Urban drainage pipe network clogging real-time analysis method based on street monitoring

Abstract

The invention provides an urban drainage pipe network clogging real-time analysis method based on street monitoring, and aims to solve the problems that urban drainage pipe network clogging is difficult to find in time, the clogging position and degree are difficult to determine, and professional pipe network monitoring equipment is high in cost and difficult to analyze and diagnose in real time. According to the method, the submerging condition is calculated in real time in combination with existing urban street monitoring and a convolutional neural network (CNN), and real-time judgment and analysis of the pipe network clogging condition and position are carried out by utilizing an urban rainfall flood model, a Monte Carlo method and a back propagation neural network (BP) coupling model; according to the city drainage pipe network clogging real-time analysis method based on street monitoring, professional equipment is not needed, hardware equipment does not need to be newly added, and real-time calculation of city rainwater pipe network clogging position and degree prediction can be achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of urban rain and flood simulation, and in particular relates to a real-time analysis method for clogging of an urban drainage network based on street monitoring. Background Art

[0002] Urban rainwater pipe network systems are prone to clogging during operation, which leads to a decrease in the drainage capacity of the pipe network, and then causes problems such as poor drainage during the flood season, road flooding and even urban waterlogging.

[0003] Existing methods for detecting blockage in rainwater pipe networks mainly include closed-circuit television (CCTV) detection, quick video detection (QV) detection, sonar detection, etc., but these methods require professional equipment and operators, and the cycle is long and the cost is high for comprehensive detection of long-distance or large-area pipe networks. In addition, these methods rely on regular inspections, cannot monitor the dynamic changes of pipes in real time, and are not flexible enough to respond to sudden siltation and blockage. The blockage of rainwater pipes is a dynamic process. Especially during heavy rainfall, dead branches and leaves, domestic garbage, mud and stones may cause the pipe network to be quickly blocked, and even the pipe structure may become unstable and collapse, resulting in a complete loss of drainage function. Traditional detection methods make it difficult to analyze and diagnose the location and degree of pipe network siltation in real time. Summary of the invention

[0004] The purpose of the present invention is to provide a real-time analysis method for urban drainage network blockage based on street monitoring, in the hope of providing a new idea for solving the real-time prediction and analysis of rainwater network blockage.

[0005] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0006] A real-time analysis method for urban drainage network blockage based on street monitoring includes the following steps:

[0007] S1. Organize basic data and establish a pipe network-surface double-layer urban stormwater model;

[0008] S2. Use the measured rainfall accumulation to complete the calibration of the pipe network-surface double-layer urban rainwater model;

[0009] S3, setting the pipe network blockage and rainfall condition parameters, and using the Monte Carlo method to generate different blockage locations and rainfall data sets for the pipe network-surface double-layer urban stormwater model;

[0010] S4. Input the data set into the pipe network-surface double-layer urban stormwater model to simulate the surface water accumulation when the pipe network and stormwater inlets at different locations are blocked;

[0011] S5. Using the pipe network blockage and surface water accumulation data set, train the BP neural network;

[0012] S6. Use surveillance videos of street waterlogging during heavy rains and the elevation and width of benchmark points to train a CNN neural network.

[0013] S7, input the street surveillance video into the CNN neural network to obtain the flooding conditions at different surface locations;

[0014] S8. Input the surface flooding situation into the BP neural network to obtain the siltation situation of different nodes.

[0015] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0016] As a preferred technical solution of the present invention: in step S1, the basic data includes rainwater pipe network data, surface elevation data, land use data, remote sensing image data and designed rainfall data.

[0017] As a preferred technical solution of the present invention: in step S1, the pipe network-surface double-layer urban stormwater model includes an underground rainwater pipe network module, a surface street and flood-prone point module, a rainwater pumping station module, and a rainwater storage tank module.

[0018] As a preferred technical solution of the present invention: in step S2, the current data of siltation and blockage of the rainwater pipe network is collected and measured, and reflected in the urban stormwater model; the corresponding measured rainfall process and surface inundation conditions are collected and observed, and the parameters of the urban stormwater model are calibrated according to the measured data, so that the urban stormwater model can restore the urban surface waterlogging conditions during heavy rain with high precision.

[0019] As a preferred technical solution of the present invention: in step S3, the pipe network and stormwater inlet blockage conditions are set, the flow section blockage parameter values ​​are set to four conditions of 0%, 30%, 50%, 70%, and 90%, and the probability of occurrence of each parameter value is set to 60%, 20%, 10%, 5%, and 5%, respectively; at the same time, a 2-hour rainfall process of six Chicago rain types of P=1a, 3a, 5a, 10a, 20a, and 50a is generated; and a large number of parameter sample data sets are randomly generated according to the above data distribution law using the Monte Carlo method.

[0020] As a preferred technical solution of the present invention: in step S4, according to the different blockage locations and rainfall data sets generated in step S3, different blockage conditions of each pipe network and storm water outlet of the urban storm water model are set, and simulation calculations are performed to generate surface water accumulation data corresponding to the pipe network blockage. Multi-core parallel technology is used in the calculation, and the generated surface water accumulation data is merged with the pipe network and storm water outlet blockage data of the previous step to form a combined data set of blockage conditions and corresponding water accumulation conditions.

[0021] As a preferred technical solution of the present invention: in step S5, the combined data set generated in step S4 is used to train n independent BP neural networks, where n corresponds to the total number of pipelines that are the focus of attention in the urban flood model; the input data of the neural network are: the water depth and rainfall intensity of the surrounding m waterlogging points; the output data of the neural network is: the blockage coefficient of the pipeline.

[0022] As a preferred technical solution of the present invention: in step S6, the input data of the CNN neural network are: a single-frame image of the surveillance video when water accumulates on the street during a rainstorm, the image is denoised, enhanced, and normalized before being input into the model to improve the quality; reference points of known heights in the manually marked image; the output data of the CNN neural network are: water depth information of the single-frame image of the surveillance video.

[0023] As a preferred technical solution of the present invention: in step S7, street monitoring information of key focus areas is collected and analyzed in real time, and public monitoring equipment from municipal administration, transportation, and public security and civil monitoring equipment from merchants and residents are monitored; in order to improve the working efficiency of the model and reduce system resource consumption, key frames of the collected monitoring video are captured and analyzed, specifically one frame is captured every 1 to 10 minutes, and the corresponding time interval should be associated with rainfall information, and the greater the rainfall, the shorter the interval.

[0024] As a preferred technical solution of the present invention: in step S8, the water depth data output by the CNN model is analyzed, the water depth data is input into the BP neural network of the corresponding pipe or stormwater inlet, and the blockage situation of the corresponding node is calculated.

[0025] The present invention provides a real-time analysis method for urban drainage pipe network blockage based on street monitoring, aiming to solve the problems that urban rainwater pipe network blockage is difficult to be discovered in time, the location and degree of blockage are difficult to determine, and professional pipe network monitoring equipment is expensive and difficult to analyze and diagnose in real time. The present invention combines existing urban street monitoring and convolutional neural network (CNN) to calculate flooding in real time, and uses urban rain flood model, Monte Carlo method, back propagation neural network (BP) coupling model to perform real-time judgment and analysis of pipe network blockage situation and location. The method proposed by the present invention does not require professional equipment or additional hardware equipment, and can achieve real-time calculation of urban rainwater pipe network blockage location and degree prediction.

[0026] The basic principle of the present invention is that the blockage of rainwater pipes will increase the scope and depth of surface flooding. Generally, as the degree of pipe blockage increases, the degree of surface water accumulation in the vicinity of the corresponding blockage location increases accordingly. That is, under different rainfall scenarios, there is a nonlinear functional relationship between surface water accumulation and the degree of pipe network blockage at the corresponding location. The present invention constructs a large sample parameter data set of different pipe network blockage conditions through the urban rain and flood model and the Monte Carlo method, and uses the BP neural network to learn and extract the correlation between the degree of pipe network blockage and surface water accumulation in the data set, and uses the CNN neural network to extract the flood depth information in street monitoring during heavy rain. When the system is running, the extracted surface flooding information is input into the trained BP neural network to obtain the blockage situation of the pipe section at the corresponding location.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] (1) Dynamic and real-time analysis of the location and degree of blockage in the rainwater pipe network.

[0029] Traditional rainwater pipe network blockage detection relies on regular inspections, which requires professional personnel and equipment, is expensive and time-consuming. During the heavy rainfall in the flood season, pipe networks are prone to blockage. Blockages are sudden and hidden, but have a huge impact on the drainage capacity of the pipe network, causing or exacerbating urban waterlogging during heavy rainfall. Traditional methods make it difficult to perceive and analyze the location and situation of blockages in real time.

[0030] The present invention can analyze the location and degree of pipe network blockage points in real time by using neural networks and urban stormwater models based on information such as the degree of inundation at waterlogging points and rainfall intensity, and is highly innovative.

[0031] (2) No need for new hardware equipment, low cost.

[0032] Traditional methods rely on robots equipped with image sensors or sonar and other professional equipment to detect the rainwater pipe network, or deploy water level and flow sensors to monitor the drainage pipe network. The above solutions all require the use of professional equipment and have a certain risk of damage, and the cost is relatively high. The present invention uses existing urban street monitoring, does not require the purchase of additional hardware equipment, and is low-cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a flowchart of the real-time analysis method for urban drainage network blockage based on street monitoring provided by the present invention.

[0034] Figure 2 A diagram showing the steps of the real-time analysis method for urban drainage network blockage based on street monitoring provided by the present invention. DETAILED DESCRIPTION

[0035] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0036] The present invention provides a real-time analysis method for urban drainage network blockage based on street monitoring, comprising the following steps:

[0037] S1: Organize basic data and establish a “pipeline network-surface” two-layer urban stormwater model.

[0038] The basic data should include rainwater pipe network data, surface elevation data, land use data, remote sensing image data and design rainfall data. The established two-layer urban stormwater model includes necessary modules such as underground rainwater pipe network, surface streets and flood-prone points, rainwater pumping stations, and rainwater storage tanks.

[0039] S2: Use the measured rainfall to complete the model calibration.

[0040] Collect and measure the current data of siltation and blockage of rainwater pipe network, and reflect it in the urban stormwater model. Collect and observe the corresponding measured rainfall process and surface inundation situation, and calibrate the parameters of the urban stormwater model according to the measured data, so that the urban stormwater model can restore the urban surface water accumulation situation during heavy rain with high accuracy.

[0041] S3: Set the pipe network blockage and rainfall conditions parameters, and use the Monte Carlo method to generate different blockage locations and rainfall data sets for the urban stormwater model.

[0042] Set the pipe network and stormwater inlet clogging conditions, such as setting the flow section blockage parameter value to 0%, 30%, 50%, 70%, 90% four conditions, set the probability of each parameter value to 60%, 20%, 10%, 5%, 5%. At the same time, generate six Chicago rain types of P = 1a, 3a, 5a, 10a, 20a, 50a 2-hour rainfall process. Use the Monte Carlo method to randomly generate a large number of parameter sample data sets according to the above data distribution law, such as a data set of 300,000 parameters.

[0043] S4: Input the data set into the urban stormwater model to simulate the surface water accumulation when the pipe network and stormwater inlets at different locations are blocked.

[0044] According to the data set generated in the previous step, different siltation conditions of each pipe network and rainwater outlet of the urban stormwater model are set, and simulation calculations are performed to generate 300,000 pieces of surface water accumulation data corresponding to pipe network siltation. Multi-core parallel technology is used in the calculation. The generated surface water accumulation data is merged with the pipe network and rainwater outlet siltation data in the previous step to form a combined data set of 300,000 siltation conditions and corresponding water accumulation conditions.

[0045] S5: Use the pipeline blockage and surface water data sets to train the BP neural network.

[0046] Using the combined data set generated in the previous step, n independent BP neural networks are trained, where n corresponds to the total number of pipelines that are the focus of attention in the urban flood model.

[0047] The input data of the neural network are: the water depth and rainfall intensity of the surrounding m water accumulation points; the output data of the neural network is: the blockage coefficient of the pipeline.

[0048] S6: Using surveillance videos of street waterlogging during heavy rains and the elevation and width of benchmark points, the CNN neural network is trained so that the CNN neural network can output the street flooding depth with high accuracy.

[0049] The input data of the CNN neural network are: 1. Single-frame images of surveillance video when water accumulates on streets during heavy rain. Before entering the model, the images are denoised, enhanced, and normalized to improve quality; 2. Reference points of known heights in manually annotated images (such as curbstones, road signs, vehicle tires, etc.).

[0050] The output data of the CNN neural network is: water depth information of a single frame image of the monitoring video.

[0051] S7: During heavy rain, the system starts running analysis, inputting street surveillance videos into the CNN neural network to obtain the flooding conditions at different surface locations.

[0052] When the system is running, it will collect and analyze street monitoring information of key areas in real time. The monitoring can come from public monitoring equipment such as municipal administration, transportation, and public security, as well as civilian monitoring equipment such as merchants and residents. In order to improve the efficiency of the model and reduce system resource consumption, key frames of the collected monitoring videos are captured and analyzed. Specifically, one frame is captured every 1 to 10 minutes. The corresponding time interval should be associated with rainfall information. The greater the rainfall, the shorter the interval.

[0053] After intercepting the interval frames of the surveillance video, the image is pre-processed, and the processed image is input into the CNN neural network to output the flooding depth.

[0054] S8: During heavy rain, the system starts to run the analysis, inputs the surface flooding situation into the BP neural network, and obtains the siltation situation of different nodes.

[0055] Analyze the water depth data output by the CNN model, input the water depth data into the BP neural network of the corresponding pipe or stormwater inlet, and calculate the blockage situation of the corresponding node.

[0056] The above-mentioned specific implementation methods are used to explain the present invention and are only preferred embodiments of the present invention, rather than limiting the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A real-time analysis method for urban drainage network blockage based on street monitoring, characterized in that: The following steps are involved: S1. Organize basic data and establish a pipe network-surface double-layer urban stormwater model; S2. Use the measured rainfall accumulation to complete the calibration of the pipe network-surface double-layer urban rainwater model; S3, setting the pipe network blockage and rainfall condition parameters, and using the Monte Carlo method to generate different blockage locations and rainfall data sets for the pipe network-surface double-layer urban stormwater model; S4. Input the data set into the pipe network-surface double-layer urban stormwater model to simulate the surface water accumulation when the pipe network and stormwater inlets at different locations are blocked; S5. Using the pipe network blockage and surface water accumulation data set, train the BP neural network; S6. Use surveillance videos of street waterlogging during heavy rains and the elevation and width of benchmark points to train a CNN neural network. S7, input the street surveillance video into the CNN neural network to obtain the flooding conditions at different surface locations; S8. Input the surface flooding situation into the BP neural network to obtain the siltation situation of different nodes.

2. The method according to claim 1, characterized in that In step S1, the basic data includes rainwater pipe network data, surface elevation data, land use data, remote sensing image data and design rainfall data.

3. The method according to claim 1, characterized in that In step S1, the pipe network-surface double-layer urban stormwater model includes an underground rainwater pipe network module, a surface street and flood-prone point module, a rainwater pumping station module, and a rainwater storage tank module.

4. The method according to claim 1, characterized in that: In step S2, the current data on siltation and blockage of the rainwater pipe network are collected and measured, and reflected in the urban stormwater model; the corresponding measured rainfall process and surface inundation conditions are collected and observed, and the parameters of the urban stormwater model are calibrated according to the measured data, so that the urban stormwater model can restore the urban surface waterlogging conditions during heavy rain with high precision.

5. The method according to claim 1, characterized in that: In step S3, the pipe network and stormwater inlet blockage conditions are set, the flow section blockage parameter values ​​are set to four conditions of 0%, 30%, 50%, 70%, and 90%, and the probability of occurrence of each parameter value is set to 60%, 20%, 10%, 5%, and 5%, respectively; at the same time, a 2-hour rainfall process of six Chicago rain types of P=1a, 3a, 5a, 10a, 20a, and 50a is generated; and a large number of parameter sample data sets are randomly generated according to the above data distribution law using the Monte Carlo method.

6. The method according to claim 1, characterized in that In step S4, according to the different blockage locations and rainfall data sets generated in step S3, different blockage conditions of each pipe network and storm water outlet of the urban storm water model are set, and simulation calculations are performed to generate surface water accumulation data corresponding to the pipe network blockage. Multi-core parallel technology is used in the calculation, and the generated surface water accumulation data is merged with the pipe network and storm water outlet blockage data of the previous step to form a combined data set of blockage conditions and corresponding water accumulation conditions.

7. The method according to claim 1, characterized in that In step S5, the combined data set generated in step S4 is used to train n independent BP neural networks, where n corresponds to the total number of pipelines that are the focus of attention in the urban flood model; the input data of the neural network are: the water depth and rainfall intensity of the surrounding m waterlogging points; the output data of the neural network is: the blockage coefficient of the pipeline.

8. The method according to claim 1, characterized in that In step S6, the input data of the CNN neural network are: a single frame image of the surveillance video when water accumulates on the street during a rainstorm, the image is denoised, enhanced, and normalized before being input into the model to improve the quality; reference points of known heights in the manually annotated image; the output data of the CNN neural network are: water depth information of a single frame image of the surveillance video.

9. The method according to claim 1, characterized in that: In step S7, street monitoring information of key focus areas is collected and analyzed in real time, and the monitoring comes from public monitoring equipment of municipal administration, transportation, and public security, and civil monitoring equipment of merchants and residents. In order to improve the working efficiency of the model and reduce system resource consumption, key frames of the collected monitoring video are captured and analyzed. Specifically, one frame is captured every 1 to 10 minutes. The corresponding time interval should be associated with the rainfall information. The greater the rainfall, the shorter the interval.

10. The method according to claim 1, characterized in that In step S8, the water depth data output by the CNN model is analyzed, and the water depth data is input into the BP neural network of the corresponding pipe or stormwater inlet to calculate the blockage status of the corresponding node.

Images