CN113191012B - Water supply pipe network pipe burst detection method based on LSSVM interactive prediction - Google Patents
Water supply pipe network pipe burst detection method based on LSSVM interactive prediction Download PDFInfo
- Publication number
- CN113191012B CN113191012B CN202110535527.4A CN202110535527A CN113191012B CN 113191012 B CN113191012 B CN 113191012B CN 202110535527 A CN202110535527 A CN 202110535527A CN 113191012 B CN113191012 B CN 113191012B
- Authority
- CN
- China
- Prior art keywords
- prediction
- lssvm
- data
- pipe network
- monitoring
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/14—Pipes
Abstract
The invention discloses a water supply pipe network pipe burst signal abnormity detection method, and belongs to the field of urban water supply pipe network safety. The traditional data prediction-based method of the invention mostly predicts the monitoring value at the current moment according to the historical monitoring data of single-point flow and pressure, and judges pipe explosion when the difference between the predicted value and the monitoring value exceeds a threshold value. However, practical experience shows that the loss and error of the monitoring data can seriously affect the single-point prediction result, thereby causing frequent false reports and false reports. According to the method, a multi-input single-output LSSVM interactive prediction model is constructed for the monitoring data at different positions in the pipe network according to the spatial correlation of the actual water consumption and the monitoring data, and one-time standard deviation is selected as a threshold value to carry out pipe explosion detection. The LSSVM interactive prediction model can reduce the influence of data loss and errors on the prediction result, is more sensitive to smaller tube explosion response, and further effectively improves the tube explosion detection performance based on data prediction.
Description
Technical Field
The invention belongs to the urban water supply pipe network, and particularly relates to a water supply pipe network pipe burst detection method based on LSSVM interactive prediction.
Background
Pipe explosion of the water supply network can cause water resource waste and affect the normal operation of the system. With more and more water supply networks in China being provided with SCADA systems, a pipe burst detection method based on data driving is widely concerned. Most of traditional data prediction-based methods predict a current monitoring value according to historical monitoring data of single-point flow and pressure, and pipe explosion is judged when the difference between the predicted value and the monitoring value exceeds a threshold value. However, practical experience shows that the loss and error of the monitoring data can seriously affect the single-point prediction result, thereby causing frequent false alarm and missed alarm. In addition, the traditional data prediction-based method only considers single-point monitoring data and lacks the research of combining a plurality of monitoring data;
in view of the above, the invention provides a water supply pipe network pipe burst detection method based on LSSVM interactive prediction, which is used for constructing a multi-input single-output LSSVM interactive prediction model according to monitoring data at different positions in a pipe network and aims to reduce the influence of data loss and errors on prediction and further improve the pipe burst detection performance based on data prediction.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a water supply pipe network pipe burst detection method based on LSSVM interactive prediction.
The technical scheme of the invention is as follows:
a water supply network pipe burst detection method based on LSSVM interactive prediction comprises the following steps:
(1) Performing correlation analysis on monitoring data at different positions in a pipe network;
(2) According to the spatial correlation of actual water consumption and monitoring data, constructing a multi-input single-output LSSVM interactive prediction model for the monitoring data with strong correlation in the pipe network, and predicting a normal pressure value;
(3) Monitoring tube explosion according to the prediction residual error (the difference between the predicted value and the measured value);
in the step (1), correlation analysis is performed on monitoring data at different positions in a pipe network by using Pearson correlation coefficients, and correlation degree among the data is measured, wherein a calculation formula is as follows:
in the formula cov (x) 1 ,x 2 ) Is x 1 、x 2 The covariance of (a).
In the step (2), an LSSVM interactive prediction model is adopted to predict the normal pressure value, and an interactive prediction model is constructed.
(3.1) the calculation formula of the LSSVM prediction model is as follows:
wherein ω is a weight and b is a deviation; according to the principle of minimizing the structural risk, the omega and b values are sought to be minimized, namely, the optimization problem exists
Where J (ω, ξ) represents the structure risk, ξ i The prediction error of the model to the training sample, C the regularization parameter for controlling the punishment degree, and the regularization parameter is solved by Lagrange
Wherein α = [ α ] 1 ,α 2 ,…,α n ]Is a lagrange multiplier. The optimization problem is converted into a linear equation system by adopting a Lagrange function, and a kernel function K (x) exists according to a Mercer condition i ,x j ) So that
The radial basis kernel function has a wider convergence domain and stronger generalization capability and is an ideal kernel function, so the radial basis kernel function is selected as the LSSVM kernel function, and the expression of the radial basis function is
Where σ is a kernel parameter, representing the width of the radial basis function. Then the final LSSVM prediction function is obtained as
(3.2) constructing a two-input one-output water pressure monitoring value interactive prediction model as follows:
in practical application, when the number of the water pressure monitoring points arranged in the pipe network exceeds 3, the monitoring points at different positions can be grouped in advance through spatial correlation, and then a two-input one-output prediction model is constructed for the monitoring data with strong correlation. Therefore, when one monitoring point data is lost or wrong, the other monitoring point data is adopted to predict the monitoring point data, so as to form interactive prediction among different monitoring point data in the pipe network, and further avoid the influence of single-point monitoring data errors on prediction results.
In the step (3), a method based on a prediction residual error is adopted to detect tube explosion, and the method comprises the following specific steps: 1) Based on historical pressure monitoring data, interactively predicting data of a pressure monitoring point at the next moment by using an LSSVM model; 2) Calculating a burst detection threshold value based on the pressure prediction residual error in the normal time period; 3) And checking whether the prediction residual exceeds a threshold value, and if the prediction residual is smaller than the threshold value, judging that the tube burst is carried out at the moment. The threshold calculation formula is as follows:
in the formula R i The residual error between the predicted value and the measured value is obtained; n is the number of test set samples, where N =14 × 96=1344;is the average of the residuals of the normal period,is calculated by the formula
The tube explosion detection success rate and the false alarm rate are in a trade-off relation, the detection rate of the tube explosion can be improved by a smaller tube explosion detection threshold value, but the higher false alarm rate can be caused, and vice versa. The sum of the prediction residual error and 1 time standard deviation in the normal time period is used as a threshold value in the primary detection, the threshold value can be adjusted according to specific conditions in practice, and when the false alarm rate needs to be reduced, the detection threshold value of the pipe explosion can be properly reduced.
Compared with the prior art, the invention has the following advantages and beneficial effects:
the water supply network pipe explosion detection method based on LSSVM interactive prediction has important significance for improving the comprehensive performance of pipe explosion detection based on data prediction. Compared with the traditional pipe explosion detection method for analyzing data of a single monitoring point and a single moment, the method disclosed by the invention utilizes the spatial correlation of actual water consumption and monitoring data, effectively weakens the influence of data loss and errors on a prediction result, and further remarkably improves the comprehensive performance of pipe explosion detection based on data prediction
Drawings
FIG. 1 is a flow chart of a method for detecting pipe burst in a water supply pipe network according to the present invention;
FIG. 2 is a water supply network topology;
FIG. 3 is a pipe explosion detection result based on Kalman filtering and LSSVM under the condition that data are not abnormal;
FIG. 4 shows a detection result based on Kalman filtering and LSSVM tube explosion under the condition of data abnormality.
Detailed Description
For better understanding of the present invention, the present invention is described in detail below with reference to examples.
The invention takes SCADA monitoring data of L-Town-C area pipe network 2018 provided by a second inter-water system simulation and control integrated international conference (CCWI/WDSA 2020) as an example for simulation verification, and the data can be obtained at (www.ccwi-WDSA 2020.com). As shown in FIG. 2, the water supply network has 92 nodes and 3 pressure monitoring points, each of which takes 15min as an interval and records the data of instantaneous flow, pressure and the like for 140 days in total. Theoretically, actually measured data including the occurrence time, duration and burst pressure of the burst are collected to verify the feasibility of the method provided by the text, but in practice, burst monitoring data are not recorded. Therefore, the water supply network hydraulic model is constructed in EPANET, and the performance of detection is carried out by adopting simulation contrast Kalman filtering and LSSVM pipe explosion on the basis of original SCADA monitoring data.
Fig. 3 shows the pipe explosion detection results based on kalman filtering and LSSVM, and pipe explosion with 5%, 10%, 15% and 20% of total water added for 3 rd, 6 th, 9 th and 12 th days is performed to compare the pipe explosion detection performances of different degrees. As can be seen from FIG. 3, 4 pipe explosion signals can be detected by both Kalman filtering and LSSVM methods, but the LSSVM method is more sensitive to smaller pipe explosion response and has lower false alarm rate in non-pipe explosion time period, so that the overall detection effect of the LSSVM method is better. Specifically, the detection threshold values of Kalman filtering and LSSVM squibs are-0.53 m and-0.31 m respectively (shown by red lines in the figure), for the LSSVM algorithm, 5% of squibs cause the pressure drop to exceed the threshold value by 0.48m, and for the Kalman filtering algorithm, the pressure drop caused by the same squibs only exceeds the threshold value by 0.16m, which shows that the LSSVM can better detect smaller squibs.
TABLE 1 comparative analysis of detection rate and false alarm rate of tube explosion
Table 1 shows the pipe explosion detection rate and the false alarm rate of kalman filtering and LSSVM under the condition of no abnormal data, wherein the pipe explosion detection rate is the ratio of the number of detected pipe explosion signals to the total number of pipe explosion signals, and the false alarm rate is the ratio of the detected normal signals to the total number of normal signals. As can be seen from Table 1, the detection rates of Kalman filtering and LSSVM for pipe explosion are 59.11% and 87.24%, respectively, and the false alarm rates are 16.67% and 8.96%, respectively. Therefore, compared with Kalman filtering, the detection rate of LSSVM tube explosion is improved by 28.13%, the false alarm rate is reduced by 7.71%, and the tube explosion detection performance is remarkably improved.
In order to contrastively analyze the influence of data abnormality on the pipe explosion detection, the pipe explosion detection results based on Kalman filtering and LSSVM are discussed and analyzed when data loss and data abnormality are wrong. As shown in fig. 4, data anomalies were set 2 times for 6h, with data set missing on day 2 and-5% detection error added on day 8. The detection result is shown in fig. 4, in which the data abnormal period is indicated by a green dotted line.
As can be seen from fig. 4, the data anomaly has a significant effect on the detection effect of the burst in the kalman filter, and has a small effect on the detection effect of the LSSVM. Specifically, when data are lost, continuous false alarm signals for 6 hours appear in Kalman filtering, and false alarms only occur in a short time of 5 minutes in the LSSVM; when the data is wrong, the Kalman filtering false alarm duration is 4h, and the LSSVM does not generate false alarm.
TABLE 2 detection rate of tube explosion and comparison analysis of false alarm rate
And table 2 counts the pipe explosion detection rate and the false alarm rate of Kalman filtering and LSSVM under the condition of data abnormity. As can be seen from the table, the detection rates of pipe explosion based on Kalman filtering and LSSVM are respectively 56.20% and 87.24%, and the false alarm rates are respectively 22.92% and 9.06%. In addition, as can be seen from comparison of table 1, when data is abnormal, the detection rates of kalman filtering and LSSVM tube explosion are reduced by 3.93% and 0.09% respectively, and the false alarm rates are increased by 6.25% and 0.1% respectively. Therefore, the influence of data abnormity on the detection effect of the Kalman filtering tube explosion is obvious.
As can be seen from the above analysis, the data abnormality has a significant influence on the detection effect of the Kalman filtering tube explosion, continuous false alarm occurs and the false alarm rate is high; and the influence on the detection effect of the LSSVM tube explosion is small, and the false alarm rate is far lower than that of Kalman filtering. This is because the burst detection effect is affected by the prediction accuracy, and the higher the accuracy, the better the detection effect. Kalman filtering estimates a current state value by carrying out weighted average on a current state measurement value and a previous state estimation value, so that the prediction precision of a nonlinear sample is low, and particularly when an abnormal data sample exists, the prediction precision is obviously reduced; the LSSVM converts the nonlinear inseparable problem of a low-dimensional space into the linear separable problem of a high-dimensional space through a radial basis kernel function, the nonlinear sample prediction precision is high, and a relaxation variable and a penalty factor are introduced to remove abnormal data, so that the prediction precision is further improved.
Claims (4)
1. A water supply network pipe burst detection method based on LSSVM interactive prediction is characterized by comprising the following steps:
(1) Performing correlation analysis on monitoring data at different positions in a pipe network;
(2) According to the spatial correlation of actual water consumption and monitoring data, constructing a multi-input single-output LSSVM interactive prediction model for the monitoring data with strong correlation in the pipe network, and predicting a normal pressure value;
(3) And monitoring tube explosion according to the prediction residual error, wherein the prediction residual error is the difference between the predicted value and the measured value.
2. The method for detecting water supply pipe network pipe explosion based on LSSVM interactive prediction as recited in claim 1, wherein in the step (1), pearson correlation coefficient is adopted to perform correlation analysis on monitoring data at different positions in the pipe network and measure the correlation degree between the data, and the calculation formula is that
In the formula, cov (x) 1 ,x 2 ) Is x 1 、x 2 The covariance of (a).
3. The method for detecting water supply pipe network pipe burst based on LSSVM interactive prediction as recited in claim 1, wherein in the step (2), an LSSVM interactive prediction model is adopted to predict the normal pressure value, and the interactive prediction model is constructed as follows:
(3.1) the calculation formula of the LSSVM prediction model is as follows:
wherein, omega is weight, b is deviation; according to the principle of minimizing the structural risk, the minimization of the omega and b values is found, namely, an optimization problem exists
Where J (ω, ξ) represents the structure risk, ξ i The prediction error of the model to the training sample, C is the regularization parameter for controlling the punishment degree, and the regularization parameter is solved by Lagrange
Wherein a = [ a ] 1 ,a 2 ,…,a n ]For Lagrange multipliers, the optimization problem is converted into a linear equation system by adopting a Lagrange function, and a kernel function K (x) exists according to Mercer conditions i ,x j ) So that
The radial basis kernel function has a wider convergence domain and stronger generalization capability and is an ideal kernel function, so the radial basis kernel function is selected as the LSSVM kernel function, and the expression of the radial basis function is
In the formula, sigma is a kernel function parameter and represents the width of a radial basis function, and then the final LSSVM prediction function is obtained
(3.2) constructing a two-input one-output water pressure monitoring value interactive prediction model as follows:
in practical application, when the number of the water pressure monitoring points arranged in the pipe network exceeds 3, the monitoring points at different positions can be grouped in advance through spatial correlation, and then a two-input one-output prediction model is constructed for the monitoring data with strong correlation, so that interactive prediction among the data of different monitoring points in the pipe network is formed, and further the influence of single-point monitoring data errors on prediction results is avoided.
4. The water supply pipe network pipe burst detection method based on LSSVM interactive prediction as claimed in claim 1, wherein in step (3), a method based on prediction residual error is adopted to detect pipe burst, and the specific steps are as follows: 1) Based on historical pressure monitoring data, interactively predicting data of a pressure monitoring point at the next moment by using an LSSVM model; 2) Calculating a burst detection threshold value based on the pressure prediction residual error in the normal time period; 3) Checking whether the prediction residual exceeds a threshold, if so, judging that the tube burst occurs at the moment, wherein the threshold calculation formula is as follows:
in the formula R i The residual error between the predicted value and the measured value is obtained; n is the number of test set samples, where N =14 × 96=1344;is the average of the residuals of the normal period,is calculated by the formula
The sum of the prediction residual error and 1 time standard deviation in the normal time period is used as a threshold value in the primary detection, the threshold value can be adjusted according to specific conditions in practice, and when the false alarm rate needs to be reduced, the detection threshold value of the pipe explosion can be properly reduced.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110535527.4A CN113191012B (en) | 2021-05-17 | 2021-05-17 | Water supply pipe network pipe burst detection method based on LSSVM interactive prediction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110535527.4A CN113191012B (en) | 2021-05-17 | 2021-05-17 | Water supply pipe network pipe burst detection method based on LSSVM interactive prediction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113191012A CN113191012A (en) | 2021-07-30 |
CN113191012B true CN113191012B (en) | 2022-12-20 |
Family
ID=76982032
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110535527.4A Active CN113191012B (en) | 2021-05-17 | 2021-05-17 | Water supply pipe network pipe burst detection method based on LSSVM interactive prediction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113191012B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117370906A (en) * | 2023-08-21 | 2024-01-09 | 长江生态环保集团有限公司 | Tube explosion detection and performance evaluation method based on single-point and time sequence anomaly detection |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102174994B (en) * | 2011-03-11 | 2012-11-07 | 天津大学 | Pipe burst accident on-line positioning system for urban water supply pipeline network |
CN103574291B (en) * | 2013-07-02 | 2016-11-23 | 同济大学 | Localization of bursted pipe method based on artificial immune system |
CN104933303A (en) * | 2015-06-10 | 2015-09-23 | 上海大学 | Method for predicting fluctuating wind speed based on optimization-based multiple kernel LSSVM (Least Square Support Vector Machine) |
EP3327206B1 (en) * | 2016-11-25 | 2020-03-25 | Tata Consultancy Services Limited | Ranking pipes for maintenance in pipe networks using approximate hydraulic metrics |
US20200003652A1 (en) * | 2018-06-29 | 2020-01-02 | Arizona Board Of Regents On Behalf Of The University Of Arizona | Spatio-temporal analytics for burst detection and location in water distribution |
CN110043808B (en) * | 2019-05-29 | 2020-05-19 | 浙江大学 | Water supply network leakage monitoring and early warning method based on time series analysis |
CN110119853B (en) * | 2019-05-29 | 2020-12-25 | 浙江大学 | Water supply network leakage alarm threshold value selection method based on time series monitoring data analysis |
CN110516883B (en) * | 2019-08-30 | 2022-07-15 | 哈尔滨工业大学 | Water supply pipe network region leakage prediction method based on space metering model |
CN110939870B (en) * | 2019-12-27 | 2021-04-27 | 天津大学 | Water supply network pressure monitoring point arrangement method for pipe burst monitoring |
CN111578154B (en) * | 2020-05-25 | 2021-03-26 | 吉林大学 | LSDR-JMI-based water supply network multi-leakage pressure sensor optimal arrangement method |
CN112393127B (en) * | 2021-01-19 | 2021-03-26 | 浙江和达科技股份有限公司 | Urban water supply network leakage management and control system |
-
2021
- 2021-05-17 CN CN202110535527.4A patent/CN113191012B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN113191012A (en) | 2021-07-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106872657B (en) | A kind of multivariable water quality parameter time series data accident detection method | |
CN111737909B (en) | Structural health monitoring data anomaly identification method based on space-time graph convolutional network | |
CN109359698A (en) | Leakage loss recognition methods based on long Memory Neural Networks model in short-term | |
CN112036075A (en) | Abnormal data judgment method based on environmental monitoring data association relation | |
CN109710983B (en) | Diesel engine cylinder layered fault diagnosis method based on key performance indexes | |
CN112799898B (en) | Interconnection system fault node positioning method and system based on distributed fault detection | |
CN113191012B (en) | Water supply pipe network pipe burst detection method based on LSSVM interactive prediction | |
CN110658308B (en) | Method for evaluating safety and reliability of online flue gas monitoring system by considering common cause failure | |
CN112187528B (en) | Industrial control system communication flow online monitoring method based on SARIMA | |
CN108038044A (en) | A kind of method for detecting abnormality towards continuous monitored target | |
CN110133400A (en) | A kind of dynamic power system method for detecting abnormality merging recursive state estimation | |
CN103646013B (en) | Multiple fault reconstruction method based on covariance matrix norm approximation | |
CN112861350A (en) | Temperature overheating defect early warning method for stator winding of water-cooled steam turbine generator | |
CN115372816A (en) | Power distribution switchgear operation fault prediction system and method based on data analysis | |
CN107371175A (en) | A kind of self-organizing network fault detection method using cooperation prediction | |
CN114401145A (en) | Network flow detection system and method | |
CN108508860A (en) | A kind of process industry production system data monitoring method based on coupled relation | |
Li et al. | Meteorological radar fault diagnosis based on deep learning | |
CN114357667B (en) | Engine starting state fault detection method based on RESID recursive structure identification | |
CN109522657B (en) | Gas turbine anomaly detection method based on correlation network and SVDD | |
CN103995985A (en) | Fault detection method based on Daubechies wavelet transform and elastic network | |
CN115017143A (en) | Data cleaning method for intelligent high-voltage switch | |
CN113418632B (en) | Concept drift detection method for oil temperature prediction of oil immersed transformer | |
CN115905869A (en) | Ultrasonic water meter fault early warning method | |
CN115171362A (en) | Early warning method and system for prevention and control of key areas |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |