Maintenance decision-making method for unmanned boats, drainage pipe siltation disease diagnosis and waterlogging warning

Abstract

The present invention relates to the field of drainage pipe network siltation diagnosis, and specifically to an unmanned boat, a drainage pipe siltation disease diagnosis method, and a maintenance decision-making method for waterlogging warning. The unmanned boat provided by the present invention is composed of an in-pipe integrated hull and a manhole box. The in-pipe integrated hull includes a Doppler ultrasound module, a three-axis gyroscope, a high-definition infrared camera, a main controller, a deep learning development board, a bulkhead, a hatch, a cabin, a wing, a towing ring, and an aviation plug interface. The Doppler ultrasound module, three-axis gyroscope, high-definition infrared camera, aviation plug interface, and deep learning development board are respectively connected to the main controller, while the wing, towing ring, hatch, and bulkhead are connected to the cabin. The main controller is mounted on the cabin, and the deep learning development board is loaded with a drainage pipe siltation disease diagnosis model, a waterlogging warning model, and a pipeline maintenance optimization decision-making model based on multi-objective optimization and waterlogging damage assessment methods. The present invention is suitable for drainage pipe network siltation disease diagnosis, waterlogging warning, and maintenance decision-making.

Description

Technical Field

[0001] The present invention relates to the field of drainage network siltation disease diagnosis and maintenance decision-making, and specifically to an unmanned boat, a drainage pipeline siltation disease diagnosis and waterlogging early warning maintenance decision-making method. Background Art

[0002] In recent years, urban flooding has become a frequent occurrence in Chinese cities, posing a serious threat to residents' daily lives and the safety of their lives and property. Flood risk prevention and control presents significant challenges. Existing research, in developing urban flood warning and disaster prevention decision-making systems, has comprehensively considered surface runoff simulations and pipeline network hydrodynamic transmission processes under heavy rainfall with varying return periods. However, this approach assumes that drainage networks are unobstructed, ignoring the impact of pipeline defects on flow capacity and velocity distribution. In reality, China's urban development has long emphasized aboveground development over underground development, resulting in a serious lack of maintenance and management of underground drainage networks.

[0003] According to a survey conducted by the Chinese Academy of Engineering's key consulting project, "Research on Urban Underground Facility Safety Management and Control Strategies," major cities across China experience an average of over 12 functional defects per kilometer of drainage pipes, with siltation and scaling accounting for over 80%. Poor drainage caused by functional defects like pipe siltation is a key factor in urban waterlogging. Furthermore, urban drainage network maintenance has long been a patchwork of addressing the root causes, with a lack of a multi-objective intelligent decision-making platform for network maintenance that considers factors such as pipe siltation diagnosis and waterlogging early warning.

[0004] The current pipeline siltation disease diagnosis and maintenance industry faces the following pain points:

[0005] First, inaccurate identification. Actual urban drainage networks often face harsh environments characterized by high corrosion and low visibility, making traditional pipeline defect detection methods unreliable. Current flow measurement equipment has an error rate ranging from 8% to 11%, resulting in poor measurement accuracy. Research on neural networks specifically tailored to extract complex pipeline defect characteristics is lacking.

[0006] Second, measurement is difficult. When pipelines are full and clogged, traditional pipeline robots are unable to navigate through silt and scale deposits, making them incapable of conducting full-segment diagnostic surveys. Research on algorithms for accurately detecting disease boundaries is insufficient, and methods for verifying the performance of deep learning models need to be improved.

[0007] Third, inefficiency. Traditional pipeline robots require segment-by-segment diagnostics, resulting in delayed and inefficient diagnosis. They also fail to predict blockages and provide early warning. There is an urgent need for a new AI-powered pipeline disease detection technology. The loss function optimization process in existing intelligent pipeline siltation diagnosis models lacks theoretically informed control equations and boundary constraints, leading to distorted output results. Summary of the Invention

[0008] Based on the needs of natural science research, experimental development and standardization services, the purpose of the present invention is to overcome the shortcomings of the existing technology and provide an unmanned boat, a maintenance decision-making method for drainage pipe siltation disease diagnosis and waterlogging warning, so as to realize the automation and intelligence of drainage pipe flow real-time monitoring, siltation disease diagnosis, waterlogging warning and pipeline maintenance decision-making functions, thereby improving the efficiency and accuracy of measurement.

[0009] The present invention adopts the following technical solutions to achieve the above-mentioned objectives. In a first aspect, the present invention provides an unmanned boat, which is composed of an in-pipe integrated hull and a manhole box. The in-pipe integrated hull includes a Doppler ultrasonic module, a three-axis gyroscope, a high-definition infrared camera, a main controller, a deep learning development board, a bulkhead, a hatch, a cabin, an auxiliary wing, a towing ring, and an aviation plug interface. The Doppler ultrasonic module, three-axis gyroscope, high-definition infrared camera, aviation plug interface, and deep learning development board are respectively connected to the main controller, the auxiliary wing, towing ring, hatch, and bulkhead are connected to the cabin, the main controller is arranged on the cabin, and the deep learning development board is loaded with a drainage pipe siltation disease diagnosis model, a waterlogging early warning model, and a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods;

[0010] The manhole box includes a floating towing cable, a lithium battery, a pipeline contactless liquid level gauge, a wireless communication module, a whip transmitting antenna, a box body, a box cover and a fixed claw. The floating towing cable, lithium battery, wireless communication module, pipeline contactless liquid level gauge and whip transmitting antenna are all installed and fixed in the box body. The whip transmitting antenna is connected to the wireless communication module, the fixed claw and the box cover are connected to the box body, the pipeline contactless liquid level gauge, lithium battery and wireless communication module are connected to the floating towing cable, the floating towing cable is connected to the aerial plug interface, and the wireless communication module is connected to an external server.

[0011] In a second aspect, the present invention provides a method for diagnosing siltation in drainage pipes, which is applied to the unmanned boat described above, comprising:

[0012] S1. Screening infrared images of drainage pipe inlets and outlets captured by unmanned vessels. Image screening criteria include that the pipe walls on both sides of the horizontal surface of the drainage pipe are visible within the image frame, the air-water interface is clear, and the images include factors such as light intensity, noise intensity, and missing points, and that the images are shot from multiple angles.

[0013] S2. Annotate the selected infrared image using an image annotation tool to obtain an infrared image of the drainage pipe inlet and outlet with the air and water surface interfaces annotated;

[0014] S3. Based on infrared images of drainage pipe inlets and outlets with air and water surface interfaces annotated, combined with pre-measured Doppler ultrasound data, drainage pipe inlet and outlet flow data, and drainage pipe sedimentation data, construct training sets, validation sets, and test sets for the drainage pipe sedimentation disease diagnosis model;

[0015] S4. Train the drainage pipe siltation disease diagnosis model using the training set, validation set, and test set of the model. The hyperparameters optimized for training include the decoupling weight decay size, batch size, maximum number of iterations, learning rate, and momentum coefficient. After training, the optimal drainage pipe siltation disease diagnosis model is obtained, and the diagnosis results of drainage pipe siltation diseases in the study area are output.

[0016] The deep learning algorithm in the drainage pipe siltation disease diagnosis model consists of Gram's angle sum field, convolutional neural network, bidirectional long short-term memory neural network and attention mechanism. The pipeline siltation disease diagnosis model embeds a deep learning algorithm guided by physical knowledge. The physical knowledge is extracted from the hydraulic model of the flow changes at the inlet and outlet of the drainage pipe under different siltation degrees.

[0017] Furthermore, the drainage pipe siltation disease diagnosis model is composed of a pipe flow intelligent measurement module and a pipe siltation intelligent diagnosis module;

[0018] The measurement process of the pipeline flow intelligent measurement module includes:

[0019] S21. Based on the radius of the drainage pipe, use Doppler ultrasonic radar to collect the average flow velocity of the water flow in the pipe and the water level above the interface between the solid and liquid;

[0020] S22, using a high-definition infrared camera to capture image information of the interface between the air and the water surface in the pipeline;

[0021] S23. Improve the depth curve estimation algorithm using a multi-scale depth separable convolution that can automatically map low-light images to their normal-light counterparts, thereby obtaining an image low-light enhancement algorithm based on a multi-scale depth convolution curve. The image low-light enhancement algorithm based on the multi-scale depth convolution curve is then used to perform feature enhancement processing on the image information of the interface between the air and water surface of the pipeline.

[0022] S24. Add a multi-scale convolutional neural network to the backbone network of YOLO v8 and add an adaptive genetic algorithm to the front end of the network detection head of YOLO v8 to obtain an improved target detection algorithm based on the adaptive genetic algorithm and the multi-scale convolutional neural network. Use the improved target detection algorithm to perform target recognition on the pipeline air-water interface image and identify the target area of the pipeline air-water interface in the image;

[0023] S25. Replacing the input layer and output layer of the U-shaped convolutional neural network algorithm structure with a stacked autoencoder structure to obtain a U-shaped convolutional neural network algorithm guided by the stacked autoencoder. When the improved target detection algorithm is used to identify the image of the pipeline air and water surface interface, the U-shaped convolutional neural network algorithm guided by the stacked autoencoder performs image segmentation on the target area of the pipeline air and water surface interface to obtain a segmentation line of the pipeline air and water surface interface, and together with the pipe wall and the measured water level, constitutes a pipeline water flow cross section;

[0024] S26. Input the pipeline water flow cross-sectional data into the multidimensional classification error adaptive boosting regression algorithm to obtain the cross-sectional area of the pipeline water flow, multiply the cross-sectional area of the pipeline water flow by the measured average flow velocity of the water flow cross-sectional area to obtain the measurement result of the output pipeline flow.

[0025] Furthermore, the diagnostic process of the pipeline siltation intelligent diagnostic module includes:

[0026] S31. Using a combination of manual inspection of drainage pipe diseases and unmanned boats, collect a training sample set of pipe inlet and outlet flow rates with pipe siltation disease labels;

[0027] S32. Use generative adversarial networks to expand the training sample set with pipeline siltation disease labels;

[0028] S33. The training sample set with pipeline siltation disease labels is divided into a training set and a validation set, and an intelligent diagnosis algorithm for pipeline siltation is constructed by integrating Gram's angle and field, convolutional neural network, bidirectional long short-term memory neural network and attention mechanism. Physical knowledge constraints are introduced into the loss function of the intelligent diagnosis algorithm for pipeline siltation. The physical knowledge is extracted from the hydraulic model of the flow changes at the inlet and outlet of the drainage pipe under different siltation degrees. The Gram's angle and field and convolutional neural network are responsible for data preprocessing of the input sample set, the bidirectional long short-term memory neural network is responsible for siltation feature extraction and dimensionality reduction, and the attention mechanism is responsible for the classification of pipeline siltation results.

[0029] In a third aspect, the present invention provides a maintenance decision-making method for drainage pipe waterlogging warning, which is applied to the unmanned boat described above, comprising:

[0030] S1. Based on the diagnosis results of drainage pipe siltation in the study area output by the drainage pipe siltation disease diagnosis model, a numerical model of drainage pipe siltation is constructed to obtain waterlogging simulation data;

[0031] S2. Using the diagnostic results of drainage pipe siltation in the study area output by the pipe siltation disease diagnosis model and the simulated waterlogging data, construct the training set, validation set, and test set of the waterlogging early warning model;

[0032] S3. Train the waterlogging warning model using the training set, validation set, and test set of the waterlogging warning model to obtain the optimal waterlogging warning model;

[0033] The deep learning algorithm in the waterlogging early warning model is composed of one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The model is embedded with a deep learning algorithm guided by physics knowledge extracted from a hydraulic model of flow changes in drainage pipe inlets and outlets under different levels of siltation and an improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground.

[0034] S4. Utilizing the outputs of the optimal drainage pipe siltation disease diagnosis model and the optimal waterlogging early warning model, construct a pipeline maintenance optimization decision model based on multi-objective optimization and a waterlogging damage assessment method, wherein the waterlogging damage assessment method comprises triangular type II fuzzy sets and a dynamic proportional substitution method;

[0035] S5. Deploy unmanned boats at the entrances and exits of each drainage pipeline in the study area. The unmanned boats deployed at the entrances and exits of each drainage pipeline upload the collected data and the output results of the drainage pipeline siltation disease diagnosis model to the server. The server generates the output results of the drainage pipeline siltation disease diagnosis model of the unmanned boats deployed at the entrances and exits of each drainage pipeline in the study area into a pipeline siltation database.

[0036] S6. Select an unmanned boat deployed at the entrance and exit of the drainage pipe closest to the server, use the generated pipeline sedimentation database as input, use the optimal waterlogging warning model and the pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods, and output the waterlogging warning results and optimization decision results respectively.

[0037] Furthermore, the construction process of the waterlogging early warning model includes:

[0038] S41. Construct hydraulic models of drainage pipes with different siltation levels;

[0039] S42. Numerical simulation of the hydraulic model of the drainage pipe under the silted state is performed to calibrate the hydraulic parameters related to the flow velocity and flow rate indicators;

[0040] S43. Based on the construction of hydraulic models of drainage pipes with different siltation levels and the calibration of hydraulic parameters, the urban surface-underground two-dimensional coupled connection model is improved from the perspectives of surface horizontal connection and surface-underground vertical connection;

[0041] S44. Using the improved one-dimensional and two-dimensional coupled connection model of urban surface and underground, generate sample sets of different pipeline sedimentation conditions with waterlogging result labels;

[0042] S45. Construct a deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The one-dimensional and two-dimensional convolutional neural networks are responsible for extracting multi-dimensional and multi-scale features of pipeline inlet and outlet flow data. The long-short-term memory network performs time series feature fusion and dimensionality reduction on the features extracted by the one-dimensional and two-dimensional convolutional neural networks. The attention mechanism then classifies and outputs the final waterlogging warning results.

[0043] S46. The hydraulic models of drainage pipes with different degrees of siltation are introduced into the loss function of the deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long short-term memory networks and attention mechanisms, as well as the physical knowledge constraints of the improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. The parameters in the model are trained and calibrated using sample sets of different pipe siltation conditions with waterlogging result labels to obtain a waterlogging early warning model.

[0044] Furthermore, the output of the optimization decision results specifically includes:

[0045] S51. By introducing triangular type II fuzzy sets and the dynamic proportional substitution method, we migrated the sample data of multiple cities to the loss rate sample matrix of the study area. Combined with the output of the waterlogging early warning model, we obtained a waterlogging loss assessment method based on triangular type II fuzzy sets and the dynamic proportional substitution method. The waterlogging loss rate in the waterlogging loss assessment method is formulated as follows:

[0046] Among them, f(s) is the flood loss rate, s is the corresponding flood depth, k and b are dynamic proportional substitution parameters, χ f The main membership function of the triangular type II fuzzy set of the property density of the study area is plotted based on the historical property density survey data of the study area. The three vertex coordinates (γ1,m * ), (γ2,1), (γ3,n * ), α is the cutoff value, The triangular type II fuzzy set representing the property density of the study area is: Necessity measure and the principal membership function χ f It can be expressed as:

[0047]

[0048] Among them, sup represents the supremum in mathematical operations;

[0049] S52. With the constraints of minimizing waterlogging damage caused by poor pipeline drainage, minimizing drainage pipeline maintenance costs, and maximizing economic benefits, a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods was constructed. The calculation method for waterlogging damage is as follows:

[0050] Damage from flooding x i = Property density (yuan / m 2 )×flooded area (m 2 )×waterlogging damage rate (%);

[0051] Drainage pipe maintenance cost i The calculation method is as follows:

[0052] y i = siltation length * siltation degree * maintenance cost base * maintenance cost fuzzy coefficient;

[0053] Maintenance cost fuzzy coefficient M cb The calculation of historical maintenance experience data is used to construct a triangular type II fuzzy set Combined with the three vertex coordinates of the triangular type-2 fuzzy set (γ1,m * ), (γ2,1), (γ3,n * ) and the cutoff value α, then Credibility measure and maintenance cost fuzzy coefficient M cb Expressed as:

[0054]

[0055] S53. Calculate whether maintenance measures are needed for the i-th section of pipeline while ensuring no risk of casualties, and calculate the target benefit function. M represents the total number of drainage pipes, a i Indicates whether the i-th section of the drainage pipe is silted up, b i Indicates whether maintenance measures are taken for the i-th section of drainage pipe. If so, If not taken

[0056] S54, the constraint condition is to maximize the target profit function That is, max(Tar), which outputs the optimal decision results for each pipe section in the area.

[0057] Furthermore, the maintenance decision-making method for drainage pipe waterlogging warning also includes:

[0058] S7. The output of the waterlogging warning and optimization decision results from the unmanned boat deployed at the drainage pipe entrance and exit closest to the server, as well as the pipeline sedimentation database, are visualized and displayed on the drainage pipe full-process decision-making platform deployed on the server, thereby realizing an integrated drainage pipe decision-making system, specifically including:

[0059] Multi-sensor data is obtained through the unmanned boat's Doppler ultrasonic module, three-axis gyroscope, high-definition infrared camera, and contactless liquid level meter. The multi-sensor data is uploaded and stored to the drainage pipeline full-process decision-making platform through the wireless communication module. The drainage pipeline full-process decision-making platform obtains a visual display of the output results of the unmanned boat's drainage pipeline sedimentation disease diagnosis model, waterlogging warning model, and pipeline maintenance optimization decision-making model. Based on the browser / server access control architecture, an integrated drainage pipeline decision-making system consisting of an unmanned boat, multi-sensor data, and a drainage pipeline full-process decision-making platform is developed and built to realize real-time monitoring of drainage pipelines, pipeline sedimentation disease diagnosis, waterlogging warning, and pipeline maintenance optimization decision-making functions.

[0060] The beneficial effects of the present invention are:

[0061] This invention uses a drainage pipe siltation disease diagnosis model based on a deep learning network, a waterlogging early warning model, and a pipeline maintenance optimization decision-making model based on multi-objective optimization and waterlogging damage assessment methods. It only requires demarcating the study area and deploying unmanned boat equipment, without the need for subsequent human intervention, to complete the entire process of drainage pipe siltation disease diagnosis, waterlogging early warning, and maintenance decision-making in the study area, which is convenient and fast.

[0062] The present invention develops and builds an integrated drainage pipeline decision-making system consisting of the unmanned boat, multi-sensor data, and a drainage pipeline full-process decision-making platform, which realizes the functions of real-time monitoring of drainage pipelines, diagnosis of pipeline siltation diseases, waterlogging warning, and optimization decision-making for pipeline maintenance;

[0063] The present invention integrates the unmanned boat hardware structure, visualization result display module, sensor equipment data receiving module, and intelligent data processing algorithm into the same architecture, and develops a full-process decision-making platform for urban drainage pipelines, realizing the full process, intelligentization, and precision of real-time monitoring of drainage pipelines, diagnosis of pipeline siltation diseases, early warning of urban waterlogging, and optimization of pipeline maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a structural diagram of the drainage pipe unmanned boat provided by the present invention;

[0065] Figure 2 This is a hardware structure diagram of the integrated hull in a tube provided by the present invention;

[0066] Figure 3 This is a diagram showing the connection relationship between modules in the in-tube integrated hull hardware structure provided by the present invention;

[0067] Figure 4 This is a hardware structure diagram of the manhole box provided by the present invention;

[0068] Figure 5 This is a diagram showing the connection relationship between modules in the hardware structure of the manhole box provided by the present invention;

[0069] Figure 6 This is a diagram showing the connection between the in-pipe integrated hull and the manhole box provided by the present invention;

[0070] Figure 7 It is a server structure diagram provided by the present invention;

[0071] In the accompanying drawings, 100 is a Doppler ultrasonic module, 101 is a three-axis gyroscope, 102 is a high-definition infrared camera, 103 is an aviation plug-in interface, 104 is a deep learning development board, 105 is a main controller, 106 is a hatch, 107 is a partition, 108 is an auxiliary wing, 109 is a towing ring, 110 is a cabin, 200 is a floating towing cable, 201 is a whip transmitting antenna, 202 is a pipeline contactless liquid level meter, 203 is a box body, 204 is a fixed claw, 205 is a wireless communication module, 206 is a box cover, 207 is a lithium battery, 300 is a manhole box, and 301 is an integrated hull in the pipe. DETAILED DESCRIPTION

[0072] To make the objectives, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0073] The present invention provides an unmanned ship, such as Figure 1 As shown, the unmanned boat consists of an in-pipe integrated hull and a manhole box. The in-pipe integrated hull has a built-in drainage pipe siltation disease diagnosis model based on a deep learning network, a waterlogging early warning model, and a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods. The integrated hull and the manhole box are interactively connected by a floating towing cable.

[0074] In this embodiment, if Figure 2 As shown in the figure, the hardware structure of the integrated hull in the tube includes a Doppler ultrasonic module 100, a three-axis gyroscope 101, a high-definition infrared camera 102, a main controller 105, a deep learning development board 104, a bulkhead 107, a hatch 106, a cabin 110, an auxiliary wing 108, a towing ring 109, and an aviation plug interface 103. The connection relationship is as follows Figure 3 As shown, the Doppler ultrasound module 100, the three-axis gyroscope 101, the high-definition infrared camera 102, the aviation plug-in interface 103, and the deep learning development board 104 are respectively connected to the main controller 105 and integrated into the cabin 110; the auxiliary wing 108, the traction ring 109, the cabin cover 106, and the partition 107 are connected to the cabin 110.

[0075] In this embodiment, if Figure 4As shown, the hardware structure of the manhole box includes a floating towing cable 200, a lithium battery 207, a pipeline contactless liquid level meter 202, a wireless communication module 205, a whip transmitting antenna 201, a box body 203, a box cover 206, and a fixing claw 204; the connection relationship is as follows Figure 5 As shown, the whip transmitting antenna 201 is connected to the wireless communication module 205 to realize the wireless data upload and download function; the fixed claw 204 and the box cover 203 are connected to the box body 203; the floating towing cable 200, the lithium battery 207, the wireless communication module 205, the pipeline contactless liquid level meter 202, and the whip transmitting antenna 201 are all installed and fixed in the box body 203; the pipeline contactless liquid level meter 202, the lithium battery 207, and the wireless communication module 205 are connected to the floating towing cable 200.

[0076] like Figure 6 As shown, the in-tube integrated hull 301 and the manhole box 300 are fixed at the bottom and the middle of the manhole chamber respectively, and the floating towing cable 200 is connected to the navigation plug interface 103 of the in-tube integrated hull to realize data interaction between the manhole box 300 and the in-tube integrated hull 301.

[0077] The method performed by the unmanned boat of the present invention is described in detail below.

[0078] The present invention provides a method for diagnosing siltation in drainage pipes, which is applied to the unmanned boat described above, comprising:

[0079] S1. Select clear infrared images of drainage pipe inlets and outlets taken by unmanned boats. The image selection criteria are that the pipe walls on both sides of the horizontal surface of the drainage pipe are visible within the image frame, the air-water interface is clear, and the images include factors such as light intensity, noise intensity, and missing points in the image, and that the images include multiple shooting angles.

[0080] S2. Annotate the selected infrared image using an image annotation tool to obtain an infrared image of the drainage pipe inlet and outlet with the air and water surface interfaces annotated;

[0081] S3. Based on infrared images of drainage pipe inlets and outlets with air and water surface interfaces marked, combined with pre-measured Doppler ultrasound data, drainage pipe inlet and outlet flow data, and drainage pipe sedimentation data, a training set, validation set, and test set for the drainage pipe sedimentation disease diagnosis model were constructed with a ratio of 4:3:3.

[0082] S4. Train the drainage pipe siltation disease diagnosis model using the training set, validation set, and test set of the model. The hyperparameters optimized for training include the decoupling weight decay size, batch size, maximum number of iterations, learning rate, and momentum coefficient. After training, the optimal drainage pipe siltation disease diagnosis model is obtained, and the diagnosis results of drainage pipe siltation diseases in the study area are output.

[0083] The deep learning algorithm in the drainage pipe siltation disease diagnosis model consists of Gram's angle sum field, convolutional neural network, bidirectional long short-term memory neural network and attention mechanism. The pipeline siltation disease diagnosis model is embedded with a deep learning algorithm guided by physical knowledge. The physical knowledge is extracted from the hydraulic model of the flow changes in the inlet and outlet of the drainage pipe under different siltation degrees.

[0084] In one embodiment of the present invention, the drainage pipe siltation disease diagnosis model is composed of a pipe flow intelligent measurement module and a pipe siltation intelligent diagnosis module;

[0085] The measurement process of the pipeline flow intelligent measurement module includes:

[0086] S21. Based on the radius of the drainage pipe, use Doppler ultrasonic radar to collect the average flow velocity of the water flow in the pipe and the water level above the interface between the solid and liquid;

[0087] S22, using a high-definition infrared camera to capture image information of the interface between the air and the water surface in the pipeline;

[0088] S23. Improve the depth curve estimation algorithm using 1×3, 3×1, 5×1, and 1×5 multi-scale depth-separable convolutions that can automatically map low-light images to their normal-light counterparts, and obtain an image low-light enhancement algorithm based on a multi-scale depth convolution curve. Then, use the image low-light enhancement algorithm based on a multi-scale depth convolution curve to perform feature enhancement processing on the image information of the interface between the air and water surface of the pipeline;

[0089] S24. Add 1×3, 7×1, 5×1, and 1×7 multi-scale convolutional neural networks to the backbone network in YOLO v8 to improve the feature extraction accuracy of the backbone network. Add an adaptive genetic algorithm to the front end of the network detection head in YOLO v8 to obtain an improved target detection algorithm based on the adaptive genetic algorithm and multi-scale convolutional neural network. Use the improved target detection algorithm to perform target recognition on images of the pipeline air-water interface and identify the target area of the pipeline air-water interface in the image.

[0090] S25. Replacing the input layer and output layer of the U-shaped convolutional neural network algorithm structure with a stacked autoencoder structure to obtain a U-shaped convolutional neural network algorithm guided by the stacked autoencoder. When the improved target detection algorithm is used to identify the image of the pipeline air and water surface interface, the U-shaped convolutional neural network algorithm guided by the stacked autoencoder performs image segmentation on the target area of the pipeline air and water surface interface to obtain a segmentation line of the pipeline air and water surface interface, and together with the pipe wall and the measured water level, constitutes a pipeline water flow cross section;

[0091] S26. Input the pipeline water flow cross-sectional data into the multidimensional classification error adaptive boosting regression algorithm to obtain the cross-sectional area of the pipeline water flow, multiply the cross-sectional area of the pipeline water flow by the measured average flow velocity of the water flow cross-sectional area to obtain the measurement result of the output pipeline flow.

[0092] In one embodiment of the present invention, the diagnostic process of the pipeline siltation intelligent diagnostic module includes:

[0093] S31. Using a combination of manual inspection of drainage pipe diseases and unmanned boats, a large number of pipeline inlet and outlet flow training sample sets with pipeline siltation disease labels are collected;

[0094] S32. Construct a hydraulic model of flow changes in drainage pipe inlets and outlets under different siltation levels, and use a generative adversarial network to expand the training sample set with pipe siltation disease labels;

[0095] S33. The training sample set with pipeline siltation disease labels is divided into a training set and a validation set in a ratio of 4:1. An intelligent diagnosis algorithm for pipeline siltation is constructed by integrating Gram's angle and field, convolutional neural network, bidirectional long short-term memory neural network and attention mechanism. Physical knowledge constraints are introduced into the loss function of the intelligent diagnosis algorithm for pipeline siltation. The physical knowledge is extracted from the hydraulic model of the flow change at the inlet and outlet of the drainage pipe under different siltation degrees. The Gram's angle and field and convolutional neural network are responsible for data preprocessing of the input sample set, the bidirectional long short-term memory neural network is responsible for siltation feature extraction and dimensionality reduction, and the attention mechanism is responsible for the classification of pipeline siltation results.

[0096] The present invention provides a maintenance decision-making method for drainage pipe waterlogging warning, which is applied to the unmanned boat described above, comprising:

[0097] S1. Based on the diagnosis results of drainage pipe siltation in the study area output by the drainage pipe siltation disease diagnosis model, a numerical model of drainage pipe siltation is constructed to obtain waterlogging simulation data;

[0098] S2. Using the diagnostic results of drainage pipe siltation in the study area output by the pipe siltation disease diagnosis model and the simulated waterlogging data, the training set, validation set, and test set of the waterlogging early warning model were constructed in a ratio of 4:3:3;

[0099] S3. Train the waterlogging warning model using the training set, validation set, and test set of the waterlogging warning model to obtain the optimal waterlogging warning model;

[0100] The deep learning algorithm in the flood warning model is composed of one- and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The model is embedded with a deep learning algorithm guided by physics knowledge extracted from a hydraulic model of flow changes in drainage pipe inlets and outlets under different levels of siltation and an improved one- and two-dimensional coupled connection model of the urban surface and underground.

[0101] S4. Utilizing the outputs of the optimal drainage pipe siltation disease diagnosis model and the optimal waterlogging early warning model, construct a pipeline maintenance optimization decision model based on multi-objective optimization and a waterlogging damage assessment method, wherein the waterlogging damage assessment method comprises triangular type II fuzzy sets and a dynamic proportional substitution method;

[0102] S5. Deploy unmanned boats at the entrances and exits of each drainage pipeline in the study area. The unmanned boats deployed at the entrances and exits of each drainage pipeline upload the collected data and the output results of the drainage pipeline siltation disease diagnosis model to the server. The server generates the output results of the drainage pipeline siltation disease diagnosis model of the unmanned boats deployed at the entrances and exits of each drainage pipeline in the study area into a pipeline siltation database.

[0103] S6. Select an unmanned boat deployed at the drainage pipe entrance or exit closest to the server, use the generated pipe siltation database as input, utilize the optimal waterlogging warning model and the pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods, and output waterlogging warning results and optimization decision results respectively;

[0104] S7. The output of the waterlogging warning and optimization decision results from the unmanned boat deployed at the drainage pipe entrance and exit closest to the server, as well as the pipeline sedimentation database, are visualized and displayed on the drainage pipe full-process decision-making platform deployed on the server, thereby realizing an integrated drainage pipe decision-making system, specifically including:

[0105] Multi-sensor data is obtained through the unmanned boat's Doppler ultrasonic module, three-axis gyroscope, high-definition infrared camera, and contactless liquid level meter. The multi-sensor data is uploaded and stored to the drainage pipeline full-process decision-making platform through the wireless communication module. The drainage pipeline full-process decision-making platform obtains a visual display of the output results of the unmanned boat's drainage pipeline sedimentation disease diagnosis model, waterlogging warning model, and pipeline maintenance optimization decision-making model. Based on the browser / server access control architecture, an integrated drainage pipeline decision-making system consisting of an unmanned boat, multi-sensor data, and a drainage pipeline full-process decision-making platform is developed and built to realize real-time monitoring of drainage pipelines, pipeline sedimentation disease diagnosis, waterlogging warning, and pipeline maintenance optimization decision-making functions.

[0106] In one embodiment of the present invention, the process of constructing the waterlogging early warning model includes:

[0107] S41. Construct hydraulic models of drainage pipes with different siltation levels based on Infoworks ICM software;

[0108] S42. Use Fluent software to perform numerical simulations on the hydraulic model of the drainage pipe under siltation conditions and calibrate the hydraulic parameters related to flow velocity and flow rate indicators;

[0109] S43. Based on the construction of hydraulic models of drainage pipes with different siltation levels and the calibration of hydraulic parameters, the urban surface-underground two-dimensional coupled connection model is improved from the perspectives of surface horizontal connection and surface-underground vertical connection;

[0110] S44. Using the improved one-dimensional and two-dimensional coupled connection model of urban surface and underground, generate sample sets of different pipeline sedimentation conditions with waterlogging result labels;

[0111] S45. Construct a deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The one-dimensional and two-dimensional convolutional neural networks are responsible for extracting multi-dimensional and multi-scale features of pipeline inlet and outlet flow data. The long-short-term memory network performs time series feature fusion and dimensionality reduction on the features extracted by the one-dimensional and two-dimensional convolutional neural networks. The attention mechanism then classifies and outputs the final waterlogging warning results.

[0112] S46. The hydraulic models of drainage pipes with different degrees of siltation are introduced into the loss function of the deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long short-term memory networks and attention mechanisms, as well as the physical knowledge constraints of the improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. The parameters in the model are trained and calibrated using sample sets of different pipe siltation conditions with waterlogging result labels to obtain a waterlogging early warning model.

[0113] In one embodiment of the present invention, outputting the optimization decision result specifically includes:

[0114] S51. By introducing triangular type II fuzzy sets and the dynamic proportional substitution method, we migrated the sample data of multiple cities to the loss rate sample matrix of the study area. Combined with the output of the waterlogging early warning model, we obtained a waterlogging loss assessment method based on triangular type II fuzzy sets and the dynamic proportional substitution method. The waterlogging loss rate in the waterlogging loss assessment method is formulated as follows:

[0115] Among them, f(s) is the flood loss rate, s is the corresponding flood depth, k and b are dynamic proportional substitution parameters, χ f The main membership function of the triangular type II fuzzy set of the property density of the study area is plotted based on the historical property density survey data of the study area. The three vertex coordinates (γ1,m *), (γ2,1), (γ3,n * ), α is the cutoff value, The triangular type II fuzzy set representing the property density of the study area is: Necessity measure and the principal membership function χ f It can be expressed as:

[0116]

[0117] Among them, sup represents the supremum in mathematical operations;

[0118] S52. With the constraints of minimizing waterlogging damage caused by poor pipeline drainage, minimizing drainage pipeline maintenance costs, and maximizing economic benefits, a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods was constructed. The calculation method for waterlogging damage is as follows:

[0119] Damage from flooding x i = Property density (yuan / m 2 )×flooded area (m 2 )×waterlogging damage rate (%);

[0120] Drainage pipe maintenance cost i The calculation method is as follows:

[0121] y i = siltation length * siltation degree * maintenance cost base * maintenance cost fuzzy coefficient;

[0122] Maintenance cost fuzzy coefficient M cb The calculation of historical maintenance experience data is used to construct a triangular type II fuzzy set Combined with the three vertex coordinates of the triangular type-2 fuzzy set (γ1,m * ), (γ2,1), (γ3,n * ) and the cutoff value α, then Credibility measure and maintenance cost fuzzy coefficient M cb Expressed as:

[0123]

[0124] S53. Calculate whether maintenance measures are needed for the i-th section of pipeline while ensuring no risk of casualties, and calculate the target benefit function. M represents the total number of drainage pipes, a i Indicates whether the i-th section of the drainage pipe is silted up, b i Indicates whether maintenance measures are taken for the i-th section of drainage pipe. If so, If not taken

[0125] S54, the constraint condition is to maximize the target profit function That is, max(Tar), which outputs the optimal decision results for each pipe section in the area.

[0126] The hardware and software structural function parameters of the unmanned boat of the present invention are shown in Table 1.

[0127] Table 1 Software and hardware structure and function parameters of unmanned ship

[0128]

[0129]

[0130] The functional parameters of the drainage pipeline full-process decision-making platform of the present invention are shown in Table 2.

[0131]

[0132] The following is a detailed explanation using the underground drainage network and surface waterlogging warning system at the Huizhou University waterlogging test site as an example.

[0133] The actual values of the inlet and outlet flow rate, pipeline siltation status (whether there is siltation, siltation length, siltation height), and waterlogging inundation status (inundation depth, inundation range) of 200 groups of underground drainage pipelines at Huizhou University were collected and compared with the maintenance decision-making method for unmanned boat, drainage pipeline siltation disease diagnosis, and waterlogging early warning proposed in this invention. The performance indicators are shown in Table 3.

[0134] Table 3 Performance index statistics of the algorithm of the present invention in the embodiment

[0135] index Mean absolute error Root mean square error Pearson correlation coefficient Drainage pipe inlet and outlet flow 0.028 0.231 0.953 Drain pipe siltation diagnosis 0.019 0.152 0.975 Flood warning 0.046 0.301 0.936

[0136] In summary, the present invention achieved mean absolute error, root mean square error, and Pearson correlation coefficient of 0.028, 0.231, and 0.953 for flow measurement, 0.019, 0.152, and 0.975 for sedimentation diagnosis, and 0.046, 0.301, and 0.936 for waterlogging warning. This demonstrates that the deep learning algorithm embedded in the unmanned vessel offers high accuracy and reliability, facilitating the realization of high-precision, intelligent decision-making for drainage network maintenance.

[0137] The foregoing description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be construed as excluding other embodiments. Rather, the present invention can be used in various other combinations, modifications, and environments and can be modified within the scope of the concept described herein through the above teachings or techniques or knowledge in the relevant field. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention are intended to be protected by the appended claims.

Claims

1. Unmanned ship, characterized by: The unmanned boat is composed of an in-tube integrated hull and a manhole box. The in-tube integrated hull includes a Doppler ultrasonic module, a three-axis gyroscope, a high-definition infrared camera, a main controller, a deep learning development board, a bulkhead, a hatch, a cabin, an auxiliary wing, a towing ring, and an aviation plug interface. The Doppler ultrasonic module, three-axis gyroscope, high-definition infrared camera, aviation plug interface, and deep learning development board are respectively connected to the main controller, and the auxiliary wing, towing ring, hatch, and bulkhead are connected to the cabin. The main controller is set on the cabin, and the deep learning development board is loaded with a drainage pipe siltation disease diagnosis model, a waterlogging early warning model, and a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods; The manhole box includes a floating towing cable, a lithium battery, a pipeline contactless liquid level gauge, a wireless communication module, a whip transmitting antenna, a box body, a box cover and a fixing claw. The floating towing cable, lithium battery, wireless communication module, pipeline contactless liquid level gauge and whip transmitting antenna are all installed and fixed in the box body. The whip transmitting antenna is connected to the wireless communication module, the fixing claw and the box cover are connected to the box body, the pipeline contactless liquid level gauge, lithium battery and wireless communication module are connected to the floating towing cable, the floating towing cable is connected to the aerial plug interface, and the wireless communication module is connected to an external server; The pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment method is used for maintenance decision-making of drainage pipe waterlogging warning, specifically including: Based on the diagnosis results of drainage pipe siltation in the study area output by the drainage pipe siltation disease diagnosis model, a numerical model of drainage pipe siltation was constructed to obtain waterlogging simulation data. The training, validation, and test sets of the waterlogging early warning model were constructed using the diagnostic results of waterlogging in the study area, which were output by the pipeline siltation disease diagnosis model, and simulated waterlogging data. The waterlogging warning model is trained through its training set, validation set, and test set to obtain the optimal waterlogging warning model; The deep learning algorithm in the waterlogging early warning model is composed of one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The model is embedded with a deep learning algorithm guided by physics knowledge extracted from a hydraulic model of flow changes in drainage pipe inlets and outlets under different levels of siltation and an improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. Using the outputs of the optimal drainage pipe siltation disease diagnosis model and the optimal waterlogging early warning model, a pipeline maintenance optimization decision model based on multi-objective optimization and a waterlogging damage assessment method was constructed. The method consisted of triangular type II fuzzy sets and a dynamic proportional substitution method. Unmanned boats are deployed at the entrances and exits of each drainage pipeline in the study area. The unmanned boats deployed at the entrances and exits of each drainage pipeline upload the collected data and the output results of the drainage pipeline siltation disease diagnosis model to the server. The server generates the output results of the drainage pipeline siltation disease diagnosis model of the unmanned boats deployed at the entrances and exits of each drainage pipeline in the study area into a pipeline siltation database; An unmanned boat is deployed at the entrance and exit of the drainage pipe closest to the server. The generated pipeline siltation database is used as input. The optimal waterlogging warning model and the pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods are used to output the waterlogging warning results and optimization decision results respectively.

2. A method for diagnosing siltation in drainage pipes, applied to the unmanned boat according to claim 1, characterized in that: include: S1. Screening infrared images of drainage pipe inlets and outlets captured by unmanned vessels. Image screening criteria include that the pipe walls on both sides of the horizontal surface of the drainage pipe are visible within the image frame, the air-water interface is clear, and the images include factors such as light intensity, noise intensity, and missing points, and that the images are shot from multiple angles. S2. Annotate the selected infrared image using an image annotation tool to obtain an infrared image of the drainage pipe inlet and outlet with the air and water surface interfaces annotated; S3. Based on infrared images of drainage pipe inlets and outlets with air and water surface interfaces annotated, combined with pre-measured Doppler ultrasound data, drainage pipe inlet and outlet flow data, and drainage pipe sedimentation data, construct training sets, validation sets, and test sets for the drainage pipe sedimentation disease diagnosis model; S4. Train the drainage pipe siltation disease diagnosis model using the training set, validation set, and test set of the model. The hyperparameters optimized for training include the decoupling weight decay size, batch size, maximum number of iterations, learning rate, and momentum coefficient. After training, the optimal drainage pipe siltation disease diagnosis model is obtained, and the diagnosis results of drainage pipe siltation diseases in the study area are output. The deep learning algorithm in the drainage pipe siltation disease diagnosis model is composed of Gram angle sum fields, convolutional neural networks, bidirectional long short-term memory neural networks, and attention mechanisms. The pipeline siltation disease diagnosis model is embedded with a deep learning algorithm guided by physical knowledge extracted from a hydraulic model of flow changes at the inlet and outlet of drainage pipes under different siltation levels. S5. Based on the diagnosis results of drainage pipe siltation in the study area output by the drainage pipe siltation disease diagnosis model, a numerical model of rainwater flooding caused by drainage pipe siltation is constructed to obtain waterlogging simulation data; S6. Use the diagnostic results of drainage pipe siltation in the study area output by the pipe siltation disease diagnosis model and the simulated waterlogging data to construct the training set, validation set, and test set of the waterlogging early warning model; S7. training the waterlogging warning model using the training set, validation set, and test set of the waterlogging warning model to obtain an optimal waterlogging warning model; The deep learning algorithm in the waterlogging early warning model is composed of one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The model is embedded with a deep learning algorithm guided by physics knowledge extracted from a hydraulic model of flow changes in drainage pipe inlets and outlets under different levels of siltation and an improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. S8. Utilizing the outputs of the optimal drainage pipe siltation disease diagnosis model and the optimal waterlogging early warning model, construct a pipeline maintenance optimization decision model based on multi-objective optimization and a waterlogging damage assessment method, wherein the waterlogging damage assessment method comprises triangular type II fuzzy sets and a dynamic proportional substitution method; S9. Deploy unmanned boats at the entrances and exits of each drainage pipeline in the study area. The unmanned boats deployed at the entrances and exits of each drainage pipeline upload the collected data and the output results of the drainage pipeline siltation disease diagnosis model to the server. The server generates the output results of the drainage pipeline siltation disease diagnosis model of the unmanned boats deployed at the entrances and exits of each drainage pipeline in the study area into a pipeline siltation database. S10. Select an unmanned boat deployed at the entrance and exit of the drainage pipe closest to the server, use the generated pipeline siltation database as input, use the optimal waterlogging warning model and the pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods, and output the waterlogging warning results and optimization decision results respectively.

3. The drainage pipe siltation disease diagnosis method according to claim 2, characterized in that: The drainage pipe siltation disease diagnosis model consists of a pipe flow intelligent measurement module and a pipe siltation intelligent diagnosis module; The measurement process of the pipeline flow intelligent measurement module includes: S21. Based on the radius of the drainage pipe, use Doppler ultrasonic radar to collect the average flow velocity of the water flow in the pipe and the water level height at the interface between the solid and liquid; S22, using a high-definition infrared camera to capture image information of the interface between the air and the water surface in the pipeline; S23. Improve the depth curve estimation algorithm using a multi-scale depth separable convolution that can automatically map low-light images to their normal-light counterparts, thereby obtaining an image low-light enhancement algorithm based on a multi-scale depth convolution curve. The image low-light enhancement algorithm based on the multi-scale depth convolution curve is then used to perform feature enhancement processing on the image information of the interface between the air and water surface of the pipeline. S24. Add a multi-scale convolutional neural network to the backbone network of YOLO v8 and add an adaptive genetic algorithm to the front end of the network detection head of YOLO v8 to obtain an improved target detection algorithm based on the adaptive genetic algorithm and the multi-scale convolutional neural network. Use the improved target detection algorithm to perform target recognition on the pipeline air-water interface image and identify the target area of the pipeline air-water interface in the image; S25. Replacing the input layer and output layer of the U-shaped convolutional neural network algorithm structure with a stacked autoencoder structure to obtain a U-shaped convolutional neural network algorithm guided by the stacked autoencoder. When the improved target detection algorithm is used to identify the image of the pipeline air and water surface interface, the U-shaped convolutional neural network algorithm guided by the stacked autoencoder performs image segmentation on the target area of the pipeline air and water surface interface to obtain a segmentation line of the pipeline air and water surface interface, and together with the pipe wall and the measured water level, constitutes a pipeline water flow cross section; S26. Input the pipeline water flow cross-sectional data into the multidimensional classification error adaptive boosting regression algorithm to obtain the cross-sectional area of the pipeline water flow, multiply the cross-sectional area of the pipeline water flow by the measured average flow velocity of the water flow cross-sectional area to obtain the measurement result of the output pipeline flow.

4. The method for diagnosing drainage pipe siltation disease according to claim 3, characterized in that: The diagnostic process of the pipeline siltation intelligent diagnostic module includes: S31. Using a combination of manual inspection of drainage pipe diseases and unmanned boats, collect a training sample set of pipe inlet and outlet flow rates with pipe siltation disease labels; S32. Use generative adversarial networks to expand the training sample set with pipeline siltation disease labels; S33. The training sample set with pipeline siltation disease labels is divided into a training set and a validation set, and an intelligent diagnosis algorithm for pipeline siltation is constructed by integrating Gram's angle and field, convolutional neural network, bidirectional long short-term memory neural network and attention mechanism. Physical knowledge constraints are introduced into the loss function of the intelligent diagnosis algorithm for pipeline siltation. The physical knowledge is extracted from the hydraulic model of the flow changes at the inlet and outlet of the drainage pipe under different siltation degrees. The Gram's angle and field and convolutional neural network are responsible for data preprocessing of the input sample set, the bidirectional long short-term memory neural network is responsible for siltation feature extraction and dimensionality reduction, and the attention mechanism is responsible for the classification of pipeline siltation results.

5. A maintenance decision-making method for drainage pipe waterlogging warning, applied to the unmanned boat according to claim 1, characterized in that: include: S1. Based on the diagnosis results of drainage pipe siltation in the study area output by the drainage pipe siltation disease diagnosis model, a numerical model of drainage pipe siltation is constructed to obtain waterlogging simulation data; S2. Using the diagnostic results of drainage pipe siltation in the study area output by the pipe siltation disease diagnosis model and the simulated waterlogging data, construct the training set, validation set, and test set of the waterlogging early warning model; S3. Train the waterlogging warning model using the training set, validation set, and test set of the waterlogging warning model to obtain the optimal waterlogging warning model; The deep learning algorithm in the waterlogging early warning model is composed of one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The model is embedded with a deep learning algorithm guided by physics knowledge extracted from a hydraulic model of flow changes in drainage pipe inlets and outlets under different levels of siltation and an improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. S4. Utilizing the outputs of the optimal drainage pipe siltation disease diagnosis model and the optimal waterlogging early warning model, construct a pipeline maintenance optimization decision model based on multi-objective optimization and a waterlogging damage assessment method, wherein the waterlogging damage assessment method comprises triangular type II fuzzy sets and a dynamic proportional substitution method; S5. Deploy unmanned boats at the entrances and exits of each drainage pipeline in the study area. The unmanned boats deployed at the entrances and exits of each drainage pipeline upload the collected data and the output results of the drainage pipeline siltation disease diagnosis model to the server. The server generates the output results of the drainage pipeline siltation disease diagnosis model of the unmanned boats deployed at the entrances and exits of each drainage pipeline in the study area into a pipeline siltation database. S6. Select an unmanned boat deployed at the entrance and exit of the drainage pipe closest to the server, use the generated pipeline sedimentation database as input, use the optimal waterlogging warning model and the pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods, and output the waterlogging warning results and optimization decision results respectively.

6. The maintenance decision-making method for drainage pipe waterlogging warning according to claim 5, characterized in that: The construction process of the waterlogging early warning model includes: S41. Construct hydraulic models of drainage pipes with different siltation levels; S42. Numerical simulation of the hydraulic model of the drainage pipe under the silted state is performed to calibrate the hydraulic parameters related to the flow velocity and flow rate indicators; S43. Based on the construction of hydraulic models of drainage pipes with different siltation levels and the calibration of hydraulic parameters, the urban surface-underground two-dimensional coupled connection model is improved from the perspectives of surface horizontal connection and surface-underground vertical connection; S44. Using the improved one-dimensional and two-dimensional coupled connection model of urban surface and underground, generate sample sets of different pipeline sedimentation conditions with waterlogging result labels; S45. Construct a deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long-short-term memory networks, and an attention mechanism. The one-dimensional and two-dimensional convolutional neural networks are responsible for extracting multi-dimensional and multi-scale features of pipeline inlet and outlet flow data. The long-short-term memory network performs time series feature fusion and dimensionality reduction on the features extracted by the one-dimensional and two-dimensional convolutional neural networks. The attention mechanism then classifies and outputs the final waterlogging warning results. S46. The hydraulic models of drainage pipes with different degrees of siltation are introduced into the loss function of the deep learning algorithm that combines one-dimensional and two-dimensional convolutional neural networks, long short-term memory networks and attention mechanisms, as well as the physical knowledge constraints of the improved one-dimensional and two-dimensional coupled connection model of the urban surface and underground. The parameters in the model are trained and calibrated using sample sets of different pipe siltation conditions with waterlogging result labels to obtain a waterlogging early warning model.

7. The maintenance decision-making method for drainage pipe waterlogging warning according to claim 5, characterized in that: Output optimization decision results include: S51. By introducing triangular type II fuzzy sets and the dynamic proportional substitution method, we migrated the sample data of multiple cities to the loss rate sample matrix of the study area. Combined with the output of the waterlogging early warning model, we obtained a waterlogging loss assessment method based on triangular type II fuzzy sets and the dynamic proportional substitution method. The waterlogging loss rate in the waterlogging loss assessment method is formulated as follows: Among them, f(s) is the flood loss rate, s is the corresponding flood depth, k and b are dynamic proportional substitution parameters, χ f The main membership function of the triangular type II fuzzy set of the property density of the study area is plotted based on the historical property density survey data of the study area. The three vertex coordinates (γ1,m * ), (γ2,1), (γ3,n * ), α is the cutoff value, The triangular type II fuzzy set representing the property density of the study area is: Necessity measure and the principal membership function χ f It can be expressed as: Among them, sup represents the supremum in mathematical operations; S52. With the constraints of minimizing waterlogging damage caused by poor pipeline drainage, minimizing drainage pipeline maintenance costs, and maximizing economic benefits, a pipeline maintenance optimization decision model based on multi-objective optimization and waterlogging damage assessment methods was constructed. The calculation method for waterlogging damage is as follows: Damage from flooding x i = property density × flooded area × waterlogging loss rate. The unit of property density is yuan / m 2 , the unit of flooded area is m 2 ; Drainage pipe maintenance cost i The calculation method is as follows: y i = siltation length * siltation degree * maintenance cost base * maintenance cost fuzzy coefficient; Maintenance cost fuzzy coefficient M cb The calculation of historical maintenance experience data is used to construct a triangular type II fuzzy set Combined with the three vertex coordinates of the triangular type-2 fuzzy set (γ1,m * ), (γ2,1), (γ3,n * ) and the cutoff value α, then Credibility measure and maintenance cost fuzzy coefficient M cb Expressed as: S53. Calculate whether maintenance measures are needed for the i-th section of pipeline while ensuring no risk of casualties, and calculate the target benefit function. M represents the total number of drainage pipes, a i Indicates whether the i-th section of the drainage pipe is silted up, b i Indicates whether maintenance measures are taken for the i-th section of drainage pipe. If so, If not taken S54, the constraint condition is to maximize the target profit function That is, max(Tar), which outputs the optimal decision results for each pipe section in the area.

8. The maintenance decision-making method for drainage pipe waterlogging warning according to claim 5, characterized in that: The method further includes: S7. The output of the waterlogging warning and optimization decision results from the unmanned boat deployed at the drainage pipe entrance and exit closest to the server, as well as the pipeline sedimentation database, are visualized and displayed on the drainage pipe full-process decision-making platform deployed on the server, thereby realizing an integrated drainage pipe decision-making system, specifically including: Multi-sensor data is obtained through the unmanned boat's Doppler ultrasonic module, three-axis gyroscope, high-definition infrared camera, and contactless liquid level meter. The multi-sensor data is uploaded and stored to the drainage pipeline full-process decision-making platform through the wireless communication module. The drainage pipeline full-process decision-making platform obtains a visual display of the output results of the unmanned boat's drainage pipeline sedimentation disease diagnosis model, waterlogging warning model, and pipeline maintenance optimization decision-making model. Based on the browser or server access control architecture, an integrated drainage pipeline decision-making system consisting of an unmanned boat, multi-sensor data, and a drainage pipeline full-process decision-making platform is developed and built to realize real-time monitoring of drainage pipelines, pipeline sedimentation disease diagnosis, waterlogging warning, and pipeline maintenance optimization decision-making functions.

Images