A method and apparatus for simulating drainage pipe networks by integrating physical mechanisms and data-driven approaches.

Abstract

This invention provides a method and apparatus for simulating drainage pipe networks by integrating physical mechanisms and data-driven approaches. The method includes: acquiring a dataset of operating conditions for several pipes in a drainage pipe network within a target area during a prediction period; processing each operating condition dataset using a target prediction model to obtain the target local head loss coefficient for each pipe, wherein the target prediction model is trained based on samples of several operating condition datasets and their corresponding labels; setting the local head loss coefficients for each pipe in the drainage model as the target local head loss coefficients; and simulating the drainage situation of the drainage pipe network during the prediction period using the drainage model. This invention obtains accurate target local head loss coefficients for each pipe in the drainage pipe network through a data-driven target prediction model. After setting the local head loss coefficients for each pipe in the physical mechanism-based drainage model, the drainage model more accurately simulates the drainage situation of the drainage pipe network during the prediction period.

Description

Technical Field

[0001] This invention relates to the field of drainage pipe network engineering technology, and in particular to a method and apparatus for simulating and calculating drainage pipe networks that integrates physical mechanisms and data-driven approaches. Background Technology

[0002] With the advancement of global climate change and urbanization, coupled with the frequent occurrence of extreme rainfall events, urban flooding and drainage issues continue to attract attention. Drainage models are important tools for urban flooding early warning and forecasting, drainage network design and renovation, and other aspects.

[0003] During water transport in drainage pipe networks, head loss originates from friction head loss along pipe sections and local head loss caused by local structures such as pipe nodes. In areas with complex pipe network topology and dense pipe nodes, the impact of local head loss is particularly significant, often becoming the main component of the total head loss.

[0004] Currently, drainage models mainly use local head loss coefficients calculated through empirical formulas to simulate local head loss, which ultimately leads to inaccurate drainage simulations. Summary of the Invention

[0005] This invention provides a method and apparatus for simulating drainage networks by integrating physical mechanisms and data-driven approaches. This method addresses the shortcomings of inaccurate drainage simulations in existing technologies. By using a target prediction model to determine the local head loss coefficient of each pipe within the prediction period, the drainage model can more accurately simulate the drainage situation of the drainage network using the local head loss coefficient of each pipe within the prediction period.

[0006] This invention provides a method for simulating and calculating drainage pipe networks by integrating physical mechanisms and data-driven approaches, comprising the following steps: Obtain the operational data set of several pipes in the drainage network of the target area during the prediction period; The target prediction model is used to process the datasets of each operating condition to obtain the target local head loss coefficient of each pipeline. The target prediction model is trained based on several datasets of operating conditions and their corresponding labels. The local head loss coefficients for each of the aforementioned pipes in the drainage model are respectively set as the target local head loss coefficients; The drainage situation of the drainage network during the predicted period is obtained by simulation calculation using the drainage model.

[0007] According to the present invention, a method for simulating and calculating drainage pipe networks that integrates physical mechanisms and data-driven approaches is provided. The working condition data includes the inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle, and / or the flow depth.

[0008] According to the present invention, a method for simulating and calculating drainage pipe networks by integrating physical mechanisms and data-driven approaches is provided. The target prediction model includes several regression trees. The target prediction model is used to process the datasets of each operating condition to obtain the target local head loss coefficient for each pipe, including: Perform the following operations on each of the aforementioned pipes: The inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle, and the flow depth are used as inputs to the regression trees to obtain the target local head loss coefficient of the pipe.

[0009] According to the present invention, a method for simulating drainage pipe networks that integrates physical mechanisms and data-driven approaches is provided, wherein the target prediction model is obtained through training as follows: Obtain the local head loss coefficients from simulations of a 3D simulation model under several working condition datasets. The local head loss coefficients obtained from the simulation are used as the labels corresponding to the samples of each working condition dataset. The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

[0010] According to the drainage network simulation calculation method integrating physical mechanism and data-driven approach provided by the present invention, before using the simulated local head loss coefficients as labels corresponding to the samples of each working condition dataset, the training method further includes the following steps: Obtain the actual local head loss coefficient of the physical test model under a partial sample of the aforementioned working condition dataset; The simulated local head loss coefficients are corrected using the actual local head loss coefficients.

[0011] According to the present invention, a method for simulating and calculating drainage pipe networks by fusing physical mechanisms and data-driven approaches is provided. The target prediction model includes several regression trees. The target prediction model is trained based on samples from each of the aforementioned operating condition datasets and the corresponding labels of those samples. The training process includes: Obtain the parameter set of the target prediction model to be trained; Set the hyperparameters of the target prediction model; The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

[0012] According to the present invention, a drainage network simulation calculation method integrating physical mechanisms and data-driven approaches is provided. This method utilizes the drainage model to simulate and calculate the drainage situation of the drainage network during a predicted period, including: Obtain the rainwater inflow rate of each rainwater collection device in the drainage network of the target area during the predicted period; The drainage model is used to simulate the drainage situation of the drainage network during the predicted period based on the rainwater inflow.

[0013] This invention also provides a drainage network simulation computing device that integrates physical mechanisms and data-driven approaches, comprising the following modules: The working condition dataset acquisition module is used to acquire the working condition dataset of several pipes in the drainage network of the target area during the prediction period. The local head loss coefficient determination module is used to process each of the said working condition datasets using a target prediction model to obtain the target local head loss coefficient for each of the pipes. The target prediction model is trained based on several working condition dataset samples and their corresponding labels. The loss coefficient setting module is used to set the local head loss coefficient of each pipe in the drainage model to the target local head loss coefficient. The simulation calculation module is used to simulate and calculate the drainage situation of the drainage network during the predicted period using the drainage model.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches as described above.

[0017] This invention provides a method and apparatus for simulating drainage pipe networks by integrating physical mechanisms and data-driven approaches. First, a target prediction model is trained using several different operating condition datasets and their corresponding labels. This model learns the mapping relationship between different operating condition datasets and their corresponding local head loss coefficients. Therefore, after processing the operating condition datasets of each pipe within the prediction period, the target prediction model can obtain accurate local head loss coefficients. Then, the local head loss coefficients of each pipe in the drainage model are set, making the drainage situation of the drainage pipe network simulated by the drainage model more accurate during the prediction period. This integration of physical mechanisms and data-driven approaches allows the drainage model to retain the physical logic of hydraulic transmission while possessing the flexible adjustment capability to adapt to complex actual operating conditions, enabling more accurate simulation of the drainage situation of the drainage pipe network. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches provided by this invention.

[0020] Figure 2 This is a schematic diagram of the training process of the target prediction model provided by the present invention.

[0021] Figure 3 This is one of the schematic diagrams illustrating the performance evaluation results obtained by evaluating the target prediction model using the training set provided by the present invention.

[0022] Figure 4 This is one of the schematic diagrams of the performance evaluation results obtained by evaluating the target prediction model using a validation set, as provided by the present invention.

[0023] Figure 5 This is the second schematic diagram of the performance evaluation results obtained by evaluating the target prediction model using the training set provided by the present invention.

[0024] Figure 6 This is the second schematic diagram of the performance evaluation results obtained by evaluating the target prediction model using a validation set, as provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the drainage pipe network for the target area provided by the present invention.

[0026] Figure 8This is a schematic diagram comparing the predicted values ​​and measured values ​​obtained in two scenarios provided by the present invention.

[0027] Figure 9 This is a schematic diagram of the structure of the drainage network simulation computing device that integrates physical mechanisms and data-driven approaches provided by the present invention.

[0028] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] Considering that the drainage models currently built based on physical mechanisms have large errors in the local head loss coefficients determined by empirical formula estimation methods during the simulation of drainage pipe networks, the simulated drainage conditions of the target area are inaccurate.

[0031] To address this issue, this invention proposes a method for simulating drainage networks that integrates physical mechanisms and data-driven approaches. By utilizing the accurate local head loss coefficients of each pipe determined by a data-driven target prediction model, the local head loss coefficients of each pipe in the drainage model are set. Then, the drainage model is used to simulate and calculate the drainage situation of the drainage network during the predicted period. This integration of physical mechanisms and data-driven approaches allows the drainage model to retain the physical logic of hydraulic transmission while possessing the flexible adjustment capability to adapt to complex actual working conditions, enabling a more accurate simulation of the drainage situation of the drainage network.

[0032] The following is combined Figures 1 to 8 This invention describes a drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches. Figure 1 This is one of the flowcharts illustrating the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches provided by this invention. Figure 1 As shown, the method includes the following steps: Step 101: Obtain the operating condition dataset of several pipes in the drainage network of the target area during the prediction period.

[0033] The drainage network simulation calculation method integrating physical mechanisms and data-driven approaches provided by this invention can be applied to any electronic device with computing capabilities. Specifically, this invention can be applied to the drainage network design phase, as well as the flood prevention phase after the drainage network has been constructed.

[0034] The target area is any specific geographical area requiring drainage network planning and design or flood prevention forecasting, such as a specific administrative district of a city, a municipal road, or any area with drainage needs. The drainage network refers to the system within the target area used to collect and transport rainwater, and may include, but is not limited to, rainwater harvesting components and pipes. Rainwater harvesting components refer to the terminal facilities in the drainage network that collect surface rainwater, and may include, but are not limited to, storm drains, storm inlets, and infiltrated rainwater harvesting devices. Some pipes may include pipe sections and pipe joints. Several may be at least one.

[0035] The operating condition datasets for different pipes in a drainage network can be the same or different. The pipeline operating condition dataset includes several operating parameters of the pipe during the prediction period, each of which can affect the magnitude of the local head loss coefficient of the pipe. For example, the operating condition dataset may include pipe specifications and / or flow depth, etc. Specification parameters can be parameters used to characterize the pipe's structural features, such as including but not limited to at least one of the following: pipe inner diameter, pipe bottom elevation difference, pipe node connection type, and pipe intersection angle. Pipe specifications can be obtained from relevant design data of the drainage network. Flow depth can be the average flow depth within the pipe, and the flow depth can be obtained in ways including but not limited to: first, simulation by the drainage model based on data from the drainage network during the prediction period; second, simulation by the drainage model based on data from historical prediction periods adjacent to the prediction period; third, data collected by sensors installed within the pipe.

[0036] Step 102: Process the datasets for each operating condition using the target prediction model to obtain the target local head loss coefficient for each pipeline.

[0037] The target prediction model refers to a data-driven model that takes a working condition dataset as input and outputs the target local head loss coefficient of the pipeline. The type of target prediction model can include, but is not limited to, extreme gradient boosting (XGBoost), random forest, neural network, or support vector machine models. The target prediction model is trained on several working condition datasets and their corresponding labels. The composition of each working condition dataset can be the same as the aforementioned working condition datasets. The label corresponding to each working condition dataset sample can be the local head loss coefficient corresponding to that sample. The target prediction model is put into use after reaching a preset accuracy requirement through prior training. This embodiment uses an extreme gradient boosting model as an example.

[0038] Step 103: Set the local head loss coefficients for each pipe in the drainage model to the target local head loss coefficients respectively.

[0039] A drainage model is a mathematical model based on physical mechanisms used to simulate the hydraulic characteristics of drainage networks. For example, drainage models may include, but are not limited to, stormwater management models or other network hydraulic models, capable of simulating the transport of rainwater within the network based on the laws of conservation of mass, conservation of momentum, and pipe resistance formulas.

[0040] The drainage model needs to use the local head loss coefficient of each pipe in the process of simulating the drainage network. Therefore, the local head loss coefficient of each pipe in the drainage model is set as the corresponding target local head loss coefficient to ensure that the local head loss coefficient of each pipe matches the actual working conditions.

[0041] Step 104: Use the drainage model to simulate and calculate the drainage situation of the drainage network during the predicted period.

[0042] Drainage conditions can be used to characterize the hydraulic operation of the drainage network in the target area during the prediction period. For example, drainage conditions can include the flow rate, velocity, water level of each pipe section in the drainage network, the flow rate of each drainage outlet and / or outflow process line, etc.

[0043] In the above scheme, the target prediction model is first trained using several types of working condition datasets and corresponding labels. It can learn the mapping relationship between different working condition datasets and corresponding local head loss coefficients. Therefore, after processing the working condition datasets of each pipeline in the prediction period, the target prediction model can obtain accurate local head loss coefficients. Then, the local head loss coefficients of each pipeline in the drainage model are set so that the drainage network simulated by the drainage model in the prediction period is more accurate. This fusion of physical mechanism and data-driven approach makes the drainage model retain the physical logic of hydraulic transmission and has the flexible adjustment capability to adapt to complex actual working conditions, and can more accurately simulate the drainage situation of the drainage network.

[0044] In one possible embodiment, the operating condition data set includes the pipe's inner diameter, pipe bottom elevation difference, pipe node connection type, pipe intersection angle, and / or flow depth.

[0045] The target prediction model can be input by taking at least some of the following factors: pipe inner diameter, pipe bottom elevation difference, pipe node connection type, pipe intersection angle, and flow depth. The target local head loss coefficient of the pipe can be obtained from the output of the target prediction model.

[0046] Considering that local head loss can lead to energy loss due to pipe connection methods, the pipe operating data can be categorized by pipe node connection type. These connection types can include, but are not limited to, two-way nodes, three-way nodes, four-way nodes, irregularly shaped nodes, and special-function nodes. Pipes with two-way node connections are called two-way pipes, and pipes with three-way node connections are called three-way pipes.

[0047] The inner diameter of the pipe can include the inner diameter of all pipe branches within the pipe. Optionally, the pipe with the largest inflow can be used as the reference pipe, and its inner diameter value is numbered as follows: The inner diameter values ​​of other pipe branches within the pipeline are sequentially numbered according to a preset direction (such as clockwise or counterclockwise). , This allows the target prediction model to easily identify the pipe branch to which each inner diameter value belongs.

[0048] The pipe bottom elevation difference includes the difference between the bottom of each branch pipe and the main pipe within a pipeline system. For example, the pipe bottom elevation difference in a two-way pipeline is the height difference between the bottoms of the two branch pipes, which can be numbered as follows: A tee pipe system can use the pipe with the largest inflow as the main pipe, and then sequentially designate each branch pipe as the first branch pipe and the second branch pipe according to the preset direction. The difference in pipe bottom height between the first branch pipe and the main pipe is numbered as follows. The pipe bottom difference between the second branch pipe and the main pipe is numbered as follows: The pipe bottom difference numbering for other node connection types of pipes follows the same pattern.

[0049] The pipe junction angle includes the angle between each branch pipe and the main pipe at the junction. For example, the junction angle of a two-way pipe is the angle between the axes of the two branch pipes, which can be numbered as follows: A tee pipe can use the largest inlet pipe as the main pipe and the remaining pipe branches as branch pipes. The pipe intersection angle includes the angle between each branch pipe and the axis of the main pipe. Specifically, using the main pipe as a reference, each pipe branch is sequentially designated as the first branch pipe and the second branch pipe according to this preset direction, and the angle between the first branch pipe and the main pipe is numbered as follows: The angle between the second branch pipe and the main pipe is numbered as follows: .

[0050] The depth of water flow can be the depth of water flowing through a pipe.

[0051] In the above scheme, by acquiring data from multiple aspects of the pipeline, the target local head loss coefficient of the pipeline can be determined more accurately.

[0052] In one possible embodiment, the target prediction model includes several regression trees, where a regression tree refers to a base model for regression prediction constructed based on a decision tree algorithm. Optionally, the regression trees in the target prediction model can be serially integrated. Step 102 above may include performing the following operations on each pipe: taking the pipe's inner diameter, pipe bottom elevation difference, pipe node connection type, pipe intersection angle, and flow depth as inputs to each regression tree to obtain the target local head loss coefficient of the pipe.

[0053] Each regression tree is input to the pipe's inner diameter, pipe bottom elevation difference, pipe node connection type, pipe intersection angle, and / or flow depth. In some applications, the data in the operating condition dataset can be normalized before being input into the target prediction model.

[0054] Rapid inference and prediction can be performed based on the optimal decision tree ensemble (target prediction model) selected during the training process. That is, based on the splitting rules of the optimal ensemble tree, the target local head loss coefficient of the pipeline under any operating condition can be obtained.

[0055] In one possible embodiment, step 104 may include the following steps: obtaining the rainwater inflow of each rainwater collection device in the drainage network of the target area during the predicted period; and using a drainage model to simulate based on each rainwater inflow to obtain the drainage situation of the drainage network during the predicted period.

[0056] One method for obtaining the rainwater inflow of each rainwater collection device in the drainage network of the target area during the prediction period is to obtain the operating condition dataset of each rainwater collection device in the drainage network during the prediction period, and then determine the rainwater inflow of each rainwater collection device during the prediction period based on the operating condition dataset of the rainwater collection device during the preset period.

[0057] In the above scheme, by setting the local head loss coefficient for each pipe in the drainage model according to the target local head loss coefficient for each pipe, the drainage situation simulated and calculated by the drainage model can be more accurate.

[0058] In one possible embodiment, the training method of the target prediction model may include, for example: Figure 2 The following steps are shown: Step 201: Obtain the local head loss coefficients obtained by simulating the three-dimensional simulation model under several working condition datasets.

[0059] The 3D simulation model can include a 3D simulation model corresponding to each type of pipe node connection. Considering that in actual engineering, two-way nodes and three-way nodes are the two most widely used connection types, together forming the main connection structure of the drainage pipe network. In contrast, other types such as four-way nodes, irregular nodes, and special function nodes are only deployed in small quantities under specific working conditions, and their overall application rate is much lower than the former two. Therefore, in this embodiment, we focus on two-way and three-way node types. The simulation approach for four-way pipes can be developed with reference to three-way pipes. The main focus is on constructing 3D simulation models of pipe node connections with two-way and three-way nodes.

[0060] Optionally, before obtaining the local head loss coefficients from the 3D simulation model under several working condition datasets, the value ranges of each input variable can be determined first. For example, the value of the pipe inner diameter can take into account scenarios of small-diameter branch pipes and large-diameter main pipes, covering the commonly used engineering range of 0.1 m to 10 m to ensure the applicability of the results. The pipe bottom elevation difference can cover the commonly used engineering range, such as 0.0 m to 5.0 m. The junction angle of the two-way pipe is in the range of 45° to 90°, and the junction angle of the three-way branch pipe and the main pipe is in the range of 30° to 90°. The flow depth is based on the pipe inner diameter D, and the water depth gradient is set to the range of 0.1D to D. All variables are taken according to the set gradient to ensure that the influence of a single variable on the output results can be accurately captured. This embodiment consists of a total of 380 working condition dataset samples. The pipe inner diameter, pipe bottom elevation difference, pipe junction angle, flow depth, etc. can be parameterized.

[0061] In some application scenarios, commercial or open-source software can be selected, and a 3D simulation model can be built using the software's built-in modeling tools based on the pipe design parameters in the drainage network. This model can recreate the pipe structure for different pipe node connection methods, such as the inner diameter, pipe intersection angle, and pipe bottom elevation difference of two-way and three-way pipes. Irrelevant and redundant structures can be eliminated, simplifying the model complexity and improving computational efficiency. Furthermore, a variable adjustment interface can be reserved during the modeling process to facilitate the replacement and iteration of different input parameters later.

[0062] Then, a computational mesh for the 3D simulation model is generated using a polyhedral mesh generation tool. For example, a combination of structured and unstructured meshes can be used to divide the computational mesh of the 3D simulation model, discretizing the 3D simulation model into computational cells to balance computational accuracy and efficiency. Boundary conditions are set, such as defining the inlet, outlet, and wall conditions for the flow. For example, wall conditions can be: controlling the near-wall mesh size to be less than 1 / 100 of the pipe diameter, setting 5 to 8 near-wall boundary layer meshes, and setting the mesh growth rate to a reasonable range of 1.2 to 1.5, ensuring that the mesh orthogonality is greater than 0.7 and the aspect ratio is less than 5.

[0063] Then, a steady-state turbulence solver based on a semi-implicit algorithm of pressure coupling equations was selected for steady-state turbulence calculations. A renormalized swarm turbulent kinetic energy-turbulent dissipation rate turbulence model was adopted, and the physical properties of the water were set. The inlet velocity was set according to different operating conditions, and the turbulent kinetic energy was configured. U represents the average flow velocity and dissipation rate at the inlet cross-section of the main pipe (such as a trunk line) in the pipeline. Here, D represents the pipe's inner diameter. The iteration count is set to be no less than 1000 steps, and the residual convergence criterion is set to a pressure residual not exceeding [a certain value]. The velocity residual is no greater than The control parameters are solved by executing calculation instructions through the terminal, while the residual convergence status and inlet / outlet flow deviation (not exceeding 1%) are monitored.

[0064] Step 202: Use the local head loss coefficients obtained from the simulation as the labels for the samples of each working condition dataset.

[0065] According to the preset variable gradient, simulation calculations corresponding to different working condition dataset samples are started sequentially. The local loss coefficients obtained by the 3D simulation model under each working condition dataset sample are obtained through iterative solution. For example, after the 3D simulation model converges, the core output parameters are extracted for each working condition dataset sample, the local head loss is calculated, and then the local head loss coefficient obtained by simulation is calculated. The calculation method of the local head loss coefficient can refer to formula (1): Formula (1) in, This represents the local head loss coefficient. This represents the acceleration due to gravity, with a value of 9.81. , Indicates local head loss. This indicates the average flow velocity at the inlet section of the main pipe (e.g., trunk line) in a pipeline.

[0066] In the above scheme, the local head loss coefficient under various working condition datasets is simulated by a three-dimensional simulation model. This can improve the efficiency of obtaining the local head loss coefficient under the working condition datasets while ensuring the accuracy of the experiment, and provide a large number of learning samples for the target prediction model.

[0067] In one possible embodiment, before using the simulated local head loss coefficients as labels for the samples in the working condition dataset, the training method may further include the following steps: obtaining the actual local head loss coefficients measured by the physical test model under a subset of the working condition dataset samples; and correcting the simulated local head loss coefficients using the actual local head loss coefficients.

[0068] Considering that the local head loss coefficient obtained from the 3D simulation model may differ from the actual local head loss coefficient, in order to further ensure the accuracy of the labels corresponding to the working condition dataset samples, a physical test model can be constructed, and the local head loss coefficient obtained from the 3D simulation model under each working condition dataset sample can be corrected based on the local head loss coefficient of the physical test model under some working condition dataset samples.

[0069] For example, a two-way pipe with an inner diameter of 0.6m, a bottom height difference of 0.3m, a junction angle of 90°, and a flow depth of 0.3m is used as the test object. The total head of the pipe inlet and outlet sections is measured to calculate the head loss coefficient, calibrate and verify the simulation results of the three-dimensional simulation model, and ensure that the simulation error is controlled within 10%.

[0070] One way to construct a physical test model is to conduct experiments using a prototype. For example, a two-way pipe specimen with an inner diameter of 0.6m is fabricated, requiring the diameter deviation between the main pipe and the branch pipe to be less than ±1%, the intersection angle error to be less than ±0.5°, and the bottom height difference to be controlled at 0.3m. Optionally, a piezometer with an accuracy of 0.1mm or an ultrasonic level gauge is used, with three piezometer points (uniformly distributed across the cross-section) arranged at the top centerline of the straight sections at the inlet and outlet of the test section, and the average value is taken to reduce measurement error. An electromagnetic flowmeter (accuracy ≥0.5%) is installed at the inlet of the main pipe to monitor the flow rate in real time; simultaneously, a flow velocity meter is used to measure the velocity distribution at the outlet cross-section to verify the flow data. The inlet flow rate is controlled by adjusting the pump valve opening to stabilize the water depth in the pipe at 0.3m. After the water flow reaches a steady flow state (water level fluctuation ≤ ±0.5mm, flow rate fluctuation ≤ ±1%), it is maintained for 10 minutes before measurement begins to avoid the influence of transient water flow on the results. The head obtained by the piezometer at the inlet cross-section is recorded. And the head obtained from the piezometer at the inlet section Each working condition can be measured three times, and the average value is taken as the valid data.

[0071] Then, comparative tests can be conducted separately on straight pipe sections of the same diameter to determine the head loss along the pipe. Therefore, the local head loss of the two-way pipe is calculated. = - - Therefore, based on the above formula (1), the actual local head loss coefficient of the two-way pipe is obtained.

[0072] Among them, the local head loss coefficient under different scenarios was obtained through physical experiments. A total of 30 scenario simulations with different combinations of variables were conducted, and the local head loss coefficients were obtained from approximately 23 of these scenarios. The error between the values ​​obtained from the 3D simulation and the actual values ​​is within 10%, and the error for the seven sets of values ​​is within 20%. The method described above for correcting the simulated local head loss coefficients using the actual local head loss coefficients can be as follows: obtain the difference between the actual local head loss coefficients of each working condition dataset sample in the physical experiment and the local head loss coefficients obtained from the 3D simulation model, and adjust the simulated local head loss coefficients for all working condition dataset samples based on these differences.

[0073] By using physical test calculations to correct the local head loss coefficient obtained from the three-dimensional simulation model, it is possible to ensure that the target prediction model can better establish the mapping relationship between the datasets of various working conditions and the local head loss coefficient.

[0074] Step 203: Train the target prediction model based on samples from each working condition dataset and the corresponding labels for each working condition dataset.

[0075] One method for training the target prediction model based on samples from each working condition dataset and their corresponding labels is as follows: Divide the samples from each working condition dataset and their corresponding labels into a training set and a validation set according to a certain ratio. Use the training set to train the target prediction model, and use the validation set to validate the target prediction model.

[0076] In one possible embodiment, the target prediction model includes several regression trees. The target prediction model is trained based on samples from each working condition dataset and the corresponding labels of each working condition dataset. The training process includes: obtaining the parameter set of the target prediction model to be trained; configuring the hyperparameters of the target prediction model; and training the hyperparameter-configured target prediction model based on samples from each working condition dataset and the corresponding labels of each working condition dataset.

[0077] In some application scenarios, the parameter set includes the maximum depth of each regression tree, the range of the number of regression trees, and / or early stopping parameters. The maximum depth of a regression tree refers to the maximum number of branch layers allowed per tree, which can be used to control the complexity of a single regression tree. The range of the number of regression trees is the total number of regression trees used to build the target prediction model. Too few trees will lead to underfitting, while too many will increase computational costs. In this embodiment, the maximum number of regression trees is determined to be 150 through pre-testing. Early stopping parameters are used to determine whether to terminate model training prematurely. For example, early stopping parameters may include the number of epochs in which the validation set error does not decrease continuously and / or the maximum number of training epochs, used to balance model accuracy and training efficiency. For instance, early stopping parameters could be triggered when the validation set loss does not decrease for 20 consecutive epochs or when the training iteration count reaches a preset number.

[0078] The hyperparameter configuration of the target prediction model can be achieved by: setting the maximum depth of each regression tree according to the maximum depth of each regression tree in the parameter group, setting the number of regression trees included in the target prediction model according to the range of the number of regression trees in the parameter group, and / or setting the monitoring logic corresponding to early training stop according to the early training stop parameter in the parameter group.

[0079] For example, the target prediction model can be configured according to this parameter set, such as limiting the maximum depth of each regression tree to 4 layers, setting the number of regression trees to 150, and setting monitoring logic for early training termination. In some application scenarios, different learning rates (0.03~0.15) can be tested to control the scaling factor of the gradient correction amount of the contribution magnitude of each tree to prevent over-correction. Finally, the learning rate is determined to be 0.05, the minimum child node weight is set to 5, the split gain threshold is set to 0.5, and nodes are split only when the split gain is greater than the threshold. The ratio of the regularization parameter constrains the gradient sum.

[0080] Then, based on the samples in each working condition dataset and the corresponding labels, the target prediction model after hyperparameter configuration is iteratively trained until the training early stopping parameters are met. During the training of each regression tree, the generalization ability of the target prediction model can be evaluated using the validation set, and the model with the minimum loss on the validation set is continuously retained as the candidate optimal model.

[0081] Among them, when training the t-th regression tree, the objective function is... To optimize the objective, where the objective function is... , The objective function is the loss function between the predicted value obtained from the target prediction model and the local head loss coefficient obtained from the simulation, used to ensure prediction accuracy. By minimizing this objective function, the optimal regression tree that balances accuracy and generalization ability is obtained. For example, for the t-th training round, the first-order gradient and second-order gradient are obtained based on the loss function between the predicted value output by the target prediction model in the previous round (t-1 round) and the local head loss coefficient obtained from the simulation. The objective function is then guided by the first-order gradient and second-order gradient. Minimize.

[0082] In some application scenarios, samples from 380 publicly available datasets are divided into training and validation sets in an 8:2 ratio for iterative training of the target prediction model. Please refer to [link / reference]. Figures 3 to 6 , Figures 3 to 6 The horizontal axis represents the actual value (the local head loss coefficient obtained from the simulation), and the vertical axis represents the predicted value determined by the target prediction model. Figures 3 to 6 The red dashed lines (y=x) represent perfect prediction lines. Figure 3 and Figure 5 The blue dots represent the predicted values ​​obtained by the target prediction model. Figure 4 and Figure 6 The yellow dots represent the predicted values ​​obtained by the target prediction model.

[0083] Among them, the comparison between the predicted value and the true value of the two-way tube output by the target prediction model for the training set samples can be as follows: Figure 3 As shown, the comparison between the predicted and actual values ​​of the two-way tube output by the target prediction model for the validation set samples can be illustrated as follows: Figure 4 As shown. Figure 3 As shown, the coefficient of determination (R²) obtained by evaluating the target prediction model using the training set is... 2 The mean absolute error (MAE) reached 0.9976, and the mean absolute error (MAE) was 0.0045; Figure 4 It can be seen that the determination coefficient (R²) obtained by evaluating the target prediction model using the validation set is... 2 The mean absolute error (MAE) was 0.9666, and the mean absolute error (MAE) was 0.0319.

[0084] The comparison between the predicted and true values ​​of the T-junction output by the target prediction model for the training set samples can be as follows: Figure 5 As shown, the comparison between the predicted and actual values ​​of the T-junction pipe output by the target prediction model for the validation set samples can be illustrated as follows: Figure 6 As shown. Figure 5 As shown, the coefficient of determination (R²) obtained by evaluating the target prediction model using the training set is... 2 The mean absolute error (MAE) reached 0.9968, and the mean absolute error (MAE) was 0.0127; Figure 6 It can be seen that the determination coefficient (R²) obtained by evaluating the target prediction model using the validation set is... 2 The mean absolute error (MAE) was 0.9066, and the mean absolute error (MAE) was 0.0598.

[0085] This shows that the target prediction model has achieved good results in predicting the local head loss coefficient of two-way and three-way pipes.

[0086] In some application scenarios, the impact of different methods for determining local head loss on the outlet flow rate of drainage networks can be simulated based on drainage models. This embodiment uses... Figure 7 Taking the target area and its drainage network as an example, Figure 7 Black dots represent rainwater wells, black triangles represent drain outlets, black lines connecting two rainwater wells or connecting a rainwater well to a drain outlet represent pipes, and blue areas represent sub-catchment areas. The total area of ​​the target area is 224,605 ​​m². 2This includes 65 storm drain grates, 65 pipes, and 185 sub-catchment areas. Two scenarios were set up for comparative analysis: Scenario 1 uses the local pipe loss coefficient obtained from empirical values ​​as the parameter of the drainage model; Scenario 2 uses the local head loss coefficient obtained by the method provided in this invention as the parameter of the drainage model. The simulation results of the outlet flow rate of the drainage zones under the two scenarios were compared and analyzed.

[0087] Comparison of drainage zone outlet flow and measured outlet flow under scenarios 1 and 2 Figure 8 As shown. Figure 8 The medium gray curve represents the measured value of the outlet flow rate, the orange curve represents the predicted value of the outlet flow rate obtained from the drainage model of scenario one, and the blue curve represents the predicted value of the outlet flow rate obtained from the drainage model of scenario two. Figure 8 The horizontal axis represents time. For example, it can be used to obtain the flow rate of the drain outlet predicted in Scenario 1, the flow rate of the drain outlet predicted in Scenario 2, and the measured flow rate of the drain outlet at several time points between 1:00 AM and 9:00 AM in the target area. Figure 8 The vertical axis represents the flow rate of the drain outlet, in units of... The accuracy of the prediction was quantified using the Mean Absolute Error (MAE) and Nash Coefficient (NSE). A smaller MAE value indicates a smaller average deviation between the predicted and measured values. Compared with the measured outlet flow rate, the MAE for Scenario 1 was 0.71 and the NSE was 0.847; the MAE for Scenario 2 was 0.25 and the NSE was 0.902, indicating that Scenario 2 had higher simulation accuracy and was more precise in simulating the peak flow rate at the pipeline outlet.

[0088] Therefore, the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches provided by this invention can accurately simulate the drainage situation of the drainage network.

[0089] The following describes the drainage network simulation calculation device that integrates physical mechanisms and data-driven approaches provided by this invention. The drainage network simulation calculation device described below can be referred to in correspondence with the drainage network simulation calculation method described above. Figure 9 As shown, the drainage network simulation computing device 700, which integrates physical mechanisms and data-driven approaches, includes the following modules: The working condition dataset acquisition module 701 is used to acquire the working condition dataset of several pipes in the drainage network of the target area during the prediction period. The local head loss coefficient determination module 702 is used to process the data sets of each operating condition using the target prediction model to obtain the target local head loss coefficient for each of the pipes. The loss coefficient setting module 703 is used to set the local head loss coefficient of each pipe in the drainage model to the target local head loss coefficient, respectively. The target prediction model is trained based on several working condition dataset samples and corresponding labels. The simulation calculation module 704 is used to simulate and calculate the drainage situation of the drainage network during the predicted period using the drainage model.

[0090] According to the present invention, a drainage pipe network simulation calculation device 700 integrating physical mechanism and data-driven approach is provided, wherein the working condition data includes the inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle and / or the flow depth.

[0091] According to the present invention, a drainage pipe network simulation computing device 700 integrating physical mechanism and data-driven approach is provided. The target prediction model includes several regression trees. A local head loss coefficient determination module 702 processes each of the operating condition datasets using the target prediction model to obtain the target local head loss coefficient for each pipe, including: Perform the following operations on each of the aforementioned pipes: The inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle, and the flow depth are used as inputs to the regression trees to obtain the target local head loss coefficient of the pipe.

[0092] According to the present invention, a drainage network simulation computing device 700 integrating physical mechanisms and data-driven approaches is provided, wherein the target prediction model is obtained by training in the following manner: Obtain the local head loss coefficients from simulations of a 3D simulation model under several working condition datasets. The local head loss coefficients obtained from the simulation are used as the labels corresponding to the samples of each working condition dataset. The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

[0093] According to the drainage network simulation computing device 700 that integrates physical mechanisms and data-driven approaches provided by the present invention, before using the simulated local head loss coefficients as labels corresponding to the samples of each working condition dataset, the training method may further include the following steps: Obtain the actual local head loss coefficient of the physical test model under a partial sample of the aforementioned working condition dataset; The simulated local head loss coefficients are corrected using the actual local head loss coefficients.

[0094] According to the present invention, a drainage network simulation computing device 700 integrating physical mechanisms and data-driven approaches is provided. The target prediction model includes several regression trees. A training module trains the target prediction model based on samples from each of the aforementioned working condition datasets and the corresponding labels of those samples. The training module includes: Obtain the parameter set of the target prediction model to be trained; Configure the hyperparameters of the target prediction model; The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

[0095] According to the present invention, a drainage network simulation calculation device 700 integrating physical mechanisms and data-driven approaches is provided. The simulation calculation module 704 uses the drainage model to simulate and calculate the drainage situation of the drainage network during a predicted period, including: Obtain the rainwater inflow rate of each rainwater collection device in the drainage network of the target area during the predicted period; The drainage model is used to simulate the drainage situation of the drainage network during the predicted period based on the rainwater inflow.

[0096] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 10 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches. This method includes: acquiring a set of operating condition datasets for several pipes in the drainage network of the target area during a prediction period; processing each set of operating condition datasets using a target prediction model to obtain the target local head loss coefficient for each pipe, wherein the target prediction model is trained based on several sets of operating condition dataset samples and their corresponding labels; setting the local head loss coefficients for each pipe in the drainage model as the target local head loss coefficients; and using the drainage model to simulate and calculate the drainage situation of the drainage network during the prediction period.

[0097] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches provided by the above methods. The method includes: acquiring a set of operating condition datasets for several pipes in the drainage network of a target area during a prediction period; processing each set of operating condition datasets using a target prediction model to obtain a target local head loss coefficient for each pipe, wherein the target prediction model is trained based on several sets of operating condition dataset samples and corresponding labels; setting the local head loss coefficients for each pipe in the drainage model as the target local head loss coefficients; and using the drainage model to simulate and calculate the drainage situation of the drainage network during the prediction period.

[0099] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches, as provided by the methods described above. The method includes: acquiring a set of operating condition datasets for several pipes in a drainage network in a target area during a prediction period; processing each set of operating condition datasets using a target prediction model to obtain a target local head loss coefficient for each pipe, wherein the target prediction model is trained based on several sets of operating condition dataset samples and their corresponding labels; setting the local head loss coefficients for each pipe in the drainage model as the target local head loss coefficients; and using the drainage model to simulate and calculate the drainage situation of the drainage network during the prediction period.

[0100] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for simulating and calculating drainage pipe networks that integrates physical mechanisms and data-driven approaches, characterized in that, include: Obtain the operational data set of several pipes in the drainage network of the target area during the prediction period; The target prediction model is used to process the datasets of each operating condition to obtain the target local head loss coefficient of each pipeline. The target prediction model is trained based on several datasets of operating conditions and their corresponding labels. The local head loss coefficients for each of the aforementioned pipes in the drainage model are respectively set as the target local head loss coefficients; The drainage situation of the drainage network during the predicted period is obtained by simulation calculation using the drainage model.

2. The method according to claim 1, characterized in that, The operating condition data set includes the inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle, and / or the flow depth.

3. The method according to claim 2, characterized in that, The target prediction model includes several regression trees. The target prediction model is used to process the datasets for each operating condition to obtain the target local head loss coefficient for each pipeline, including: Perform the following operations on each of the aforementioned pipes: The inner diameter of the pipe, the difference in pipe bottom elevation, the pipe node connection type, the pipe intersection angle, and the flow depth are used as inputs to the regression trees to obtain the target local head loss coefficient of the pipe.

4. The method according to any one of claims 1 to 3, characterized in that, The target prediction model is obtained by training as follows: Obtain the local head loss coefficients from simulations of a 3D simulation model under several working condition datasets. The local head loss coefficients obtained from the simulation are used as the labels corresponding to the samples of each working condition dataset. The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

5. The method according to claim 4, characterized in that, Before using the local head loss coefficients obtained from the simulation as labels for the samples in each of the operating condition datasets, the training method further includes the following steps: Obtain the actual local head loss coefficient of the physical test model under a partial sample of the aforementioned working condition dataset; The simulated local head loss coefficients are corrected using the actual local head loss coefficients.

6. The method according to claim 4, characterized in that, The target prediction model includes several regression trees. Based on samples from each of the aforementioned work condition datasets and the corresponding labels for each sample, the target prediction model is trained, including: Obtain the parameter set of the target prediction model to be trained; Configure the hyperparameters of the target prediction model; The target prediction model is trained based on the samples of each of the aforementioned working condition datasets and the corresponding labels of each of the aforementioned working condition datasets.

7. The method according to any one of claims 1 to 3, characterized in that, The drainage situation of the drainage network during the predicted period is simulated and calculated using the drainage model, including: Obtain the rainwater inflow rate of each rainwater collection device in the drainage network of the target area during the predicted period; The drainage model is used to simulate the drainage situation of the drainage network during the predicted period based on the rainwater inflow.

8. A drainage pipe network simulation computing device integrating physical mechanisms and data-driven approaches, characterized in that, include: The working condition dataset acquisition module is used to acquire the working condition dataset of several pipes in the drainage network of the target area during the prediction period. The local head loss coefficient determination module is used to process each of the said working condition datasets using a target prediction model to obtain the target local head loss coefficient for each of the pipes. The target prediction model is trained based on several working condition dataset samples and their corresponding labels. The loss coefficient setting module is used to set the local head loss coefficient of each pipe in the drainage model to the target local head loss coefficient. The simulation calculation module is used to simulate and calculate the drainage situation of the drainage network during the predicted period using the drainage model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the drainage network simulation calculation method that integrates physical mechanisms and data-driven approaches as described in any one of claims 1 to 7.

Images