A method for detecting abnormal events in water supply network pressure arrays

Abstract

This invention discloses a method for detecting abnormal events in a water supply network pressure array. The invention utilizes multivariate empirical mode decomposition (MEMD) and the k-neighbor mean algorithm to perform mode decomposition and reconstruction of the pressure array signal, and then analyzes the reconstructed signal to detect abnormal events in the water supply network pressure array. Specifically, the method includes steps such as multivariate empirical mode decomposition of the pressure array signal, IMF component reconstruction, Hilbert transform of the reconstructed signal, and anomaly detection based on k-neighbor mean. By applying MEMD multivariate empirical mode decomposition, this invention can integrate and analyze the data of the entire pressure measuring point array, effectively utilizing the data correlation between measuring points, making the mode decomposition results more reliable. Furthermore, by using the k-neighbor mean algorithm, multivariate data can be analyzed simultaneously. By calculating the average distance of the k nearest neighbors, abnormal events deviating from the normal pattern can be effectively identified.

Description

Technical Field

[0001] This invention belongs to the field of urban water supply networks, and specifically relates to a method for detecting abnormal events in water supply network pressure arrays. Background Technology

[0002] A transient wave is a rapidly moving elastic shock that propagates at relatively high speeds in fluid piping systems and is typically triggered by predetermined or unexpected events within the system, including rapid opening and closing of valves, pump start-up and shutdown, pipe bursts, and many other possibilities. [1] These sudden system changes can trigger severe or even catastrophic pressure fluctuations in the pipelines. Currently, water companies primarily use smart pressure gauges to monitor the water supply network to ensure its normal operation. [2] .

[0003] To identify abnormal events in water supply networks, existing technologies include: [Literature / Reference] [3] Empirical Mode Decomposition (EMD) / Ensemble Empirical Mode Decomposition (EEMD) techniques were employed to perform hierarchical decomposition of the signal. Subsequently, kurtosis and Z-filtering techniques were used to automatically select key Intrinsic Mode Functions (IMFs). Further analysis of these selected IMF components demonstrated that this method can effectively identify anomalous events. (References) [4] Wavelet Transform (WT) methods are used to detect singularities in pressure signals, and abnormal events are identified by analyzing these singularities. These methods have shown significant effectiveness in analyzing water pressure signals at individual measuring points in water supply networks. However, with the rapid development of IoT technology, a large number of pressure measuring points are deployed in water supply networks. Analyzing each individual measuring point separately would not only increase time costs but also easily lead to false alarms at single points.

[0004] Compared to Empirical Mode Decomposition (EMD) / Ensemble Empirical Mode Decomposition (EEMD) methods for single-point signals, Multivariate Empirical Mode Decomposition (MEMD) technology offers the ability to process multiple related signals simultaneously, enabling the revelation of common characteristics and interactions among these signals. This method achieves more comprehensive data analysis by considering the dynamic changes of the signal system as a whole, rather than analyzing individual signals in isolation. Applying MEMD decomposition allows for the integration and analysis of data from the entire pressure measurement array, facilitating a more accurate understanding and identification of the nature of events. Furthermore, the k-Nearest Neighbors (KNN) algorithm is suitable for processing data in multidimensional feature spaces. Using the KNN algorithm, detailed analysis of pressure data at various time points can be performed, effectively identifying anomalous events that deviate from the norm.

[0005] References:

[0006] [1]Duan HF, Pan B, Wang M, et al. State-of-the-art review on the transient flow modeling and utilization for urban water supply system (UWSS) management [J]. Journal of Water Supply: Research and Technology-AQUA, 2020, 69 (12). DOI: 10.2166 / aqua.2020.048.

[0007] [2] Wang Haitao, Tu Zhengqin, Zhang Kun, et al. Test on burst monitoring of water supply network based on intelligent fire hydrant [J]. China Water & Wastewater, 2021(037-021).

[0008] [3]Yusop HM,Ghazali MF,Yusof MFM,et al.The use of transmissionline modeling to test the effectiveness of I-kaz as autonomous selection ofintrinsic mode function[J].IOP Conference Series:Materials Science andEngineering,2017,257:012070-.DOI:10.1088 / 1757-899X / 257 / 1 / 012070.

[0009] [4]Srirangarajan S,Allen M,Preis A,et al.Wavelet-based Burst EventDetection and Localization in Water Distribution Systems[J].Journal of SignalProcessing Systems,2013,72(1):1-16.DOI:10.1007 / s11265-012-0690-6. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention proposes a method for detecting anomalies in water supply network pressure array signals. By utilizing multivariate empirical mode decomposition (MEMD) and k-neighbor mean algorithm, the pressure array signal is decomposed and reconstructed, and the reconstructed signal is analyzed to achieve the detection of abnormal events in the water supply network pressure array.

[0011] In a first aspect, the present invention provides a method for detecting abnormal events in a water supply network pressure array, the method comprising the following steps:

[0012] Step 1: Multivariate Empirical Mode Decomposition of Pressure Array Signal

[0013] The pressure array signal is adaptively decomposed into intrinsic mode functions using multivariate empirical mode decomposition.

[0014] Step 2, IMF component reconstruction

[0015] Kurtosis values ​​are used to describe anomalous signals that may exist in IMF components;

[0016] The IMF components are sorted according to their kurtosis values. The two IMF components with the largest kurtosis values ​​are normalized and then reconstructed to obtain the reconstructed signal c(t).

[0017] After reconstructing the pressure array signal sequence C(t) by performing IMF component reconstruction at each measuring point, the reconstructed signal sequence C(t) is obtained.

[0018] Step 3, Hilbert transform of the reconstructed signal

[0019] The instantaneous amplitude information of the reconstructed signal at each measuring point in the reconstructed signal sequence C(t) of the pressure array is extracted by Hilbert transform; the instantaneous amplitude A(t) of the pressure array is obtained.

[0020] Step 4, Anomaly Detection Based on k-Neighborhood Mean

[0021] Anomaly detection is performed on the instantaneous amplitude A(t) of the pressure array based on the k-neighbor mean algorithm.

[0022] Secondly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted to be loaded and executed by a processor of a terminal device as described in the method for detecting abnormal events in a water supply network pressure array.

[0023] Thirdly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, which are adapted to be loaded and executed by the processor to provide a method for detecting abnormal events in a water supply network pressure array.

[0024] The beneficial effects of this invention are as follows: By applying MEMD multivariate empirical mode decomposition, this invention can integrate and analyze the data of the entire pressure measurement point array, effectively utilizing the data correlation between measurement points, making the mode decomposition results more reliable. Furthermore, by using the k-neighbor mean algorithm, multivariate data can be analyzed simultaneously. By calculating the average distance of the k nearest neighbors, abnormal events deviating from the normal pattern can be effectively identified. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention;

[0026] Figure 2 Here is a diagram of the example pipeline network structure;

[0027] Figure 3 This is a schematic diagram of MEMD decomposition of the pressure array signal;

[0028] Figure 4 A schematic diagram of the instantaneous amplitude signal of the Hilbert transform of the reconstructed signal at measurement point 0008. Detailed Implementation

[0029] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the scope and application of the present invention.

[0030] This application provides a method for detecting abnormal events in a water supply network pressure array, including the following steps:

[0031] Step 1: Multivariate Empirical Mode Decomposition of Pressure Array Signal

[0032] Multivariate empirical mode decomposition is used to adaptively decompose the pressure array signal into intrinsic mode functions (IMFs).

[0033] Definition: The input signal V(t) of the water supply network pressure array is defined as {v1(t), v2(t), ..., v...}. s (t)} T (T represents the signal length, and s represents the total number of measurement points) Indicates the direction angle The direction vectors are given by g = 1, 2, ..., G, where G is the number of direction vectors.

[0034] Based on the above definitions, the main steps of MEMD multivariate empirical mode decomposition provided in this application embodiment include:

[0035] S1.1 Calculate the input signal V(t) in each direction vector X θ mapping of g

[0036] S1.2 obtains all mappings using the multivariate spline interpolation method. G envelope lines

[0037] S1.3 The mean m(t) is calculated using G envelopes, as shown in equation (1);

[0038]

[0039] S1.4 Obtain the intermediate signal b(t) = V(t) - m(t). If the intermediate signal b(t) satisfies the two constraints of the IMF: (1) the number of times it crosses zero is equal to or at most differs from the number of extreme points by 1; (2) the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is 0, then the intermediate signal b(t) is taken as a new IMF component, and the input signal V(t) = V(t) - b(t) is taken as the input of S1.1. Repeat steps S1.1 to S1.4. Otherwise, take the input signal V(t) = b(t) as the input of S1.2. Repeat steps S1.1 to S1.4 until the number of extreme points on all direction vectors is less than 3. The IMF component is filtered out, and the remaining residual component R(t) = V(t) is left.

[0040] Step 2, IMF component reconstruction

[0041] In the IMF components decomposed in step 1, the pipe burst event usually causes obvious extreme points in certain specific IMFs because these components capture high-frequency fluctuations in the signal, which is consistent with the sudden pressure change generated during the pipe burst. Kurtosis is used to describe the abnormal signals that may exist in the IMF components. The kurtosis is calculated as shown in equation (2).

[0042]

[0043] In equation (2), u i The vibration amplitude corresponding to a point in the discrete sequence of time-domain waveforms. denoted as the average amplitude of the discrete sequence, and n is the number of points in the discrete sequence.

[0044] The IMF components are sorted according to their kurtosis values. Since the amplitudes of the IMF components are different, the two IMF components with the largest kurtosis values ​​are normalized before reconstruction to obtain the reconstructed signal c(t). After reconstructing the IMF components for each measuring point, the pressure array reconstructed signal sequence C(t) = {c1(t), c2(t), ..., c s (t)}, where s represents the total number of measurement points.

[0045] Step 3, Hilbert transform of the reconstructed signal

[0046] The instantaneous amplitude information of the reconstructed signal c(t) at each measuring point in the pressure array reconstructed signal sequence C(t) is extracted by Hilbert transform.

[0047] In a particular preferred example, this step is as follows:

[0048] S3.1 defines the Hilbert transform y(t) of the reconstructed signal c(t) as...

[0049]

[0050] In equation (3), PV is the Cauchy principal value of the integral.

[0051] S3.2 Couples c(t) and y(t) to obtain a complex conjugate pair, giving the analytic complex signal z(t) of c(t) as:

[0052] z(t)=c(t)+iy(t) (4)

[0053] S3.3 The formula for calculating the instantaneous amplitude a(t) of the analyzed signal is shown in equation (5):

[0054]

[0055] After obtaining the instantaneous amplitude at each measuring point, the instantaneous amplitude sequence of the pressure array A(t) = {a1(t), a2(t), ..., a s (t)}, where s represents the total number of measurement points.

[0056] Step 4, Anomaly Detection Based on k-Neighborhood Mean

[0057] Anomaly detection is performed on the instantaneous amplitude sequence A(t) obtained after Hilbert transform using the k-neighborhood mean algorithm.

[0058] In a preferred example, the specific process of the k-neighborhood mean algorithm is as follows:

[0059] S4.1 Given a data point o and a set of data points D, find the k-nearest neighbor distance d of point p in D. k (p) satisfies:

[0060] (1) There are at least k points o∈D\{p}, d(p,o)≤d k (p);

[0061] (2) There are at most k-1 points o∈D\{p}, d(p,o) <d k (p);

[0062] Then d is calledk (p) is the k-th distance of p.

[0063] S4.2 k-Nearest Neighbor Average Distance Calculation

[0064]

[0065] In equation (6), K(q) represents the distance of object q in dataset D that does not exceed d. k The set of all points of (p).

[0066] The distribution density of point q is reflected by calculating the average distance between a sample point within the k-th distance and its k nearest samples. If the average distance is large, it indicates that point q is more likely to be a local outlier.

[0067] The formula for calculating the distance between sample points is shown in Equation (7), which uses Euclidean distance calculation.

[0068]

[0069] In equation (7), s represents the total number of measuring points.

[0070] Normal operation data of the pipeline network in adjacent time periods are used as the basis for threshold calculation. Then, the k-nearest neighbor model is used for anomaly detection. The average distance between the k-nearest neighbors of the sample points is calculated, and the average distance between the k-nearest neighbors of the sample points in 95% of the interval is selected as the threshold. Anomalies are judged based on this threshold. When the average distance between the k-nearest neighbors of a sample point exceeds this threshold, the point is marked as an anomaly, and the time corresponding to the point is the time when the abnormal event occurred.

[0071] This application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of a method for detecting abnormal events in a water supply network pressure array.

[0072] This application also provides a terminal device, including a processor and a computer-readable storage medium. The processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, which are adapted to be loaded and executed by the processor to provide a method for detecting abnormal events in a water supply network pressure array.

[0073] Application example:

[0074] like Figure 1 As shown, this example illustrates a method for detecting abnormal events in a pressure array, which specifically includes the following steps:

[0075] Step 1: Multivariate empirical mode decomposition of the pressure array.

[0076] Based on the pipeline network of a specific test area, a total of 11 monitoring points were established, with a data acquisition frequency of 0.1Hz. Pressure signals (a total of 8640 pressure data points per monitoring point) for August 24, 2020, were analyzed. The geographical locations and corresponding numbers of the pressure monitoring points are as follows: Figure 2 As shown, there is no sampling data at measurement point 0007, and the sampling frequency of measurement point 0011 is different from that of other measurement points, so it will not be analyzed.

[0077] The pressure data curve at each measuring point is decomposed into 13 IMF components and one residual component. Figure 3 In the diagram, a represents the MEMD decomposition diagram of measurement point 0001, b represents the MEMD decomposition diagram of measurement point 0002, c represents the MEMD decomposition diagram of measurement point 0003, d represents the MEMD decomposition diagram of measurement point 0004, e represents the MEMD decomposition diagram of measurement point 0005, f represents the MEMD decomposition diagram of measurement point 0006, g represents the MEMD decomposition diagram of measurement point 0007, h represents the MEMD decomposition diagram of measurement point 0008, i represents the MEMD decomposition diagram of measurement point 0009, and j represents the MEMD decomposition diagram of measurement point 0010.

[0078] Step 2, IMF component reconstruction

[0079] The kurtosis values ​​of each IMF component at each measuring point were calculated according to equation (2) and are shown in Table 1.

[0080] Table 1. Kurtosis values ​​corresponding to IMF components at each measuring point.

[0081] 0001 4.52 2.81 11.10 8.76 ... 2.47 0002 4.67 2.55 9.46 8.50 ... 2.05 0003 6.02 2.65 12.58 8.65 ... 2.73 ... ... ... ... ... ... ... 0010 19.38 6.89 17.94 9.34 ... 2.43

[0082] For each pressure measurement point signal, the two IMF components with the highest kurtosis values ​​are selected, and the selected IMF components are normalized and then reconstructed to obtain the reconstructed signal.

[0083] Step 3, Hilbert transform of the reconstructed signal

[0084] The instantaneous amplitude signal obtained by performing Hilbert transform on the reconstructed signal according to equations (3)-(5), taking the 0008 measuring point as an example, is as follows: Figure 4 As shown, the signal generated by the anomaly has a higher amplitude, representing a possible abnormal event in the original signal.

[0085] Step 4, Anomaly Detection Based on k-Neighborhood Mean

[0086] When the pipeline network in a certain test area is running normally and stably, the pressure trend does not change much during the sampling interval. Therefore, the value of k is set to 5 to ensure both anomaly detection capability and computing power. First, after performing steps 1 to 3, k-neighbor mean anomaly detection is performed on the data from August 23, 2022. The average distance of the k-nearest neighbor of the sample points at all times can be obtained as shown in Table 2.

[0087] Table 2. Average distance between k-nearest neighbor points for normal data sample points.

[0088] Average distance 0.513 0.402 0.446 0.382 ... 0.318 0.702

[0089] Based on the 95% interval, the average distance between the k-nearest neighbors of the abnormal sample values ​​can be obtained as 0.540. Therefore, the threshold range for normal events is set as [0, 0.540], and events outside this range are judged as abnormal events. After performing steps 1 to 3 on the data to be detected, k-nearest neighbor anomaly detection is performed, and the average k-nearest neighbor distances of the sample points are shown in Table 3.

[0090] Table 3 Average distance between k-nearest neighbors of the sample points to be tested

[0091] Average distance 0.425 0.226 0.218 0.302 ... 0.488 0.463

[0092] The abnormal time points of the sample to be detected can be determined based on the threshold calculated in Table 1. After merging the neighboring time points of the abnormal time points, the final disturbance event can be obtained. The known disturbance events on August 24, 2024 are compared with the disturbance events detected by the method of this invention, as shown in Table 4.

[0093] Table 4 Comparison of disturbance events detected by the method of the present invention with known disturbance events.

[0094]

[0095]

[0096] The wavelet singularity detection method, the single-point EMD anomaly detection method, and the method of this invention are compared and contrasted, as shown in Table 5.

[0097] Table 5 Comparison of the detection effects of the three methods for detecting disturbance events.

[0098]

[0099] As can be seen from Table 5, the method of the present invention can effectively reduce the total number of anomalies compared with wavelet singularity detection method and single measurement point EMD anomaly detection, reduce false alarms without missing reports, and improve the accuracy of disturbance event detection.

[0100] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for detecting abnormal events in a water supply network pressure array, characterized in that... The method includes the following steps: Step 1: Multivariate Empirical Mode Decomposition of Pressure Array Signal The pressure array signal is adaptively decomposed into intrinsic mode functions using multivariate empirical mode decomposition. Step 2, IMF component reconstruction Kurtosis values ​​are used to describe anomalous signals that may exist in IMF components; The IMF components are sorted according to their kurtosis values. The two IMF components with the largest kurtosis values ​​are normalized and then reconstructed to obtain the reconstructed signal c(t). After reconstructing the pressure array signal sequence C(t) by performing IMF component reconstruction at each measuring point, the reconstructed signal sequence C(t) is obtained. Step 3, Hilbert transform of the reconstructed signal The instantaneous amplitude information of the reconstructed signal at each measuring point in the reconstructed signal sequence C(t) of the pressure array is extracted by Hilbert transform; the instantaneous amplitude A(t) of the pressure array is obtained. Step 4, Anomaly Detection Based on k-Neighborhood Mean Anomaly detection is performed on the instantaneous amplitude A(t) of the pressure array based on the k-neighbor mean algorithm.

2. The method for detecting abnormal events in a water supply network pressure array according to claim 1, characterized in that: Step 1 is as follows: S1.1 Calculate the input signal V(t) in each direction vector mapping S1.2 Get all mappings G envelope lines S1.3 Calculate the mean m(t) using G envelopes; S1.4 Obtain the intermediate signal b(t) = V(t) - m(t). If the intermediate signal b(t) satisfies the two constraints of the IMF, then the intermediate signal b(t) is used as a new IMF component, and the input signal V(t) = V(t) - b(t) is used as the input of S1.

1. Repeat steps S1.1 to S1.

4. Otherwise, the input signal V(t) = b(t) is used as the input of S1.

1. Repeat steps S1.1 to S1.4 until the number of extreme points on all direction vectors is less than 3. The IMF component filtering is completed, and the remaining residual component R(t) = V(t) is obtained.

3. The method for detecting abnormal events in a water supply network pressure array according to claim 2, characterized in that: The G envelopes are obtained by multivariate spline interpolation.

4. A method for detecting abnormal events in a water supply network pressure array according to claim 2 or 3, characterized in that: The two constraints of the IMF are as follows: (1) The number of times the zero point is crossed is equal to or at most differs from the number of extreme points by 1; (2) The mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is 0.

5. The method for detecting abnormal events in a water supply network pressure array according to claim 1, characterized in that: Step 3 specifically involves: S3.1 Define the Hilbert transform y(t) of the reconstructed signal c(t); S3.2 Couple c(t) and y(t) to obtain complex conjugate pairs, and give the analytic complex signal z(t) of c(t); S3.3 Calculate the instantaneous amplitude a(t) based on the analytical complex signal z(t), and then obtain the instantaneous amplitude A(t) of the pressure array.

6. The method for detecting abnormal events in a water supply network pressure array according to claim 1, characterized in that: Step 4, specifically the k-neighborhood mean algorithm, is as follows: S4.1 Given a data point o and a set of data points D, find the k-nearest neighbor distance d of point p in D. k (p) satisfies: (1) There are at least k points o∈D\{p}, d(p,o)≤d k (p); (2) There are at most k-1 points o∈D\{p}, d(p,o) <d k (p); Then d is called k (p) is the k-th distance of p; S4.2 k-Nearest Neighbor Average Distance Calculation In the formula, K(q) represents the distance of object q in dataset D that does not exceed d. k The set of all points of (p); The distribution density of object q is reflected by calculating the average distance between a sample point within the k-th distance and its k nearest samples.

7. The method for detecting abnormal events in a water supply network pressure array according to claim 6, characterized in that: The distance between sample points is calculated using Euclidean distance.

8. A method for detecting abnormal events in a water supply network pressure array according to claim 6 or 7, characterized in that: Step 4 specifically involves using normal operating data from adjacent periods of the pipeline network as the basis for threshold calculation, then using the k-nearest neighbor algorithm for anomaly detection, calculating the average distance between the k nearest neighbors of the sample points, and selecting the average distance between the k nearest neighbors of 95% of the sample points as the threshold. Anomalies are judged based on this threshold. When the average distance between the k nearest neighbors of a sample point exceeds this threshold, the point is identified as an anomaly, and the time corresponding to this point is the time when the anomaly event occurred.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by the processor of the terminal device as described in any one of claims 1-8, a method for detecting abnormal events in a water supply network pressure array.

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor implements instructions; and the computer-readable storage medium stores multiple instructions, characterized in that... The instructions are adapted to be loaded and executed by a processor as described in any one of claims 1-8, for a method of detecting abnormal events in a water supply network pressure array.

Images