Water supply pump fault early warning method and device

Abstract

The present application discloses a water supply pump fault early warning method and device, including: obtaining historical data of water supply pump fault related measuring points; preprocessing the historical data by principal component analysis to obtain a training matrix; using an improved DBSCAN algorithm to process the training matrix to obtain a memory matrix, and constructing an MSET early warning model according to the memory matrix; obtaining an observation matrix X in a time period to be estimated. obs , input MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ; Get the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs ‑X est , set the fault warning threshold, and perform fault warning through the sliding window method. The water pump fault warning method provided by this application can achieve early and accurate warning, real-time monitoring, conform to actual operation, high sensitivity of fault warning, etc., and can effectively reduce the occurrence of water supply pump accidents in power plants.

Description

Technical Field

[0001] The present application relates to the technical field of feedwater pumps for gas-steam combined cycle power plants, and in particular to a feedwater pump fault early warning method and device. Background Art

[0002] The feedwater pump is an important auxiliary equipment in a gas-steam combined cycle power plant. The feedwater pump system of the waste heat boiler is the heart of the steam-water cycle. Ensuring its safe and stable operation is a key link in ensuring the safe and stable operation of the unit.

[0003] The working condition of the feedwater pump will affect the entire steam-water system. Once a failure occurs, it will cause the unit load to decrease, which will not only reduce the power generation of the power plant, but also increase the maintenance cost of the power plant. When a serious failure occurs, the power plant will have to shut down, causing huge economic losses to the power plant.

[0004] In the prior art, a warning model is constructed based on a mechanism for fault warning of a water supply pump. However, the working environment of a water supply pump is relatively complex, and the internal working mechanism will change over time. The warning model constructed based on a mechanism is highly complex and cannot detect potential faults of the water supply pump in a timely manner. Summary of the invention

[0005] To this end, the present application provides a water supply pump fault warning method and device to solve the problems of high complexity and inability to timely detect potential faults of the water supply pump in the prior art.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] In a first aspect, a water supply pump failure early warning method comprises:

[0008] Obtain historical data of measurement points related to water supply pump failure;

[0009] Preprocessing the historical data by principal component analysis to obtain a training matrix;

[0010] The training matrix is ​​processed by an improved DBSCAN algorithm to obtain a memory matrix, and an MSET early warning model is constructed according to the memory matrix;

[0011] Get the observation matrix X within the time period to be estimated obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ;

[0012] Get the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est, set the fault warning threshold, and perform fault warning through the sliding window method.

[0013] Furthermore, the historical data is preprocessed by principal component analysis to obtain a training matrix, which specifically includes:

[0014] The historical data is subjected to Z-score normalization processing by the first formula to obtain a normalized matrix Z;

[0015] Calculate the correlation coefficient matrix R of the standardized matrix Z according to the second formula;

[0016] Calculate the eigenvalue λ and eigenvector ν of the correlation coefficient matrix R, and sort the eigenvalues ​​in ascending order;

[0017] Calculate the contribution rate C of each monitoring parameter variable of the water supply pump according to the third formula;

[0018] The cumulative contribution rate of each monitoring parameter variable is calculated according to the fourth formula;

[0019] The first formula is:

[0020]

[0021] in, S j are the mean and standard deviation of the j-th state variable for one month,

[0022]

[0023]

[0024] The second formula is:

[0025]

[0026] The third formula is:

[0027]

[0028] The fourth formula is:

[0029]

[0030] Furthermore, the number of principal elements of the training matrix is ​​5, and the cumulative contribution rate is 96.763%.

[0031] Furthermore, the improved DBSCAN algorithm is used to process the training matrix to obtain a memory matrix, which specifically includes:

[0032] Query the Eps neighborhood of each node and calculate the density in each neighborhood according to the fifth formula;

[0033] The average density and variance of each node are calculated according to the sixth and seventh formulas respectively;

[0034] Calculate the density variation coefficient cv value of each node according to the eighth formula;

[0035] The data is partitioned according to the equal depth block method;

[0036] Set the Eps value for each partition;

[0037] Run the DBSCAN algorithm on each data partition;

[0038] Get the clustering results and form the memory matrix D with the results;

[0039] The fifth formula is:

[0040]

[0041] Where Eps is the radius; Pts(i) is the point set with center i and radius Eps; |Pts(i)| is the number of elements; ρ i is the density within the neighborhood;

[0042] The sixth formula is:

[0043]

[0044] The seventh formula is:

[0045]

[0046] Where n = |Pts(i)|;

[0047] The eighth formula is:

[0048]

[0049] Furthermore, the setting of the Eps value of each partition specifically includes:

[0050] Determine the value of minPts;

[0051] Calculate the number of nodes in the Eps neighborhood of each point according to the ninth formula;

[0052] Calculate the radius of each point in each partition according to the tenth formula;

[0053] Calculate the sum of Eps(i) according to the eleventh formula and take the average;

[0054] The ninth formula is:

[0055] |Pts(i)|=ρi *π*Eps 2 ;

[0056] The tenth formula is:

[0057]

[0058] The eleventh formula is:

[0059]

[0060] Furthermore, when determining the value of minPts, when the cv value is smaller, it is taken as 4; when the cv value is larger, it is taken as 2.

[0061] Furthermore, the memory matrix is ​​a 5×1308 memory matrix.

[0062] Furthermore, when the fault warning is performed by the sliding window method, the warning coefficient k is 1.2 and the warning threshold is 0.4.

[0063] Furthermore, the historical data of the measurement points related to the water supply pump failure are obtained, including 22 state variables.

[0064] In a second aspect, a water supply pump failure warning device comprises:

[0065] State variable module, used to obtain historical data of measurement points related to water supply pump failure;

[0066] A principal component analysis module, used for preprocessing the historical data by principal component analysis to obtain a training matrix;

[0067] An MSET early warning model construction module is used to process the training matrix using an improved DBSCAN algorithm to obtain a memory matrix, and to construct an MSET early warning model according to the memory matrix;

[0068] The estimation matrix calculation module is used to obtain the observation matrix X within the time period to be estimated. obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ;

[0069] Fault warning module, used to obtain the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est , set the fault warning threshold, and perform fault warning through the sliding window method.

[0070] Compared with the prior art, this application has at least the following beneficial effects:

[0071] The present application provides a water supply pump fault early warning method and device, including: obtaining historical data of water supply pump fault related measurement points; preprocessing the historical data by principal component analysis to obtain a training matrix; using an improved DBSCAN algorithm to process the training matrix to obtain a memory matrix, and constructing an MSET early warning model according to the memory matrix; obtaining an observation matrix X in a time period to be estimated obs , input MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ; Get the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est , set the fault warning threshold, and perform fault warning through the sliding window method. The water pump fault warning method provided by this application can achieve early and accurate warning, real-time monitoring, conform to actual operation, high sensitivity of fault warning, etc., and can effectively reduce the occurrence of water supply pump accidents in power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more intuitively illustrate the prior art and the present application, several exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing the present application; for example, those skilled in the art are capable of easily making conventional adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components) based on the technical concepts and exemplary drawings disclosed in the present application.

[0073] Figure 1 A first flow chart of the water supply pump failure early warning method provided in this application;

[0074] Figure 2 A second flow chart of the water supply pump failure early warning method provided in this application;

[0075] Figure 3 The estimated and actual values ​​of the seal water outlet temperature provided for this application;

[0076] Figure 4 The estimated and actual values ​​of the feedwater pump outlet pressure provided for this application;

[0077] Figure 5 Sliding residual diagram of the early warning model provided for this application. DETAILED DESCRIPTION

[0078] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0079] In the description of this application: unless otherwise specified, the meaning of "plurality" is two or more. The terms "first", "second", "third", etc. in this application are intended to distinguish the objects referred to, and do not have special meanings in terms of technical connotations (for example, they should not be understood as emphasizing the importance or order, etc.). Expressions such as "including", "comprising", "having", etc. also mean "not limited to" (certain units, components, materials, steps, etc.).

[0080] The terms such as "upper", "lower", "left", "right", "middle", etc. used in this application are usually used to facilitate intuitive understanding by comparing with the accompanying drawings, and are not absolute limitations on the positional relationship in the actual product. Without departing from the technical concept disclosed in this application, changes in these relative positional relationships should also be regarded as the scope of this application.

[0081] See also Figure 1 and Figure 2 The present application provides a water supply pump failure early warning method, comprising:

[0082] S1: Obtain historical data of relevant measuring points of water supply pump failure;

[0083] Historical data of measurement points related to feedwater pump failures are selected from the power plant SIS system, and abnormal data are eliminated. The remaining data are constructed into an initial matrix. The size of the initial matrix is ​​m*n, where m represents the number of operating parameters related to feedwater pump failures, and n represents the total number of operating states. Because the sampling periods of different measurement points are different, this application selects a sampling period of 5 minutes.

[0084] Specifically, the operating parameters related to feedwater pump failure include feedwater pump inlet pressure, feedwater pump inlet temperature, feedwater pump outlet main pipe pressure, recirculation valve position, feedwater pump inlet flow, inlet filter differential pressure, feedwater pump speed, feedwater pump throttle opening, unit load, and 22 state variables such as vibration and temperature signals of each bearing.

[0085] Table 1: Main operating parameters of water supply pump

[0086]

[0087]

[0088] S2: preprocessing the historical data by principal component analysis to obtain a training matrix K;

[0089] Because there are many measuring points for the water supply pump, if 22 state variables are used for modeling, the speed of the algorithm will be affected, and some measuring points have almost no effect on the failure of the water supply pump. Therefore, principal component analysis (PCA) is used to extract fault features from the data, thereby reducing the complexity of the early warning model, achieving data dimensionality reduction, and improving the efficiency of the early warning model.

[0090] Specifically, data preprocessing includes the following steps:

[0091] S21: Perform Z-score normalization on the historical data according to formula (1) to obtain a normalized matrix Z;

[0092]

[0093] in,

[0094]

[0095]

[0096] S j are the mean and standard deviation of the j-th state variable for one month, respectively;

[0097] S22: Calculate the correlation coefficient matrix R of the standardized matrix Z according to formula (2);

[0098]

[0099] Specifically, the correlation coefficient matrix R can reflect the correlation coefficient between each measuring point;

[0100] S23: Calculate the eigenvalue λ and eigenvector v of the correlation coefficient matrix R, and sort the eigenvalues ​​in ascending order;

[0101] Specifically, λ1>λ2>…λ n ;

[0102] S24: Calculate the contribution rate C of each monitoring parameter variable of the water supply pump according to formula (3);

[0103]

[0104] S25: Calculate the cumulative contribution rate of each monitoring parameter variable according to formula (4);

[0105]

[0106] Table 2: PCA contribution rate

[0107]

[0108] In this way, the main monitoring parameters of the water supply pump are determined. When the cumulative contribution rate is ≥85%, it means that the k principal components contain most of the information of the original data. After simulation, k is 5, and the cumulative variance contribution rate is 96.763%, which means that through the change of PCA coordinates, the status of the water supply pump can be reflected by these five state parameters, achieving the purpose of dimensionality reduction, and obtaining a 5×7200 training matrix K.

[0109] S3: Using the improved DBSCAN algorithm to process the training matrix K, obtaining the memory matrix D, and constructing the MSET early warning model according to the memory matrix D;

[0110] Although the number of state parameters has been reduced through principal component analysis, the inventor selected the water supply pump state for one month, and the amount of historical data of the water supply pump is still large.

[0111] Constructing the memory matrix D is the most important part in the multi-state technology. The most ideal way is to select all historical data to construct the memory matrix D, but computers cannot handle such a large amount of data. If a part of the historical data is selected, it will not be able to cover all normal working conditions. Therefore, the inventor uses an improved DBSCAN algorithm to cluster the historical data, reduce the amount of historical data, and make the clustered data reflect all states under normal working conditions of the water pump.

[0112] DBSCAN is a density-based clustering algorithm that can cluster data based on the closeness between them. By dividing samples with close distances into one category, a cluster category is obtained. The closeness is generally determined by calculating the Euclidean distance between sample points. All samples are divided into different categories by setting the radius Eps and the minimum number of samples in the domain minPts. The setting of these two parameters has a great influence on the clustering effect, so setting appropriate Eps and minPts is the key to this algorithm.

[0113] However, the traditional DBSCAN will produce a large number of outliers when processing high-dimensional density-inhomogeneous data. Therefore, the improvement of DBSCAN specifically introduces the idea of ​​density detection, which includes the following steps:

[0114] S31: query the Eps neighborhood of each node and calculate the density in each neighborhood according to formula (5);

[0115]

[0116] Where Eps is the radius; Pts(i) is the point set with center i and radius Eps; |Pts(i)| is the number of elements; ρ i is the density within the neighborhood;

[0117] The other points in the circular neighborhood of node i are denoted as k, and their densities are calculated respectively;

[0118] S32: Calculate the average density and variance of each node according to equations (6) and (7);

[0119]

[0120]

[0121] Where n = |Pts(i)|;

[0122] S33: Calculate the density variation coefficient cv value of each node according to formula (8), and substitute the variance calculated in S32 into formula (2);

[0123]

[0124] S34: partitioning the data according to the equal depth block method;

[0125] Specifically, for example, there are 100 data points, with an average of 5 in each box, for a total of 20 boxes. Define a threshold λ, divide the range of one box by the range of the next box, and split it if it is greater than λ. The split point is the maximum of this box or the minimum of the next box.

[0126] S35: Setting the Eps value of each partition;

[0127] The Eps value of each partition is defined as the average distance between each point and the first |Pts(i)| points in its neighborhood.

[0128] S351: Determine the value of minPts. When the cv value is small, it is set to 4; when the cv value is large, it is set to 2;

[0129] S352: Calculate the number of nodes in the Eps neighborhood of each point according to formula (9);

[0130] |Pts(i)|=ρ i *π*Eps 2 (9)

[0131] S353: Calculate the radius of each point in each partition according to formula (10);

[0132]

[0133] S354: Calculate the sum of Eps(i) according to formula (11) and take the average;

[0134]

[0135] This average radius is the Eps value of this partition.

[0136] S36: Run the DBSCAN algorithm on each data partition;

[0137] S37: Obtain clustering results, and form a memory matrix D with the results;

[0138] After clustering with the improved DBSCAN algorithm, a 5×1308 matrix is ​​finally obtained, and this memory matrix contains the normal historical data of the water pump for this month, so an early warning model can be established through multivariate state technology.

[0139] S4: Get the observation matrix X within the time period to be estimated obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est :

[0140] For a new observation vector X of a process or device at a certain moment obs MSET compares this state with the constructed memory matrix and calculates the estimated matrix X est , the estimated vector is the product of the process memory matrix and the weight vector, and the formula is:

[0141] X est =D*W=D*[w1,w2…w m ] T (12)

[0142] The weight vector W represents a similarity measure between this state and the state in the process memory matrix.

[0143] S5: Obtain the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est , set the fault warning threshold and perform fault warning through the sliding window method;

[0144] The weight vector can be obtained by minimizing the residual ε vector, constructing the residual between the new observation vector and the estimated vector;

[0145] ε=X est -X obs (13)

[0146] The smaller ε is, the more accurate the prediction result is. The minimum square error for ε is calculated as:

[0147] minε=min(|X est -X obs |)

[0148] =min(|Xobs -D*W|)

[0149] minε 2 =min[(X obs -D*W) T *(X obs -D*W)]

[0150] make get:

[0151] W=(D T *D) -1 *(D T *X obs )

[0152] But (D T *D) There is an irreversible situation because there may be a certain correlation between the various measuring points of the water supply pump. The point multiplication in the above formula can be replaced by a nonlinear sub-operation. The nonlinear sub-operation used here can be calculated using the Euclidean distance.

[0153]

[0154] A i and B j is the column vector in matrices A and B, a ik and b jk Yes A i , B j The elements in .

[0155] Substituting formula (9) into formula (6), the estimated vector corresponding to the state of the water pump on that day can be obtained as:

[0156]

[0157] The state of the equipment during operation can be reflected by the residual between the input state vector and the output estimation vector. When the residual exceeds the specified threshold, the water pump may have failed.

[0158] However, the working environment of the water supply pump is complex and will be affected by external factors. By using the sliding window method to reasonably select the window width, the continuous changes of the residual statistical characteristics can be obtained in a timely and rapid manner, thereby eliminating these influences. Assume that within a certain period of time, there are n residual sequences based on MSET.

[0159] ε=[ε1ε2……ε n ]

[0160] Take a sliding window with a width of N for the residual sequence and calculate the average value of N consecutive residuals in the window:

[0161]

[0162] Let the maximum value of the sliding window mean sequence be E max , then the warning threshold

[0163] E w =k*E max

[0164] In the formula, k is the warning threshold coefficient, generally k>1

[0165] The value of the warning coefficient k is generally determined by on-site operating experience, and is usually 1.2. If the value of k is too large, the threshold will be set too large, which will reduce the sensitivity of the fault warning and make it impossible to give a sufficient early warning; if the value of k is too small, the threshold will be set too low, which will reduce the accuracy of the fault warning.

[0166] In order to verify the method, the inventor selected sample data from February 1 to February 25, 2022 to construct a memory matrix. The sample data is a 22×7200 matrix. The data from February 26 to February 28, 2022 is used to verify the feasibility of the early warning model.

[0167] See also Figure 3 and Figure 4 , is a comparison of the observed and estimated values ​​of the sealing water outlet temperature and the water pump outlet pressure. It can be seen that the observed and estimated values ​​of the water pump outlet pressure are almost the same, and the sealing water outlet temperature has a certain deviation from the observed and estimated values ​​only at a few moments, but the point with the largest deviation does not exceed 0.5℃, which shows that the estimation effect is good, and the improved multivariate state estimation has a higher warning accuracy.

[0168] See also Figure 5 , Figure 5 This is the sliding residual diagram of the early warning model from February 26 to February 28 (a total of 864 sample points). The early warning threshold selected in this paper is 0.4. At the 302nd sample point (1 a.m. on February 27), the sliding residual exceeded the early warning threshold, indicating that the water supply pump had failed at this time. At the 342nd sample point (4:30 a.m. on February 27), the operation and maintenance personnel found that the water pump was blocked, resulting in insufficient flow, and carried out maintenance to eliminate the fault. The water supply pump fault early warning method using improved DBSCAN and MSET can detect faults 3.5 hours earlier than manual inspections, which can realize early fault warning of the water supply pump and reduce the economic losses of the power plant when a fault occurs.

[0169] The present application provides a water supply pump failure early warning device, comprising:

[0170] State variable module, used to obtain historical data of measurement points related to water supply pump failure;

[0171] A principal component analysis module, used for preprocessing the historical data by principal component analysis to obtain a training matrix;

[0172] An MSET early warning model construction module is used to process the training matrix using an improved DBSCAN algorithm to obtain a memory matrix, and to construct an MSET early warning model according to the memory matrix;

[0173] The estimation matrix calculation module is used to obtain the observation matrix X within the time period to be estimated. obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ;

[0174] The fault warning module is used to obtain the residual matrix RL=Xobs-Xest between the estimation matrix Xest and the observation matrix Xobs, set the fault warning threshold, and perform fault warning by a sliding window method.

[0175] For the specific limitations of the water supply pump fault warning device, please refer to the limitations of the water supply pump fault warning method mentioned above, which will not be repeated here.

[0176] The present invention has the following advantages over the prior art:

[0177] (1) Using multivariate state estimation technology to establish a fault warning model for the water supply pump, it can provide real-time warning of potential faults in the water supply pump. When the equipment has a hidden fault, the warning model compares the residual change trend with the set threshold to achieve accurate fault warning in the early stage of the fault.

[0178] (2) The PCA algorithm was used to preprocess the data, and 5 principal components were used to replace the original 22 variable parameters. The cumulative contribution rate reached 96.763%, which achieved data dimensionality reduction, extracted fault features, and reduced the complexity of modeling.

[0179] (3) The memory matrix is ​​constructed through the improved DBSCAN algorithm. The improved DBSCAN algorithm introduces the basic idea of ​​density detection, which solves the problem that the DBSCAN algorithm will generate a large number of outliers when processing high-dimensional density uneven data. It also sets two important parameters, namely the optimal radius and the minimum number of nodes, which greatly reduces the amount of historical data and improves the speed and accuracy of the early warning model.

[0180] The technical features of the above embodiments may be arbitrarily combined (as long as there is no contradiction in the combination of these technical features). To make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

[0181] The above is a relatively specific and detailed description of the present application through general description and specific embodiments. It should be understood that, based on the technical concept of the present application, several conventional adjustments or further innovations can be made to these specific embodiments; however, as long as they do not deviate from the technical concept of the present application, the technical solutions obtained by these conventional adjustments or further innovations also fall within the scope of protection of the claims of the present application.

Claims

1. A water pump failure early warning method, characterized in that: include: Obtain historical data of measurement points related to water supply pump failure; Preprocessing the historical data by principal component analysis to obtain a training matrix; The training matrix is ​​processed by an improved DBSCAN algorithm to obtain a memory matrix, and an MSET early warning model is constructed according to the memory matrix; Get the observation matrix X within the time period to be estimated obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ; Get the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est , set the fault warning threshold and perform fault warning through the sliding window method; The improved DBSCAN algorithm is used to process the training matrix to obtain a memory matrix, which specifically includes: Query the Eps neighborhood of each node and calculate the density in each neighborhood according to the fifth formula; The average density and variance of each node are calculated according to the sixth and seventh formulas respectively; Calculate the density variation coefficient cv value of each node according to the eighth formula; The data is partitioned according to the equal depth block method; Set the Eps value for each partition; Run the DBSCAN algorithm on each data partition; Get the clustering results and form the memory matrix D with the results; The fifth formula is: Where Eps is the radius; Pts(i) is the point set with center i and radius Eps; |Pts(i)| is the number of elements; ρ i is the density within the neighborhood; The sixth formula is: The seventh formula is: Where n = |Pts(i)|; The eighth formula is: The setting of the Eps value of each partition specifically includes: Determine a value of MinPts; wherein the value of MinPts is related to the cv value; Calculate the number of nodes in the Eps neighborhood of each point according to the ninth formula; Calculate the radius of each point in each partition according to the tenth formula; Calculate the sum of Eps(i) according to the eleventh formula and take the average; The ninth formula is: Pts(i)|=ρ i *π*Eps 2 ; The tenth formula is: The eleventh formula is:

2. The water supply pump failure early warning method according to claim 1 is characterized in that: The preprocessing of the historical data by principal component analysis to obtain a training matrix specifically includes: The historical data is subjected to Z-score normalization processing by the first formula to obtain a normalized matrix Z; Calculate the correlation coefficient matrix R of the standardized matrix Z according to the second formula; Calculate the eigenvalue λ and eigenvector v of the correlation coefficient matrix R, and sort the eigenvalues ​​in ascending order; Calculate the contribution rate C of each monitoring parameter variable of the water supply pump according to the third formula; The cumulative contribution rate of each monitoring parameter variable is calculated according to the fourth formula; The first formula is: in, S j are the mean and standard deviation of the j-th state variable for one month, The second formula is: The third formula is: The fourth formula is:

3. The water supply pump failure early warning method according to claim 2 is characterized in that: The number of principal elements in the training matrix is ​​5, and the cumulative contribution rate is 96.763%.

4. The water supply pump failure early warning method according to claim 1 is characterized in that: The memory matrix is ​​a 5×1308 memory matrix.

5. The water supply pump failure early warning method according to claim 1 is characterized in that: When the fault warning is performed by the sliding window method, the warning coefficient k is 1.2 and the warning threshold is 0.

4.

6. The water supply pump failure early warning method according to claim 1, characterized in that: The historical data of the measurement points related to the water supply pump failure are obtained, including 22 state variables.

7. A water pump failure warning device, characterized in that: include: State variable module, used to obtain historical data of measurement points related to water supply pump failure; A principal component analysis module, used for preprocessing the historical data by principal component analysis to obtain a training matrix; An MSET early warning model construction module is used to process the training matrix using an improved DBSCAN algorithm to obtain a memory matrix, and to construct an MSET early warning model according to the memory matrix; The estimation matrix calculation module is used to obtain the observation matrix X within the time period to be estimated. obs , input the MSET early warning model and use multivariate state estimation to obtain the estimation matrix X est ; Fault warning module, used to obtain the estimated matrix X est With the observation matrix X obs The residual matrix RL between obs -X est , set the fault warning threshold and perform fault warning through the sliding window method; The improved DBSCAN algorithm is used to process the training matrix to obtain a memory matrix, which specifically includes: Query the Eps neighborhood of each node and calculate the density in each neighborhood according to the fifth formula; The average density and variance of each node are calculated according to the sixth and seventh formulas respectively; Calculate the density variation coefficient cv value of each node according to the eighth formula; The data is partitioned according to the equal depth block method; Set the Eps value for each partition; Run the DBSCAN algorithm on each data partition; Get the clustering results and form the memory matrix D with the results; The fifth formula is: Where Eps is the radius; Pts(i) is the point set with center i and radius Eps; |Pts(i)| is the number of elements; ρ i is the density within the neighborhood; The sixth formula is: The seventh formula is: Where n = |Pts(i)|; The eighth formula is: The setting of the Eps value of each partition specifically includes: Determine the value of MinPts; the value of MinPts is related to the cv value; calculate the number of nodes in the Eps neighborhood of each point according to the ninth formula; calculate the radius of each point in each partition according to the tenth formula; Calculate the sum of Eps(i) according to the eleventh formula and take the average; The ninth formula is: |Pts(i)|=ρ i *π*Eps 2 ; The tenth formula is: The eleventh formula is:

Images