Rainfall classification method and system based on urban drainage network inflow characteristics

Abstract

The invention discloses a rainfall classification method and system based on urban drainage pipe network inflow features, and the method comprises the steps: carrying out the re-division of historical rainfall sessions, and extracting the session rainfall feature parameters corresponding to all obtained target rainfall sessions; calculating drainage pipe network overflow pollution core index parameters based on the session rainfall characteristic parameters; determining a target feature vector of each target rainfall session based on the session rainfall feature parameters and the overflow pollution core index parameters; and based on the target feature vector, determining the number of types of rainfall classification through a clustering algorithm, and constructing a corresponding rainfall classification model. According to the method, rainfall sessions are scientifically divided by combining the rainfall interval time and the pipe network emptying time, rainfall features and overflow pollution indexes are extracted to construct feature vectors, classification categories are determined through clustering optimization, the rainfall classification model is built, accurate classification of rainfall runoff is achieved, and reliable foundation support can be provided for runoff control and operation and maintenance of the urban drainage pipe network.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a rainfall classification method and system based on the inflow characteristics of urban drainage pipe networks. Background Technology

[0002] Urban flooding, also known as urban waterlogging, is a disaster phenomenon caused by rainwater runoff from short-term heavy rainfall or continuous rainfall exceeding the carrying capacity of the urban drainage system, resulting in water accumulation and inundation of surface areas such as roads, streets, and underground spaces. Essentially, it reflects the drastic changes in surface runoff conditions and the imbalance in drainage capacity under the background of rapid urbanization. It differs from traditional floods caused by river overflows or breaches, but its harm is equally severe.

[0003] Urban flooding poses comprehensive and severe hazards. Direct harms include road inundation causing traffic paralysis and disrupting normal urban operations; massive economic losses due to flooded vehicles, houses, and shops; and potential incidents such as electrical leaks and drownings, seriously threatening citizens' lives. Indirect and long-term harms are even more profound: floodwater mixed with sewage and garbage pollutes the environment, posing public health risks and damaging critical infrastructure such as electricity, communications, and water supply. Post-disaster reconstruction costs are enormous, business shutdowns cause losses, and damage to the city's image leads to decreased investment, severely impacting social stability and long-term economic development.

[0004] In the process of using information technologies such as the Internet of Things (IoT) to monitor river pollution, rainfall can significantly interfere with pollutant concentration monitoring data. On the one hand, rainfall increases river flow, and even if the total amount of pollutants in the river remains unchanged, the pollutant concentration will decrease due to water dilution. On the other hand, rainfall triggers surface runoff, washing pollutants such as agricultural fertilizers and heavy metals and organic matter from urban runoff into rivers, causing an increase in pollutant concentrations within the rivers. This dual impact significantly reduces the reliability of monitoring data during rainfall periods, making it prone to obvious anomalies. Existing technologies typically identify and remove such outliers through anomaly detection methods. However, the river's ecological environment is a continuously changing process, and a complete monitoring data sequence is crucial for accurately understanding the evolution of the river ecosystem and assessing pollution trends. Summary of the Invention

[0005] The purpose of this invention is to address the problem that existing technologies are unable to indicate the degree of impact of rainfall on overflow pollution and cannot meet the actual needs of precise treatment of overflow pollution in urban drainage networks. This invention proposes a rainfall classification method based on the inflow characteristics of urban drainage networks.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a rainfall classification method based on the inflow characteristics of urban drainage pipe networks is provided, including a model construction method for the rainfall classification model, comprising the following steps: Based on the rainfall interval between each historical rainfall event and the corresponding pipeline drainage time for each historical rainfall event, the rainfall events are reclassified to obtain several target rainfall events. The pipeline drainage time is the time required for the water accumulated in the pipeline to be drained to the design safe water level after the rainfall ends. Based on rainfall monitoring data, the corresponding rainfall characteristic parameters and overflow pollution core index parameters for each target rainfall event are extracted. The rainfall characteristic parameters include rainfall amount parameters, intensity parameters, and antecedent condition parameters, wherein the antecedent condition parameters include the corrected drought time and the average filling degree of the pipeline network before rainfall. The overflow pollution core index parameters include the total overflow amount of each type of pollutant, as well as one or more of the overflow risk index, pipeline hydraulic response coefficient, and comprehensive overflow pollution load. Based on the rainfall characteristic parameters and overflow pollution core indicator parameters, the target feature vector of each target rainfall event is determined. Based on the target feature vector, a clustering algorithm is used to determine the number of rainfall categories and construct a corresponding rainfall classification model. The rainfall classification model takes the target feature vector as input and outputs the corresponding rainfall category, which is used to indicate the degree of impact of rainfall on overflow pollution.

[0007] As one possible implementation method, the pipeline drainage time is compared with a preset basic interval time, and the maximum value between the two is taken as the shortest rainfall interval time for the corresponding historical rainfall events. Historical rainfall events with intervals less than or equal to the corresponding shortest intervals are merged, and several target rainfall events are obtained based on the merging results.

[0008] As one possible implementation method: Based on the characteristic parameters of each rainfall event, the core indicators of overflow pollution in the drainage network are calculated. These core indicators include the total overflow of each type of pollutant, as well as the overflow risk index, the hydraulic response coefficient of the network, and the comprehensive overflow pollution load. The entropy weight method is used to assign weights to the core indicators of overflow pollution in the drainage network, and the core indicators of overflow pollution in the drainage network with weights greater than the preset weight threshold are used as the corresponding core indicators of overflow pollution.

[0009] As one possible implementation method, based on the characteristic parameters of each rainfall event, the hydraulic-water quality coupling simulation is used to obtain the node water level process, pipe section filling process, overflow trigger time and overflow duration corresponding to each target rainfall event, as well as the overflow frequency and the total overflow of each type of pollutant corresponding to each target rainfall event. The comprehensive overflow pollution load for the corresponding target rainfall event is calculated based on the total overflow volume of each type of pollutant. Based on the obtained overflow duration, the overflow frequency is used to calculate the overflow risk index of the corresponding target rainfall event. The overflow risk index is used to reflect the frequency and duration of overflow occurrence. The hydraulic response coefficient of the pipeline network is calculated based on the pipeline segment filling process for the corresponding target rainfall event. The hydraulic response coefficient of the pipeline network reflects the impact intensity of rainfall on the hydraulic state of the pipeline network.

[0010] As one possible implementation method, one selects a core indicator parameter of overflow pollution as the evaluation indicator, and calculates the contribution of each rainfall characteristic parameter to the evaluation indicator based on a feature screening model. The input of the feature screening model is the rainfall characteristic parameter of each event, and the output is the evaluation indicator. Based on preset filtering rules, the corresponding target rainfall parameters are obtained by filtering from the characteristic parameters of each rainfall event based on the contribution. The target feature vector corresponding to the target rainfall event is constructed based on the target rainfall parameters and the core overflow pollution index parameters.

[0011] As one possible implementation, based on the feature vectors of each target, the elbow rule and the silhouette coefficient are used to jointly determine the optimal number of clusters; based on the optimal number of clusters, the corresponding rainfall classification is configured, and the rainfall classification corresponding to each target rainfall event is determined; The specific steps for determining the optimal number of clusters based on each target feature vector, using the elbow rule and silhouette coefficient in combination, are as follows: The feature vectors of each target are standardized to obtain several samples; K-means selects cluster centers using a roulette wheel method; Calculate the Euclidean distance from each sample to each cluster center, assign the sample to the nearest category, update the cluster centers, and repeat the iteration until the SSE change rate is less than the preset change rate threshold or the maximum number of iterations is reached. The number of clusters corresponding to the SSE drop inflection point and the maximum silhouette coefficient is selected as candidate solutions, and 5-fold cross-validation is used to verify the clustering stability. When the overflow pollution feature similarity of samples of the same category in the candidate solutions is greater than the preset similarity threshold, and the difference between samples of different categories is greater than the preset difference threshold, the optimal solution that has completed cross-validation is output. The optimal solution is used to indicate the number of optimal classification categories.

[0012] As one possible implementation, a high-dimensional classification space for judging the importance level of rainfall is constructed using the target rainfall parameters as coordinate axes, and a support vector machine (SVM) is used for fitting to generate the boundary conditions and boundary functions of the target rainfall parameters corresponding to each rainfall category. Configure overflow pollution contribution thresholds corresponding to the core overflow pollution indicator parameters for each rainfall category, and generate corresponding level judgment constraints; Based on the target rainfall parameter boundary conditions, boundary functions, and level judgment constraints, the judgment conditions corresponding to each category are generated to obtain the corresponding rainfall classification model.

[0013] As one possible implementation method, a method for classifying rainwater runoff based on the constructed rainfall classification model is also included, specifically: The rainfall classification model determines the rainfall category of the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

[0014] Secondly, a rainfall classification system based on the inflow characteristics of urban drainage networks is provided, including a model building module for constructing a corresponding rainfall classification model. The model building module includes: The division unit is used to re-divide the rainfall events based on the rainfall interval between each historical rainfall event and the corresponding pipe network drainage time for each historical rainfall event, and obtain several target rainfall events. The pipe network drainage time is the time required for the water in the pipe network to be drained to the design safe water level after the rainfall ends. The feature extraction unit is used to extract the corresponding event-specific rainfall characteristic parameters and overflow pollution core indicator parameters for each target rainfall event based on rainfall monitoring data; it is also used to determine the target feature vector for each target rainfall event based on the event-specific rainfall characteristic parameters and overflow pollution core indicator parameters; the event-specific rainfall characteristic parameters include rainfall parameters, intensity parameters, and antecedent condition parameters, wherein the antecedent condition parameters include the corrected drought time and the average filling degree of the pipeline network before rainfall; the overflow pollution core indicator parameters include the total overflow amount of each type of pollutant, as well as one or more of the overflow risk index, pipeline network hydraulic response coefficient, and comprehensive overflow pollution load; The clustering analysis unit is used to determine the number of rainfall categories based on the target feature vector using a clustering algorithm, and to construct a corresponding rainfall classification model. The rainfall classification model takes the target feature vector as input and outputs the corresponding rainfall category, which is used to indicate the degree of impact of rainfall on overflow pollution.

[0015] As one possible implementation, it also includes a classification unit; The classification unit is used to enable the rainfall classification model to determine the rainfall category corresponding to the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

[0016] This invention, by adopting the above technical solutions, has significant technical effects: This invention reclassifies rainfall events by combining rainfall intervals and pipe network emptying times, resulting in target rainfall events that closely reflect the actual operating conditions of the pipe network and avoiding classification biases caused by unreasonable rainfall event segmentation. It extracts rainfall characteristics and core overflow pollution indicators for the target rainfall events, providing reliable foundational data for subsequent feature vector construction and classification model building. Based on relevant parameters, it determines the target feature vector, effectively integrating the correlation information between rainfall and overflow pollution, providing comprehensive feature support for clustering and classification. By using clustering algorithms to determine the number of rainfall categories and constructing a classification model, it enables the scientific and systematic classification of stormwater runoff, providing fundamental model support for the management and maintenance of stormwater runoff in urban drainage pipe networks. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to the present invention. Detailed Implementation

[0019] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] Example 1: A rainfall classification method based on the inflow characteristics of urban drainage pipe networks, including a method for constructing a rainfall classification model, such as... Figure 1 As shown, it includes the following steps: S100. Reclassify the historical rainfall events and extract the event rainfall characteristic parameters corresponding to each target rainfall event. S110. Based on the rainfall interval between each historical rainfall event and the pipeline drainage time corresponding to each historical rainfall event, the rainfall events are re-divided to obtain several target rainfall events. The pipeline drainage time is the time required for the accumulated water in the pipeline to be drained to the designed safe water level after the rainfall ends. In this embodiment, based on the dual constraints of rainfall interval time and pipeline emptying time, rainfall events that meet the shortest rainfall interval time are classified into the same event. Specifically: S111. Determine the draining time of the pipeline network corresponding to each historical rainfall event, and determine the shortest rainfall interval time corresponding to the historical rainfall event based on the draining time of the pipeline network; in this embodiment, the time required for the water in the pipeline network to be drained to the preset safe water level after the rainfall of the corresponding historical rainfall event is determined by pipeline hydraulic simulation or on-site monitoring, and the corresponding draining time T0 of the pipeline network is obtained. In this embodiment, the stormwater management model (SWMM model) is used as a tool to import the pipeline network GIS data into the model and input the pipe segment parameters (pipe diameter, slope, roughness coefficient), node parameters (elevation, overflow outlet setting elevation) and catchment area underlying surface parameters (impermeability, permeability coefficient). Select a typical rainfall event and simulate the water level change process in the pipe network after the rainfall ends; A pipe section filled to less than 30% is considered a safe water level, ensuring that the pipe network has 70% flow space to prevent overflow caused by subsequent rainfall; The time from the end of rainfall to the first stable drop of the average filling degree of the pipeline to below 30% is the pipeline emptying time T0. The average filling degree of the pipeline refers to the average filling degree of all pipe sections at the corresponding time. The filling degree of the pipe section is the ratio of the water depth h in the pipeline to the inner diameter d of the pipeline.

[0021] As one possible implementation method, the draining time of the pipeline network is taken as the corresponding shortest rainfall interval time; As another possible implementation method, the obtained pipeline emptying time T0 is compared with the preset basic interval time T. base By comparing the two values, the maximum value is taken as the shortest rainfall interval T for the corresponding historical rainfall events. The specific calculation formula is as follows: T=max(T0,T base ); In this embodiment, to ensure that the initial scouring effect is independent, the basic interval time T is... base The interval is set to 5 hours, and those skilled in the art can set the basic interval time based on previous runoff pollution.

[0022] S112. Merge historical rainfall events with rainfall intervals less than or equal to the corresponding shortest rainfall interval, and obtain several target rainfall events based on the merging results.

[0023] S120. Based on rainfall monitoring data, extract the corresponding rainfall characteristic parameters for each target rainfall event; Based on the identified target rainfall events, and according to the corresponding rainfall monitoring data, three types of characteristic parameters are extracted for each rainfall event: rainfall amount, intensity, and antecedent conditions. Specifically: Rainfall parameters include: rainfall amount per event P (mm), rainfall duration D (min), and the coefficient of variation (CV) of total rainfall, which reflects the spatiotemporal uniformity of rainfall distribution. P ; Intensity parameter: 5-minute peak intensity I 5max (mm / h), average strength I avg (mm / h), time t from the start of rainfall to the peak intensity p (min); Preliminary condition parameters: Corrected drought time t' d The average filling degree F0 of the pipeline network before rainfall, calculated from pipeline network liquid level monitoring data; To reflect the impact of pollutant degradation during drought on initial accumulation, the drought time t' was corrected. d The calculation method is as follows: ; Among them, t d This represents the number of rainless days prior to the rainfall event; k is the pollutant degradation coefficient, calibrated through on-site monitoring; and · indicates a multiplication operation.

[0024] In this embodiment, the time step of the rainfall monitoring data is 1 minute.

[0025] Those skilled in the art may add rainfall characteristic parameters for each event as needed, and this specification does not impose detailed limitations on such additions.

[0026] S200: Based on the characteristic parameters of each rainfall event, calculate the core indicator parameters of overflow pollution in the drainage network; The core indicators of pollution from overflow in the drainage network include: The total overflow volume of each pollutant, overflow risk index, hydraulic response coefficient of the pipeline network, and comprehensive overflow pollution load.

[0027] S210. Based on the characteristic parameters of each rainfall event, the core indicator parameters of overflow pollution in drainage pipe networks are obtained through hydraulic-water quality coupling simulation.

[0028] In this embodiment, a refined coupled model of the drainage network in the study area is constructed based on SWMM. The refined coupled model includes a network topology module, a hydraulic simulation module, and a water quality simulation module. Specifically: Pipeline topology module: Input basic pipeline topology parameters, including pipe segment parameters and node parameters, divide the catchment area and match the underlying surface type, specifically: Pipeline parameters include: pipe diameter, slope, and roughness coefficient; Node parameters include: elevation, overflow outlet settings, and interception ratio; The types of underlying surfaces include: impermeability and permeability coefficient.

[0029] Hydraulic simulation module: Uses Saint-Venant's equations to solve unsteady flow in pipe networks, specifically: Input the basic parameters of the pipeline network topology and the characteristic parameters of each rainfall event; Output the node water level process, pipe section filling process, and overflow trigger time t for each rainfall event. overflow and overflow duration D overflow .

[0030] Water quality simulation module: Uses a coupled pollutant flushing model (such as an existing publicly available improved exponential model) to perform hydraulic-water quality coupled simulations, specifically: Input the initial pollutant concentration C0 of the pipeline network, the pollutant flushing model calibration parameters, and the hydraulic parameters output by the hydraulic simulation module; Output the total overflow pollution L for each rainfall event. total (kg), overflow pollution load per unit rainfall (L) unit (kg / mm), overflow frequency N overflow (Number of rainfalls / events); Among them, the total amount of overflow pollution L total (kg) includes the total overflow of a single pollutant for each type of pollutant. In this embodiment, the pollutants include suspended solids (SS), chemical oxygen demand (COD), total nitrogen (TN), total phosphorus (TP), ammonia nitrogen (NH3-N), and five-day biochemical oxygen demand (BOD5).

[0031] Based on the data output from the refined coupled model (data from the hydraulic simulation module and the water quality simulation module), the overflow risk index of various pollutants, the hydraulic response coefficient of the pipe network, and the comprehensive overflow pollution load are calculated. Specifically: The comprehensive overflow pollution load L is calculated based on the total overflow volume of each type of pollutant. com The comprehensive overflow pollution load L com It directly reflects the total contribution of rainfall to overflow pollution, and the specific calculation method is as follows: ; Among them, L iw represents the total overflow of the i-th pollutant. i The water environment impact weight of this pollutant is determined in accordance with GB3838-2022 "Surface Water Environmental Quality Standard".

[0032] Based on the obtained overflow duration D overflow Overflow frequency N overflow Calculate the overflow risk index R based on (number of rainfalls / events). overflow Overflow risk index R overflow The calculation method reflects the frequency and duration of overflows as follows: ; Where D represents the duration of rainfall.

[0033] Calculation of the hydraulic response coefficient R of the pipe network based on the pipe segment filling process hyd The hydraulic response coefficient R of the pipeline network hyd The impact intensity of rainfall on the hydraulic state of the pipe network is reflected in the following calculation method: ; Among them, F max F0 represents the maximum average filling degree of the pipeline network during the rainfall period, P represents the initial filling degree, and P represents the rainfall amount of the target event.

[0034] Those skilled in the art may add core indicator parameters for drainage network overflow pollution based on actual needs, and this specification does not impose detailed limitations on them.

[0035] S220. Feature screening is performed on the core indicator parameters of overflow pollution in drainage pipe networks to obtain the core indicator parameters of overflow pollution. In this embodiment, the entropy weight method is used to assign weights to the core index parameters of the overflow pollution of the drainage pipe network, and the core index parameters of the overflow pollution of the drainage pipe network with weights greater than a preset weight threshold (such as 0.2) are used as the core index parameters of overflow pollution. In this embodiment, the specific steps for calculating using the entropy weight method are as follows: First, the core indicator parameters of overflow pollution in the drainage network are normalized to eliminate the influence of dimensions. Second, calculate the entropy value E of the core indicator parameters of overflow pollution in each drainage pipe network. j The specific calculation method is as follows: ; Where, p ij , where represents the standardized value of the j-th core indicator parameter for drainage network overflow pollution in the i-th rainfall event, and n represents the total number of target rainfall events; Third, calculate the weights w of the core indicators of overflow pollution in each drainage network. j The specific calculation method is as follows: ; Where m is the number of parameters.

[0036] S300. Based on the rainfall characteristic parameters of the aforementioned events and the core index parameters of overflow pollution, determine the target feature vector for each target rainfall event; As one possible implementation method, a target feature vector is constructed based on the rainfall characteristic parameters of the said event and the overflow pollution core indicator parameters; To reduce the data dimensionality of the target feature vector and improve the efficiency of subsequent clustering, as another possible implementation method, target rainfall parameters are selected from the rainfall feature parameters of the said rainfall events, and a target feature vector is constructed based on the target rainfall parameters and the overflow pollution core indicator parameters. Specifically: S310. Calculate the contribution of each rainfall characteristic parameter to the core index parameter of overflow pollution, and select the target rainfall parameter from the rainfall characteristic parameters based on the contribution. S311. Select a core indicator parameter of overflow pollution as the evaluation indicator, and calculate the contribution of each rainfall characteristic parameter to the evaluation indicator based on the feature screening model (GBDT model). The input of the feature screening model is the rainfall characteristic parameter of each event, and the output is the evaluation indicator.

[0037] In this embodiment, the comprehensive overflow pollution load L is used. com As an evaluation metric, i.e., the output variable of the feature selection model, the feature selection model is iteratively trained to calculate the contribution of each rainfall characteristic parameter to the evaluation metric. In this embodiment, to avoid model overfitting, the following GBDT model parameters are set: The learning rate, which controls the contribution weight of each tree and balances fitting accuracy and stability, is set to 0.05. The number of decision trees was determined using 5-fold cross-validation to ensure the model fully learns the data patterns; it was set to 200. Maximum tree depth, to limit the complexity of a single tree and avoid overfitting, is set to 6; The loss function uses the mean squared error (MSE), which is suitable for continuous output variables.

[0038] S312. Based on preset filtering rules, select the corresponding target rainfall parameters from the rainfall characteristic parameters of each rainfall event based on the contribution degree. In this embodiment, the filtering rule is based on a preset number of feature filters (5), and selects several rainfall feature parameters with the greatest contribution as target rainfall parameters.

[0039] S320. Based on the target rainfall parameters and the overflow pollution core indicator parameters, construct the target feature vector corresponding to the target rainfall event; In this embodiment, the target feature vector is an 8-dimensional feature vector composed of 5 target rainfall parameters and 3 core overflow pollution indicator parameters.

[0040] S400. Based on the target feature vector, the number of rainfall categories is determined by improving the K-means clustering algorithm, and a corresponding rainfall classification model is constructed. The specific steps are as follows: S410. Based on the feature vectors of each target, the elbow rule and silhouette coefficient are used together to determine the optimal number of clusters k. The specific steps are as follows: Data preprocessing: Standardize each target feature vector to eliminate the differences in the units of different parameters and obtain several samples; Initializing cluster centers: K-means selects initial centers using a roulette wheel algorithm (samples farther away from the already selected centers have a higher probability of being selected), avoiding the local optima problem of traditional K-means; Iterative convergence: Calculate the Euclidean distance from each sample to each cluster center, assign the sample to the nearest class, update the cluster centers, and repeat the iteration until the SSE change rate is less than the preset change rate threshold (10). -4 (or reach the maximum number of iterations (500 times);) Cross-validation: The number of clusters corresponding to the inflection point of SSE drop and the maximum value of the silhouette coefficient is selected as the candidate solution, and 5-fold cross-validation is used to verify the cluster stability to ensure that the similarity (profile coefficient) of overflow pollution features of samples of the same category is greater than the preset similarity threshold (85%), and the difference between samples of different categories is greater than the preset difference threshold (70%). Output the optimal solution after cross-validation, which indicates the number of optimal classification categories.

[0041] S420. Configure the corresponding rainfall classification based on the optimal number of clusters, and determine the rainfall classification corresponding to each target rainfall event; The corresponding rainfall classification is configured based on the optimal solution, and the number of rainfall classification categories is the same as that of the optimal solution; Based on the aggregation results of the optimal solution, the rainfall classification corresponding to each target rainfall event is determined. That is, the target rainfall events whose target feature vectors are located in the same cluster belong to the same class. The rainfall classification corresponding to each target rainfall event can be determined based on the rainfall classification corresponding to each cluster.

[0042] As an example, the optimal number of clusters k=3, corresponding to rainfall categories of high-impact rainfall, medium-impact rainfall, and low-impact rainfall. Specifically: High-impact rainfall contributes the most to overflow pollution, medium-impact rainfall contributes a moderate amount, and low-impact rainfall contributes the least.

[0043] S430. Based on the target feature vectors and rainfall classifications corresponding to each target rainfall event, construct the corresponding rainfall classification model; This embodiment constructs a high-dimensional classification space based on the rainfall classification corresponding to each target rainfall event, defines the rainfall importance level and judgment threshold, and obtains a rainfall classification model. The specific steps are as follows: Constructing a high-dimensional classification space: Using the target rainfall parameters as coordinate axes, construct a high-dimensional classification space for judging the importance level of rainfall; Fitting classification boundaries: Support vector machine (SVM) is used to fit the target rainfall parameters of each category in the high-dimensional classification space, generating boundary surfaces of different levels of rainfall, and converting them into quantifiable boundary functions to obtain the target rainfall parameter boundary adjustment and boundary function; Level Judgment Constraints: Configure overflow pollution contribution thresholds corresponding to the core overflow pollution indicator parameters for each rainfall category to generate corresponding level judgment constraints; Rainfall importance classification criteria: Based on the boundary conditions, boundary functions, and level judgment constraints of the target rainfall parameters, the judgment conditions corresponding to each category are generated to obtain the corresponding rainfall classification model; This embodiment introduces an overflow pollution contribution threshold as a constraint for level determination, clarifying the quantitative determination rules for each rainfall importance level, for example: High-impact rainfall must meet the integrated overflow pollution load L com ≥100kg and overflow risk index R overflow ≥0.3; Low-impact rainfall must meet the integrated overflow pollution load L com ≤20kg / mm ​​and overflow risk index R overflow ≤0.1; The impact of rainfall falls somewhere in between.

[0044] As one possible implementation method, the rainfall classification model also includes a multi-source data fusion judgment rule for fuzzy regions. That is, based on the high-dimensional classification space, the fuzzy regions are discriminated, and the specific steps are as follows: Defining fuzzy regions: Rainfall samples near the cluster boundary surface with a profile coefficient in the range of 0.5 to 0.7 are defined as fuzzy region samples. The key rainfall parameters or core indicators of overflow pollution in this type of sample are in the transition range between different impact categories.

[0045] Fusion judgment of multi-source data in fuzzy areas: By fusing actual monitoring data of the pipeline network with the scenario simulation results of the rainstorm management model, and by fine-tuning the rainfall parameters by ±10%, the rate of change of the core indicators of overflow pollution is analyzed. If the rate of change is greater than 20%, the sample is classified into a higher-level impact category; otherwise, it is classified into a lower-level impact category.

[0046] Classification standard validation: Rainfall and overflow monitoring data for 12 consecutive months in the study area were selected, and valid rainfall samples were substituted into the classification standard for validation. The results showed that the classification accuracy was >88%, indicating that the classification standard has good applicability and stability.

[0047] In this embodiment, the above steps are illustrated using actual data: Step 1: Select a combined sewer network subsystem (service area 5.2km) 2 ); The basic parameters of the pipeline network topology are as follows: Pipeline network scale: 128 pipe sections, pipe diameter 300~1200mm, average slope 0.35%, roughness coefficient n=0.013 (concrete pipe). Nodes and overflow outlets: There are 96 nodes, including 8 overflow outlets (located at the end of the pipeline network, with the overflow outlet elevation corresponding to a pipe section fill degree of 0.8). Underlying surface: Industrial land (85% impermeability) accounts for 40%, residential land (70% impermeability) accounts for 50%, and green space (10% impermeability) accounts for 10%.

[0048] Simulated rainfall event: The rainfall event on June 15, 2022 was selected (rainfall duration 120 min, total rainfall 48 mm, average intensity 24 mm / h, peak intensity 38 mm / h). Hydraulic processes in the pipe network after rainfall: At the end of rainfall (14:00), the average filling degree of the pipe network (F=0.72) (some pipe sections were close to full flow); subsequently, the water level gradually decreased due to interception at the downstream sewage treatment plant and gravity drainage in the pipe sections. 14:30 (30 minutes after rainfall): Average fill degree F=0.58; 15:30 (90 min after rainfall): Average fill degree F = 0.41; 16:15 (135 min after rainfall): Average fill factor F=0.30, and remained stable between 0.28 and 0.30 for the next hour; The final T0 is determined as follows: the pipeline reaches a safe water level 135 minutes (2.25 hours) after rainfall, therefore T0 = 2.25 hours.

[0049] To ensure the initial scouring effect is independent, the base interval time T base Set to 5 hours.

[0050] According to the dual constraint mechanism: the shortest rainfall interval T = max(T0, T... base=max(2.25h,5h)=5h; that is, the time interval between two independent rainfall events in this area must be ≥5h. If a new rainfall event occurs within 5h after the previous rainfall event ends, they are merged into one rainfall event; if the interval is ≥5h, they are divided into two independent rainfall events.

[0051] Rainfall events were divided into sessions: Continuous rainfall data (time step 1 minute) from July 2022 (flood season) were selected and divided into sessions according to the above criteria. The results are as follows: Raw rainfall data: A total of 12 rainfall events were recorded in July, with a cumulative rainfall duration of 3800 minutes; After dividing the rainfall into events: there were a total of 8 independent rainfall events, and details are shown in Table 1.

[0052] Table 1 As can be seen from the table above, the two rainfall events on July 12 (interval 3.3h < 5h) were merged into one event, while the remaining rainfall events were divided into independent events due to intervals ≥ 5h. This ensured that the initial filling degree of the pipeline network corresponding to each rainfall event was ≤ 0.3 (safe water level), avoiding interference from residual water from previous rainfall on subsequent overflow pollution simulations.

[0053] Extracting rainfall characteristic parameters: Based on the 8 independent rainfall events after division, combined with rainfall radar monitoring data (time step 1 min) and pipeline online monitoring data, 12 rainfall characteristic parameters were extracted (Table 2) to construct the basic parameter system required for classification.

[0054] Table 2 1. Correcting the drought time t' d : Number of rainless days before the rain in session 1 (t) d =3d, pollutant degradation coefficient k=0.08d -1 (Calibrated through indoor degradation tests of pipeline sediments), then: t' d =3e-0.08×3=3×0.7866=2.45d. This parameter reflects the impact of pollutant degradation on the initial accumulation during the drought period, and is more in line with the actual pollution load than the traditional "number of rainless days before rain".

[0055] 2. Average filling degree of the pipeline network F0 before rainfall: The filling degree of the pipeline sections corresponding to the eight overflow outlets within 1 hour before rainfall in Session 1 were 0.22, 0.24, 0.26, 0.28, 0.23, 0.25, 0.27, and 0.25, respectively, with an average value F0=0.25, which meets the "safe water level" requirement and verifies the rationality of the session division.

[0056] 3. Coefficient of variation (CV) of total rainfall PStandard deviation of rainfall time series (1-minute step) for event 3 Mean μ P =20.0mm, then CV P =8.2 / 20.0=0.41, reflecting the uneven spatial and temporal distribution of the rainfall (with short-duration heavy rainfall peaks), which had a significant hydraulic impact on the pipeline network.

[0057] Step 2: Taking a combined sewer system subsystem (serving an area of ​​5.2 km²) as an example... 2 To construct an SWMM model, use the following parameters as the object: Basic geographical and pipeline parameters are shown in Table 3: Table 3 Water quality parameters (on-site calibration): Through on-site monitoring in March-April 2022 (20 sets of overflow water samples were collected, and rainfall parameters were monitored simultaneously), the scour model parameters and weights of 6 pollutants were calibrated as shown in Table 4: Table 4 Rainfall parameters (from the 8 rainfall events extracted in step 1): Taking event 1 (July 5th, 02:10-04:30, P=32.5mm, t') as an example... d =2.45d, F0=0.25), Session 3 (July 8, 15:30-17:50, P=45.2mm, t' d =2.68d, F0=0.26), Session 7 (July 20, 03:00-05:10, P=42.1mm, t' d =4.12d, F0=0.27) is a typical case, and is input into the SWMM model for simulation.

[0058] Simulation process and output results: 1. Hydraulic simulation results (taking session 3 as an example): Rainfall event 3 lasted 140 minutes, with a peak intensity of I at 5 minutes. 5max =35.8mm / h, the simulated hydraulic response process of the pipe network is as follows: 20 minutes after the start of rainfall (15:50): the fill rate of some downstream pipe sections reached 0.6, and the flow velocity increased from 0.8 m / s to 1.2 m / s; 45 minutes after the start of rainfall (16:15): Overflow outlets No. 3 and No. 5 were triggered for the first time (water level exceeded the overflow elevation), with a peak overflow flow rate of 0.5m. 3 / s; 30 minutes after the rainfall ended (18:20): all overflow outlets stopped overflowing, and the average filling degree of the pipeline network dropped to 0.35.

[0059] 2. Key hydraulic parameter output: Overflow frequency N overflow =2 (two overflow processes from 16:15 to 16:40 and from 16:55 to 17:10). Total overflow duration D overflow =45min; Maximum average fill degree F max =0.78; Hydraulic response coefficient of the pipeline network: 3. Water quality simulation results (complete data from 8 rainfall events): Calculate the total rainfall from the 8 rainfall events and correct the drought duration t' d The total overflow of the six pollutants is shown in Table 5 below: Table 5 Example of calculating the comprehensive overflow pollution load Lcom for event 7: The total overflow amounts of the six pollutants in Session 7 are as follows: SS=92.6kg, COD=60.5kg, TN=11.5kg, TP=4.2kg, NH3-N=5.1kg, and BOD5=26.8kg. The comprehensive load calculated by weighting wi is: Lcom=(92.6×0.25)+(60.5×0.22)+(11.5×0.18)+(4.2×0.15)+(5.1×0.12)+(26.8×0.08)=23.15+13.31+2.07+0.63+0.61+2.14=80.3kg.

[0060] Core indicator selection (entropy weight method): To reduce parameter redundancy, the entropy weight method is used to select the six parameters L in Table 4. SS L COD L TN L TP R overflow R hyd Weights were calculated, and core indicators (weight > 0.2) were selected. The weight calculation results are shown in Table 6. Table 6 Core Indicators Determined: Three core indicators were ultimately selected (cumulative weight 0.75): Comprehensive overflow pollution load L com (Weight 0.28): Directly reflects the total contribution of rainfall to overflow pollution; Overflow Risk Index R overflow (Weight 0.25): Reflects the frequency and duration of overflows; The hydraulic response coefficient R of the pipeline network hyd(Weight 0.22): Reflects the impact intensity of rainfall on the hydraulic state of the pipe network.

[0061] Step 3: Based on the 12 characteristic parameters of the 8 rainfall events extracted in Step 1 and the 3 core overflow pollution indicators calculated in Step 2, organize them into a model input matrix; GBDT Feature Filtering: Model parameter settings: Considering the characteristics of the small sample data (8 rainfall events), the GBDT model parameters are set to avoid overfitting: Learning rate: 0.05 (to control the contribution weight of each tree and balance fitting accuracy and stability); Number of decision trees: 200 (determined through 5-fold cross-validation to ensure the model fully learns data patterns); Maximum tree depth: 6 (to limit the complexity of a single tree and avoid overfitting); Loss function: Mean Squared Error (MSE), applicable to continuous output variables L com .

[0062] Feature importance calculation and ranking: Using the 12 rainfall parameters in Table 2 as input, L com As the output, these scores are substituted into the GBDT model for training to obtain the importance scores of each feature, as shown in Table 7. Table 7 Key parameter filtering results: The top 5 key rainfall parameters were selected (cumulative importance score 0.91): 1. Rainfall amount per event P (0.25): directly determines the total runoff and affects the flushing capacity of the pipeline network; 2.5 min peak intensity I 5max (0.21): Reflects the instantaneous impact of short-duration heavy rainfall on the pipeline network, which is prone to triggering overflow; 3. Correcting the drought time t' d (0.18): Determines the amount of surface pollutant accumulation and affects scour load; 4. Average filling degree of the pipeline network before rainfall F0 (0.15): reflects the initial bearing capacity of the pipeline network; a lower filling degree indicates stronger impact resistance. 5. Coefficient of variation (CV) of total rainfall P (0.12): Reflects the uneven spatial and temporal distribution of rainfall; a high value can easily lead to localized overflow in the pipe network.

[0063] Improved K-means clustering: Target feature vector construction: The five key rainfall parameters selected by GBDT are integrated with the three core overflow pollution indicators from step 2 to form the target classification feature vector for each rainfall event (Table 8): Table 8 Determining the optimal number of clusters: The 8-dimensional feature vectors were standardized (to eliminate dimensional differences), and the SSE and silhouette coefficients were calculated for the number of clusters (k=2, 3, 4) respectively (Table 9): Table 9 Elbow rule: SSE decreases as the number of clusters increases. When k=3, SSE drops sharply from 1.85 to 1.02. After that, the rate of decrease slows down. When k=4, it drops to only 0.89, forming the "elbow inflection point". Silhouette coefficient: The silhouette coefficient is the largest (0.78) when k=3, indicating that the sample has high similarity with the same class and large difference with the opposite class.

[0064] In summary, the optimal number of clusters is 3, corresponding to the following rainfall categories: high-impact rainfall (contributing the most to overflow pollution), medium-impact rainfall (contributing moderately), and low-impact rainfall (contributing little).

[0065] Improved K-means clustering process and results: Initial cluster center optimization: The K-means roulette wheel method is used to select the initial centers. The three samples with the largest feature vector differences (event 2, event 3, and event 7) are selected as the initial centers from the eight rainfall events to avoid the local optimum problem of traditional K-means.

[0066] Iterative convergence calculation: Calculate the Euclidean distance from each sample to the three initial centers, assign categories according to the "nearest neighbor principle", update the cluster centers, and repeat the iteration until the SSE change rate is <10. -4 (Converged after 12 iterations).

[0067] Clustering Results and Validation: The final classification results for the 8 rainfall events were obtained (Table 10), and the stability was evaluated using 5-fold cross-validation. Table 10 Clustering results rationality analysis: High-impact rainfall (events 3 and 7): Event 7, L com =80.3kg (highest in 8 games), R overflow =0.38 (frequent overflow and long duration), corresponding to P=42.1mm, I 5max =38.5mm / h, which meets the characteristics of high pollution contribution from "large rainfall + high peak intensity + long drought duration"; Low-impact rainfall (events 2 and 4): Event 2's L com =12.3kg (minimum), R overflow=0.05 (only a brief overflow), corresponding to P=21.8mm, t' d =0.22d (low pollutant accumulation), which meets the characteristics of low pollution contribution; The impact of rainfall on events 1, 6, and 8: all indicators fall between the two, such as L in event 6. com =56.7kg, R overflow =0.25, reflecting the transitional characteristics.

[0068] Step 4: Using the clustering results from Step 3 as the core (the 8 rainfall events are divided into high, medium, and low impact categories), compile the statistical characteristics of key parameters and core overflow pollution indicators for each type of rainfall event (Table 11), and construct the constructed rainfall classification model: Rainfall classification models include importance classification criteria and rules for handling fuzzy regions. Table 11 Statistics of key parameters and overflow pollution indicators for various types of rainfall (New event 5 is the rainfall on July 18, used for verification). Table 11 I. Quantification of Classification Criteria (Boundary Functions and Thresholds): High-dimensional boundary fitting (taking the core two-dimensional plane as an example): Select the two key parameters that contribute most to overflow pollution - the rainfall amount P (x-axis) and the corrected drought time t. d (y-axis) Construct a two-dimensional classification plane, and use SVM to fit the boundary functions of each category of rainfall (kernel function is RBF, goodness of fit R). 2 All > 0.89), as shown in Table 12: Table 12 Boundary function application example: If a rainfall event P = 40 mm, substituting it into the high-medium influence boundary function yields... If the actual t' d If t' > 4.5d, it is considered high-impact rainfall; if t' d If 3.0d < 4.14d, then further determination is needed by combining other parameters.

[0069] Threshold setting for overflow pollution contribution: Based on the water environment quality target of Guangming District, Shenzhen (surface water bodies reach the Class IV standard of GB3838-2022), the thresholds for the core indicators of overflow pollution for three types of rainfall are set as shown in Table 13.

[0070] Table 13 A complete classification standard system: integrating boundary functions and thresholds to form a three-in-one classification standard of "parameter-threshold-boundary". Table 14 Classification criteria for rainfall importance of a city's pipeline subsystem II. Handling of blurred areas: Fuzzy region definition: In the clustering results, rainfall samples with a silhouette coefficient between 0.5 and 0.7 are defined as fuzzy region samples. Table 15 Multi-source data fusion judgment: Taking match 6 as an example, the following steps are used to optimize the judgment: 1. Online monitoring data verification: During the 6th rainfall event, the average COD concentration at the 8 overflow outlets was 45 mg / L, and the peak overflow flow rate was 0.35 m³ / h. 3 / s, actual overflow pollution load L com =56.7kg < 65kg (high impact threshold); 2. SWMM scenario simulation: With a 10% increase in the P disturbance of field 6 (resulting in 42.6 mm), the simulated L is obtained. com =68.2kg≥65kg, but the actual P=38.7mm did not reach the value after disturbance; 3. Final judgment: Based on the monitoring data and simulation results, the actual overflow pollution load of event 6 did not reach the high impact threshold, and the original classification (medium impact rainfall) is maintained.

[0071] Similarly, the classification of rainfall events in session 8, after verification, remains in effect.

[0072] Classification Criterion Validation: Historical Data Back-substitution Validation: Substitute the 8 rainfall data from step 3 into the classification criteria of the above rainfall classification model to verify the classification accuracy. The results are shown in Table 16. Table 16 New data testing and verification: Four new rainfall events in August 2022 (not included in cluster modeling) were selected, categorized according to the established criteria, and compared with actual overflow monitoring data. The results are shown in Table 17. Table 17 Verification results: The classification accuracy of the four new rainfall events reached 100%, indicating that the standard has good stability and applicability.

[0073] This invention achieves an innovative breakthrough in the coupling and correlation mechanism, incorporating "rainfall characteristics, pipeline hydraulic state, and overflow pollution" into a unified analysis framework. By correcting for drought time and introducing key parameters such as the initial filling degree of the pipeline, it quantifies the nonlinear impact mechanism of rainfall on pipeline overflow pollution, providing a scientific basis for the accurate identification of high-risk rainfall.

[0074] This invention achieves a dual improvement in the efficiency and scientific rigor of classification methods. On the one hand, it relies on GBDT feature screening to identify the coefficient of variation (CV) of total rainfall, which is ignored by traditional methods. P By combining key parameters such as K-means multi-indicator analysis to determine the optimal number of clusters and avoid subjective bias in human judgment, the consistency between the classification results and the actual contribution of overflow pollution is improved by more than 20% compared with traditional methods. On the other hand, by reducing the dimensionality of multidimensional high-dimensional rainfall data to 5-dimensional key parameters, the clustering calculation time is shortened by 40%. Combined with 5-fold cross-validation to ensure the stability of the model, it can be adapted to the rainfall classification work of large-scale drainage pipe network systems. Relying on machine learning to achieve automatic identification of key rainfall features, dimensionality reduction of high-dimensional data and objective clustering, the scientificity and adaptability of the classification are further enhanced.

[0075] Furthermore, this invention possesses significant engineering application value and strong universality. In terms of precise monitoring, high-impact rainfall events account for only 15%–20% of all events, yet contribute over 75% of the overflow pollution load. Therefore, developing targeted monitoring schemes can reduce monitoring costs by 60% and achieve full coverage of major pollution sources. In terms of efficient management, a differentiated and precise overflow pollution auxiliary treatment system is formed based on classification standards. Before rainfall, rainfall pre-classification can be completed based on weather forecasts, triggering early warnings and guiding pre-emptive drainage of pipelines and pre-activation of storage facilities. During rainfall, operation and maintenance strategies are dynamically adjusted based on real-time monitoring data. After rainfall, the model and pipeline are optimized through evaluation data feedback. This control strategy, coupled with differentiated operation and maintenance strategies such as pre-emption of high-impact rainfall pipe networks and energy consumption optimization of low-impact rainfall storage facilities, can increase the overflow pollution load reduction rate by 18% to 25%, providing core technical support for the refined and intelligent operation and maintenance of drainage pipe networks. Furthermore, this method can flexibly adapt to different drainage pipe network types, such as co-current and diversion systems, by adjusting pipe network parameters such as interception ratio and pipe diameter, as well as pollutant weights. It can provide quantifiable and scalable standardized technical tools for overflow pollution control in drainage pipe networks of different climate zones and cities of different sizes, realizing a shift from "uniform effort" to "targeted management," and contributing to the long-term governance of urban drainage systems.

[0076] Example 2: A rainfall classification method based on the inflow characteristics of urban drainage pipe networks, which classifies rainwater runoff based on the rainfall classification model constructed in Example 1, includes the following steps: The rainfall classification model determines the rainfall category of the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

[0077] As can be seen from Example 1, there are several features that make up the target feature vector. Extracting these features based on rainfall monitoring data is existing technology or has been disclosed in Example 1, so it will not be described again here.

[0078] Example 3: A rainfall classification system based on the inflow characteristics of urban drainage pipe networks, including a model building module for constructing a corresponding rainfall classification model. The model building module includes: The division unit is used to re-divide the rainfall events based on the rainfall interval between each historical rainfall event and the corresponding pipe network drainage time for each historical rainfall event, and obtain several target rainfall events. The pipe network drainage time is the time required for the water in the pipe network to be drained to the design safe water level after the rainfall ends. The feature extraction unit is used to extract the corresponding event rainfall feature parameters and overflow pollution core indicator parameters for each target rainfall event based on rainfall monitoring data; it is also used to determine the target feature vector for each target rainfall event based on the event rainfall feature parameters and overflow pollution core indicator parameters. The clustering analysis unit is used to determine the number of rainfall categories based on the target feature vector using a clustering algorithm, and to construct a corresponding rainfall classification model.

[0079] Example 4: A rainfall classification system based on the inflow characteristics of urban drainage pipe networks, including classification units; The classification unit is used to enable a preset rainfall classification model to determine the rainfall category corresponding to the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

[0080] The rainfall classification model described is the rainfall classification model constructed in Examples 1 and 3.

[0081] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0083] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] This invention is described with reference to flowchart illustrations and / or block diagrams of the method, terminal device (system), and computer program product according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0085] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0087] It should be noted that: The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0088] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0089] Furthermore, it should be noted that the shapes and names of the parts and components described in the specific embodiments described in this specification may differ. All equivalent or simple variations made to the structure, features, and principles described in this patent concept are included within the protection scope of this patent. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to replace them, as long as they do not depart from the structure of this invention or exceed the scope defined in these claims, they should all fall within the protection scope of this invention.

Claims

1. A rainfall classification method based on the inflow characteristics of urban drainage pipe networks, characterized in that, The model building method for rainfall classification models includes the following steps: Based on the rainfall interval between each historical rainfall event and the corresponding pipeline drainage time for each historical rainfall event, the rainfall events are reclassified to obtain several target rainfall events. The pipeline drainage time is the time required for the water accumulated in the pipeline to be drained to the design safe water level after the rainfall ends. Based on rainfall monitoring data, the corresponding rainfall characteristic parameters and overflow pollution core index parameters for each target rainfall event are extracted. The rainfall characteristic parameters include rainfall amount parameters, intensity parameters, and antecedent condition parameters, wherein the antecedent condition parameters include the corrected drought time and the average filling degree of the pipeline network before rainfall. The overflow pollution core index parameters include the total overflow amount of each type of pollutant, as well as one or more of the overflow risk index, pipeline hydraulic response coefficient, and comprehensive overflow pollution load. Based on the rainfall characteristic parameters and overflow pollution core indicator parameters, the target feature vector of each target rainfall event is determined. Based on the target feature vector, a clustering algorithm is used to determine the number of rainfall categories and construct a corresponding rainfall classification model. The rainfall classification model takes the target feature vector as input and outputs the corresponding rainfall category, which is used to indicate the degree of impact of rainfall on overflow pollution.

2. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 1, characterized in that: The pipeline drainage time is compared with the preset basic interval time, and the maximum value between the two is taken as the shortest rainfall interval time for the corresponding historical rainfall event. Historical rainfall events with intervals less than or equal to the corresponding shortest intervals are merged, and several target rainfall events are obtained based on the merging results.

3. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 1, characterized in that: Based on the characteristic parameters of each rainfall event, the core indicators of overflow pollution in the drainage network are calculated. These core indicators include the total overflow of each type of pollutant, as well as the overflow risk index, the hydraulic response coefficient of the network, and the comprehensive overflow pollution load. The entropy weight method is used to assign weights to the core indicators of overflow pollution in the drainage network, and the core indicators of overflow pollution in the drainage network with weights greater than the preset weight threshold are used as the corresponding core indicators of overflow pollution.

4. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 3, characterized in that: Based on the characteristic parameters of each rainfall event, the hydraulic-water quality coupled simulation was used to obtain the node water level process, pipe section filling process, overflow trigger time and overflow duration corresponding to each target rainfall event, as well as the overflow frequency and the total overflow of each type of pollutant corresponding to each target rainfall event. The comprehensive overflow pollution load for the corresponding target rainfall event is calculated based on the total overflow volume of each type of pollutant. Based on the obtained overflow duration, the overflow frequency is used to calculate the overflow risk index of the corresponding target rainfall event. The overflow risk index is used to reflect the frequency and duration of overflow occurrence. The hydraulic response coefficient of the pipeline network is calculated based on the pipeline segment filling process for the corresponding target rainfall event. The hydraulic response coefficient of the pipeline network reflects the impact intensity of rainfall on the hydraulic state of the pipeline network.

5. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 3, characterized in that: Select a core indicator parameter of overflow pollution as the evaluation indicator, and calculate the contribution of each rainfall characteristic parameter to the evaluation indicator based on the feature screening model. The input of the feature screening model is the rainfall characteristic parameter of each event, and the output is the evaluation indicator. Based on preset filtering rules, the corresponding target rainfall parameters are obtained by filtering from the characteristic parameters of each rainfall event based on the contribution. The target feature vector corresponding to the target rainfall event is constructed based on the target rainfall parameters and the core overflow pollution index parameters.

6. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 5, characterized in that: Based on the feature vectors of each target, the elbow rule and the silhouette coefficient are used to jointly determine the optimal number of clusters; Based on the optimal number of clusters, the corresponding rainfall classification is configured, and the rainfall classification corresponding to each target rainfall event is determined; The specific steps for determining the optimal number of clusters based on each target feature vector, using the elbow rule and silhouette coefficient in combination, are as follows: The feature vectors of each target are standardized to obtain several samples; K-means selects cluster centers using a roulette wheel method; Calculate the Euclidean distance from each sample to each cluster center, assign the sample to the nearest category, update the cluster centers, and repeat the iteration until the SSE change rate is less than the preset change rate threshold or the maximum number of iterations is reached. The number of clusters corresponding to the SSE drop inflection point and the maximum silhouette coefficient is selected as candidate solutions, and 5-fold cross-validation is used to verify the clustering stability. When the overflow pollution feature similarity of samples of the same category in the candidate solutions is greater than the preset similarity threshold, and the difference between samples of different categories is greater than the preset difference threshold, the optimal solution that has completed cross-validation is output. The optimal solution is used to indicate the number of optimal classification categories.

7. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to claim 6, characterized in that: Using the target rainfall parameters as coordinate axes, a high-dimensional classification space for judging the importance level of rainfall is constructed, and a support vector machine (SVM) is used for fitting to generate the boundary conditions and boundary functions of the target rainfall parameters corresponding to each rainfall category; Configure overflow pollution contribution thresholds corresponding to the core overflow pollution indicator parameters for each rainfall category, and generate corresponding level judgment constraints; Based on the target rainfall parameter boundary conditions, boundary functions, and level judgment constraints, the judgment conditions corresponding to each category are generated to obtain the corresponding rainfall classification model.

8. The rainfall classification method based on the inflow characteristics of urban drainage pipe networks according to any one of claims 1 to 7, characterized in that, It also includes methods for classifying stormwater runoff based on the constructed rainfall classification model, specifically: The rainfall classification model determines the rainfall category of the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

9. A rainfall classification system based on the inflow characteristics of urban drainage pipe networks, characterized in that, The model building module includes a model building module for constructing a corresponding rainfall classification model. The model building module includes: The division unit is used to re-divide the rainfall events based on the rainfall interval between each historical rainfall event and the corresponding pipe network drainage time for each historical rainfall event, and obtain several target rainfall events. The pipe network drainage time is the time required for the water in the pipe network to be drained to the design safe water level after the rainfall ends. The feature extraction unit is used to extract the corresponding event-specific rainfall characteristic parameters and overflow pollution core indicator parameters for each target rainfall event based on rainfall monitoring data; it is also used to determine the target feature vector for each target rainfall event based on the event-specific rainfall characteristic parameters and overflow pollution core indicator parameters; the event-specific rainfall characteristic parameters include rainfall parameters, intensity parameters, and antecedent condition parameters, wherein the antecedent condition parameters include the corrected drought time and the average filling degree of the pipeline network before rainfall; the overflow pollution core indicator parameters include the total overflow amount of each type of pollutant, as well as one or more of the overflow risk index, pipeline network hydraulic response coefficient, and comprehensive overflow pollution load; The clustering analysis unit is used to determine the number of rainfall categories based on the target feature vector using a clustering algorithm, and to construct a corresponding rainfall classification model. The rainfall classification model takes the target feature vector as input and outputs the corresponding rainfall category, which is used to indicate the degree of impact of rainfall on overflow pollution.

10. The rainfall classification system based on the inflow characteristics of urban drainage pipe networks according to claim 9, characterized in that, It also includes classification units; The classification unit is used to enable the rainfall classification model to determine the rainfall category corresponding to the rainfall event to be analyzed based on the target feature vector corresponding to the rainfall event to be analyzed.

Images