A pump device characteristic curve correction method based on cluster analysis and a confidence domain algorithm

Abstract

The application discloses a water pump device characteristic curve correction method based on clustering analysis and a trust region algorithm, and comprises the following steps: clustering steady-state working condition data of a water pump device by using a K-means clustering method, a large number of similar working conditions are combined through iteration, and typical working condition points of the water pump device are screened out; based on the typical working condition points of the water pump device, an original characteristic curve of the water pump device is obtained by using a moving least square method or cubic spline or polynomial fitting; a three-dimensional affine transformation is realized by introducing parameter modification to the original characteristic curve of the water pump device, and the characteristic curve correction of the water pump device is realized based on a Dogbox algorithm in the trust region algorithm. The method has the advantages that: the method does not need to carry out model tests or field tests, but is corrected according to basic operation monitoring data of general large axial flow / mixed flow pump stations, is relatively strong in universality, can still maintain a certain accuracy in a working condition interval with insufficient data, and can provide scientific basis and technical support for actual scheduling of the pump station.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large water pumping station optimal scheduling, and particularly relates to a water pump device characteristic curve correction method based on clustering analysis and a trust region algorithm. BACKGROUND

[0002] As an important regulating building in inter-basin water transfer projects, the pump station will change the traditional scheduling mode dominated by artificial experience into scientific, efficient and intelligent decision-making in the process of promoting digital twin wisdom construction, so as to achieve the goal of cost reduction and efficiency improvement. The flow-lift-blade angle characteristic curve of the pump device is a curve that describes the relationship and change law between the three performance parameters of pump flow, lift and blade setting angle. It reflects the running state of the pump, and its accuracy directly affects the scheduling decision-making.

[0003] At present, the flow-lift-blade angle characteristic curve of the pump device used in the scheduling work is mostly obtained through model test when it leaves the factory. In the running process, due to the influence of factors such as installation condition, inflow condition, equipment degradation and scale effect, there are differences between the actual running state and the ideal working condition, the characteristic curve is offset, and the pump station optimal scheduling model based on this cannot calculate the actual running scheme. In order to ensure the accuracy of the pump station scheduling decision, it is necessary to correct the flow-lift-blade angle characteristic curve of the pump device. However, considering the engineering operation safety, the actual running condition of the pump station is generally concentrated in a small design running interval, and it is difficult to obtain the measured data under the remaining stable working condition through on-site unit test.

[0004] After consulting relevant literature, there are many studies on the prediction and fitting of pump device characteristic curve, but there are few studies on the offset and correction of characteristic curve in engineering practice. The existing methods cannot provide technical support for correcting the flow-lift-blade angle characteristic curve of the pump device. SUMMARY

[0005] The purpose of the present application is to provide a water pump device characteristic curve correction method based on clustering analysis and a trust region algorithm, thereby solving the aforementioned problems in the prior art.

[0006] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0007] A water pump device characteristic curve correction method based on clustering analysis and a trust region algorithm,

[0008] S1, selecting typical working condition points through clustering analysis:

[0009] The K-means clustering method is used to cluster the steady state working condition data of the pump device, a large number of similar working conditions are merged through iteration, and the typical working condition points of the pump device are selected;

[0010] S2, original characteristic curve extraction:

[0011] Based on the typical working points of the water pump device, the original characteristic curve of the water pump device is obtained by using the moving least square method or cubic spline or polynomial fitting;

[0012] S3, characteristic curve correction based on typical working points:

[0013] The original characteristic curve of the water pump device is introduced to realize three-dimensional affine transformation, and the Dogbox algorithm in the trust region algorithm is used to realize the characteristic curve correction of the water pump device.

[0014] Preferably, the K-means clustering method in step S1 specifically includes the following contents,

[0015] S11, selecting K discrete initial clustering centers by K-means++ algorithm;

[0016] S12, assigning the data set to the cluster corresponding to the clustering center by classical K-means++ algorithm;

[0017] S13, calculating the mean value of each class cluster and taking this mean value as the new clustering center;

[0018] S14, repeating steps S12 to S13 until the clustering center is unchanged or the change is less than the change threshold;

[0019] S15, using the elbow method to determine whether it is the best K value, if not, go to step S11, if yes, divide the steady state working condition data of the water pump device into K clustering centers, and select the typical working points of the water pump device.

[0020] Preferably, step S2 specifically includes,

[0021] The flow-head-blade angle characteristic curve of the water pump device is fitted by using the method of binary quadratic polynomial, and the original characteristic curve equation of the water pump device is obtained, which has the following form:

[0022] θ=p00+p10*Q+p01*H+p20*Q 2 +p11*Q*H+p02*H 2

[0023] Wherein, θ is the blade angle of the water pump device; Q is the flow of the water pump device; H is the head of the water pump device; p00, p10, p01, p20, p11, p02 are respectively the coefficients of each term in the fitted polynomial equation.

[0024] Preferably, step S3 specifically includes the following contents,

[0025] S31, introduce parameters a, b, c, r to modify the original characteristic curve equation obtained in step S2 to realize three-dimensional affine transformation, and the linear transformed equation is as follows:

[0026]

[0027] Wherein, a, b, c, r are new parameters introduced by the modified equation; Q' is the flow value of the corrected water pump device; H' is the head value of the corrected water pump device; θ' is the blade installation angle value of the corrected water pump device;

[0028] Suppose that the normal operation condition interval of the water pump device is Q∈[Q min ,Q max ], H∈[H min ,H max ], θ∈[θ min ,θ max ]; and the search domains are respectively a∈[-0.3*|Q min -Q max |, 0.3*|Q min -Q max |], b∈[-0.3*|H min -H max |, 0.3*|H min -H max |], c∈[-0.3*|θ min -θ max |, 0.3*|θ min -θ max |]; Q min and Q max are the minimum flow and the maximum flow of the water pump device respectively; H min and H max are the minimum head and the maximum head of the water pump device respectively; θ min and θ max are the minimum blade installation angle and the maximum blade installation angle of the water pump device respectively.

[0029] S32, based on the Dogbox algorithm in the confidence domain algorithm, the fitting parameters are determined by minimizing the residual sum of squares, that is, the characteristic curve is translated and rotated in the three-dimensional space concept, so that the characteristic curve approximates the selected typical working condition points, and the characteristic curve correction of the water pump device is completed.

[0030] Preferably, step S1 further comprises,

[0031] S0, data collection and pretreatment: according to the actual engineering scheduling operation of different time scales, the water regime, the work regime related data of the water pump device are statistically summarized, the data are processed by using the data cleaning method, and the steady state working condition data of the water pump device are screened out by the self-adaptive gaussian filtering method and the confidence method.

[0032] Preferably, the water regime, the work regime related data include the station water level, the station water level, the flow monitoring value of each unit, the blade angle monitoring value of each unit of the water pump device;

[0033] The data processing includes eliminating null, outlier and abnormal value.

[0034] Preferably, in step S0, the adaptive gaussian filtering method is used to filter noise, and the calculation formula is,

[0035] S(t,σ)=f(t)*g(t,σ)

[0036] Wherein, S(t,σ) is the smoothing result of the signal or data after gaussian filtering; f(t) is the original signal or data sequence, which is a function of time t; t is the time variable; g(t,σ) is the gaussian kernel function; * is the convolution operator; σ is the standard deviation of the gaussian function;

[0037]

[0038] Wherein, m is the sampling time point; f(m) is the historical data signal; f k+1 (m) is the output value after k+1 iteration smoothing; f k (m+j) is the value of signal or data at position m+j in the kth iteration; w k (m+j,σ) is the weight value of each point in the window; j is the offset in the convolution window, which represents the time offset relative to m; M is the width parameter of the convolution window.

[0039] Preferably, the confidence method is,

[0040] After the data is denoised by the adaptive gaussian filtering method, the error probability distribution function of the direct monitoring index steady state data is obtained, so that the change threshold of the index is determined under the premise of a certain confidence interval, as the basis for judging the steady state; when all the indexes are in steady state, it is considered that the working condition point is the steady state working condition point.

[0041] Preferably, after step S3, it further includes,

[0042] S4, extraction of corrected characteristic curve: draw the corrected flow-head-blade angle characteristic curve through origin software.

[0043] Preferably, step S4 is,

[0044] Discretize the corrected characteristic curve equation obtained in step S3 using the origin software to draw a corrected integer angle contour map, that is, a corrected flow-head-blade angle characteristic curve map.

[0045] The beneficial effects of the present application are: 1. The present application screens stable state working condition data from long sequence monitoring data by adaptive Gaussian filtering method combined with confidence method, which can retain the original signal change information to the greatest extent while filtering out noise. 2. The present application performs clustering processing on long sequence historical data based on K-means clustering method, which greatly reduces the data amount, improves the calculation speed, makes the data points cover as many working conditions as possible, and avoids the curve overfitting phenomenon that may occur in the linear regression process due to the excessive density of point set in the partial common working condition interval during the operation of the same pump station. 3. The present application introduces a new parameter to transform the original characteristic curve equation by affine transformation, which makes the curve as close to the data points as possible without destroying the original curve spatial structure, avoids abnormal deformation of the characteristic curve caused by direct fitting which does not conform to the hydraulic characteristics of the pump, and thus improves the curve correction quality. 4. The present application performs parameter optimization based on the Dogbox algorithm in the confidence domain algorithm, which has good robustness, stability and convergence characteristics when solving the high-dimensional nonlinear parameter optimization problem that may occur due to the introduction of a new parameter. 5. The present application has strong universality, and can be used to process and generate corrected pump device characteristic curves for axial flow / mixed flow pump stations with long sequence historical monitoring data. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a flowchart of the method in the embodiment of the present application;

[0047] Figure 2 is a corrected pump flow-head-blade angle characteristic curve map (vertical line) in the embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application clearer and more apparent, the present application will be further described in detail below with reference to the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0049] As shown in Figure 1 , in the embodiment, a pump device characteristic curve correction method based on clustering analysis and confidence domain algorithm is provided, which includes the following five parts,

[0050] I. Data collection and preprocessing

[0051] According to the actual engineering scheduling operation of different time scales, the water regime, work regime and other related data of the water pump device are statistically summarized, the data is processed by using data cleaning method, and the steady state working condition data of the water pump device is screened out by using self-adaptive Gaussian filtering method and confidence method.

[0052] Specifically: according to the actual engineering scheduling operation of different time scales, the water regime, work regime and other related data are statistically summarized. The data mainly includes the water level below the pump station, the water level above the pump station, the flow monitoring value of each unit of the pump station, the blade angle monitoring value of each unit of the pump station and other water / working condition data. After removing the null value, outlier value and abnormal value by data cleaning, the original signal change information is retained to the greatest extent by using self-adaptive Gaussian filtering method while filtering noise; the filtering process of self-adaptive Gaussian filtering method is,

[0053] S(t,σ)=f(t)*g(t,σ)

[0054] Wherein, S(t,σ) is the smoothing result of the signal or data after Gaussian filtering; f(t) is the original signal or data sequence, which is a function of time t; t is the time variable; g(t,σ) is the Gaussian kernel function; * is the convolution operator; σ is the standard deviation of Gaussian function.

[0055]

[0056] Wherein, m is the sampling time point; f(m) is the historical data signal; f k+1 (m) is the output value after k+1 times iteration smoothing; f k (m+j) is the value of signal or data at position m+j in the kth iteration; w k (m+j,σ) is the weight value of each point in the window; j is the offset in the convolution window, which represents the time offset relative to m; M is the width parameter of the convolution window.

[0057] The confidence method is as follows: after removing the noise, the error probability distribution function of the steady state data of the direct monitoring indicators such as flow Q and water level H is obtained, so that the change threshold of the indicators is determined under the premise of a certain confidence interval, which is used as the basis for judging the steady state. When all the indicators are in steady state, it can be considered that the working condition point is the steady state working condition point of the water pump device.

[0058] II. Cluster analysis to screen typical working condition points

[0059] The steady state working condition data of the water pump device is clustered by using K-means clustering method, a large number of similar working conditions are merged by iteration, and the typical working condition points of the water pump device are screened out, so that the influence of the excessive density of working condition points in some interval on the correction effect of the whole is avoided.

[0060] Specifically: because the same pump station has similar data points and high similarity in the unit operation process, the point set density is too large in the common working condition interval, and the density is too small in the rare working condition interval, which may cause overfitting in the correction process; therefore, the K-means clustering method is used to cluster the long sequence of historical data, and a large number of similar working conditions are combined through iteration, so that the subsequent water pump device characteristic curve correction is faster and more accurate. The K-means clustering method specifically includes the following steps,

[0061] 2.1, selecting K discrete initial cluster centers by K-means++ algorithm.

[0062] 2.2, assigning the data set to the cluster center corresponding to the cluster by the classic K-means++ algorithm.

[0063] 2.3, calculating the mean of each cluster and taking the mean as the new cluster center.

[0064] 2.4, repeating steps 2.2 and 2.3 until the cluster center is unchanged or changes very little (such as less than a set threshold).

[0065] 2.5, using the elbow method to determine whether it is the best K value, if not, go to step 2.1, if yes, divide the typical working condition points into K cluster centers to form the correction training set (i.e. typical working condition points) of the unit.

[0066] Three, original characteristic curve extraction

[0067] Based on the typical working condition points of the water pump device, the original characteristic curve of the water pump device is obtained by using the moving least square method or the cubic spline or the polynomial fitting.

[0068] Specifically: the water pump device characteristic curve is numerically processed, and various methods such as moving least square method, cubic spline and polynomial can be used, and the method of binary quadratic polynomial is used to fit the flow-head-blade angle characteristic curve of the water pump device, which is as follows,

[0069] θ=p00+p10*Q+p01*H+p20*Q 2 +p11*Q*H+p02*H 2

[0070] Wherein, θ is the blade setting angle of the water pump device (unit °); Q is the flow of the water pump device (unit m 3 / s); H is the head of the water pump device (unit m). p00, p10, p01, p20, p11, p02 are the coefficients of each term in the fitted polynomial equation.

[0071] Four, characteristic curve correction based on typical working condition points

[0072] The original characteristic curve of the water pump device is modified by introducing parameters to realize three-dimensional affine transformation, and the characteristic curve of the water pump device is corrected based on the Dogbox algorithm in the trust region algorithm.

[0073] Specifically, the initial characteristic curve equation is modified by introducing appropriate parameters to realize three-dimensional affine transformation. The characteristic curve equation obtained by introducing parameters a, b, c, and r is modified, and the linearly transformed equation is as follows:

[0074]

[0075] Wherein, a, b, c, and r are new parameters introduced in the modified equation; Q' is the flow value of the corrected water pump device; H' is the head value of the corrected water pump device; and θ' is the blade installation angle value of the corrected water pump device.

[0076] Suppose that the normal operation condition interval of the pump group is Q∈[Q min ,Q max ], H∈[H min ,H max ], and θ∈[θ min ,θ max ]; then the search domains are defined as a∈[-0.3*|Q min -Q max |, 0.3*|Q min -Q max |], b∈[-0.3*|H min -H max |, 0.3*|H min -H max |], and c∈[-0.3*|θ min -θ max |, 0.3*|θ min -θ max |]; Q min and Q max are the minimum flow and maximum flow of the water pump device, respectively; H min and H max are the minimum head and maximum head of the water pump device, respectively; and θ min and θ max are the minimum blade installation angle and maximum blade installation angle of the water pump device, respectively.

[0077] To solve the high-dimensional nonlinear optimization problem caused by three-dimensional affine transformation, the Dogbox algorithm in the trust region algorithm is used to determine the fitting parameters by minimizing the sum of squares of residuals, that is, to make the characteristic curve translate and rotate in the three-dimensional space concept, so that the characteristic curve approximates the selected typical working condition points, and the characteristic curve correction is completed.

[0078] V. Corrected characteristic curve extraction

[0079] The corrected flow-head-blade angle characteristic curve is drawn by origin software.

[0080] Specifically, the corrected characteristic curve equation obtained in the fourth part is discretized by origin software to draw the corrected integer angle contour map, that is, the corrected flow-head-blade angle characteristic curve map.

[0081] Example two

[0082] In order to better illustrate the execution process of the method of the present application, this embodiment takes a 3# water pump unit of a certain pump station as an example, carries out water pump device blade angle characteristic curve correction and result display based on the method, and specifically includes the following contents:

[0083] I. Data collection and pretreatment

[0084] The water regime, working condition and other related data of the 3# water pump unit of the pump station during the dispatch period from October 21, 2022 to April 6, 2023 are collected and statistically summarized, and the steady state working condition data is screened from the long sequence monitoring data by adaptive Gaussian filtering method and confidence method. The data requirements are shown in Table 1.

[0085] Table 1 Data requirement table

[0086]

[0087] II. Cluster analysis to screen typical working condition points

[0088] The screened monitoring data of the 3# water pump unit of the pump station is clustered by K-means clustering method to cluster the long sequence historical data, and a large number of similar working conditions are combined through iteration, so as to correct the water pump device characteristic curve subsequently. The specific steps are as follows,

[0089] (1) K discrete initial clustering centers are screened by K-means++ algorithm.

[0090] (2) The data set is assigned to the cluster corresponding to the clustering center by classical K-means++ algorithm.

[0091] (3) The mean value of each class cluster is calculated and taken as the new clustering center.

[0092] (4) Go to (2) and (3) until the clustering center is constant or changes very little.

[0093] (5) Whether it is the best K value is judged by elbow method, if not, go to (1).

[0094] The optimal K value of unit clustering is finally determined as 200, i.e. the previous 10276 working points are divided into 200 cluster centers by clustering to form the correction training set of the unit.

[0095] III. Extraction of original characteristic curve

[0096] The characteristic curve of the water pump device is numerically processed, and a binary quadratic polynomial method is used to fit the flow-head-blade angle characteristic curve of the water pump device, which has the following form:

[0097] θ = -20.694 + 0.267*Q - 0.625*H + 0.0025*Q 2 *H + 0.127*H 2

[0098] where θ is the blade angle of the water pump device (°), Q is the flow rate of the water pump device (m 3 / s), and H is the head of the water pump device (m). The correlation coefficient R 2 ≥ 0.98 has met the relevant accuracy requirements.

[0099] IV. Correction of characteristic curve based on typical working points

[0100] (1) Introduce appropriate parameters and search domain: introduce parameters a, b, c, r to transform the characteristic curve equation obtained in the third part, and the linearly transformed equation is as follows:

[0101]

[0102] Let the normal operation condition interval of the pump group be Q ∈ [25m 3 / s, 40m 3 / s], H ∈ [3.8m, 6.5m], and θ ∈ [-7°, -1°]; then the search domain is defined as a ∈ [-7.5, 7.5], b ∈ [-2, 2], and c ∈ [-2, 2].

[0103] (2) Based on the Dogbox algorithm in the trust region algorithm, the fitting parameters are determined by minimizing the sum of squared residuals, i.e. the characteristic curve is translated and rotated in the three-dimensional space concept to make the characteristic curve approximate the selected typical working points, thereby completing the correction of the characteristic curve.

[0104] V. Extraction of corrected characteristic curve

[0105] The corrected curve equation obtained in the fourth part is discretized using Origin software, and the corrected integer angle contour map is drawn, i.e. the flow-head-blade angle characteristic curve after correction is obtained, as shown in Table 2. Figure 2

[0106] ​Table 2 Corrected corner calculation accuracy index

[0107]

[0108] By adopting the above technical solutions disclosed in the application, the following beneficial effects are obtained:

[0109] The application provides a water pump device characteristic curve correction method based on clustering analysis and a confidence domain algorithm, and the stable state working condition data is screened from long sequence monitoring data by using a self-adaptive Gaussian filtering method combined with a confidence method, the self-adaptive Gaussian filtering method can retain the original signal change information to the greatest extent while filtering noise. The long sequence historical data is clustered by using a K-means clustering method, so that the data amount is greatly reduced, the calculation speed is improved, the data points cover as many working conditions as possible, and the curve overfitting phenomenon that may occur in the linear regression process due to the excessive density of the point set in the partial common working condition interval during the operation of the same pump station is avoided. The original characteristic curve equation is modified by introducing a new parameter to perform affine transformation, the curve is as close to the data points as possible without destroying the original curve space structure, direct fitting is avoided, abnormal deformation of the characteristic curve that does not conform to the hydraulic characteristics of the water pump is avoided, and thus the curve correction quality is improved. The Dogbox algorithm in the confidence domain algorithm is used for parameter optimization, and when solving the high-dimensional nonlinear parameter optimization problem that may occur due to the introduction of the new parameter, the Dogbox algorithm has good robustness, stability and convergence characteristics. The application has strong universality, and can be used to process and generate the corrected water pump device characteristic curve for the axial flow / mixed flow pump station with long sequence historical monitoring data.

[0110] The above only describes the preferred embodiments of the application, and it should be noted that, for ordinary skilled persons in the technical field, some improvements and refinements can be made without departing from the principles of the application, and these improvements and refinements should also be considered as the protection scope of the application.

Claims

1. A method for correcting characteristic curve of water pump device based on cluster analysis and trust region algorithm, characterized in that: Comprising the following steps, S1, typical working condition point screening by cluster analysis: The K-means clustering method is used to cluster the steady state working condition data of the water pump device, a large number of similar working conditions are combined through iteration, and the typical working condition points of the water pump device are screened out; S2, original characteristic curve extraction: Based on the typical working condition points of the water pump device, the moving least squares method or cubic spline or polynomial fitting is used to obtain the original characteristic curve of the water pump device; Step S2 is specifically, The flow-head-blade angle characteristic curve of the water pump device is fitted by using a binary quadratic polynomial method, and the original characteristic curve equation of the water pump device is obtained, which has the following form: θ = p00 + p10 * Q + p01 * H + p20 * Q 2 + p11 * Q * H + p02 * H 2 Wherein, θ is the blade angle of the water pump device; Q is the flow of the water pump device; H is the head of the water pump device; p00, p10, p01, p20, p11, p02 are the coefficients of each term in the fitted polynomial equation; S3, characteristic curve correction based on typical working condition points: The original characteristic curve of the water pump device is transformed by introducing parameters to realize three-dimensional affine transformation, and the Dogbox algorithm in the trust region algorithm is used to realize the characteristic curve correction of the water pump device; Step S3 specifically includes the following contents, S31, introduce parameters a, b, c, r to transform the original characteristic curve equation obtained in step S2 to realize three-dimensional affine transformation, and the linear transformed equation is as follows: Wherein, a, b, c, r are new parameters introduced in the transformed equation; Q' is the flow value of the corrected water pump device; H' is the head value of the corrected water pump device; θ' is the blade angle value of the corrected water pump device; Let the normal operation condition interval of water pump device be Q∈[Q min ,Q max ], H∈[H min ,H max ], θ∈[θ min ,θ max ]; let the search domain be a∈[-0.3*|Q min -Q max |, 0.3*|Q min -Q max |], b∈[-0.3*|H min -H max |, 0.3*|H min -H max |], c∈[-0.3*|θ min -θ max |, 0.3*|θ min -θ max |]; Q min and Q max are the minimum flow and maximum flow of the water pump device respectively; H min and H max are the minimum head and maximum head of the water pump device respectively; θ min and θ max are the minimum angle and maximum angle of the blade of the water pump device respectively. S32, based on the Dogbox algorithm in the trust region algorithm, the fitting parameters are determined by minimizing the sum of squares of residuals, that is, the characteristic curve is translated and rotated in the three-dimensional space concept, so that the characteristic curve approximates the selected typical working condition points, and the characteristic curve correction of the water pump device is completed.

2. The method for correcting characteristic curve of water pump device based on cluster analysis and confidence domain algorithm according to claim 1, characterized in that: The K-means clustering method in step S1 specifically includes the following contents, S11, select K discrete initial cluster centers by K-means++ algorithm; S12, assign the data set to the cluster corresponding to the cluster center by classical K-means++ algorithm; S13, calculate the mean value of each class cluster and take this mean value as the new cluster center; S14, repeat steps S12 to S13 until the cluster center is unchanged or the change is less than the change threshold; S15, use the elbow method to determine whether it is the best K value, if not, go to step S11, if yes, divide the steady state working condition data of the water pump device into K cluster centers, and screen out the typical working condition points of the water pump device.

3. The method of claim 1, wherein the method is characterized by: Before step S1, it also includes, S0, data collection and pretreatment: according to the actual engineering scheduling operation of different time scales, the water and working condition related data of the water pump device are statistically summarized, the data are processed by using data cleaning method, and the steady state working condition data of the water pump device are selected by using self-adaptive Gaussian filtering method and confidence method.

4. The method of claim 3, wherein the method is characterized by: The water and working condition related data include the water level below the station, the water level above the station, the flow monitoring value of each unit, and the blade angle monitoring value of each unit; The data processing includes eliminating null, outlier and abnormal value.

5. The method for correcting characteristic curve of water pump device based on cluster analysis and confidence domain algorithm according to claim 3, characterized in that: In step S0, the noise is filtered by using an adaptive Gaussian filter method, and the calculation formula is, S(t, σ) = f(t) * g(t, σ) wherein S(t, σ) is the smoothing result of the signal or data after Gaussian filtering; f(t) is the original signal or data sequence, which is a function of time t; t is a time variable; g(t, σ) is a Gaussian kernel function; * is a convolution operator; and σ is a standard deviation of the Gaussian function; where m is the sampling time point; f(m) is the historical data signal; f k+1 (m) is the output value after the k+1th iteration smoothing; f k (m+j) is the value of the signal or data at position m+j in the kth iteration; w k (m+j,σ) is the weight value of each point in the window; j is the offset in the convolution window, which represents the time offset relative to m; M is the width parameter of the convolution window.

6. The method for correcting characteristic curve of water pump device based on cluster analysis and confidence domain algorithm according to claim 5, characterized in that: The confidence method is specifically, After the data is denoised by using the adaptive Gaussian filter method, an error probability distribution function of the direct monitoring index steady-state data is obtained, so that the variation threshold of the index is determined under a certain confidence interval as the basis for judging the steady state; when all the indexes are in the steady state, the working condition point is considered to be a steady-state working condition point.

7. The method for correcting characteristic curve of water pump device based on cluster analysis and confidence domain algorithm according to claim 1, characterized in that: After step S3, there is further included, S4, extracting the corrected characteristic curve: the origin software is used to draw the corrected flow-head-blade angle characteristic curve.

8. The method for correcting characteristic curve of water pump device based on cluster analysis and confidence domain algorithm according to claim 7, characterized in that: Step S4 is specifically, The origin software is used to disperse the corrected characteristic curve equation obtained in step S3, and a corrected integer angle contour map is drawn, that is, a corrected flow-head-blade angle characteristic curve diagram is obtained.

Images