A design method for an asynchronous sensor system for real-time verification of water demand at water supply network nodes

Abstract

The present invention discloses a method for designing an asynchronous pressure sensor system for real-time verification of water demand at water supply network nodes. After combining the pressure sensitivity matrix with the clustering method, an asynchronous pressure sensor system design method for real-time verification of water demand at water supply network nodes is established. The main steps are as follows: (1) Solving the sensitivity matrix of the water supply network pressure sensor; (2) Calculating the correlation coefficient between each column of the sensitivity matrix and creating a distance matrix; (3) Using the distance matrix to perform cluster analysis on the pressure sensor to obtain several sensor groups; (4) Staggering the upload time of the sensor group to form an asynchronous sensor system. The asynchronous sensor system design method proposed in the present invention can effectively improve the verification frequency of the real-time hydraulic model.

Description

Technical Field

[0001] The invention belongs to the category of urban water supply pipe networks, and specifically is a design method of an asynchronous sensor system for real-time verification of water demand at water supply pipe network nodes. Background Art

[0002] As an important part of urban public facilities, the water supply network system delivers drinking water to thousands of households to meet people's production and living needs. With the development of urbanization in my country, urban water consumption has increased year by year, but the water supply system is facing serious aging problems, which makes the water supply network face many challenges such as excessive water supply energy consumption, serious network leakage, frequent pipe bursts and pollutant invasion. These problems have prompted water companies to control and manage the water supply network system more efficiently. The hydraulic model is an important tool to improve the operation and management level of the water supply network.

[0003] The real-time model can dynamically simulate the operating conditions of the pipe network system, such as node pressure, pipeline flow and water quality status. In the pipe networks of many cities, real-time hydraulic models are used for system scheduling, leakage control and pipe burst analysis. Among the many parameters of the real-time hydraulic model, the node water demand is a time-varying parameter with high uncertainty. Due to current technical and economic limitations, the node water demand cannot be measured directly. Accurate estimation of node water demand is the key to the correct use of real-time hydraulic models. Due to the development of online monitoring technology, the use of a large amount of measured data uploaded by monitoring points arranged in the pipe network to verify (invert) the node water demand in real time has received widespread attention and application in the industry.

[0004] Over the past 30 years, with the continuous development of sensor technology, most water companies in my country have installed a large number of pressure sensors in the water supply network. However, due to the different years of arrangement, the performance of the hardware and power supply of the sensors varies. Limited by the nature of the hardware, some sensors usually upload data at a relatively low frequency. This leads to a problem that the sensor upload frequency cannot meet the needs of real-time water demand verification. For example, in order for the real-time model to meet the needs of online scheduling, the real-time model should verify the node water demand every 15 minutes, while the existing old sensors can only upload data every 60 minutes. It is not feasible to install a large number of new high-performance (high-frequency) sensors in a short period of time, or to replace all old sensors with new sensors. Therefore, how to make the best use of the existing sensors in the network and increase the frequency of water demand verification of the real-time hydraulic model is of great practical significance.

[0005] In view of this, the present invention proposes an asynchronous sensor system design method for estimating water demand of nodes in a real-time hydraulic model of a water supply network, which improves the efficiency of water demand verification by relying on existing sensors. Summary of the invention

[0006] The present invention aims to propose a method for designing an asynchronous sensor system for real-time verification of water demand at a water supply network node. To achieve the above purpose, the present invention adopts the following steps:

[0007] The present invention first discloses a method for designing an asynchronous pressure sensor system for real-time verification of water demand at a water supply network node, which comprises the following steps:

[0008] (1) Solve the sensitivity matrix of the water supply network pressure sensor;

[0009] (2) Calculate the correlation coefficient between each column of the sensitivity matrix and create a distance matrix;

[0010] (3) Use the distance matrix to perform cluster analysis on the sensors and obtain a pressure sensor grouping scheme;

[0011] (4) Set different data upload time points for each group of sensors to form an asynchronous sensor system.

[0012] As a preferred embodiment of the present invention, the step (1) is specifically:

[0013] Assume that the pipe network has n nodes and m pressure sensors; distribute the total water supply of the water plant to n nodes according to the pipe length, and obtain the benchmark water demand X = [x1, x2, ..., x n ] T ), bring the benchmark water demand into each node of the water supply network model, perform hydraulic adjustment on the model, and obtain the benchmark pressure H = [H1, H2, …, H n ] T ; Select the pressure values ​​of the nodes where m pressure sensors are located from H And the pressure value H of node i i ;

[0014] For node i, its water demand is disturbed by Δx i , get the water demand after disturbance Keep the water demand of other nodes unchanged, that is, the water demand of n nodes after disturbance is The water demand after disturbance Bring the pipe network model into the hydraulic balance to obtain the node pressure after disturbance from Select the pressure values ​​of the nodes where m pressure sensors are located And the pressure value H′ of node i i ;

[0015] The sensitivity of the node pressure of m sensors to the pressure of node i is calculated by the following formula:

[0016]

[0017] Repeat the above process, traverse all nodes i=1,2,…,n, and get,

[0018]

[0019] Then the sensitivity matrix S of the pressure sensor is obtained;

[0020]

[0021] There are m column vectors in S, corresponding to m pressure sensors, and the column vector S j (j=1,2,…,m) has n elements, representing the sensitivity of sensor j to the pressure of n nodes.

[0022] As a preferred embodiment of the present invention, the step (2) is specifically:

[0023] The correlation coefficients between each column of the sensitivity matrix S are calculated to obtain the distance matrix D between the pressure sensors:

[0024] dist i,j =corr(S(i),S(j))

[0025]

[0026] Among them, dist i,j is the distance between sensor i and sensor j; S(i) and S(j) are the sensitivity vectors of sensor i and sensor j, and corr(a,b) is the correlation between vector a and vector b.

[0027] As a preferred embodiment of the present invention, step (3) is specifically:

[0028] According to the existing sensor upload time step T1 and the set asynchronous upload time step T2, where T1>T2, the number of monitoring point groups NG is calculated; the calculation method of NG is,

[0029]

[0030] |.| represents rounding down;

[0031] Using the K-means clustering method, each sensor is initially a separate group. Each iteration calculates the distance between all groups and merges the two groups with the closest distance until the number of groups is reduced to the preset value NG. The distance between groups is calculated as follows:

[0032]

[0033] Among them, L i,jis the distance between group i and group j; n i With n j is the number of sensors in group i and group j; x i,k is the kth sensor in group i, x j,l is the lth sensor in group j; finally, the sensor groups with the set number of groups are obtained through clustering.

[0034] As a preferred embodiment of the present invention, the step (4) is specifically:

[0035] Different data upload time points are set for each group of sensors to form an asynchronous sensor system. That is, without changing the upload frequency of each sensor, the upload time of some sensor groups is delayed to increase the data upload frequency of the entire sensor system. The delay time of each group of sensors is,

[0036] t k =(k-1)*T2,k=1,2,…,NG.

[0037] For example, if the existing sensor upload time step T1 = 30 minutes, and the set asynchronous upload time step T2 = 10 minutes, the number of monitoring point groups is At this time, the delay time of the first group is (1-1)*10=0min, the delay time of the second group is (2-1)*10=10min, and the delay time of the third group is (3-1)*10=20min.

[0038] To further illustrate the asynchronous upload strategy, Figure 1 Provide explanation.

[0039] like Figure 1 As shown, Figure 1 (a) is a traditional synchronous monitoring data uploading system. The network has 6 sensors (black circles), and these sensors can only upload one data every 30 minutes at most (0:00, 0:30, 1:00, ...). In order to increase the data uploading frequency, the present invention proposes an asynchronous monitoring data uploading system, see Figure 1 (b) If Figure 1As shown in (b), steps 1-3 are used to divide the 6 sensors into 3 groups, with two sensors in each group. The first group of monitoring points (black squares) upload data using the original time points (0:00, 0:30, 1:00, ...); the second group of monitoring points (black triangles) postpone the upload time by 10 minutes and upload data at (0:10, 0:40, 1:10, ...); the third group of monitoring points (black diamonds) postpone the upload time by 20 minutes and upload data at (0:20, 0:50, 1:20, ...). The above asynchronous upload system increases the original data collection every 30 minutes to data collection every 10 minutes (0:00, 0:10, 0:20, 0:30, 0:40, 0:50, 1:00, ...) without changing the sensor upload frequency, so as to meet the requirements of online water demand verification.

[0040] 5. Use the data uploaded by the sensor asynchronous upload system to verify water demand.

[0041] In the asynchronous sensor system proposed in the present invention, sensors in different groups upload data at each time node, and the data are used to verify the water demand of the node.

[0042] From step 4, it can be seen that the advantage of the asynchronous sensor system proposed in the present invention is that the data acquisition frequency is improved, and the disadvantage is that the amount of data available for each data is reduced. Figure 1 As shown, Figure 1 (a) Each time step can collect 6 data, and Figure 1 (b) Only 2 data can be collected at each time step. The reduction of data volume may reduce the accuracy of the model. In order to solve this problem, the present invention proposes an effective sensor grouping strategy (steps 1-4) to make the data uploaded by each group of sensors more representative.

[0043] To further explain the principles of steps 1-4, Figure 2 Explain the rationale and purpose of grouping. Figure 2 As shown in the figure, 12 sensors are installed in the network, which are divided into two groups, represented by black triangles and black circles. Using the asynchronous upload strategy, data can be obtained every 15 minutes. Intuitively, Figure 2 (b) Sensor grouping strategy Figure 2 (a) is more reasonable. This is because Figure 2 In (a), adjacent sensors are grouped together, resulting in excessive hydraulic correlation among sensors in the same group and serious homogenization of measured data. On the one hand, sensors in the same group upload homogeneous measurement data, resulting in information redundancy. On the other hand, this method only provides local hydraulic information of the water supply network, making it difficult to effectively verify the node water demand of the entire water supply network. Figure 2(b) Sensors with low similarity are grouped together. In this way, sensors in the same group are distributed throughout the water supply network, ensuring that each group has sufficient information about the entire network system, improving data representativeness and making it more reasonable.

[0044] Based on the above principle, the present invention uses a sensitivity matrix to measure the similarity of monitoring data and distributes dissimilar sensors into a group to improve the data representativeness of each time step, thereby improving the accuracy of water demand verification.

[0045] It is worth pointing out that after completing the above-mentioned monitoring point grouping and asynchronous uploading strategy, high-frequency sensors can be deployed as appropriate. The mixed use of high-frequency and low-frequency sensors reduces the cost of upgrading the monitoring system and has important economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of synchronous and asynchronous uploading of monitoring system data, where (a) is a synchronous monitoring data uploading system, (b) is an asynchronous monitoring data uploading system, and the grid in the figure represents the water supply network system;

[0047] Figure 2 This is a schematic diagram for comparing different asynchronous upload grouping strategies. The grid in the figure represents the water supply network system.

[0048] Figure 3 This is the water supply network and sensor location map of H City;

[0049] Figure 4 Group graph for sensors. DETAILED DESCRIPTION

[0050] The present invention aims to propose an asynchronous sensor system design method for real-time verification of water demand at water supply network nodes. The implementation of the present invention is further described in detail below with reference to the accompanying drawings and examples. The specific steps are as follows:

[0051] Step 1. Obtain the sensitivity matrix of the pressure sensor.

[0052] like Figure 3 As shown, the water supply network in H City has 3 water plants, 4242 water demand nodes (n=4242), 4841 pipe sections, a total length of 1576.98 kilometers, 48 ​​pressure sensors (m=48), and the water output of the water plant is known. The EPANET hydraulic solver is used to perform hydraulic simulation on the model.

[0053] Table 1 Water demand of nodes in the benchmark model

[0054]

[0055]

[0056] Table 2 Pressure of sensor nodes in the benchmark model

[0057]

[0058] The node and pipe numbers are the index order of EPANET nodes and pipes. Call the EPANET programmer toolbox to adjust the water supply network and obtain the water pressure of each sensor node in the benchmark model. Then increase the flow rate of node i (the flow rates of other nodes remain unchanged) and re-adjust the network to obtain the water pressure of each sensor node.

[0059]

[0060]

[0061] Among them, S ij For sensor node A j Sensitivity to the pressure at node i; H i , is the node i and sensor node A in the benchmark model j Pressure; H′ i , The relationship between node i and sensor node A after the traffic of node i changes j The pressure of the pipeline network; n is the number of pipeline network nodes, and m is the number of pipeline network pressure sensor nodes. Substitute the simulation value into the above formula to obtain the sensitivity matrix. In this embodiment, the entire sensitivity matrix S has 4242 rows and 48 columns, and a 10×10 example is taken:

[0062]

[0063] Step 2. Get the distance matrix.

[0064] The correlation coefficients are calculated for each pair of columns of the sensitivity matrix S to obtain the distance matrix D between the pressure sensors. Since the entire distance matrix has 48 rows and 48 columns, the following is an example of the first 10 rows and columns:

[0065]

[0066] Step 3. Perform cluster analysis to obtain several sensor groups.

[0067] Using the agglomerative clustering method, each sensor is initially a separate group, and the distance between all groups is calculated each time, and the two groups with the closest distance are merged until the number of clusters is reduced to a preset value. The distance between groups is calculated as follows:

[0068]

[0069] Among them, L i,j is the distance between group i and group j; n i With n j is the number of sensors in group i and group j; x i,k is the kth sensor in group i, x j,l is the lth sensor in group j.

[0070] The 48 sensors in the pipe network are clustered into two categories. The final result is that monitoring points 1, 2, 3, 5, 7, 9, 11, 12, 13, 15, 17, 19, 20, 22, 23, 26, 28, 29, 31, 33, 34, 36, 37, 41, 42 belong to category 1; monitoring points 4, 6, 8, 10, 14, 16, 18, 21, 24, 25, 27, 30, 32, 35, 38, 40, 43, 44, 45, 46, 47, 48 belong to category 2. Figure 4 Shows how sensors are grouped.

[0071] Step 4. Stagger the upload time to form an asynchronous sensor system.

[0072] The sensors in the water supply network of H City were originally set to upload data every 30 minutes, that is, at 00:00, 00:30, 1:00, etc. in a day, so the calibration frequency of the real-time hydraulic model is once every 30 minutes. Figure 2 The upload time of sensor group 2 (black circle in Figure 2) remains unchanged. The upload time of sensor group 2 (black triangle in Figure 2) is delayed by 15 minutes, that is, data is uploaded at 00:15, 00:45, 1:15, etc. in a day. After this setting, a batch of sensors will upload monitoring data every 15 minutes, so the calibration frequency of the real-time hydraulic model can be increased to once every 15 minutes.

[0073] Step 5. Use measured data to verify node water demand

[0074] The monitoring data of a certain day in H city were selected for verification. After the above grouping and adjustment of the upload time, the real-time hydraulic model was calibrated. Table 3 shows the average calibration error of 48 pressure monitoring points in 24 hours. It can be seen from Table 3 that the average error of most monitoring points is less than 0.5 meters, and the maximum error is 0.9 meters. The average error of all monitoring points is much less than 2 meters (international standard).

[0075] Table 3 Average calibration error of 48 pressure sensors

[0076]

[0077] This shows that the asynchronous sensor system design method for real-time verification of water demand at water supply network nodes proposed in the present invention can effectively improve the verification frequency of the real-time hydraulic model while also ensuring the accuracy of the verification, which is much higher than the industry standard.

[0078] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A method for designing an asynchronous pressure sensor system for real-time verification of water demand at water supply network nodes, characterized in that: The steps include: (1) Solve the sensitivity matrix of the water supply network pressure sensor; (2) Calculate the correlation coefficient between each column of the sensitivity matrix and create a distance matrix; (3) Use the distance matrix to perform cluster analysis on the sensors and obtain a pressure sensor grouping scheme; (4) Setting different data upload time points for each group of sensors to form an asynchronous sensor system; The step (1) is specifically: Assume that the pipe network has n nodes and m pressure sensors; distribute the total water supply of the water plant to n nodes according to the pipe length, and obtain the benchmark water demand X = [x1, x2, ..., x n ] T , bring the benchmark water demand into each node of the water supply network model, perform hydraulic adjustment on the model, and obtain the benchmark pressure H = [H1, H2, ..., H n ] T ; Select the pressure values ​​of the nodes where m pressure sensors are located from H And the pressure value H of node i i ; For node i, its water demand is disturbed by Δx i , get the water demand after disturbance Keep the water demand of other nodes unchanged, that is, the water demand of n nodes after disturbance is The water demand after disturbance Bring the pipe network model into the hydraulic balance to obtain the node pressure after disturbance from Select the pressure values ​​of the nodes where m pressure sensors are located And the pressure value H of node i i ′ ; The sensitivity of the node pressure of m sensors to the pressure of node i is calculated by the following formula: Repeat the above process, traverse all nodes i=1,2,…,n, and get, Then the sensitivity matrix S of the pressure sensor is obtained; There are m column vectors in S, corresponding to m pressure sensors, and the column vector S j There are n elements, representing the sensitivity of sensor j to the pressure of n nodes; The step (2) is specifically: The correlation coefficients between each column of the sensitivity matrix s are calculated to obtain the distance matrix D between the pressure sensors: dist i,j =corr(S(i),S(j)) Among them, dist i,j is the distance between sensor i and sensor j; S(i) and S(j) are the sensitivity vectors of sensor i and sensor j, corr(a,b) is the correlation between vector a and vector b; Step (3) is specifically as follows: According to the existing sensor upload time step T1 and the set asynchronous upload time step T2, where T1>T2, the number of monitoring point groups NG is calculated; the calculation method of NG is, |.| represents rounding down; Using the K-means clustering method, each sensor is initially a separate group. Each iteration calculates the distance between all groups and merges the two groups with the closest distance until the number of groups is reduced to the preset value NG. The distance between groups is calculated as follows: Among them, L i,j is the distance between group i and group j; n i With n j is the number of sensors in group i and group j; x i,k is the kth sensor in group i, x j,l is the lth sensor in group j; finally clustering obtains a set number of sensor groups; The step (4) is specifically: Different data upload time points are set for each group of sensors to form an asynchronous sensor system. That is, without changing the upload frequency of each sensor, the upload time of some sensor groups is delayed to increase the data upload frequency of the entire sensor system. The delay time of each sensor group is, t k =(k-1)*T2,k=1,2,…,NG.

Images