A method for identifying key pressure control points in a water supply network

Abstract

The present invention relates to the field of optimization of pressure measurement points, and discloses a method for identifying key pressure control points of a water supply network, including the following steps: Step S1, input the historical data of each pressure measurement point of the water supply network, eliminate abnormal pressure patterns and extract significant pressure patterns from the pressure measurement point data, and construct classification features for the original sample data according to different operating conditions of the intermediate pressure pumping stations in the water supply network; Step S2, establish a pressure change pattern clustering model, perform iterative clustering on the sample data of all pressure measurement points in the water supply network under different classification feature data, and determine the most suitable number of clustering clusters according to the weighted pattern significance rate; Step S3, for different scheduling objects, sort according to the target similarity in each category, and select the pressure measurement point with the highest target similarity in each category as its pressure main control point. The present invention can accurately identify the pressure control points representing the pressure change characteristics of the area to be measured.

Description

Technical Field

[0001] The present invention relates to the technical field of piezometric point optimization, and in particular to a method for identifying key pressure control points in a water supply network. Background Art

[0002] The SCADA pressure monitoring system of a water supply network is an important reference for the operation managers of a water supply company in making dispatching decisions. For a system without pressure monitoring points arranged, the layout of pressure monitoring points in a water supply network is a multi-objective decision-making optimization problem, which requires using as few pressure monitoring points as possible to collect in the greatest degree the details reflecting the real-time changes in the pressure distribution state of the network. Existing pressure point layout technologies usually adopt technical methods such as empirical method, sensitivity matrix analysis, fuzzy clustering analysis method, and multi-objective optimization method, but all have deficiencies.

[0003] The empirical method means that engineers arrange pressure monitoring points at the most unfavorable points, high-pressure areas, pressure change-sensitive areas, large water users, etc. in the network according to the network layout and operation management experience to guide the operation and dispatching of the network. This method is simple but cannot guarantee the rationality and accuracy of the layout plan.

[0004] The sensitivity matrix analysis method applies the basic principles of hydraulics and topology to establish a sensitivity matrix and equation for the node pressure of a water supply network, and arranges pressure monitoring points according to the sensitivity ranking to reflect the change situation of the network state, but it is easy to cause the aggregation of piezometric points.

[0005] The fuzzy clustering analysis method proposes the concept of influence coefficient according to the situation of the water pressure of a certain node affected by the water pressure fluctuations of other nodes, and conducts fuzzy clustering analysis on the influence coefficient matrix, groups the nodes, and selects the most representative node in each group as the piezometric point. However, this method has problems of unstable clustering results and poor interpretability of clustering results.

[0006] The multi-objective optimization method uses an optimization search algorithm to solve an optimization model to obtain the optimal objective function solution that meets the constraint conditions. However, this method will have two objective functions that are mutually exclusive, there are multiple non-dominated solutions, and it needs to be selected manually based on experience.

[0007] In a water supply network with piezometric points already arranged, the dispatching operator cannot take into account the states of all piezometric points during the actual dispatching process. How to identify the key pressure control points in a water supply network with pressure monitoring points already arranged is a technology that still needs to fill the gap. Summary of the Invention

[0008] To overcome the deficiencies of the above-mentioned existing technologies, the purpose of the present invention is to provide a method for identifying key pressure control points in a water supply network, introducing DTW (Dynamic Time Warping) as a measure of similarity distance, and analyzing the large historical pressure monitoring data by combining Kmeans clustering to identify the pressure control points that can represent the pressure change characteristics of the area to be measured.

[0009] To achieve the above object, the present invention proposes a method for identifying key pressure control points in a water supply network, including the following steps:

[0010] Step S1, input the historical data of each pressure measurement point in the water supply network, eliminate the abnormal pressure patterns and extract the significant pressure patterns from the pressure measurement point data, and construct classification features for the original sample data according to the different operating conditions of the intermediate pressure pumping stations in the water supply network;

[0011] As a preferred embodiment of the present invention, the elimination of abnormal pressure patterns further includes the following steps:

[0012] Step S111, determine a suitable time sliding window size sz and set abnormal conditions;

[0013] Furthermore, the abnormal conditions include:

[0014] The first abnormal condition:

[0015] For any point within the time sliding window where μ P(j,j+sz) represents the average value of the pressure data within any time sliding window, and σ P(j,j+sz) represents the standard deviation of the pressure data within any time sliding window. When any sample data meets the first abnormal condition, the data is eliminated;

[0016] The second abnormal condition:

[0017] All data σ P(j,j+sz) = 0 within the time window. When any sample data meets the second abnormal condition, the data is eliminated.

[0018] Step S112, eliminate the abnormal P err in the sample data according to the abnormal value condition;

[0019] Furthermore, the abnormal P err of the eliminated sample is where and are the abnormal data of the first abnormal condition and the second abnormal condition respectively.

[0020] As a preferred embodiment of the present invention, the significant pressure mode extraction further includes the following steps:

[0021] Step S121, perform normalization processing on the piezometric point data, where P origin is the original sample data, and P new is the sample data after data normalization;

[0022] Step S122, perform data downsampling, P upsample = upsample(P new ), where P upsample is the sample data after downsampling.

[0023] As a preferred embodiment of the present invention, the operating conditions of the intermediate booster pumping station include the superimposed water supply condition, the bypass water supply condition, and the clear water tank water supply condition.

[0024] Step S2, establish a pressure change mode clustering model, perform iterative clustering on the sample data of all piezometric points in the water supply pipe network under different classification feature data, and determine the most suitable number of clustering clusters according to the weighted mode significance rate;

[0025] As a preferred embodiment of the present invention, it includes the following steps:

[0026] Step S21, select a suitable clustering cluster interval [c min , c max , where c min is the minimum number of clustering clusters, and c max is the maximum number of clustering clusters;

[0027] Step S22, use dynamic time warping as the similarity measure between different pressure change modes, and use the Kmeans clustering algorithm to traverse the clustering cluster interval, and determine the number of clustering clusters and the corresponding clustering results according to the weighted mode significance rate.

[0028] Furthermore, the calculation of the weighted mode significance rate includes the following steps:

[0029] Step S221, calculate the mode significance rate of each classification feature, and the mode significance rate R stable is expressed as:

[0030] R stable = k stable / k,

[0031] where k is the number of any clustering clusters, and k stable is the number of clustering clusters whose clustering results have not changed after N iterations of randomly initializing the positions of the clustering clusters;

[0032] Step S222: Calculate the weighted mode significance rate. The weighted mode significance rate weights the mode significance rates under different operating conditions according to the proportion of sample data under different operating conditions of the intermediate booster pumping station. Specifically:

[0033] where w 1 , w 2 and w 3 represent different weight values, and represent the significance rates of three operating conditions respectively.

[0034] Step S3: For different scheduling objects, sort according to the target similarity in each category, and select the piezometric point with the highest target similarity in each category as its pressure main control point.

[0035] As a preferred embodiment of the present invention, it includes the following steps:

[0036] Step S31: Calculate the target similarity between each pressure monitoring point and the target scheduling object. The target similarity R os is specifically: R os = 1 / DTW(P i , P obj ),

[0037] where P i is the pressure change pattern of any piezometric point, P obj is the pressure change pattern of the target scheduling object, and DTW(a, b) represents calculating the DTW distance between sequences a and b;

[0038] Step S32: Sort according to the within-class target similarity, and select the piezometric point with the highest target similarity as the within-class main control point.

[0039] Compared with the prior art, the beneficial effects of one aspect of the present invention disclosed are as follows:

[0040] (1) The present invention uses the Kmeans clustering algorithm based on DTW (Dynamic Time Warping) to cluster the pressure patterns of different piezometric points, avoiding the influence of the time delay effect generated during pressure propagation on the clustering result, and ensuring the accuracy of the clustering result;

[0041] (2) The present invention proposes a complete set of mode significance rate indicators to evaluate the stability of the clustering result, avoiding unreasonable pressure pattern clustering results due to the clustering algorithm falling into a local optimum;

[0042] (3) The present invention selects the main control point according to the consistency between the within-class pressure change pattern and the pressure change pattern of the scheduling object, and can adaptively adjust the main control point according to time changes and different scheduling objects, having high application potential;

[0043] (4) The present invention can help the dispatching operator efficiently and conveniently master the overall dispatching situation of the dispatching area, and has high engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flowchart of the steps of a method for identifying key pressure control points in a water supply pipe network according to the present invention;

[0045] Figure 2 is a logic diagram of a method for identifying key pressure control points in a water supply pipe network according to the present invention;

[0046] Figure 3 is a schematic diagram of three working conditions of an embodiment disclosed by the present invention;

[0047] Figure 4 is a spatial visualization diagram of clustering results under three working conditions of an embodiment disclosed by the present invention;

[0048] Figure 5 is a distribution diagram of main control points of an embodiment disclosed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The following describes the embodiments of the present invention through specific specific examples in combination with the accompanying drawings. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0050] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0051] A large number of monitoring data show that the pressure change patterns of different pressure measurement points show different correlations with different spatial positions and topological relationships. Therefore, identifying key pressure control points in a water supply pipe network with pressure monitoring points already arranged can greatly facilitate the operator to master the operation status of the pipe network in the entire dispatching area.

[0052] Figure 1 is a flowchart of the steps of a method for identifying key pressure control points in a water supply pipe network according to the present invention. As Figure 1 shown, a method for identifying key pressure control points in a water supply pipe network according to the present invention includes the following steps:

[0053] Step S1, load data and perform data preprocessing: Input the historical data of each pressure measurement point in the water supply network, eliminate abnormal pressure patterns and extract significant pressure patterns from the pressure measurement point data, and construct classification features for the original sample data according to the different operating conditions of the intermediate pressure boosting pump stations in the water supply network.

[0054] Step S2, time series clustering: Establish a clustering model for pressure change patterns, perform iterative clustering on the sample data of all pressure measurement points in the water supply network under different classification features, and determine the most appropriate number of clustering clusters according to the weighted pattern significance rate.

[0055] Step S3, optimal selection of main control points: For different dispatching objects, sort according to the target similarity in each category, and select the pressure measurement point with the highest target similarity in each category as its pressure main control point.

[0056] Figure 2 This is the logic diagram of a method for identifying key pressure control points in a water supply network according to the present invention. Refer to Figure 2 , and further illustrate the above method, which specifically includes the following steps:

[0057] 1) Load the historical data of each pressure measurement point in the water supply network;

[0058] 2) Preprocess the historical data of each pressure measurement point in the water supply network, including eliminating the abnormal pressure module, extracting the significant pressure module, and constructing classification features according to the working conditions of superimposed pressure water supply, over-supplied water, and clear water tank water supply;

[0059] 2.1) When eliminating the abnormal pressure module, it includes:

[0060] Determine the appropriate time sliding window size sz and set the abnormal conditions, where the abnormal conditions include:

[0061] The first abnormal condition: For any point within the time sliding window where μ P(j,j+sz) represents the average value of the pressure data within any time sliding window, and σ P(j,j+sz) represents the standard deviation of the pressure data within any time sliding window. When any sample data satisfies the first abnormal condition, the data is eliminated;

[0062] The second abnormal condition: All data σ within the time window P(j,j+sz) = 0. When any sample data satisfies the second abnormal condition, the data is eliminated.

[0063] Eliminate the abnormal P in the sample data according to the outlier condition err , and eliminate the abnormal P of the sample err Specifically:

[0064] Among them, and represent the abnormal data of the first abnormal condition and the second abnormal condition respectively.

[0065] 2.2) When performing significant pressure pattern extraction, it includes:

[0066] Normalize the piezometric point data, and the expression is: where P origin represents the original sample data, and P new represents the sample data after data normalization;

[0067] Perform data downsampling, and the expression is: P upsample = upsample(P new ), where P upsample is the sample data after downsampling.

[0068] 2.3) Classify the characteristics of the original sample data according to the superimposed pressure water supply condition, the over - supply water condition, and the clear water tank water supply condition of the intermediate pressure pumping station in the water supply network.

[0069] 3) Determine the appropriate initial clustering cluster number interval [a, b];

[0070] 4) Use DTW as the similarity measure between different pressure change patterns, and traverse the clustering cluster interval through the kmeans algorithm;

[0071] 5) In the loop process, determine the clustering cluster number and the corresponding clustering result according to the weighted pattern significance rate. The calculation of the weighted pattern significance rate includes:

[0072] Calculate the pattern significance rate of each classification feature. The pattern significance rate R stable is expressed as:

[0073] R stable = k stable / k

[0074] where k is the number of any clustering clusters, and k stable is the number of clustering clusters whose clustering results have not changed after N iterations of randomly initializing the positions of the clustering clusters;

[0075] Calculate the weighted pattern significance rate. The weighted pattern significance rate weights the pattern significance rates under different operating conditions according to the proportion of sample data under different operating conditions of the intermediate pressure pumping station. The expression is:

[0076]

[0077] where w 1, w 2 and w 3 represent different weight values, and represent the significance rates of three working conditions respectively;

[0078] 6) Obtain the result after kmeans clustering;

[0079] 7) Calculate the target similarity between each pressure monitoring point and the target scheduling object. The target similarity R os The expression is:

[0080] R os = 1 / DTW(P i , P obj ), where P i is the pressure change pattern of any pressure measurement point, and P obj is the pressure change pattern of the standard scheduling object. DTW(a, b) represents calculating the DTW distance between sequences a and b;

[0081] 8) Sort the target similarities within the class, select the pressure measurement point with the highest target similarity as the master control point within the class, and generate the master control point distribution plan.

[0082] To verify the effectiveness of the method of the present invention, an actual operation experiment is now carried out based on an embodiment with the actual data of the input water supply network in a certain area. The operation process includes:

[0083] 1. Data preprocessing

[0084] Take the calculation period of a certain area from October 1 of a certain year to November 17 of the following year, and input the monitoring data of all pressure measurement points in the scheduling area during this period. Set an appropriate sliding window size, and filter out abnormal patterns (outliers, invalid values) in the pressure monitoring data according to abnormal conditions. Then, remove the scale effect of the pressure data through data standardization, and reduce the dimension of the data through downsampling to improve the efficiency of clustering calculation.

[0085] Refer to Figure 3 , classify the sample data according to different operating conditions, divide them into over-supplying water, superimposed pressure water supply, and clear water tank water supply, and construct classification features according to different operating conditions.

[0086] 2. Time series clustering

[0087] The DTW algorithm and the Kmeans clustering algorithm are implemented using Python, and the traditional Euclidean distance in the Kmeans clustering algorithm is replaced with the DTW distance. The initial number of clustering clusters is set in the range of [3, 20], and the pressure change patterns of all pressure measurement points under different working conditions are clustered by traversing the initial number of clustering clusters. For each given initial number of clustering clusters k, 5 clustering operations are randomly performed, and the maximum number of iterations for each clustering is 100. The pattern significance rate and weighted significance rate under three working conditions corresponding to different numbers of clustering clusters k are shown in Table 1:

[0088] Table 1: Pattern significance rate under three working conditions corresponding to different numbers of clustering clusters k

[0089]

[0090] Select the initial number of clustering clusters 12 with the largest weighted significance rate, and the spatial visualization of the clustering results under three working conditions is as Figure 4 shown.

[0091] 3. Optimization of master control points

[0092] Calculate the target similarity of the pressure measurement points in each category and sort them. The point with the highest target similarity in each category is used as the master control point. The distribution of the master control points calculated by the model is as Figure 5 shown.

[0093] It can be seen that for the method for identifying key pressure control points of a water supply network in the present invention, DTW (Dynamic Time Warping) is introduced as a measure of similarity distance to solve the time delay problem caused by pressure propagation, and Kmeans clustering is combined to analyze the large historical pressure monitoring data to identify pressure monitoring points with similar pressure change trends, and the corresponding key pressure control points are dynamically identified according to the correlation between the pressure change trends of the pressure monitoring points and the water plant / pumping station water outlet pressure change trends.

[0094] The above embodiments only illustratively explain the principles and effects of the present invention, rather than limiting the present invention. Any person skilled in the art can modify and change the above embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of the rights protection of the present invention shall be as listed in the claims.

Claims

1. A method for identifying key pressure control points in a water supply pipe network, comprising the following steps: Step S1, input the historical data of each pressure measurement point in the water supply pipe network, eliminate abnormal pressure patterns and extract significant pressure patterns from the pressure measurement point data, and construct classification features for the original sample data according to the different operating conditions of the intermediate pressure boosting pump stations in the water supply pipe network; The elimination of the abnormal pressure pattern further includes the following steps: Step S111, determine a suitable time sliding window size and set abnormal conditions; Step S112, eliminate the abnormalities in the sample data according to the abnormal value conditions; Step S2, establish a pressure change pattern clustering model, perform iterative clustering on the sample data of all pressure measurement points in the water supply pipe network under different classification features, and determine the most suitable number of clustering clusters according to the weighted pattern significance rate; Step S3, for different dispatching objects, sort according to the target similarity in each category, and select the pressure measurement point with the highest target similarity in each category as its pressure main control point.

2. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, in step S111, the abnormal conditions include: First abnormal condition: For any point within the time sliding window where sz is the time sliding window size, μ P(j,j+sz) is the average value of the pressure data within any time sliding window, and σ P(j,j+sz) is the standard deviation of the pressure data within any time sliding window. When any sample data meets the first abnormal condition, the data is excluded; Second abnormal condition: All data σ within the time window P(j,j+sz) = 0, when any one of the sample data meets the second abnormal condition, the data is excluded.

3. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, In step S112, the abnormal P of the rejection sample err is as follows: Wherein, and are the abnormal data of the first abnormal condition and the second abnormal condition respectively.

4. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, in step S1, the extraction of the significant pressure pattern further includes the following steps: Step S121, perform standardization processing on the pressure measurement point data, specifically: Among them, P origin is the original sample data, and P new is the sample data after data standardization; Step S122, perform data downsampling, specifically: P upsample = upsample(P new ), where P upsample is the downsampled sample data.

5. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, in step 1, the operating conditions of the intermediate pressure boosting pump stations include the superimposed pressure water supply condition, the bypass water supply condition and the clear water tank water supply condition.

6. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, in step S2, it further includes the following steps: Step S21, select a suitable clustering cluster interval [c min , c max , where c min is the minimum number of clustering clusters, and c max is the maximum number of clustering clusters; Step S22, use dynamic time warping as the similarity measure between different pressure change patterns, and use the Kmeans clustering algorithm to traverse the clustering cluster interval, and determine the number of clustering clusters and the corresponding clustering results according to the weighted pattern significance rate.

7. A method for identifying key pressure control points in a water supply pipe network according to claim 6, characterized in that, in step S22, the calculation of the weighted pattern significance rate further includes the following steps: Step S221, calculate the pattern significance rate of each classification feature, and the pattern significance rate is R stable Specifically: R stable = k stable / k, where k is the number of any clusters, and k stable is the number of clusters whose clustering results do not change after iterating N times with randomly initialized cluster positions; Step S222, calculate the weighted pattern significance rate, and weight the pattern significance rates under different operating conditions according to the proportion of the sample data under different operating conditions of the intermediate pressure boosting pump stations, specifically: Among them, w 1 , w 2 and w 3 represent different weight values, and respectively represent the significance rates of three working conditions.

8. A method for identifying key pressure control points in a water supply pipe network according to claim 1, characterized in that, in step S3, it further includes the following steps: Step S31, calculate the target similarity between each pressure monitoring point and the target scheduling object, and the target similarity is R os Specifically: R os = 1 / DTW(P i , P obj ), Among them, P i is the pressure change pattern of any pressure measurement point, and P obj is the pressure change pattern of the target to be calibrated. DTW(a, b) is the DTW distance between sequences a and b; Step S32, sort according to the in-class target similarity, and select the pressure measurement point with the highest target similarity as the in-class main control point.

Images