theory to calculate the weight coefficient w(k) of the k-th state vector in the standard state vector matrix of the gas turbine exhaust temperature operation reference:

[0058]

[0059] (c) Calculate the estimated value

[0060]

[0061] of the real-time state vector according to w(k), where S is a diagonal matrix composed of the standard deviations of the exhaust temperature data of the gas turbine exhaust temperature measurement points, h is the width coefficient of the Mercer kernel function, and n is the number of state vectors in the standard state vector matrix of the gas turbine exhaust temperature operation reference.

[0062] Step 5: Calculate the dispersion of all measurement points of the gas turbine exhaust temperature, and perform real-time processing using Kalman filtering to obtain the filtered dispersion estimate value; the dispersion is greatly affected by the exhaust temperature mean value, and sometimes irregular fluctuations occur, even leading to false alarms. The Kalman filtering method is used to perform real-time smoothing processing on the fluctuations of the dispersion, and continuously update its own model following time, thereby improving the reliability of the dispersion index. For example, Figure 3 As shown in the trend chart of the estimated value of the dispersion after Kalman filtering, it can be seen that the Kalman filtering estimated value of the dispersion exceeds the limit after the 145th time point and shows a trend of rising. According to this situation, it can be judged that the gas turbine state is abnormal and further fault separation is required.

[0063] In the present invention, the calculation process of the dispersion S of all measurement points of the gas turbine exhaust temperature is as follows:

[0064]

[0065] wherein, is the average exhaust temperature of all measurement points, is the temperature at the compressor outlet, and the temperature unit is ℉. (100) indicates that it is added only when the gas turbine operating condition changes.

[0066] In the present invention, the calculation process of the dispersion estimate value is as follows:

[0067]

[0068] wherein, is the estimated value of the measurement point dispersion at the (k - 1)th moment, S is the measurement point dispersion, H is the unit matrix, and K k is the first update matrix, is the P matrix at the previous moment, and P K is the second update matrix, R is the noise matrix.

[0069] Step 6: Determine whether the smoothed residual exceeds the limit three times continuously through the threshold of the normal operation interval. At the same time, determine whether the dispersion estimate value exceeds the dispersion threshold, so as to realize the monitoring of the gas turbine exhaust temperature. Since the multivariate state autoregressive estimation residual has high sensitivity, here it is judged by whether it exceeds the limit three times continuously, and combined with the filtered estimation of the dispersion, the result is made more reliable while improving the sensitivity; the specific process is as follows:

[0070] (i) If the smoothed residual exceeds the limit three times continuously and the residual of the dispersion estimate value exceeds the threshold, it is determined that the measurement point corresponding to the gas turbine exhaust temperature is abnormal, and it is judged that there may be a fault in the relevant combustion chamber part;

[0071] (ii) If the number of times the smoothed residual exceeds the limit is less than three, but the residual of the dispersion estimate value exceeds the threshold, it is considered that there is a fault in the exhaust temperature measurement point, and calibration needs to be checked;

[0072] (iii) If the smoothed residual does not exceed the limit three times continuously and the residual of the dispersion estimate does not exceed the dispersion threshold, it is considered that the current operating state of the gas turbine exhaust temperature belongs to the normal range.

[0073] The process of obtaining the threshold of the normal operating range in the present invention is as follows: Obtain the historical operating exhaust temperature data of the normal operating range, combine the standard state vector matrix for the reference of the gas turbine exhaust temperature operation, apply the multivariate state autoregressive estimation to calculate the estimated value of the normal historical operating exhaust temperature data, and conduct statistical analysis on the residual between the historical operating exhaust temperature data and the estimated value to obtain the residual distribution characteristics, and use the median of the residuals as the threshold of the normal operating range.

[0074] The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering in the present invention can be applied to the real-time monitoring of the operating state of the gas turbine. Since the residual of the multivariate state autoregressive estimation has high sensitivity, early warning prompts can be given in the early stage of minor deterioration of equipment faults. While the Kalman filtering estimation of the dispersion has high reliability. By combining the two methods, a comprehensive judgment of the exhaust temperature of the gas turbine is carried out, which improves the monitoring sensitivity of the gas turbine exhaust temperature while ensuring the reliability of abnormal diagnosis. On the other hand, the gas turbine condition monitoring method of the present invention can also distinguish and diagnose the exhaust temperature measurement point fault and the combustion equipment fault, providing more detailed and accurate guiding information for on-site personnel.

[0075] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering, characterized in that, Specifically, it includes the following steps: Step 1: Collect the historical operating exhaust gas temperature data of all measuring points of the gas turbine in the full load operating range; the historical operating exhaust gas temperature data consists of a set of measurement point vectors [x j , where x j is the exhaust gas temperature measurement value of the j-th measuring point; Step 2: Preprocess the collected historical exhaust temperature data of the operation, eliminate outliers and duplicate values, and obtain a preprocessing vector set; Step 3: According to the method for determining the operation status vector library, obtain the standard status vector matrix for the operation reference of the gas turbine exhaust temperature from the preprocessing vector set; it includes the following sub-steps: Step 31: Find the minimum value and the maximum value of the exhaust temperature data in the preprocessing vector set, equally divide the entire change interval according to the minimum value and the maximum value, and at the same time arrange the preprocessing vector set from small to large; Step 32: Establish a status vector matrix with the preprocessing vector corresponding to the minimum value as the starting boundary and the preprocessing vector corresponding to the maximum value as the ending boundary; Step 33: Select the corresponding preprocessing vector according to the divided distance in the sorted preprocessing vector set; Step 34: Calculate the dissimilarity degree inside the selected preprocessing vector. If the dissimilarity degree is qualified, add the selected preprocessing vector to the status vector status matrix; otherwise, randomly select within the neighborhood of the selected preprocessing vector, and repeat Step 34; Step 35: Repeat Steps 33 - 34 until the divided distance reaches the ending boundary to obtain the standard status vector matrix for the operation reference of the gas turbine exhaust temperature; Step 4: Obtain the real-time status vector of the gas turbine exhaust temperature, calculate the estimated value of the real-time status vector according to the multivariate state autoregressive estimation and the standard status vector matrix for the operation reference of the gas turbine exhaust temperature, calculate the residual between the real-time status vector and the estimated value, and use the moving average method to smooth the residual between the real-time status vector and the estimated value to obtain the smoothed residual; Step 5: Calculate the dispersion degree of all measuring points of the gas turbine exhaust temperature, and perform real-time processing using the Kalman filter to obtain the estimated value of the filtered dispersion degree; Step 6: Determine whether the smoothed residual exceeds the limit three times continuously through the threshold of the normal operation interval. At the same time, determine whether the estimated value of the dispersion degree exceeds the dispersion degree threshold, so as to realize the monitoring of the gas turbine exhaust temperature.

2. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that, The process for judging that the dissimilarity degree is qualified is as follows: If the difference between the exhaust temperature data of more than 2 / 3 of the measuring points in the selected preprocessing vector exceeds 0.1 °C, it is considered that the dissimilarity degree is qualified.

3. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that The calculation process of the estimated value of the real-time status vector is as follows: (a) The standard state vector matrix for reference of the exhaust gas temperature operation of the gas turbine and the real-time state vector Obtain the distance operator through the Mahalanobis distance operator: (b) Combine the distance operator with the Mercer kernel theory to calculate the standard state vector matrix for the operating reference of the gas turbine exhaust temperature The weight coefficient w(k) of the k-th state vector in (2) (c) Calculate the estimated value of the real-time state vector based on w(k) where S is a diagonal matrix composed of the standard deviations of the exhaust temperature data of the gas turbine exhaust temperature measuring points, h is the width coefficient of the Mercer kernel function, and n is the number of status vectors in the standard status vector matrix for the operation reference of the gas turbine exhaust temperature.

4. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that The calculation process of the dispersion degree S of all measuring points of the gas turbine exhaust temperature is as follows: Among them, is the average exhaust gas temperature of all measurement points, is the compressor outlet temperature, with the temperature unit being ℉, and (100) indicates that it is added only when the operating conditions of the gas turbine change.

5. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that, The calculation process of the estimated value of the dispersion degree is as follows: Among them, is the estimated value of the measurement point dispersion at time k-1, S is the measurement point dispersion, H is the identity matrix, K k is the first update matrix, is the P matrix at the previous moment, P K is the second update matrix, R is the noise matrix.

6. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that The process for obtaining the threshold of the normal operation interval is as follows: Obtain the historical exhaust temperature data of the normal operation interval, combine the standard status vector matrix for the operation reference of the gas turbine exhaust temperature, apply the multivariate state autoregressive estimation to calculate the estimated value of the normal historical exhaust temperature data, and perform statistical analysis on the residual between the historical exhaust temperature data and the estimated value to obtain the residual distribution characteristics, and use the median of the residual as the threshold of the normal operation interval.

7. The gas turbine condition monitoring method based on multivariate state autoregression and Kalman filtering according to claim 1, characterized in that, The implementation process of Step 6 is as follows: (i) If the smoothed residual exceeds the limit three times consecutively and the residual of the dispersion estimate value exceeds the threshold, it is determined that the measuring point corresponding to the exhaust gas temperature of the gas turbine is abnormal, and it is judged that there may be a fault in the relevant combustion chamber part; (ii) If the number of times the smoothed residual exceeds the limit is less than three, but the residual of the dispersion estimate value exceeds the threshold, it is considered that there is a fault in the exhaust gas temperature measuring point, and calibration needs to be checked; (iii) If the smoothed residual does not exceed the limit three times consecutively and the dispersion residual does not exceed the dispersion threshold, it is considered that the operating state of the current gas turbine exhaust gas temperature belongs to the normal range.

Images