Railway locomotive diesel engine operation parameter clustering anomaly detection method

Abstract

The invention relates to the technical field of railway locomotive diesel engine operation parameter clustering anomaly detection methods, in particular to a railway locomotive diesel engine operation parameter clustering anomaly detection method based on a proxy model optimization technology, which comprises the following steps: constructing a clustering anomaly detection model parameter sample point set; and constructing an approximation model of the clustering anomaly detection algorithm parameters, inputting the initial clustering anomaly detection algorithm parameters to the approximation model, and estimating the detection effect of the input clustering anomaly detection algorithm parameters by using the approximation model. Algorithm parameter updating is carried out in real-time anomaly detection of a clustering algorithm, and the accuracy of a clustering anomaly detection model is improved. Besides, the agent model technology not only makes full use of the prediction capability of the agent model for data, but also has excellent performance of self-optimization, and solves the problems that a traditional clustering method depends too much on artificial experience and the precision is reduced along with time in the detection process.

Description

Technical Field

[0001] The invention relates to the technical field of a method for detecting anomaly clustering of operating parameters of a railway locomotive diesel engine, and in particular to a method for detecting anomaly clustering of operating parameters of a railway locomotive diesel engine. Background Art

[0002] Under the background of safe and stable operation of railway locomotives, it is of great significance to use big data science to mine the development laws of railway locomotive data operation. As a popular direction of data science, abnormal data detection technology can be used for data cleaning to reduce the interference of abnormal data on data analysis or for early warning protection to improve the safety of railway locomotive operation. Therefore, establishing effective anomaly detection technology for railway locomotive operation and maintenance has become one of the key issues for the safety and reliability of railway locomotives. Clustering algorithm technology can analyze the outlier state of core data items and emphasize the value of data analysis. At present, locomotives only have parameter alarm functions, but do not have fault diagnosis and operation and maintenance suggestion prompt functions. Data analysis and processing need to be returned to the locomotive depot external computer for viewing and solving. Therefore, the development of a railway locomotive diesel engine operating parameter anomaly detection method based on clustering algorithm is used on locomotives to provide fault diagnosis and operation and maintenance prompt information, which can effectively improve the locomotive operation and maintenance efficiency and improve the availability of locomotive diesel engines.

[0003] Most of the current anomaly detection methods use traditional clustering methods such as random forests and nearest neighbor clustering algorithms. The recognition efficiency of these methods depends largely on the effectiveness of the manually designed initial algorithm parameters and algorithm selection, which limits the designability of clustering algorithms. In addition, in the actual operating environment, the operating parameters of railway locomotive diesel engines are constantly changing. Even if the existing algorithms can detect abnormal data, the recognition rate will gradually decrease over time.

[0004] Therefore, there is an urgent need for a method that can improve the accuracy of abnormal detection of operating parameters of railway locomotive diesel engines in real time. Summary of the invention

[0005] In order to overcome the existing problem that when designing cluster anomaly detection algorithm parameters, the designed model parameter scheme needs to be adjusted, and the time cost of each adjustment is high; if an abnormal problem occurs on-site in a railway diesel engine, the problem needs to be discovered in the later stage of the entire abnormality detection, and unqualified model parameters cannot be quickly discovered and determined, and the trial and error cost is high. In addition, in the process of designing cluster anomaly detection model parameters, technical personnel are required to repeatedly debug, and the whole process overly relies on the experience of technical personnel, and the accuracy of cluster anomaly detection algorithm parameters cannot be guaranteed. The present invention provides a cluster anomaly detection method for railway locomotive diesel engine operating parameters based on agent model optimization technology.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a railway locomotive diesel engine operating parameter clustering anomaly detection method based on agent model optimization technology, the method comprising the following steps:

[0007] Step 1: Construct a clustering anomaly detection model parameter sample point set;

[0008] Step 2: Based on the sample point set and proxy model technology, an approximate model of clustering anomaly detection algorithm parameters is constructed;

[0009] Step 3: Input the initial clustering anomaly detection algorithm parameters into the approximate model, and use the approximate model to estimate the detection effect of the input clustering anomaly detection algorithm parameters;

[0010] Step 4: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 does not meet the requirements, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set;

[0011] Step 5: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 meets the requirements, the optimal parameter solution is output, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set;

[0012] Step 6: According to the optimal solution, the clustering anomaly detection algorithm is used to perform anomaly analysis on the current data.

[0013] According to another embodiment of the present invention, further comprising: the clustering algorithm of step one is: there are kn railway locomotive diesel engine sensor points, kk centers are found in the railway locomotive diesel engine sensor data, and the data point distance expression is: Make the sum of the squares of the distances between the locomotive diesel engine sensor data point and its nearest center point the largest and the kk value the smallest, where x = (W kn , kk) is a sample point.

[0014] According to another embodiment of the present invention, further comprising, wherein W kn is the distance between the data points of the locomotive diesel engine sensor, x ki is the kith data center point; z i is the kjth data center point.

[0015] According to another embodiment of the present invention, further comprising, in step 2, the expression of the radial basis function proxy model in any approximate model is:

[0016] According to another embodiment of the present invention, further comprising, wherein i and n are n=10, and i can be a number from 1 to 10; To estimate the objective function value of clustering algorithm;

[0017] a i is the coefficient expression; x 0i is the radial basis center point.

[0018] According to another embodiment of the present invention, further comprising: the loss function of the approximate model system in step 2 is:

[0019] According to another embodiment of the present invention, further comprising, wherein i, j, m, n, n, m can be any positive integer greater than 10; i can be a number from 1 to n; j can be a number from 1 to m; x j -x 0i is the center point of the radial basis function, which is Scheme 1; E j is the jth solution; a i is a coefficient expression obtained by minimizing the loss function.

[0020] According to another embodiment of the present invention, the step 4 further comprises:

[0021] Step 401: Select k data from n railway locomotive diesel engine sensor data as cluster centers;

[0022] Step 402: Calculate the Euclidean distance from the remaining data to the k cluster centers, which is expressed as:

[0023]

[0024] Among them, F(x,c i ) is the remaining railway locomotive diesel engine data x to the i-th cluster center c i The Euclidean distance of

[0025] Step 403: recalculate the k cluster centers according to the above formula, and calculate the data average of all the railway locomotive diesel engine sensor data dimensions in the cluster, and reselect the data center point according to the average;

[0026] Step 404: clustering again with the newly selected railway locomotive diesel engine data center point, and repeating the process until the clustering result does not change any more;

[0027] Step 405 calculates the Mahalanobis distance of the clustering result to determine whether the clustering result is abnormal data of a railway locomotive diesel engine. The first clustering result of K-means is used as input. The Mahalanobis distance is used to calculate the cluster center point of the railway locomotive diesel engine data set, and then the distance from the cluster to the center point is calculated respectively. The data distance W is set. kn, when the result of the above formula is greater than the distance threshold, it means that the railway locomotive diesel engine data in the cluster is abnormal data, and the railway locomotive diesel engine data anomaly detection is completed. The anomaly detection error is judged according to the sample set data, and the proxy model optimization technology is used to update k k The distance threshold W between the value and the data kn .

[0028] The beneficial effect of the present invention is that it proposes to update the algorithm parameters in the real-time anomaly detection of the clustering algorithm to improve the accuracy of the clustering anomaly detection model. In addition, the proxy model technology not only makes full use of the proxy model's ability to predict data, but also has excellent self-optimization performance, which solves the problem that the traditional clustering method is too dependent on manual experience and the accuracy decreases over time during the detection process; the present invention also proposes to use the optimization efficiency of the proxy model in the anomaly detection process to achieve real-time improvement of the proxy model, that is, to use the current proxy model as the training starting point of the new proxy model for model training. This method not only shortens the model training time and improves the model recognition accuracy, but also solves the problem of difficulty in selecting the initial parameters of the proxy model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0030] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0031] like Figure 1 Schematic diagram of the structure of the present invention, a method for detecting anomaly of railway locomotive diesel engine operating parameters clustering based on agent model optimization technology, the method comprising the following steps:

[0032] Step 1: Construct a clustering anomaly detection model parameter sample point set;

[0033] Step 2: Based on the sample point set and proxy model technology, an approximate model of clustering anomaly detection algorithm parameters is constructed;

[0034] Step 3: Input the initial clustering anomaly detection algorithm parameters into the approximate model, and use the approximate model to estimate the detection effect of the input clustering anomaly detection algorithm parameters;

[0035] Step 4: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 does not meet the requirements, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set;

[0036] Step 5: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 meets the requirements, the optimal parameter solution is output, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set;

[0037] Step 6: According to the optimal solution, the clustering anomaly detection algorithm is used to perform anomaly analysis on the current data.

[0038] Specifically, compared with the prior art, the present invention is based on constructing an approximate model of cluster anomaly detection algorithm parameters, using the approximate model to estimate the cluster anomaly detection algorithm parameters to be detected, obtaining cluster anomaly detection algorithm parameter estimated values, comparing the target function estimated value with the lowest target function value in the sample point set, and using the target function value closest to 0 as the output value of the optimal solution, thereby quickly completing the preliminary screening of cluster anomaly detection algorithm parameter solutions, and finding the optimal solution from among many cluster anomaly detection algorithm parameter solution solutions, and then using anomaly detection to analyze and calculate the optimal cluster anomaly detection algorithm parameters, and further judging whether the solution meets the requirements. In this way, the agent model optimization design method is applied to the anomaly detection algorithm, which reduces the number of manual anomaly detection algorithm debugging times and improves the detection efficiency of the anomaly detection algorithm.

[0039] According to another embodiment of the present invention, further comprising: the clustering algorithm of step one is: there are kn railway locomotive diesel engine sensor points, kk centers are found in the railway locomotive diesel engine sensor data, and the data point distance expression is: Make the sum of the squares of the distances between the locomotive diesel engine sensor data point and its nearest center point the largest and the kk value the smallest, where x = (W kn , kk) is a sample point.

[0040] According to another embodiment of the present invention, further comprising, wherein W kn is the distance between the data points of the locomotive diesel engine sensor, x ki is the kith data center point; z i is the kjth data center point.

[0041] According to another embodiment of the present invention, further comprising, in step 2, the expression of the radial basis function proxy model in any approximate model is:

[0042] According to another embodiment of the present invention, further comprising, wherein i and n are n=10, and i can be a number from 1 to 10; To estimate the objective function value of clustering algorithm;

[0043] a i is the coefficient expression; x 0i is the radial basis center point.

[0044] According to another embodiment of the present invention, further comprising: the loss function of the approximate model system in step 2 is:

[0045] According to another embodiment of the present invention, further comprising, wherein i, j, m, n, n, m can be any positive integer greater than 10; i can be a number from 1 to n; j can be a number from 1 to m; x j -x 0i is the center point of the radial basis function, which is Scheme 1; E j is the jth solution; a i is a coefficient expression obtained by minimizing the loss function.

[0046] According to another embodiment of the present invention, the step 4 further comprises:

[0047] Step 401: Select k data from n railway locomotive diesel engine sensor data as cluster centers;

[0048] Step 402: Calculate the Euclidean distance from the remaining data to the k cluster centers, which is expressed as:

[0049]

[0050] Among them, F(x,c i ) is the remaining railway locomotive diesel engine data x to the i-th cluster center c i The Euclidean distance of

[0051] Step 403: recalculate the k cluster centers according to the above formula, and calculate the data average of all the railway locomotive diesel engine sensor data dimensions in the cluster, and reselect the data center point according to the average;

[0052] Step 404: clustering again with the newly selected railway locomotive diesel engine data center point, and repeating the process until the clustering result does not change any more;

[0053] Step 405 calculates the Mahalanobis distance of the clustering result to determine whether the clustering result is abnormal data of a railway locomotive diesel engine. The first clustering result of K-means is used as input. The Mahalanobis distance is used to calculate the cluster center point of the railway locomotive diesel engine data set, and then the distance from the cluster to the center point is calculated respectively. The data distance W is set. kn , when the result of the above formula is greater than the distance threshold, it means that the railway locomotive diesel engine data in the cluster is abnormal data, and the railway locomotive diesel engine data anomaly detection is completed. The anomaly detection error is judged according to the sample set data, and the proxy model optimization technology is used to update k k The distance threshold W between the value and the data kn .

[0054] Specifically, the present invention can establish a railway locomotive diesel engine abnormality analysis model, can quickly identify the abnormal conditions of railway locomotive diesel engines, and then enable engineering and technical personnel to perform maintenance and inspection to ensure the safe and stable operation of railway locomotives. The initial parameter items of the diesel engine clustering abnormality detection model can be optimized during the operation stage by combining the railway locomotive diesel engine clustering abnormality detection model with the agent model optimization technology. The agent model optimization technology includes a polynomial response surface model, a support vector machine regression model, a radial basis function model, and a Kriging regression model combined with the expected improvement criterion. Through this abnormality detection method, the time period to which the abnormal data belongs can be determined, which is convenient for railway locomotive design engineers to further view the data and provide accurate maintenance suggestions to ground maintenance personnel.

[0055] The above description is only illustrative rather than restrictive of the present invention. Those skilled in the art will understand that many modifications, changes or equivalents may be made without departing from the spirit and scope defined by the appended claims, but all will fall within the scope of protection of the present invention.

Claims

1. A method for clustering anomaly detection of railway locomotive diesel engine operating parameters based on agent model optimization technology, characterized in that: The method comprises the following steps: Step 1: Construct a clustering anomaly detection model parameter sample point set; Step 2: Based on the sample point set and proxy model technology, an approximate model of clustering anomaly detection algorithm parameters is constructed; Step 3: Input the initial clustering anomaly detection algorithm parameters into the approximate model, and use the approximate model to estimate the detection effect of the input clustering anomaly detection algorithm parameters; Step 4: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 does not meet the requirements, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set; Step 5: If the estimated detection effect of the clustering anomaly detection algorithm parameters in step 3 meets the requirements, the optimal parameter solution is output, the sample point set is updated, and the approximate model is rebuilt based on the updated sample point set; Step 6: According to the optimal solution, the clustering anomaly detection algorithm is used to perform anomaly analysis on the current data.

2. The method for detecting anomaly of railway locomotive diesel engine operating parameters by clustering according to claim 1 is characterized in that: The clustering algorithm of step 1 is: there are kn railway locomotive diesel engine sensor points, and kk centers are found in the railway locomotive diesel engine sensor data, so that the data point distance expression is: Make the sum of the squares of the distances between the locomotive diesel engine sensor data point and its nearest center point the largest and the kk value the smallest, where x = (W kn , kk) is a sample point.

3. The method for detecting anomaly of railway locomotive diesel engine operating parameters by clustering according to claim 1, characterized in that: in, W kn is the distance between the data points of the locomotive diesel engine sensor, x ki is the kith data center point; z i is the kjth data center point.

4. The method for detecting anomaly of railway locomotive diesel engine operating parameters by clustering according to claim 1, characterized in that: The expression of the radial basis function proxy model in the step 2 in any approximate model is:

5. The method for detecting anomaly of railway locomotive diesel engine operating parameters by clustering according to claim 4 is characterized in that: in, i, n is n = 10, i can be a number from 1 to 10; To estimate the objective function value of clustering algorithm; a i is the coefficient expression; x 0i is the radial basis center point.

6. The railway locomotive diesel engine operating parameter clustering anomaly detection method according to claim 1, characterized in that: The loss function of the approximate model system in step 2 is:

7. The railway locomotive diesel engine operating parameter clustering anomaly detection method according to claim 6, wherein: i, j, m, n, n, m can be any positive integer greater than 10; i can be a number from 1 to n; j can be a number from 1 to m; x j -x 0i is the center point of the radial basis function, which is Scheme 1; E j is the jth solution; a i is a coefficient expression obtained by minimizing the loss function.

8. The method for detecting anomaly of railway locomotive diesel engine operating parameters by clustering according to claim 1, characterized in that: The fourth step comprises: Step 401: Select k data from n railway locomotive diesel engine sensor data as cluster centers; Step 402: Calculate the Euclidean distance from the remaining data to the k cluster centers, which is expressed as: Among them, F(x,c i ) is the remaining railway locomotive diesel engine data x to the i-th cluster center c i The Euclidean distance of Step 403: recalculate the k cluster centers according to the above formula, and calculate the data average of all the railway locomotive diesel engine sensor data dimensions in the cluster, and reselect the data center point according to the average; Step 404: clustering again with the newly selected railway locomotive diesel engine data center point, and repeating the process until the clustering result does not change any more; Step 405 calculates the Mahalanobis distance of the clustering result to determine whether the clustering result is abnormal data of a railway locomotive diesel engine. The first clustering result of K-means is used as input. The Mahalanobis distance is used to calculate the cluster center point of the railway locomotive diesel engine data set, and then the distance from the cluster to the center point is calculated respectively. The data distance W is set. kn , when the result of the above formula is greater than the distance threshold, it means that the railway locomotive diesel engine data in the cluster is abnormal data, and the railway locomotive diesel engine data anomaly detection is completed. The anomaly detection error is judged according to the sample set data, and the proxy model optimization technology is used to update k k The distance threshold W between the value and the data kn .

Images