Method for identifying urban complex underlying surface runoff forming key driving elements

Abstract

The invention relates to a method for identifying key driving elements of runoff formation of a complex underlying surface of a city, which comprises the following steps of: firstly, defining a variable source index as a new index for measuring dynamic contribution of a city impervious area to runoff formation; simulating and calculating a variable source index under the situation of combination of different influence factors through an urban rainfall flood model; further optimizing hyper-parameters of support vector regression by using a support vector regression model in combination with a Bayesian optimization method, and establishing a relation model between the influence factors and the variable source indexes; and finally, analyzing the interpretability of the support vector regression model by using an SHAP value, and quantifying the contribution degree of each influence factor to the variable source index. According to the method, the influence of the multi-element synergistic effect on the complex underlying surface runoff process is considered, the accuracy of urban rainfall flood simulation and prediction is effectively improved through urban rainfall flood model simulation fusing physical basis and data-driven machine learning, and a new thought is provided for urban hydrological modeling and rainfall flood management.

Description

Technical Field

[0001] The invention belongs to the technical field of urban stormwater management, and in particular relates to a method for identifying key driving factors for runoff formation on complex urban underlying surfaces. Background Art

[0002] With the continuous acceleration of urbanization, the urban hydrological effect has become more obvious, and the urban hydrological cycle process has undergone profound changes. The hydrological effect of urbanization is basically an exploration of how the increase of impervious underlying surface affects the hydrological cycle process. Previous studies focused on evaluating the impact of the expansion of the total impervious area on the hydrological response. Recently, it has gradually shifted from evaluating the impact of the total impervious area to considering the structural characteristics of the impervious surface, that is, distinguishing the effectiveness and ineffectiveness of the impervious surface on the hydrological process. In the total impervious area, the total impervious surface can be divided into effective impervious surface and ineffective impervious surface according to the hydraulic connectivity between the impervious surface and the drainage system. The runoff generated by the effective impervious surface directly enters the drainage system; the ineffective impervious surface is connected to the connected permeable surface, and the runoff formed first enters the connected permeable surface, and is discharged to the pipe network after the surface runoff is generated on the permeable surface.

[0003] The distinction between effective impervious surfaces and ineffective impervious surfaces is of great significance for understanding the impact of the increase of impervious surfaces on the urban hydrological cycle. Due to the high heterogeneity of the underlying surface, the surface runoff paths of different impervious surfaces are different. Although the ineffective impervious surface retains the runoff characteristics of the impervious surface, its runoff is affected by many factors such as the storage and infiltration of the connected pervious surface. At the same time, the runoff process of the connected pervious surface is also affected by the runoff from the ineffective impervious surface. There is a process of interaction between permeable and impervious surfaces, which profoundly affects the runoff process.

[0004] Although existing studies have emphasized the identification and estimation of effective impervious surfaces and their hydrological responses, they only focus on the binary classification of effective impervious surfaces and ineffective impervious surfaces, and pay insufficient attention to the interaction between different underlying surfaces under the influence of surface runoff paths, ignoring the variability of the proportion of impervious area contribution to runoff under the synergistic influence of various factors. Some studies analyze the impact of different types of impervious surfaces on the hydrological cycle based on scenario simulation, and sometimes even produce contradictory results. In addition, with the advancement of systematic global sponge city construction, more and more low-impact development measures are used to manage surface runoff from impervious surfaces, causing many effective impervious surfaces to be transformed into ineffective impervious surfaces. This transformation not only increases the complexity and heterogeneity of urban underlying surfaces, but also further complicates the surface runoff paths, and the impact on hydrological processes becomes more obvious. Traditional runoff generation and runoff theories and calculation methods have made it difficult to identify the key driving factors for runoff formation on complex urban substrates, to reveal the formation mechanism of runoff on complex substrates, to accurately analyze the impact of surface runoff paths on the runoff process, and to accurately identify the key factors that cause changes in the runoff generation and runoff mechanism, thus making it difficult to meet the needs of urban stormwater simulation and prediction.

[0005] Therefore, in view of the above situation, there is an urgent need to develop a method to identify the key driving factors of runoff formation on complex urban underlying surfaces in order to overcome the shortcomings of current urban stormwater management practices. Summary of the invention

[0006] In view of the problems and shortcomings of the prior art, the present invention provides a method for identifying the key driving factors for the formation of runoff on complex urban surfaces. By considering the impact of the synergistic effect of multiple factors on the runoff process on complex surfaces, integrating the physical-based urban stormwater model simulation and data-driven machine learning methods, the key driving factors affecting the formation of runoff on complex urban surfaces can be accurately and quickly identified.

[0007] To achieve the above object, the present invention provides the following technical solution, comprising the following steps:

[0008] S1. Define the variable source index as a key indicator to measure the dynamic contribution of urban impervious area to runoff formation, and calculate the variable source index of the hydrological response unit to be studied;

[0009] S2. Establish an ideal catchment area runoff model based on the urban stormwater model;

[0010] S3. Determine the influencing factors of the variable source index, consider the synergistic effects of urban rainfall characteristics, soil characteristics, and sub-catchment characteristics, and set up combination scenarios of different factors;

[0011] S4, based on the constructed ideal catchment area runoff model, simulate the hydrological response under different factor combination scenarios in step S3, and calculate the corresponding variable source index to form a large data set of variable source index;

[0012] S5. Combining machine learning methods with physical models, based on the obtained variable source index large data set, using Bayesian optimization method to optimize the hyperparameters of support vector regression, and establishing an influencing factor-variable source index regression relationship model;

[0013] S6. Use SHAP (SHapley Additive exPlanations) values ​​to analyze the interpretability of the trained influencing factor-variable source index regression model, quantify and rank the contribution of each influencing factor to the variable source index, and identify and determine the key driving factors of runoff formation under complex urban underlying surface conditions.

[0014] Preferably, step S1 defines a variable source index as a key indicator to measure the variability of the contribution ratio of urban impervious area to runoff. The underlying surface of a hydrological response unit is divided into three categories: permeable surface, effective impervious surface, and non-effective impervious surface. Focusing on the dynamic contribution of impervious area to runoff, the variable source index of effective impervious surface and non-effective impervious surface is first calculated respectively, and then the variable source index of the entire hydrological response unit is obtained by weighting according to the area.

[0015] Preferably, step S2 establishes an ideal catchment runoff model based on the urban stormwater model. In order to overcome the limitations of site characteristics in the simulation of actual runoff plots, the influence of various factors such as urban rainfall characteristics, soil characteristics, and sub-catchment characteristics on the accuracy and universality of the simulation results is comprehensively considered, an ideal catchment is constructed, and these variables are controlled and standardized, so as to more accurately simulate and analyze the runoff response under different scenarios, and provide a basis for subsequent simulation and calculation of variable source indexes under massive scenarios and revealing the influence of different factors on the variable source index.

[0016] Preferably, step S3 considers the synergistic effect of multiple factors and sets different factor combination scenarios. In order to overcome the limitations of site characteristics in existing scenario simulation studies, enhance the representativeness of the combination of influencing factors, and improve the generalization ability of subsequent machine learning, in the process of considering different influencing factor combination scenarios of ideal watersheds, not only key factors such as urban rainfall characteristics, soil characteristics, and sub-watershed characteristics are covered, but also special attention is paid to how these factors interact with each other in the context of urban heterogeneity, and a large number of combination scenarios of different factors are set.

[0017] Preferably, step S4 calculates and forms a large data set of variable source indexes under massive scenarios. In order to form a large data set of variable source indexes under massive scenarios, the ideal catchment area model constructed is used to simulate the hydrological response under different factor combination scenarios in step S2, and the variable source indexes corresponding to the different factor combination scenarios are calculated, thereby constructing a comprehensive large data set of variable source indexes.

[0018] Preferably, step S5 uses the Bayesian optimization method to construct the "influencing factor-variable source index" regression relationship. Support vector regression is used to construct the relationship between the influencing factors and the variable source index in a large number of scenarios. In this process, the Bayesian optimization method is used to intelligently select parameter combinations to ensure the ability to search for the global optimal solution while reducing the number of evaluation parameter combinations. Based on the Bayesian optimization method, the hyperparameters of the support vector regression are obtained to construct the "influencing factor-variable source index" regression relationship in a large number of scenarios.

[0019] Preferably, step S6 determines the key driving factors that affect the formation of runoff based on the SHAP values ​​of different influencing factors. After the support vector regression training is completed, an interpreter object is created that can explain the predictions of the model. The SHAP value of each sample in the data set is calculated through the interpreter object. The SHAP value represents the contribution of each factor to the model prediction results. A positive value indicates that the feature has a positive contribution to the prediction results, and a negative value indicates a negative contribution. Then, the SHAP value of each factor in all scenarios is calculated, and the absolute value is taken and then averaged to obtain the average absolute SHAP value of each factor, which represents the importance of each factor on the variable source index. Finally, the average absolute SHAP values ​​of different factors are sorted from high to low, and the key driving factors for the formation of runoff under the conditions of complex urban underlying surfaces are determined according to the order of importance of different factors.

[0020] As a further solution of the present invention, a variable source index is defined as a step to measure the dynamic contribution of urban impervious area to runoff formation. First, the underlying surface of a hydrological response unit is divided into three categories: permeable surface, effective impervious surface, and non-effective impervious surface. The relationship is:

[0021] A=A P +A EI +A NEI

[0022] Where: A is the total area of ​​the hydrological response unit, m 2 ; A P , A EI , A NEI are permeable, effective impermeable, and non-effective impermeable areas, m 2 ; A EI With A NEI The sum is the total impervious area.

[0023] When the rainfall on the effective impervious surface meets the requirements of depression storage, the runoff enters the drainage system, and its variable source index VSI EI :

[0024] VSI EI =1-D EI / P

[0025] Where: DEI is the effective impervious surface depression storage capacity, mm; P is the rainfall, mm.

[0026] In addition to satisfying depression storage, the non-effective impervious surface should also deduct part of the runoff that infiltrates or is stored through the connected permeable underlying surface (RPA). Its variable source index VSI NEI for:

[0027] VSI NEI =1-D NEI / PF RPA / P

[0028] Where: D NEI is the storage volume of non-effective impervious surface depression, mm; F RPA It is the runoff that infiltrates and accumulates on the non-effective impervious surface through the connected permeable underlying surface, mm.

[0029] Then, calculate VSI respectively EI 、VSI NEI Then the variable source index VSI of all impervious underlying surfaces is obtained by weighting by area:

[0030] VSI=VSI EI α EI +VSI NEI α NEI

[0031] Where: VSI is the variable source index of the impervious underlying surface, α EI is the ratio of effective impervious area to total impervious area in the hydrological response unit, α NEI It is the ratio of the non-effective impervious area to the total impervious area in the hydrological response unit.

[0032] As a further solution of the present invention, the steps of establishing an ideal catchment area runoff model based on the urban stormwater model specifically include: the ideal catchment area runoff model should include sub-catchments, inspection wells, and pipelines, and the surface runoff generated by the catchment area enters the pipeline through the inspection wells. In order to conveniently calculate the F in the variable source index calculation formula RPA , a comparison catchment area also needs to be set up.

[0033] The only difference between the comparison subcatchment and the ideal subcatchment is that the ideal subcatchment distinguishes between effective impervious surfaces and ineffective impervious surfaces, while the runoff from all impervious areas in the comparison subcatchment directly enters the pipe. The difference in the total amount of runoff entering the inspection well between the two subcatchments is the F in the calculation formula of the variable source index. RPA .

[0034] As a further solution of the present invention, considering the synergistic effect of multiple factors, the steps of setting different factor combination scenarios specifically include: first, determining the possible factors affecting the variable source index, and the change of the variable source index is affected by the effective impervious surface variable source index VSI EI , Non-effective impervious surface variable source index VSI NEI common impact.

[0035] VSI EI The influencing factors are relatively single; VSI NEI It mainly depends on F in the variable source index calculation formula RPA , F RPA In addition to being affected by rainfall, it is also mainly affected by the storage and infiltration process of the permeable underlying surface. The storage capacity is mainly caused by the characteristics of the sub-catchment area itself, such as the proportion of non-effective impervious area, slope, and permeable underlying surface depressions. The infiltration capacity is mainly caused by soil characteristics, such as soil infiltration capacity and soil initial water content. In addition, rainfall, as a driving factor of the hydrological cycle, also has a significant impact on the variable source index, and the total rainfall and rainfall schedule distribution need to be considered.

[0036] When considering different combinations of factors, set the upper and lower limits of each parameter as well as the step size in the simulation, and set all possible combinations of different factors.

[0037] As a further solution of the present invention, the step of calculating and forming a large data set of variable source index under a massive scenario specifically includes: using the constructed ideal catchment area and the comparison sub-catchment area model to simulate the hydrological response under the massive scenario of different factor combinations set in the third step, and calculating the difference between the total runoff of the ideal catchment area and the comparison catchment area, thereby obtaining F RPA , based on which the variable source index under different parameter combination scenarios is calculated, and a large data set of variable source index under massive scenarios is constructed and formed.

[0038] As a further solution of the present invention, the steps of constructing the "influencing factor-variable source index" regression relationship using the Bayesian optimization method are as follows:

[0039] First, the objective function is determined based on the prediction performance of the support vector regression model, including indicators such as the determination coefficient, mean square error, and mean absolute error; then, a search space for hyperparameters is set for the Bayesian optimization method, including different kernel function types, kernel function parameters, and regularization parameters.

[0040] In each iteration, the Bayesian optimizer selects a new set of hyperparameters for training the support vector regression model and evaluates its performance. Based on the evaluation results, the optimizer updates its internal Gaussian process model to guide the selection of hyperparameters in the next iteration. In addition, a cross-validation mechanism is set up during the parameter optimization process to ensure the robustness of the model and prevent overfitting or underfitting.

[0041] As a further solution of the present invention, the step of determining the importance ranking of different influencing factors based on their SHAP values ​​specifically includes: first, according to the "influencing factor-variable source index" relationship obtained by training with the support vector regression method in the fifth step, the SHAP value of the influencing factor in the fourth step is calculated. Then, for each influencing factor, the average absolute value of its SHAP value is calculated to represent the average impact of different influencing factors on model prediction. Finally, the influencing factors are ranked according to the average absolute SHAP value. The larger the value, the greater the impact of the factor on the model prediction. Finally, the importance ranking of different influencing factors is obtained, and the key influencing factors affecting the variable source index are determined.

[0042] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0043] 1. Compared with the existing identification and estimation techniques of effective impervious surfaces, the method of the present invention for identifying key driving factors for runoff formation on complex urban underlying surfaces converts the variable contribution of impervious surfaces in highly heterogeneous urban underlying surfaces to runoff into a quantifiable indicator, namely the variable source index. As a key indicator for measuring the dynamic contribution of urban impervious areas to runoff formation, it can more intuitively and effectively reveal the inherent mechanism of runoff formation and change on complex underlying surfaces.

[0044] 2. The method of the present invention identifies the key driving factors for the formation of runoff on complex urban underlying surfaces, considers the impact of the synergistic effect of multiple factors on the runoff process of complex underlying surfaces, develops a modeling paradigm based on data-driven machine learning methods coupled with a physically based urban stormwater model, and constructs an "influencing factor-variable source index" regression relationship, which can accurately and quickly identify the key factors that affect the contribution of impervious area to the drainage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention is a flowchart of the method for identifying key driving factors for runoff formation on complex urban underlying surfaces. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the application is further elaborated in detail below in conjunction with the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments of other researchers in the field on this embodiment belong to the protection scope of the present invention. At the same time, for the step numbering in the embodiment of the present invention, it is only set for the convenience of explanation, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0047] The embodiment of the present invention provides a method for identifying key driving factors of runoff formation on complex urban underlying surfaces, such as Figure 1 As shown, the following steps are included:

[0048] The first step is to define the variable source index as a key indicator to measure the dynamic contribution of urban impervious area to runoff. Specifically, the underlying surface of a hydrological response unit is divided into three categories: permeable surface, effective impervious surface, and non-effective impervious surface. The variable source index of effective impervious surface and non-effective impervious surface is calculated respectively, and then the variable source index of the entire hydrological response unit is obtained by weighting according to the area.

[0049] The second step is to establish an ideal catchment runoff model based on the urban stormwater model. An ideal catchment is constructed to control and standardize variables such as urban rainfall characteristics, soil characteristics, and sub-catchment characteristics, so as to more accurately simulate and analyze the runoff response under different combinations of influencing factors.

[0050] The third step is to consider the synergistic effect of multiple factors and set different combination scenarios of influencing factors. In the process of considering different combination scenarios of influencing factors of the ideal catchment area, a large number of combination scenarios of different factors are set to enhance the representativeness of the combination of influencing factors and improve the generalization ability of subsequent machine learning.

[0051] The fourth step is to calculate and form a large dataset of variable source indexes under massive scenarios. By constructing an ideal catchment model, the hydrological response under different factor combination scenarios is simulated, and the corresponding variable source index is calculated to construct a comprehensive large dataset of variable source indexes.

[0052] The fifth step is to use the Bayesian optimization method to construct the "influencing factor-variable source index" regression relationship. Support vector regression is used to construct the relationship between influencing factors and variable source index in massive scenarios. The Bayesian optimization method is used to intelligently select the hyperparameter combination of support vector regression, while reducing the number of evaluation hyperparameter combinations and ensuring the ability to search for the global optimal solution.

[0053] Step 6: Determine the key driving factors based on the SHAP values ​​of different influencing factors. Calculate the SHAP value of each influencing factor for all scenarios in the data set, and then calculate the average absolute SHAP value of each influencing factor under all scenarios to represent the importance of each factor on the variable source index and rank them. Determine the key driving factors that affect the formation of runoff on complex urban underlying surfaces based on the average absolute SHAP value.

[0054] It should be noted that the current binary classification of effective impervious surfaces and ineffective impervious surfaces does not pay enough attention to the interaction between different underlying surfaces under the influence of surface runoff paths, and ignores the variability of the proportion of impervious area contribution to runoff under the synergistic influence of various factors.

[0055] With the advancement of systematic global sponge city construction, more and more effective impervious surfaces are transformed into ineffective impervious surfaces, which not only increases the complexity and heterogeneity of urban underlying surfaces, but also further complicates the surface runoff paths, and has a more obvious impact on hydrological processes. Traditional runoff theory and calculation methods have been difficult to reveal the runoff formation mechanism of complex underlying surfaces. The embodiments of the present invention are intended to solve the above problems.

[0056] As a preferred embodiment of the present invention, the steps of defining a variable source index are as follows:

[0057] Firstly, the underlying surface of the hydrological response unit is divided into three categories: permeable surface, effective impervious surface, and non-effective impervious surface, and their respective areas are calculated.

[0058] Next, the depression storage capacity of effective impervious surface and non-effective impervious surface, as well as rainfall, are calculated.

[0059] Finally, the variable source index of the entire hydrological response unit is calculated based on area weighting.

[0060] In this embodiment, the variable source index defined simplifies the complexity of the urban underlying surface in the runoff formation process into a quantifiable indicator, which not only helps to simulate the urban runoff process more accurately, but also provides a new perspective for urban planning and flood management. By distinguishing between permeable surfaces, effective impervious surfaces and non-effective impervious surfaces and their different hydrological response processes, the contribution of urban impervious surfaces to runoff can be captured more meticulously, thereby improving the prediction accuracy of the model.

[0061] As a preferred embodiment of the present invention, the steps of establishing an ideal catchment runoff model based on an urban stormwater model are: first, an ideal catchment runoff model is constructed, including sub-catchments, inspection wells, pipelines, etc.; then a comparative catchment area is set to calculate the runoff that infiltrates and accumulates through the connected permeable underlying surface through the non-effective impermeable surface.

[0062] Specifically, the Green-Ampt method in the urban stormwater model is used to simulate the infiltration process, and the formula is as follows:

[0063]

[0064] In the formula, f p is the soil infiltration capacity, mm / h; K S is the saturated hydraulic conductivity of soil, mm / h; θ d is the initial soil water shortage rate; ψ s is the soil moisture absorption head at the wetting front, mm; F ​​is the cumulative infiltration (mm).

[0065] In the Green-Ampt method, the parameter K S Reflects the soil infiltration capacity, parameter θ d Various factors reflecting the previous soil moisture level, rainfall characteristics and sub-catchment characteristics are set through parameters in the urban stormwater model. The urban stormwater model is used to automatically simulate the hydrological response under massive scenarios, and calculate the variable source index under various combinations of influencing factors, thus constructing a large dataset of variable source index of impervious surfaces under massive scenarios.

[0066] In this embodiment, the key to constructing an ideal catchment runoff model is to control and standardize key variables, which enables the model to perform accurate simulations under different scenarios. Through idealized methods, the impact of actual site characteristics on simulation results can be reduced, thereby improving the universality and accuracy of the model.

[0067] As a preferred embodiment of the present invention, further considering the synergistic effect of multiple factors, the steps of setting different factor combination scenarios include: determining factors that may affect the variable source index, such as rainfall characteristics, soil characteristics, sub-catchment characteristics, etc., setting the upper and lower limits of each parameter and the step size in the simulation, and forming all possible combinations of different factors.

[0068] Specifically, setting up massive scenario combinations can comprehensively consider key factors such as urban rainfall characteristics, soil characteristics, sub-catchment characteristics and their interactions. This method not only enhances the representativeness of the model, but also improves the generalization ability of subsequent machine learning models in different urban environments.

[0069] As a preferred embodiment of the present invention, the step of calculating and forming a large data set of variable source indexes in a massive scenario specifically includes:

[0070] The ideal catchment area and the contrast sub-catchment area model are used to simulate the hydrological response under different factor combination scenarios.

[0071] The difference between the total runoff of the ideal catchment area and the comparison catchment area is calculated, and the variable source index can be obtained by further calculation, and a large data set of variable source index can be constructed.

[0072] It should be noted that, in this embodiment, the key to forming a large data set is to use an ideal catchment area model to simulate hydrological responses under massive scenarios. This method can not only provide rich data support, but also provide high-quality training samples for subsequent machine learning model training.

[0073] As a preferred embodiment of the present invention, the step of constructing the “influencing factor-variable source index” regression relationship using the Bayesian optimization method specifically includes:

[0074] First, determine the objective function, which is based on the prediction performance of the support vector regression model, including indicators such as determination coefficient, mean square error, and mean absolute error;

[0075] Then, the hyperparameter search space of the Bayesian optimization method is set, including the kernel function type, kernel function parameters, and regularization parameters.

[0076] Finally, a cross-validation mechanism is used to ensure the robustness of the model.

[0077] In the embodiment of the present invention, the Bayesian optimization method can be used to intelligently search for the optimal parameter combination in the parameter space, reduce the consumption of computing resources, and ensure the prediction performance of the model. This method significantly improves the efficiency and accuracy of model training.

[0078] As a preferred embodiment of the present invention, the step of determining the importance ranking of different influencing factors based on their SHAP values ​​specifically includes: calculating the SHAP value of each influencing factor under massive scenarios, calculating the average absolute SHAP value of each factor, and ranking them to determine the key influencing factors affecting the variable source index.

[0079] In the embodiment of the present invention, using SHAP value to sort the importance of influencing factors can intuitively reveal the degree of influence of different influencing factors on the variable source index. This method can not only identify key driving factors, but also provide a scientific basis for urban stormwater management and planning.

[0080] In summary, the method for identifying the key driving factors of runoff formation on complex urban substrates provided by the present invention can take into account the impact of the synergistic effect of multiple factors on the runoff process on complex substrates, integrate physical-based urban stormwater model simulation, data-driven machine learning and other technical means, develop a modeling paradigm based on the coupling of machine learning and hydrological models, accurately and quickly identify the key driving factors that affect the formation of runoff on complex urban substrates, and form a method with universal applicability.

[0081] The above description only describes in detail the preferred embodiments of the present invention and is not intended to limit the present invention. It should be pointed out that any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying key driving factors of runoff formation on complex urban underlying surfaces, characterized in that: The following steps are involved: S1. Define the variable source index as a key indicator to measure the dynamic contribution of urban impervious area to runoff formation, and calculate the variable source index of the hydrological response unit to be studied; S2. Establish an ideal catchment area runoff model based on the urban stormwater model; S3. Determine the influencing factors of the variable source index, consider the synergistic effects of urban rainfall characteristics, soil characteristics, and sub-catchment characteristics, and set up combination scenarios of different factors; S4, based on the constructed ideal catchment area runoff model, simulate the hydrological response under different factor combination scenarios in step S3, and calculate the corresponding variable source index to form a large data set of variable source index; S5. Combining machine learning methods with physical models, based on the obtained variable source index large data set, using Bayesian optimization method to optimize the hyperparameters of support vector regression, and establishing an influencing factor-variable source index regression relationship model; S6. Use SHAP value to analyze the interpretability of the trained influencing factor-variable source index regression relationship model, quantify and rank the contribution of each influencing factor to the variable source index, and identify and determine the key driving factors of runoff formation under complex urban underlying surface conditions.

2. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 1 is characterized in that: The variable source index is defined by comprehensively considering the urban rainfall characteristics, topographic conditions, soil types, and sub-catchment characteristics, quantifying the dynamic contribution of urban impervious areas to runoff formation under different combinations of factors, and describing the runoff dynamic process under urban underlying surface conditions; It is used to represent the proportion of runoff from impervious surfaces entering the pipe network to rainfall, and also represents the proportion of the impervious area of ​​the generated runoff entering the drainage pipe network to the total impervious area in the hydrological response unit.

3. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 2, characterized in that: In step S1, the variable source index of the hydrological response unit to be studied is calculated as follows: S1.

1. The underlying surface of the hydrological response unit to be studied is divided into three categories: permeable surface, effective impervious surface, and non-effective impervious surface. The relationship is: A=A P +A EI +A NEI (1) Where A is the total area of ​​the hydrological response unit, A P , A EI , A NEI are permeable, effective impermeable, and non-effective impermeable areas, respectively. EI With A NEI The sum is the total impervious area; S1.

2. Calculate the variable source index of effective impervious surface and non-effective impervious surface respectively; When rainfall on the effective impervious surface meets the requirements of depression storage, the runoff enters the drainage system. The variable source index VSI of the effective impervious surface EI , calculated as follows: ALL EI =1-D EI / P (2) Where D EI is the effective impervious surface depression storage capacity, P is the rainfall; In addition to satisfying the depression storage, the non-effective impervious surface also deducts the part of runoff that infiltrates or is stored through the connected permeable underlying surface. The variable source index VSI of the non-effective impervious surface NEI , calculated as follows: ALL NEI =1-D NEI / PF RPA / P (3) Where D NEI is the storage capacity of non-effective impervious surface depression, F RPA It is the runoff that infiltrates and accumulates on non-effective impervious surfaces through connected permeable underlying surfaces; S1.2, based on VSI EI 、VSI NEI , weighted by area, the variable source index VSI of all impervious underlying surfaces in the entire hydrological response unit is obtained: ALL = ALL EI α EI +ALL NEI α NEI (4) In the formula, α EI is the ratio of effective impervious area to total impervious area in the hydrological response unit, α NEI It is the ratio of the non-effective impervious area to the total impervious area in the hydrological response unit.

4. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 3 is characterized in that: In step S2, an ideal catchment area runoff model is established based on the urban stormwater model, including: sub-catchment areas, inspection wells, and pipelines. The surface runoff generated by the sub-catchment areas enters the pipeline through the inspection wells. Considering the characteristics of urban rainfall, soil and sub-catchment areas, an ideal catchment area and a comparative catchment area are constructed. The ideal catchment area distinguishes between effective impervious surfaces and ineffective impervious surfaces. The runoff of all impervious areas in the comparative catchment area directly enters the pipeline. The difference in the total amount of runoff entering the inspection well between the ideal catchment area and the comparative catchment area is calculated to obtain F. RPA .

5. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 4 is characterized in that: In step S3, the influencing factors of the variable source index are determined, including: effective impervious surface variable source index VSI EI , Non-effective impervious surface variable source index VSI NEI , VSI NEI Depends on F in step S1.2 RPA , F RPA Affected by rainfall, the storage capacity and infiltration capacity of the permeable underlying surface; The storage capacity is the sub-catchment area's own characteristics, including: the proportion of non-effective impervious area, slope, and permeable water subsurface depression storage; the infiltration capacity is the soil characteristics, including: soil infiltration capacity, soil initial water content; the rainfall, including: total rainfall, rainfall schedule distribution, setting different levels of rainfall and rain peak positions according to rainfall characteristics; Set up combination scenarios of different influencing factors, set the upper and lower limits of each influencing factor parameter and the step size in the simulation to form various combinations of different factors.

6. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 5 is characterized in that: In step S4, the Green-Ampt method in the urban stormwater model is used to simulate the infiltration process, and the formula is as follows: In the formula, f p is the soil infiltration capacity, K S is the saturated hydraulic conductivity of soil, θ d is the initial soil water shortage rate; ψ s is the soil moisture absorption head at the wetting front, and F is the cumulative infiltration; In the Green-Ampt method, the parameter K S Reflects the soil infiltration capacity, parameter θ d Various factors reflecting the previous soil moisture level, rainfall characteristics and sub-catchment characteristics are set through parameters in the urban stormwater model. The urban stormwater model is used to automatically simulate the hydrological response under massive scenarios, and calculate the variable source index under various combinations of influencing factors, thus constructing a large dataset of variable source index of impervious surfaces under massive scenarios.

7. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 1, characterized in that: In step S5, an influencing factor-variable source index regression relationship model is established, which includes the following sub-steps: S5.

1. Use support vector regression to construct the regression relationship between ideal catchment influencing factors and variable source index, and analyze the influence of rainfall driving factors, sub-catchment characteristics, and soil characteristics on the variable source index on the surface runoff path; S5.

2. Determine the objective function of support vector regression, where the objective function is determined based on the prediction performance of the support vector regression model, including indicators such as determination coefficient, mean square error, and mean absolute error; S5.3, setting a search space of hyperparameters for the Bayesian optimization method, including different kernel function types, kernel function parameters, and regularization parameters, and optimizing the hyperparameters of the support vector regression model to establish a regression relationship between the influencing factors and the variable source index; S5.

4. In each iteration, the Bayesian optimizer selects a new set of hyperparameters for training the support vector regression model and evaluating the performance. S5.

5. Based on the evaluation results, update the Gaussian process model inside the optimizer to guide the selection of hyperparameters in the next iteration; and set up a cross-validation mechanism during the parameter optimization process.

8. The method for identifying key driving factors of runoff formation on complex urban underlying surfaces according to claim 1 is characterized in that: In step S6, the interpretability of the trained influencing factor-variable source index regression relationship model is analyzed using the SHAP value, which includes the following sub-steps: S6.

1. According to the influencing factor-variable source index regression relationship model trained in step S5, the SHAP value of each influencing factor under the combination scenario of different factors in step S4 is calculated to quantify the influence of each factor on the model prediction result; S6.

2. For each influencing factor, calculate the average absolute value of the SHAP value, which represents the average impact of different influencing factors on the model prediction; S6.

3. Sort the influencing factors according to the average absolute SHAP value. The larger the value, the greater the impact of the influencing factor on the model prediction. The importance ranking of different influencing factors is obtained to determine the key influencing factors affecting the variable source index. S6.

4. Evaluate key influencing factors and provide decision support tools for urban stormwater management and drainage system design.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying key driving factors for runoff formation on complex urban underlying surfaces as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the method for identifying key driving factors for runoff formation on complex urban underlying surfaces as described in any one of claims 1 to 8 are implemented.

Images