A sewer network system monitoring point optimization arrangement and liquid level real-time inversion method

Abstract

This invention relates to the field of monitoring point layout and real-time monitoring of sewage or rainwater drainage systems in urban smart drainage, and particularly to a method for optimizing the layout of monitoring points and real-time liquid level inversion in drainage pipe network systems. Using this method, on the one hand, the layout of monitoring points in the drainage system network can be optimized; on the other hand, existing liquid level monitoring equipment can be used to dynamically calculate the liquid level status of important nodes, quickly reflecting the operating status of the drainage system and the degree of flooding while ensuring high accuracy.

Description

Technical Field

[0001] This invention relates to the field of monitoring point layout and real-time monitoring of sewage or rainwater drainage systems in urban smart drainage, and particularly to a method for optimizing the layout of monitoring points and real-time liquid level inversion in drainage pipe network systems. Background Technology

[0002] With the acceleration of urbanization and the continuous expansion of urban construction, urban drainage problems are becoming increasingly prominent. The concept of a smart urban drainage system emphasizes that a comprehensive understanding of the operational status of the urban drainage system, such as liquid level, flow rate, siltation, and leakage, is crucial for ensuring the safe and efficient operation of the drainage system.

[0003] A real-world urban drainage network is a massive system. Its complexity, the interplay of system boundaries, and the uncertainty of actual rainfall events make it highly sensitive to real-time changes in its status. Therefore, to achieve a truly rapid and accurate response to the drainage system's operation, a large number of monitoring devices are required. However, in actual drainage networks, deploying sensors at all critical nodes and outlets often necessitates a large number of devices, resulting in high infrastructure and management costs. Furthermore, the large amount of redundant data reduces the online monitoring response capability of the drainage system. Conversely, reducing the number of monitoring devices to lower costs would decrease the reliability of the monitoring and assessment results.

[0004] Current methods for optimizing the layout of monitoring points in drainage pipe networks primarily employ clustering-based dimensionality reduction methods (Chinese patents with publication numbers CN116887297A, CN113221440A, and CN115329514A) and weight-based methods (Chinese patent with publication number CN106651040A). Clustering processes can lead to some information loss, and both clustering and weighting methods are highly subjective, impacting the optimized layout of monitoring points. Therefore, a comprehensive and objective method for optimizing the layout of monitoring points is lacking. Furthermore, with a limited number of liquid level monitoring devices, rapidly and accurately acquiring real-time liquid level data from key nodes in urban drainage systems is a pressing issue, crucial for guiding daily sewage system operations, urban flood control scheduling, and accelerating emergency command by relevant departments.

[0005] Therefore, a method for optimizing the layout of monitoring points and real-time liquid level inversion in drainage pipe network systems is proposed to achieve more objective and comprehensive optimization of monitoring point layout and real-time liquid level data inversion of important nodes. Summary of the Invention

[0006] The purpose of this invention is to solve the problems existing in the current drainage pipe network technology and to provide a method for optimizing the layout of monitoring points and real-time liquid level inversion in drainage pipe network systems. Using this method, on the one hand, the monitoring points of the drainage system network can be optimized, and on the other hand, existing liquid level monitoring equipment can be used to dynamically calculate the liquid level status of important nodes, so as to quickly reflect the operating status of the drainage system while ensuring high accuracy.

[0007] A method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system includes the following steps:

[0008] (1) For the drainage network system to be studied, collect regional network data and underlying surface data, and construct an SWMM model (Storm Water Management Model);

[0009] (2) Collect historical actual rainfall events with different rainfall intensities, and then generate simulated rainfall events using the rainstorm intensity formula of the study area through different combinations of rainfall pattern parameters. Combine the actual rainfall events and simulated rainfall events to form a rainfall dataset;

[0010] (3) Based on the drainage network SWMM model constructed in (1), the rainfall events in the rainfall dataset in (2) are used as inputs to the SWMM network model, and the liquid level data of each node are extracted from the calculation output results of the model to form the SWMM model result dataset.

[0011] (4) Based on the liquid level data of each node in the result dataset in (3), calculate the amount of information (VOI) contained in each node of the pipeline network and the cross-information entropy (TE) between each node to form the VOI matrix and the TE matrix.

[0012] (5) Based on the VOI matrix and TE matrix calculated in (4), the optimization objectives are to maximize the total information content of the selected nodes, minimize the cross-information entropy between nodes, and minimize the number of nodes. The Pareto front is found using the NSGA-II multi-objective intelligent optimization algorithm. A suitable number of nodes are selected to form a set of monitoring points according to the existing monitoring points in the study area and the actual engineering needs.

[0013] (6) Using the liquid level of the existing monitoring points in the set of monitoring points selected in (5) as input and the liquid level of all monitoring points in the set of monitoring points as output, construct an AutoEncoder neural network model, and use the result dataset in (3) to train the AutoEncoder model to obtain a real-time inversion model of the monitoring liquid level.

[0014] Furthermore, in step (1), collecting the pipeline network data and underlying surface data of the area specifically includes:

[0015] The study area includes the location, burial depth, diameter, location, elevation, outlet location, elevation, and land use type of the pipeline network system.

[0016] Furthermore, in step (2), simulated rainfall events are generated using the rainstorm intensity formula for the study area through different combinations of rainfall pattern parameters. The specific rainstorm intensity formula is as follows;

[0017]

[0018] Where q is the design rainfall intensity (L / (s·hm)) 2 P is the design return period (a), t is the rainfall duration (min), and A1, C, b, n are local parameters determined according to the geographical location of the study area.

[0019] Furthermore, using the Chicago rainfall pattern, and based on the storm intensity formula for the study area, simulated rainfall events are generated by combining different return periods, rainfall durations, peak rainfall coefficients, and time steps.

[0020] Simulated rainfall events were generated based on the Chicago rainfall pattern using different combinations of rainfall pattern parameters and the storm intensity formula for the study area. Specifically, these events included:

[0021] When the duration of rainfall is:

[0022] t = t a +t b ,but

[0023]

[0024] Where t b For the duration of pre-peak rainfall, t a The duration of post-peak rainfall is given by r, which is the peak rainfall coefficient of the Chicago rain pattern.

[0025] The return period P is 1 year, 2 years, 5 years, 10 years, 20 years and 50 years, the rainfall duration t is 30 minutes, 60 minutes, 120 minutes, 180 minutes, 240 minutes, 360 minutes, 720 minutes and 1440 minutes, and the rainfall peak coefficient r is a combination of 0.2, 0.5 and 0.8 to generate simulated rainfall events.

[0026] Furthermore, in step (3), the rainfall events in the rainfall dataset from (2) are used as inputs to the SWMM network model one by one, specifically including:

[0027] The selected rainfall dataset includes real rainfall events and simulated rainfall events to cover rainfall events with different characteristics as much as possible.

[0028] Furthermore, in step (4), the amount of information (VOI) contained in each node of the pipeline network and the cross-information entropy (TE) between each node are calculated to form the VOI matrix and the TE matrix, specifically including:

[0029] (1) According to Bayes' theorem and information theory, decision-makers can update their prior probability of the system based on new node flood data. Therefore, the rainfall events in the rainfall dataset are randomly divided into a prior dataset and an evidence dataset, with sizes n and m, respectively.

[0030] (2) Based on the node liquid level, the state S of each pipeline node is divided into two types: state s1 (no flooding) and state s2 (flooding). If the liquid level of the node is higher than the ground elevation of the node during a rainfall event, it is state s2; otherwise, it is state s1. The number of times each node is in state s1 and state s2 in the prior dataset and evidence dataset are counted respectively.

[0031] (3) Based on the frequency of states s1 and s2 in the prior dataset and evidence dataset, when new information appears, according to Bayes' theorem, the updated prior probability can be expressed as:

[0032]

[0033] Where P(s|m) is the prior probability updated after receiving information m, P(m|s) is the conditional probability of receiving information m when the system state is s, P(m) is the probability of receiving information m, and P(s) is the evidence probability.

[0034] Suppose that municipal managers can obtain the liquid level information at node j using a liquid level sensor placed at node i, i.e., whether flooding has occurred. Timely warnings can minimize the negative impact of flooding, while warnings issued too early or too late can lead to public panic and disrupt normal travel, respectively. Therefore, when the state at node j is s, actions (or warnings) a with different lag times can be defined, along with their associated consequences C(a,s). Based on the received information u... m and unreceived information u s The effectiveness of these actions (or warnings) can be calculated as follows:

[0035]

[0036] Where S is the total number of states s; u m Based on the operational utility of the received information, u s This refers to the operational utility of not receiving information.

[0037] Therefore, the action a selected based on the received information m m The value is u m -us A rational decision-maker will choose the action with the greatest utility; therefore, the value of the VoI (Value of Information) used by node i to determine the state detected by node j can be calculated using the following formula:

[0038]

[0039] Where M represents the total number of messages; VOI i (j) represents the information value of node i used to determine the detection state of node j.

[0040] The above calculations are performed on each node to form the VOI matrix.

[0041] (4) The cross-information entropy TE(i,j) between two selected nodes (i and j) is calculated in discrete form as follows:

[0042]

[0043] Where, P(i s ,j SS Let P(i) be the joint probability when the state of node i is s. S ) and P(j SS Let and be the probabilities of nodes i and j being in detection states s and ss, respectively. TE(i,j) is the cross-information entropy value used by node i to determine the detection state of node j. Similar to the VOI calculation, the above calculation is performed for each node, ultimately forming the TE matrix.

[0044] Furthermore, in step (5), the optimization objectives are to maximize the total information content of the selected nodes, minimize the cross-information entropy between nodes, and minimize the number of nodes. The Pareto front is found using the NSGA-II multi-objective intelligent optimization algorithm, specifically including:

[0045] Using the VOI matrix and TE matrix, construct a vector containing binary decision variables. A multi-objective optimization model is used to find the minimum number of nodes with the maximum VOI and the minimum TE. The optimization objective is expressed as follows:

[0046]

[0047] Among them, b i b j Z1, Z2, and Z3 are two-variable decision variable vectors, and Z1, Z2, and Z3 are three optimization objectives.

[0048] Furthermore, in step (6), the liquid level of existing monitoring points in the set of monitoring points selected in (5) is used as input, and the liquid level of all monitoring points in the set of monitoring points is used as output to construct an AutoEncoder neural network model, specifically including:

[0049] The constructed AutoEncoder model is an encoder-decoder structure, using an unsupervised learning approach. Both the input and output are liquid level data. Rainfall intensity is added to the input as an external variable of the decoder in the AutoEncoder model. The number of nodes in the encoder input layer is the same as the number of existing monitoring points, and the number of nodes in the decoder output layer is the same as the number of monitoring points finally selected.

[0050] For the liquid level inversion results, the coefficient of determination (R²) is used. 2 The accuracy of the algorithm completion is evaluated using the mean absolute error (MAE) and the mean absolute error (MAE). The specific calculation formula is as follows:

[0051]

[0052] Among them, y i This represents the actual monitoring value. This indicates the calculation of the inversion value. This represents the average value of the actual monitored values, and n represents the number of monitored values.

[0053] The advantages of this invention over the prior art are:

[0054] The drainage pipe network system monitoring point optimization and real-time liquid level inversion method provided by the present invention adopts an objective and comprehensive approach to achieve scientific selection and layout of monitoring points, optimizes the layout of monitoring points, and can accurately and in real-time invert the liquid level height of important nodes in the drainage pipe network through existing monitoring points, so as to quickly reflect the operating status of the drainage system and the degree of waterlogging while ensuring high accuracy. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of a drainage pipe network system in a certain urban area according to an embodiment of the present invention.

[0056] Figure 2 This is a flowchart of the monitoring point optimization layout and real-time liquid level inversion method in an embodiment of the present invention.

[0057] Figure 3 This is a normalized information graph of a node in an embodiment of the present invention.

[0058] Figure 4 This is a cross-information entropy graph (normalized) of a node in an embodiment of the present invention.

[0059] Figure 5 This is a schematic diagram showing the locations of different numbers of recommended liquid level monitoring points in embodiments of the present invention. Detailed Implementation

[0060] The specific implementation of the present invention will be described in detail below with reference to the accompanying drawings and embodiments, so as to enable those skilled in the art to better understand the present invention.

[0061] In one embodiment of the present invention, it is necessary to... Figure 1 The table shows the real-time global monitoring and inversion of the liquid level of the rainwater drainage system nodes in a certain urban area of ​​City H. The city has an average altitude of 186 meters and an average annual rainfall of 1456.6 mm. The rainfall is unevenly distributed throughout the four seasons, mainly concentrated in the spring and summer transition season, which makes the pipe network prone to overflow. Figure 1 The target area's pipe network shown has 4062 nodes, and its drainage system is a combined sewer system. This embodiment uses 162 historical rainfall events obtained from the H City Meteorological Bureau, and 144 simulated rainfall events. This invention uses GIS software for pipe network topology processing and Python for programming calculations. Specifically, it includes the following steps:

[0062] S1: For the drainage network system to be studied, collect regional network data and underlying surface data, and construct an SWMM model (Storm Water Management Model);

[0063] S2: Collect historical actual rainfall events with different rainfall intensities, and then generate simulated rainfall events using the rainstorm intensity formula of the study area through different combinations of rainfall pattern parameters. Combine the actual rainfall events and simulated rainfall events to form a rainfall dataset.

[0064] S3: Based on the drainage network SWMM model constructed in S1, the rainfall events in the rainfall dataset in S2 are used as inputs to the SWMM network model one by one. The liquid level data of each node is extracted from the model's calculation output to form the SWMM model result dataset.

[0065] S4: Based on the liquid level data of each node in the result dataset of S3, calculate the amount of information (VOI) contained in each node of the pipeline network and the cross-information entropy (TE) between each node to form the VOI matrix and the TE matrix.

[0066] S5: Based on the VOI matrix and TE matrix calculated in S4, the optimization objectives are to maximize the total information content of the selected nodes, minimize the cross-information entropy between nodes, and minimize the number of nodes. The Pareto front is found using the NSGA-II multi-objective intelligent optimization algorithm, and an appropriate number of nodes are selected to form a set of monitoring points according to the existing monitoring points in the study area and actual needs.

[0067] S6: Using the liquid level of existing monitoring points in the set of monitoring points selected in S5 as input and the liquid level of all monitoring points in the set of monitoring points as output, construct an AutoEncoder neural network model. Use the result dataset in S3 to train the AutoEncoder model to obtain a real-time inversion model of the monitored liquid level.

[0068] For the liquid level inversion results, the coefficient of determination (R²) is used. 2 The accuracy of the algorithm completion is evaluated using the mean absolute error (MAE) and the mean absolute error (MAE). The specific calculation formula is as follows:

[0069]

[0070] Among them, y i This represents the actual monitoring value. This indicates the calculation of the inversion value. This represents the average value of the actual monitored values, and n represents the number of monitored values.

[0071] The flowcharts for methods S1 to S6 above are as follows: Figure 2 As shown, the drainage pipe network system monitoring point optimization and real-time liquid level inversion method provided by this invention achieves scientific selection and layout of monitoring points using an objective and comprehensive approach, optimizes the monitoring point layout, and can accurately and in real-time invert the liquid level height of important nodes in the drainage pipe network (without the need to install monitoring equipment) using existing monitoring points, thus quickly reflecting the operating status of the drainage system and the degree of waterlogging while ensuring high accuracy.

[0072] In this embodiment, to implement the above-mentioned monitoring point optimization and real-time inversion method, monitoring point optimization and level inversion verification are performed based on the five existing level gauges in the target area's drainage network. Using the five existing level gauges (A1 to A5) as historical monitoring points, the proposed drainage network system monitoring point optimization method is used to find other recommended monitoring points that best match them.

[0073] First, 162 and 144 historical rainfall events were iteratively imported into the constructed SWMM model to obtain the result dataset. Next, the information content and cross-information entropy of each node were calculated. The 306 rainfall events were divided into a prior dataset (240 events) and an evidence dataset (66 events). The division method was to randomly select 33 events from each of the historical rainfall events as the evidence dataset, and the remaining events as the prior dataset. Then, the prior probability was calculated using the following formula:

[0074]

[0075] Where P(s|m) is the prior probability updated after receiving information m, P(m|s) is the conditional probability of receiving information m when the system state is s, P(m) is the probability of receiving information m, and P(s) is the evidence probability.

[0076] When the state at node j is s, actions (or warnings) a with different lag times and their associated consequences C(a,s) can be defined. Based on the received information u... m and unreceived information u s The effectiveness of these actions (or warnings) can be calculated as follows:

[0077]

[0078]

[0079] Where S is the total number of states s.

[0080] Therefore, the action a selected based on the received information m m The value is u m -u s A rational decision-maker will choose the action with the greatest utility; therefore, the value of the VoI (Value of Information) used by node i to determine the state detected by node j can be calculated using the following formula:

[0081]

[0082] Where M is the total number of information, VOI i (j) represents the information value of node i used to determine the detection state of node j.

[0083] The above calculations are performed on each node, ultimately forming the VOI matrix. For example... Figure 3 As shown in the figure, this graph represents the VOI value (after normalization) of a certain node.

[0084] Next, calculate the cross-information entropy TE(i,j) between the two selected nodes (i and j) in discrete form. The specific calculation formula is as follows:

[0085]

[0086] Where, P(i s ,j ss Let P(i) be the joint probability when the state of node i is s. s ) and P(j ss Let and represent the probabilities of nodes i and j exhibiting detection states s and ss, respectively. TE(i,j) is the cross-information entropy value used by node i to determine the detection state of node j. Similar to the VOI calculation, the above calculation is performed for each node, ultimately forming the TE matrix. Figure 4As shown in the figure, this graph represents the TE value (after normalization) of a certain node.

[0087] Using the VOI matrix and TE matrix, construct a vector containing binary decision variables. A multi-objective optimization model is used to find the minimum number of nodes with the maximum VOI and the minimum TE. The optimization objective is expressed as follows:

[0088]

[0089] Among them, b i b j The decision variable vector is represented by Z1, Z2, and Z3, which are three optimization objectives. The Pareto front obtained through multi-objective optimization is used to select an appropriate number of monitoring points based on actual needs, such as funding constraints and regulations regarding the number of monitoring devices. Then, with a fixed number of monitoring points, the combination of points providing the most information is selected as the recommended liquid level monitoring points. For example... Figure 5 As shown in the figure, the optimal layout of the liquid level monitoring equipment is illustrated when 10, 15, 20, and 30 monitoring points are selected (representing 200%, 300%, 400%, and 600% of the existing number of monitoring points, respectively).

[0090] Furthermore, with Figure 5 Using the liquid levels of existing monitoring points in the selected monitoring point set as input and the liquid levels of all monitoring points in the set as output, an AutoEncoder neural network model is constructed, specifically including:

[0091] The constructed AutoEncoder model is an encoder-decoder structure, employing unsupervised learning. Both input and output are liquid level data. Rainfall intensity is added to the input as an external variable in the decoder. The number of nodes in the encoder input layer is the same as the existing number of monitoring points, and the number of nodes in the decoder output layer is the same as the final selected number of monitoring points. The liquid level inversion results (test set) are shown in Table 1, indicating that the measured values ​​of the target drainage system node liquid levels in this urban area are basically consistent with the inverted calculated values, where R... 2 The accuracy can reach 0.98 to 0.99, indicating that the method has high inversion accuracy and can obtain liquid level information of important nodes without setting up real monitoring equipment.

[0092] Table 1 Evaluation of Inversion Algorithms

[0093]

[0094] The embodiments described above are merely one application of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation are within the protection scope of the present invention.

Claims

1. A method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system, comprising the following steps: (1) For the drainage network system to be studied, collect the network data and underlying surface data of the area, and construct the SWMM model; (2) Collect historical actual rainfall events with different rainfall intensities, and then generate simulated rainfall events by different combinations of rainfall pattern parameters and the formula for the intensity of rainstorms in the study area. Combine the actual rainfall events and simulated rainfall events to form a rainfall dataset; (3) Based on the drainage network SWMM model constructed in (1), the rainfall events in the rainfall dataset in (2) are used as inputs to the SWMM network model, and the liquid level data of each node are extracted from the calculation output results of the model to form the SWMM model result dataset. (4) Based on the liquid level data of each node in the result dataset in (3), calculate the amount of information VOI contained in each node of the pipeline network and the cross-information entropy TE between each node to form the VOI matrix and the TE matrix. (5) Based on the VOI matrix and TE matrix calculated in (4), the optimization objectives are to maximize the total information content of the selected nodes, minimize the cross-information entropy between nodes, and minimize the number of nodes. The Pareto front is found using the NSGA-II multi-objective intelligent optimization algorithm. A suitable number of nodes are selected to form a set of monitoring points according to the existing monitoring points in the study area and the actual engineering needs. (6) Using the liquid level of existing monitoring points in the set of monitoring points selected in (5) as input and the liquid level of all monitoring points in the set of monitoring points as output, construct an AutoEncoder neural network model, and use the result dataset in (3) to train the AutoEncoder model to obtain a real-time inversion model of monitoring liquid level.

2. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (1), the collection of regional pipeline network data and underlying surface data specifically includes: The study area includes the location, burial depth, pipe diameter, location, elevation, outlet location, elevation, and land use type of the pipeline network system.

3. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (2), a simulated rainfall event is generated based on the Chicago rain pattern. The specific formula for the intensity of the rainstorm is as follows; Where q is the design rainfall intensity L / (s·hm) 2 P is the design return period a, t is the rainfall duration min, and A1, C, b, n are local parameters determined according to the geographical location of the study area.

4. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 3, characterized in that, In step (2), simulated rainfall events are generated based on the Chicago rainfall pattern using different combinations of rainfall pattern parameters and the rainfall intensity formula for the study area. Specifically, this includes: When the duration of rainfall is: t = t a +t b ,but Where t b For the duration of pre-peak rainfall, t a The duration of post-peak rainfall is given by r, which is the peak rainfall coefficient of the Chicago rain pattern. The return period P is 1 year, 2 years, 5 years, 10 years, 20 years or 50 years, the rainfall duration t is 30 minutes, 60 minutes, 120 minutes, 180 minutes, 240 minutes, 360 minutes, 720 minutes or 1440 minutes, and the rainfall peak coefficient r is 0.2, 0.5 or 0.8 to generate simulated rainfall events.

5. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (3), the rainfall events in the rainfall dataset from (2) are used as inputs to the SWMM network model one by one, specifically including: The selected rainfall dataset includes real rainfall events and simulated rainfall events to cover rainfall events with different characteristics as much as possible.

6. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (4), the information content (VOI) of each node in the pipeline network and the cross-information entropy (TE) between each node are calculated to form the VOI matrix and the TE matrix, specifically including: (1) According to Bayes' theorem and information theory, decision-makers can update their prior probability of the system based on new node flood data. Therefore, the rainfall events in the rainfall dataset are randomly divided into a prior dataset and an evidence dataset, with sizes n and m, respectively. (2) Based on the node liquid level, the state S of each pipeline node is divided into two types: state s1 without flooding and state s2 with flooding. If the liquid level of the node is greater than the ground elevation of the node during a rainfall event, it is state s2; otherwise, it is state s1. The number of times each node is in state s1 and state s2 in the prior dataset and evidence dataset are counted respectively. (3) Based on the frequency of states s1 and s2 in the prior dataset and evidence dataset, when new information appears, according to Bayes' theorem, the updated prior probability can be expressed as: Where P(s|m) is the prior probability updated after receiving information m, P(m|s) is the conditional probability of receiving information m when the system state is s, P(m) is the probability of receiving information m, and P(s) is the evidence probability. Suppose that municipal managers can obtain the liquid level information at node j using a liquid level sensor placed at node i, i.e., whether flooding has occurred; timely warnings can minimize the negative impact of flooding, while warnings issued too early or too late can cause public panic and disrupt normal travel, respectively; therefore, when the state at node j is s, actions or warnings a with different lag times can be defined, along with their associated consequences C(a,s); the effectiveness of these actions or warnings can be calculated based on the received and unreceived information as follows: Where S is the total number of states s, u m Based on the operational utility of the received information, u s For operations where no information has been received; Therefore, the action a selected based on the received information m m The value is u m -u s A rational decision-maker will choose the action with the greatest utility; therefore, the information value (VOI) of node i used to determine the state detected by node j can be calculated using the following formula: Where M is the total number of information, VOI i (j) represents the information content of node i used to determine the detection state of node j; The above calculations are performed on each node to ultimately form the VOI matrix; (4) In discrete form, the cross-information entropy TE(i,j) between two selected nodes i and j can be calculated as follows: Where, P(i s ,j ss Let P(i) be the joint probability of node i when its state is s, and the joint probability of node j when its state is ss. s ) and P(j ss ) are the probabilities of nodes i and j being detected in states s and ss, respectively. TE(i,j) is the cross-information entropy value of node i used to determine the detection state of node j. Similar to the VOI calculation, the above calculation is performed on each node to finally form the TE matrix.

7. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (5), the optimization objectives are to maximize the total information content of the selected nodes, minimize the cross-information entropy between nodes, and minimize the number of nodes. The Pareto front is found using the NSGA-II multi-objective intelligent optimization algorithm, which specifically includes: Using the VOI matrix and TE matrix, construct a vector containing binary decision variables. A multi-objective optimization model is used to find the minimum number of nodes with the maximum VOI and the minimum TE. The optimization objective is expressed as follows: Among them, b i b j Z1, Z2, and Z3 are two-variable decision variable vectors, and Z1, Z2, and Z3 are three optimization objectives.

8. The method for optimizing the layout of monitoring points and real-time liquid level inversion in a drainage pipe network system as described in claim 1, characterized in that, In step (6), the liquid level of existing monitoring points in the set of monitoring points selected in (5) is used as input, and the liquid level of all monitoring points in the set of monitoring points is used as output to construct an AutoEncoder neural network model, specifically including: The constructed AutoEncoder model is an encoder-decoder structure, using an unsupervised learning approach. Both the input and output are liquid level data. Rainfall intensity is added to the input as an external variable of the decoder in the AutoEncoder model. The number of nodes in the encoder input layer is the same as the number of existing monitoring points, and the number of nodes in the decoder output layer is the same as the number of monitoring points finally selected.

Images