Urban inland inundation decision-making method based on agent simulation and evolution deduction

Abstract

The invention discloses an urban inland inundation decision-making method based on agent simulation and evolution deduction, and the method comprises the steps: constructing an urban inland inundation digital base model, dividing an urban region into a plurality of space grid units, and setting a static attribute and a dynamic state variable for each unit; establishing a drainage facility agent model; constructing an intelligent agent decision-making model, generating a drainage control action by taking the dynamic state of the space grid unit as input, and acting on the drainage facility agent model to adjust the operation state; a training data set is generated based on historical rainfall and ponding data, and the intelligent agent is trained by comprehensively considering the ponding depth, the threshold exceeding condition and the drainage facility operation cost through a reward function; the drainage facility is intelligently regulated and controlled by simulating the evolution process of urban inland inundation. According to the method, urban waterlogging evolution can be dynamically simulated, a drainage facility control strategy is optimized, economic losses of different land utilization types are considered, and urban global waterlogging intelligent prevention and control and decision optimization are achieved.

Description

Technical Field

[0001] This invention belongs to the field of smart city decision-making technology, specifically relating to an urban flooding decision-making method based on intelligent agent simulation and evolutionary deduction. Background Technology

[0002] With the acceleration of urbanization, urban flooding has become an increasingly prominent problem. Frequent extreme rainfall events lead to waterlogging on urban roads, building leaks, and property damage, seriously affecting urban safety and residents' lives. Existing technologies have proposed various prevention and control measures for urban flooding, including drainage facility construction, rainwater storage, and intelligent control strategies.

[0003] Traditional urban flood management typically relies on historical experience or static drainage facility scheduling rules, such as fixed pump station operating times or simple drainage strategies based on historical rainfall. These methods lack the ability to dynamically simulate the spatiotemporal evolution of urban flooding and cannot fully consider the impact of complex factors such as rainfall intensity, duration, spatial distribution, urban topography, and land use type on the water accumulation process.

[0004] Existing numerical simulation methods (such as two-dimensional hydrodynamic models or grid-based water accumulation simulations) can describe water movement, but they are computationally expensive when dealing with large-scale urban areas or high-resolution grids, and it is difficult to directly incorporate intelligent control strategies for drainage facilities into the simulation framework, making it difficult to use the simulation results for real-time or near-real-time urban flooding decision-making.

[0005] Existing urban flooding decision-making systems are mostly static scheduling systems based on rules or optimization algorithms, lacking adaptive capabilities and struggling to cope with complex and ever-changing rainfall scenarios. In particular, existing technologies struggle to provide unified consideration and intelligent control for the differences in economic losses corresponding to different land use types, the physical capacity limitations of drainage facilities, and the optimization of real-time scheduling.

[0006] In the prior art, CN117150600A discloses a rooftop water storage and drainage control method for urban flood prevention. This method includes the following steps: acquiring basic urban hydrological data; constructing a hydrological model based on the basic urban hydrological data; optimizing and coupling the urban hydrological model using rainfall data to obtain an urban flooding prediction model; performing prediction calculations based on the urban flooding prediction model to obtain urban flood prediction data; and generating autonomous decisions based on the urban flood prediction data to obtain a rooftop water storage and drainage decision strategy to control the rooftop water storage and drainage valves to perform water storage and drainage operations.

[0007] However, this method has the following limitations: it mainly controls rooftop drainage valves and lacks the ability to comprehensively simulate and regulate water accumulation across the entire city, including roads, green spaces, and other key surface types, thus failing to optimize the city's drainage system globally.

[0008] Therefore, there is an urgent need for a simulation and evolutionary model that can combine urban flooding digital base model, drainage facility proxy model and intelligent agent decision-making model to accurately simulate the dynamic evolution process of urban flooding, and intelligently regulate drainage facilities under various rainfall scenarios to optimize urban flooding prevention and control decisions, while taking into account economic losses and operating costs. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an urban flooding decision-making method based on agent simulation and evolutionary deduction.

[0010] The objective of this invention can be achieved through the following technical solutions: This invention provides an urban flooding decision-making method based on intelligent agent simulation and evolutionary inference, comprising the following steps: A digital baseline model for urban flooding is constructed, dividing the target urban area into multiple spatial grid units, and setting static attributes and dynamic state variables for each spatial grid unit; the dynamic state variable is the water depth. A proxy model for drainage facilities is established within the aforementioned digital infrastructure model for urban flooding. Construct an agent decision-making model; A training dataset for an intelligent agent decision-making model is constructed based on historical flooding data of the target city; the training dataset includes initial flooding status data and corresponding time-series rainfall data. Based on the initial water accumulation state data in the training dataset, the dynamic state variables of each spatial grid cell are initialized; The intelligent agent decision model uses the dynamic state variables of each spatial grid cell as state input, generates control actions for the drainage facility based on the state input, and applies the control actions to the drainage facility agent model to adjust the operating state of the drainage facility. Based on the operating status of the drainage facility proxy model, the static attributes of each spatial grid unit and the time-series rainfall data, the dynamic state variables in the urban waterlogging digital base model are updated to simulate the evolution process of urban waterlogging under the control action. The reward value is calculated based on the updated dynamic state variables, and the agent decision-making model is trained based on the reward value; Urban flooding decision-making is achieved based on a trained agent decision-making model.

[0011] Furthermore, the static attributes include terrain elevation, land use type, and drainage capacity; wherein, the terrain elevation is used to represent the relative topographic height of each spatial grid cell, determining the natural flow direction of water and the susceptibility to waterlogging. The land use types include rooftops, roads, green spaces, and water bodies; The drainage capacity is used to represent the maximum drainage volume of the drainage facilities covered by the spatial grid cell per unit time.

[0012] Furthermore, the construction of the urban flooding digital infrastructure model involves dividing the target urban area into multiple spatial grid units, and setting static attributes and dynamic state variables for each spatial grid unit, specifically including: Acquire topographic data, land use data, and drainage facility data for the target urban area. Then, based on a preset spatial resolution, perform grid-based discretization on the target urban area, generating a set of spatial grid cells, each denoted as . The terrain data includes digital elevation model data of the target urban area; the land use data includes spatial distribution data of roofs, roads, green spaces, water bodies and other surface cover types in the target urban area; the drainage facility data includes drainage network data, pump station location data, drainage outlet location data and corresponding drainage facility design drainage capacity parameters. Set corresponding static properties for each spatial grid cell, including the following steps: The terrain data is mapped to each spatial grid cell, and the terrain elevation of the corresponding spatial grid cell is determined by calculating the average value or the center point elevation value of the terrain elevation data within the spatial grid cell. ; The land use data is spatially overlaid with the spatial grid cells. Based on the area proportion of each land use type in the spatial grid cells, the land use type with the largest area proportion is selected as the land use type of the spatial grid cell. ; The service area of ​​the drainage facilities corresponding to each spatial grid unit is determined based on the drainage facility data, and the drainage capacity of the corresponding spatial grid unit is determined based on the design drainage capacity of the drainage facilities. ; Set dynamic state variables for each spatial grid cell Dynamic state variables Indicates time Spatiotemporal grid cells The depth of surface water accumulation was used to obtain a digital baseline model of urban flooding.

[0013] Furthermore, the drainage facility proxy model is used to represent the drainage facilities corresponding to each spatial grid unit of the target city, and each drainage facility has an adjustable operating state and physical capabilities.

[0014] Furthermore, the establishment of a drainage facility proxy model in the urban flooding digital base model specifically includes: The drainage facilities in the target city are abstracted as facility proxy points set in spatial grid cells, and each drainage facility is denoted as . ;in, Indicates the drainage facility number; For each drainage facility Record the corresponding connected spatial grid cells, including the entrance spatial grid cell. and export space grid unit The inlet space grid cell is used to represent the pumping location of the drainage facility, and the outlet space grid cell is used to represent the drainage location of the drainage facility. For each drainage facility The operating status is set, which includes a switch status and an opening degree. The switch status indicates whether the drainage facility is in operation, and the opening degree indicates the degree of adjustment of the drainage facility in operation. The opening degree is denoted as... ,and ; For each drainage facility Set physical capacity parameters, including maximum design drainage flow. Water level at facility inlet and facility outlet water level ; When the drainage facility When the switch is in the "on" state, the drainage facility At any moment Actual drainage flow The calculation formula is: in, Indicates drainage facilities Maximum design drainage flow rate; Indicates the drainage facilities at any time The degree of opening; Indicates the entrance space grid cell The water level of the accumulated water; Represents the exit space grid cell water level; The head influence coefficient is obtained by fitting the characteristic curve of the drainage facility. When the drainage facility When the switch is in the off state, the actual drainage flow rate of the drainage facility .

[0015] Furthermore, the training dataset for constructing the agent decision-making model based on historical flooding data of the target city specifically includes: Historical flooding data of the target city is obtained, including historical rainfall data and water accumulation status data corresponding to the rainfall data; the water accumulation status data is acquired through sensors. The historical rainfall data and water accumulation status data are time-aligned and spatially mapped to map the historical rainfall data to each spatial grid cell, and the water accumulation status data is converted into the initial water accumulation status data of the corresponding spatial grid cell. Time-series rainfall data is constructed based on the time-aligned historical rainfall data, and corresponding initial water accumulation status data is constructed based on the water accumulation status data. Based on the historical flooding data, multiple sets of extended data for different rainfall scenarios are generated by perturbating and expanding the rainfall intensity, duration, and spatial distribution. The initial waterlogging status data, the time-series rainfall data, and the extended data are combined to construct a training dataset containing various urban flooding scenarios.

[0016] Furthermore, the agent decision-making model includes a policy network and a value network; The state input of the agent decision-making model is represented as follows: in, Indicates time Status input; Indicates time Spatiotemporal grid cells The depth of surface water accumulation; Indicates the total number of spatial grid cells; The control actions of the agent decision-making model are represented as follows: in, Indicates time A set of control actions; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The opening degree is used to characterize the degree of adjustment of the drainage facility when it is in the open state; Indicates the total number of drainage facilities; The policy network is used to adjust based on state input. Generate control actions The value network is used to evaluate the cumulative reward of the control action in the current state and guide the training of the agent's decision-making model.

[0017] Furthermore, updating the dynamic state variables in the urban flooding digital infrastructure model based on the operational status of the drainage facility proxy model, the static attributes of each spatial grid unit, and time-series rainfall data specifically includes: At any moment to Within the time step, for each spatial grid unit in the digital baseline model of urban flooding dynamic state variables The update is performed using the following formula: in, Represents the updated spatial grid cell Dynamic state variables; Indicates the time step; Indicates time Acting on spatial grid cells The rainfall intensity was obtained based on time-series rainfall data; Represents spatial grid cells The amount of infiltration or retention loss; Representation and spatial grid cells Adjacent grid cell sets; Indicates from spatial grid cells Flow to adjacent grid cells Surface runoff; Indicates spatial grid cells The drainage flux discharged through drainage facilities; This indicates that water flows into the spatial grid cell through the drainage system. The drainage flux; The amount of infiltration or retention loss Based on the land use type in the static attributes of each spatial grid unit Confirmed, indicated as: in, Indicates land use type The corresponding infiltration coefficient is obtained by calibration based on historical observation data or empirical parameters; The surface runoff Determined based on the terrain elevation difference between spatial grid cells, expressed as: in, , Representing spatial grid cells and Terrain elevation in static attributes; The water exchange coefficient between spatial grid cells is determined by a combination of the distance between spatial grid cells, the connection relationship between them, and the surface roughness parameters. The drainage flux is based on the connection to the spatial grid unit. The set of drainage facilities is defined as follows: in, Indicates drainage facilities The entrance space grid unit; Indicates drainage facilities The exit space grid cell; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The actual drainage flow rate.

[0018] Furthermore, the training process of the intelligent agent decision-making model includes: At every moment The reward value is calculated based on the updated dynamic state variables of each spatial grid cell. ; Based on the reward value Calculate the cumulative return from the current moment. , represented as: in, Indicates from time Initial cumulative returns; Indicates the discount factor. Indicates the termination time of the simulation process; The value network of the agent decision-making model is based on the state input. Estimate the corresponding state value, and based on the cumulative reward. Update the value network parameters; The policy network of the agent decision-making model, based on the evaluation results of the value network, maximizes the cumulative reward. Update the strategy network parameters and optimize the control strategy for drainage facilities.

[0019] Furthermore, the reward value The formula is: in, Indicates time The reward value; These are the weighting coefficients; Indicates time Spatial grid unit The dynamic state variable, namely the water depth; Represents spatial grid cells At any moment An indicator function for whether water accumulation has exceeded a threshold, when The value is 1 if the condition is met, and 0 otherwise. Represents spatial grid cells The preset water depth threshold; Represents spatial grid cells Land use types; Indicates land use type The corresponding economic loss weighting coefficient is used to characterize the potential economic loss of different land use types when waterlogging occurs; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The degree of opening; Indicates drainage facilities At any moment The actual drainage flow rate; Indicates the total number of drainage facilities; This indicates the total number of spatial grid cells.

[0020] Compared with the prior art, the present invention has the following advantages: (1) Existing technologies generally rely on historical experience or static rules for drainage scheduling, lacking the ability to dynamically respond to rainfall processes and water accumulation evolution. They are difficult to accurately reflect the impact of rainfall intensity, duration, and spatial distribution changes on the urban flooding process, resulting in delayed or ineffective scheduling strategies. This invention introduces dynamic state variables into the digital basis model of urban flooding and updates the water accumulation depth of each spatial grid unit hourly based on rainfall input, infiltration process, surface runoff, and drainage facility operation status. This enables continuous simulation of the spatiotemporal evolution of urban flooding. At the same time, it combines an intelligent agent decision model to dynamically generate control actions based on real-time status, thereby significantly improving the response capability to complex rainfall processes and achieving a more refined and real-time urban flooding control effect.

[0021] (2) Although existing numerical simulation methods can characterize hydrodynamic processes, they suffer from high computational complexity in high-resolution or large-scale urban modeling and are difficult to effectively couple with drainage facility control strategies, limiting their application in real-time decision-making. This invention adopts a simplified hydrodynamic update model based on spatial grid cells and introduces a drainage facility proxy model to parameterize drainage behavior, reducing computational complexity while ensuring physical rationality. At the same time, by clearly defining the inlet and outlet grid cells of the drainage facility, the directional transfer of drainage flux between grids is realized, thereby achieving efficient coupling between hydrodynamic processes and control strategies. This makes the model not only computationally efficient but also able to support real-time training and application of intelligent decision-making.

[0022] (3) Existing urban flooding decision-making methods are mostly static scheduling methods based on rules or traditional optimization methods, which lack adaptive learning capabilities and are difficult to continuously optimize scheduling strategies under variable rainfall scenarios. This invention constructs an intelligent agent decision-making model that includes a policy network and a value network. It takes the water accumulation state of spatial grid cells as input and the opening and closing status of drainage facilities as control actions as output. Based on the reward feedback during the simulation process, the model is iteratively trained, enabling the agent to continuously learn and optimize scheduling strategies under various rainfall scenarios. This achieves adaptive decision-making capabilities for complex environments and significantly improves the generalization and robustness of scheduling strategies.

[0023] (4) Existing technologies, when conducting drainage scheduling, typically only focus on physical indicators such as water volume or water level, without considering the differences in economic value corresponding to different land use types. This results in scheduling strategies failing to reflect the actual need for "priority protection of key areas" and making it difficult to minimize overall economic losses. This invention introduces an economic loss weight coefficient corresponding to the land use type into the reward function, and weights and couples the water accumulation depth and threshold state of different spatial grid units with their economic value. This allows the agent to automatically learn during training to prioritize reducing the water accumulation risk of high-value areas, thereby achieving optimized scheduling guided by minimizing economic losses and improving the practical application value of urban flood control decisions.

[0024] (5) Existing technologies often neglect the operating costs and energy consumption of drainage facilities during drainage scheduling, which may lead to overuse of drainage equipment, reduce system operating efficiency, and increase maintenance costs. This invention introduces a drainage facility operating cost term into the reward function, and couples the on / off state, opening degree, and actual drainage flow of the drainage facilities for calculation. This ensures drainage effectiveness while constraining the intensity of equipment use, enabling the agent to learn to balance drainage effectiveness and operating costs during training, thus achieving a more economical and efficient scheduling strategy. Attached Figure Description

[0025] Figure 1 This is a flowchart of the urban flooding decision-making method according to an embodiment of the present invention; Figure 2 This is a model diagram of an urban flooding decision-making system according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] Example 1: This embodiment provides an urban flooding decision-making method based on agent simulation and evolutionary inference, such as... Figure 1 As shown, it includes the following steps: Step S1: Construct a digital baseline model for urban flooding, dividing the target urban area into multiple spatial grid units, and setting static attributes and dynamic state variables for each spatial grid unit; Among them, the dynamic state variable is the water depth; the static attributes include topographic elevation, land use type and drainage capacity; among them, topographic elevation is used to represent the relative terrain height of each spatial grid cell, which determines the natural flow direction of water and the susceptibility to waterlogging. Land use types include rooftops, roads, green spaces, and water bodies; Drainage capacity is used to represent the maximum amount of drainage that a spatial grid cell can cover per unit time.

[0028] In this embodiment, step S1 specifically includes: First, the target urban area is digitally modeled. By discretizing a continuous space into multiple spatial grid units, a structured representation of the urban flooding process is achieved. In this process, the urban area is divided into grids according to a preset spatial resolution, forming a set of multiple spatial grid units. Each spatial grid unit is denoted as […]. The choice of spatial resolution can be determined based on the trade-off between computational accuracy and computational efficiency. For example, a smaller grid scale can be used for detailed simulation requirements, while the grid scale can be appropriately increased for large-scale, fast simulation scenarios, thereby reducing computational complexity while ensuring simulation accuracy.

[0029] In terms of data acquisition, topographic data, land use data, and drainage facility data for the target urban area were comprehensively acquired. Specifically, digital elevation model (DEM) data was preferred for topographic data to reflect changes in surface elevation; land use data was used to characterize the spatial distribution of different land cover types; and drainage facility data included the drainage network topology, the locations of pumping stations and drainage outlets, and the design drainage capacity parameters of various drainage facilities. This fusion of multi-source data provides fundamental support for subsequent simulations of urban flooding evolution.

[0030] For each spatial grid cell, a static attribute parameter system is further constructed. Topographic data is mapped to each spatial grid cell, and the topographic elevation of that grid cell is determined by calculating the average elevation data within the grid area or selecting the elevation value of the center point. This method can reduce the impact of data noise while maintaining the overall topographic trend, thus more realistically reflecting the gravity-driven direction of water flow. The introduction of topographic elevation allows the model to automatically form water flow paths based on elevation differences, which is beneficial for characterizing the water accumulation and diffusion process.

[0031] By spatially overlaying land use data with spatial grid cells, and selecting the type with the largest area proportion within the grid cell, this approach preserves key surface features while maintaining classification simplicity. This facilitates differentiated modeling for subsequent infiltration capacity, retention capacity, and economic loss assessment. For example, roads and rooftops typically have low infiltration capacity, while green spaces have high infiltration capacity. Water bodies are more likely to be considered catchment areas than damaged areas; this classification provides a basis for setting subsequent model parameters.

[0032] Based on drainage facility data and the spatial distribution of the drainage network, the service area of ​​the drainage facilities corresponding to each spatial grid unit is determined. Then, according to the design drainage capacity parameters of the drainage facilities, the drainage capacity of the spatial grid unit is calculated. This parameter characterizes the maximum amount of water that a grid unit can discharge through the drainage system per unit time. Its introduction reflects the differences in drainage infrastructure conditions in different areas, thus providing realistic constraints for the model during simulation. By introducing drainage capacity, unrealistic excessive drainage phenomena can be avoided during simulation, improving the physical rationality of the model.

[0033] In terms of dynamic modeling, a dynamic state variable that changes with time is defined for each spatial grid cell to represent time intervals. t The water depth is the surface water depth of the grid cell. Water depth, as a core state variable, directly reflects the degree of flooding and serves as the state input for the subsequent agent decision-making model. By using water depth as a unified metric, comparable analysis of flooding risks across different regions can be achieved, while also providing an intuitive basis for optimizing scheduling strategies.

[0034] Through the above processing, a digital baseline model for urban flooding is constructed. This model uses a spatial grid as the basic unit, unifying the expression of static features such as topography, land use, and drainage capacity with the dynamic evolution of water depth, thereby achieving a fundamental characterization of the formation and development of urban flooding. Compared to traditional continuous hydrodynamic models, this modeling approach significantly reduces computational complexity while ensuring the representation of key physical mechanisms, and provides a clear, structured state-space representation for subsequent drainage facility regulation and agent decision-making, which is beneficial for achieving efficient simulation and decision optimization in large-scale urban areas.

[0035] Step S2: Establish a proxy model for drainage facilities in the digital infrastructure model for urban flooding; In this embodiment, step S2 specifically includes: Based on the existing digital infrastructure model of urban flooding, an abstract model of the urban drainage system is created, transforming the complex actual drainage facility system into a drainage facility proxy model that can participate in calculation and decision-making. In this way, drainage behavior can be embedded into the gridded urban flooding evolution process while preserving physical meaning, thereby achieving unified modeling of drainage regulation and water accumulation evolution.

[0036] The drainage facilities in the target city are abstracted as facility proxy points set in spatial grid cells, and each drainage facility is denoted as . ;in, Indicates the drainage facility number; For each drainage facility Record the corresponding connected spatial grid cells, including the entrance spatial grid cell. and export space grid unit The inlet space grid cell is used to represent the pumping location of the drainage facility, and the outlet space grid cell is used to represent the drainage location of the drainage facility. For each drainage facility The operating status is set, including switch status and opening degree. The switch status indicates whether the drainage facility is in operation, and the opening degree indicates the degree of adjustment of the drainage facility in operation. The opening degree is denoted as... ,and ; For each drainage facility Set physical capacity parameters, including maximum design drainage flow. Water level at facility inlet and facility outlet water level ; When drainage facilities When the switch is in the "on" state, the drainage facility At any moment Actual drainage flow The calculation formula is: in, Indicates drainage facilities Maximum design drainage flow rate; Indicates the drainage facilities at any time The degree of opening; Indicates the entrance space grid cell The water level of the accumulated water; Represents the exit space grid cell water level; The head influence coefficient is obtained by fitting the characteristic curve of the drainage facility. When drainage facilities When the switch is in the off state, the actual drainage flow rate of the drainage facility .

[0037] This calculation method comprehensively considers the impact of equipment adjustment capacity and hydraulic conditions on drainage effect, and has clear physical significance. This term describes the suppressive effect of head on drainage capacity. When the outlet water level is higher than the inlet water level, drainage needs to overcome a certain head difference, and the actual drainage capacity will be limited. When the outlet water level is lower than or equal to the inlet water level, this term is 0 and does not affect the drainage capacity.

[0038] The purpose of introducing the max function is to ensure that the suppression effect only occurs under unfavorable head conditions, thereby avoiding a non-physically meaningful "negative enhancement" situation; parameters The head influence coefficient, obtained by fitting the performance curve of drainage facilities, reflects the sensitivity of different types of drainage equipment to head difference. Introducing this term allows the model to more realistically reflect the phenomenon of "high water level suppressing drainage capacity" during actual drainage processes, thereby improving the reliability of the simulation results.

[0039] Step S3: Construct an intelligent agent decision-making model. This model is designed using a reinforcement learning framework and consists of a policy network and a value network, used to achieve adaptive control of drainage facilities in dynamic environments. By introducing a dual-network structure of policy and value networks, training stability and convergence efficiency can be improved while ensuring decision-making ability. The policy network generates control actions for the drainage facilities based on the current environmental state, while the value network evaluates the long-term benefits under the current state and policy, thus providing direction for policy optimization.

[0040] The state input of the agent decision-making model is represented as: in, Indicates time Status input; Indicates time Spatiotemporal grid cells The depth of surface water accumulation; Indicates the total number of spatial grid cells; The control actions of the agent decision-making model are represented as follows: in, Indicates time A set of control actions; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The opening degree is used to characterize the degree of adjustment of the drainage facility when it is in the open state; Indicates the total number of drainage facilities; Policy networks are used to adjust based on state inputs. Generate control actions The value network is used to evaluate the cumulative reward of control actions in the current state and guide the training of the agent's decision-making model.

[0041] Step S4: Construct a training dataset for the agent decision-making model based on historical flooding data of the target city, specifically including: First, historical flooding data for the target city is obtained. This data includes historical rainfall data and corresponding water accumulation data for each rainfall event. Water accumulation data can be obtained through water level sensors, monitoring equipment deployed in key urban areas, or remote sensing inversion. By incorporating real-world observation data, the model training can more closely reflect actual conditions, thereby improving the reliability of subsequent decision-making.

[0042] During data preprocessing, historical rainfall data and waterlogging status data are time-aligned to establish a correspondence between them at a unified time scale. Simultaneously, combined with the spatial grid partitioning results, rainfall data is mapped to each spatial grid cell, transforming the raw waterlogging observation data into the initial waterlogging status data for the corresponding grid cell. This step converts discrete observation data into a structured model input format, enabling it to directly participate in simulation and training processes.

[0043] Based on time-aligned rainfall data, a continuous time-series rainfall data sequence is constructed to drive the urban flooding evolution model. Simultaneously, the waterlogging state at corresponding moments is used as initial conditions to generate initial waterlogging state data matching the rainfall sequence. This approach introduces real historical states at the simulation starting point, thereby improving the accuracy of the initial simulation phase and avoiding error accumulation caused by biased initial conditions.

[0044] Building upon this foundation, to enhance the model's adaptability to different rainfall scenarios, historical data is perturbed and expanded. Specifically, this can be achieved by scaling rainfall intensity proportionally, stretching or compressing rainfall duration, and randomly perturbing the spatial distribution of rainfall, thereby generating multiple sets of differentiated rainfall scenario data. This expansion method effectively compensates for the limited number of historical samples, allowing the training data to cover a wider range of possible scenarios, thus enhancing the model's generalization ability.

[0045] Finally, the initial waterlogging status data, time-series rainfall data, and extended rainfall scenario data are combined to construct a training dataset containing various urban flooding evolution processes. This training dataset not only includes real historical scenarios but also various simulated extended scenarios, allowing the agent to be exposed to richer environmental change patterns during training, thereby learning more robust drainage control strategies.

[0046] Step S5: Based on the initial water accumulation state data in the training dataset, initialize the dynamic state variables of each spatial grid cell, specifically including: The initial water accumulation data corresponding to the current rainfall scenario in the training dataset are mapped to a set of spatial grid cells, and each spatial grid cell is assigned an initial water accumulation depth value. This method allows the model to reflect the actual water accumulation distribution characteristics from historical observations at the initial stage, avoiding bias problems caused by starting from ideal dry conditions or uniform initial conditions. Compared to simply setting a uniform initial water depth, this initialization method can more realistically depict the spatial differences in the city before or during rainfall, helping to improve the accuracy of subsequent simulations.

[0047] In the specific mapping process, based on the relationship between the sensor observation point location and the spatial grid, discrete observation data can be extended to all spatial grid cells using spatial interpolation methods (such as inverse distance weighted interpolation or Kriging interpolation), so that each grid cell has a corresponding initial water depth value. For areas lacking direct observation data, reasonable estimates can be made by combining the terrain elevation, land use type, and water accumulation conditions of surrounding grid cells, thereby ensuring the spatial continuity and rationality of the initialization results.

[0048] Step S6: The agent decision-making model uses the dynamic state variables of each spatial grid cell as state input, generates control actions for the drainage facility based on the state input, and applies the control actions to the drainage facility proxy model to adjust the operating state of the drainage facility. Specifically, this includes: At any moment tThe water depth of all spatial grid cells is used as the state input vector, which is then input into the policy network. This state input reflects the spatial distribution of water accumulation within the current urban area, with the degree of water accumulation, aggregation trends, and potential diffusion paths in different areas all implicit in this state representation. By using the global grid state as input, the agent can comprehensively consider the mutual influence between different areas during the decision-making process, avoiding the overall adverse effects of local optimal decisions, thereby improving the global coordination of the control strategy.

[0049] After extracting features and mapping calculations from the state inputs, the policy network outputs a set of control actions for each drainage facility. For each drainage facility, the control actions include an on / off state and an opening degree. The on / off state determines whether the drainage facility is put into operation, while the opening degree adjusts the drainage intensity. By simultaneously outputting discrete and continuous variables, joint control of "whether to operate" and "the degree of operation" can be achieved, making the scheduling strategy more flexible and refined.

[0050] During action generation, constraints are applied to the policy network output to ensure the rationality of the control results. For example, the opening degree is limited to the range of 0 to 1, which can be normalized using an activation function; the switching state can be determined using a threshold function or probability sampling, thereby ensuring that the action space meets the actual control requirements. This constraint mechanism can prevent the generation of control commands that exceed the physical limits, improving the executability of the model in practical applications.

[0051] Furthermore, by uniformly modeling and centrally deciding on all drainage facilities, the operational relationships between different facilities can be coordinated globally. This prevents multiple drainage facilities from over-pumping the same area or interfering with each other, thus achieving a more rational system-level scheduling effect. This step provides crucial control input for subsequent urban flooding evolution simulation and reward feedback, and is a core component of achieving intelligent optimization decision-making.

[0052] Step S7: Based on the operating status of the drainage facility proxy model, the static attributes of each spatial grid unit and the time-series rainfall data, update the dynamic state variables in the urban waterlogging digital base model to simulate the evolution process of urban waterlogging under the action of control. In step S7, after the agent generates drainage control actions and applies them to the drainage facility proxy model, the dynamic state variables in the urban flooding digital base model are updated based on the operating status of the drainage facilities, the static attributes of the spatial grid cells, and time-series rainfall data. This achieves a dynamic evolution simulation of urban flooding under control. Essentially, this step constructs a water balance process that takes into account rainfall input, surface runoff, infiltration loss, and artificial drainage regulation, allowing the evolution of flooding to progress gradually within a physically reasonable framework.

[0053] At any moment to Within the time step, for each spatial grid unit in the digital baseline model of urban flooding dynamic state variables The update is performed using the following formula: in, Represents the updated spatial grid cell Dynamic state variables; Indicates the time step; Indicates time Acting on spatial grid cells The rainfall intensity was obtained based on time-series rainfall data; Represents spatial grid cells The amount of infiltration or retention loss; Representation and spatial grid cells Adjacent grid cell sets; Indicates from spatial grid cells Flow to adjacent grid cells Surface runoff; Indicates spatial grid cells The drainage flux discharged through drainage facilities; This indicates that water flows into the spatial grid cell through the drainage system. The drainage flux; Infiltration or retention loss Based on the land use type in the static attributes of each spatial grid unit Determined, used to describe the reduction in water volume caused by some rainfall or accumulated water seeping into the ground or being retained on the surface, is expressed as: in, Indicates land use type The corresponding infiltration coefficient is obtained by calibration based on historical observation data or empirical parameters; Surface runoff Determined based on the terrain elevation difference between spatial grid cells, used to characterize surface runoff processes, and represented as follows: in, , Representing spatial grid cells and Terrain elevation in static attributes; The coefficient of water exchange between spatial grid cells is determined by considering the distance between the spatial grid cells, their connectivity, and the surface roughness parameters. This expression significantly simplifies the traditional hydrodynamic equations while maintaining physical meaning, reducing computational complexity and making it suitable for large-scale grid computing scenarios.

[0054] Drainage flux depends on the connection to the spatial grid cell The set of drainage facilities is defined as follows: in, Indicates drainage facilities The entrance space grid unit; Indicates drainage facilities The exit space grid cell; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The actual drainage flow rate.

[0055] By continuously iterating through the above update process, the evolutionary trajectory of urban flooding can be continuously obtained over time, and the control actions of the intelligent agent can be responded to at each time step, thus forming a closed-loop mechanism of "decision-evolution-feedback". This mechanism enables drainage control strategies to directly influence the development process of urban flooding and to be continuously optimized through subsequent reward feedback, providing key support for the intelligent control of urban flooding.

[0056] Step S8: Calculate the reward value based on the updated dynamic state variables, and train the agent decision-making model based on the reward value, specifically including: At every moment The reward value is calculated based on the updated dynamic state variables of each spatial grid cell. ; Based on reward value Calculate the cumulative return from the current moment. , represented as: in, Indicates from time Initial cumulative returns; Indicates the discount factor. Indicates the termination time of the simulation process; Reward Value The formula is: in, Indicates time The reward value; These are the weighting coefficients; Indicates time Spatial grid unit The dynamic state variable, namely the water depth; Represents spatial grid cells At any moment An indicator function for whether water accumulation has exceeded a threshold, when The value is 1 if the condition is met, and 0 otherwise. Represents spatial grid cells The preset water depth threshold; Represents spatial grid cells Land use types; Indicates land use type The corresponding economic loss weighting coefficient is used to characterize the potential economic loss of different land use types when waterlogging occurs; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The degree of opening; Indicates drainage facilities At any moment The actual drainage flow rate; Indicates the total number of drainage facilities; This indicates the total number of spatial grid cells.

[0057] First item To characterize the persistent impact of waterlogging, a weighting coefficient related to land use type is introduced. This can reflect the differences in economic losses caused by flooding in different areas. For example, commercial areas or main traffic arteries have higher weights, while green spaces or water bodies have relatively lower weights. This design allows the model to prioritize reducing flooding in high-value areas during optimization, thereby achieving a control objective oriented towards minimizing economic losses. (Second item) Used to characterize the risk of severe flooding, this term acts as an additional penalty for a "dangerous state," emphasizing control over extreme situations rather than continuous water depth. Introducing this term encourages agents to prioritize eliminating or avoiding severely flooded areas during decision-making, thereby improving overall safety and preventing catastrophic consequences in localized areas. (Third term) This term describes the operating cost or energy consumption of drainage facilities. It couples drainage intensity with operating costs in its modeling, meaning that the stronger the drainage behavior, the higher the corresponding "cost." By introducing this term, we can prevent agents from adopting an "unrestricted drainage" strategy, thereby reducing water accumulation while taking into account energy consumption and equipment operating burden, achieving a balance between control effectiveness and operating costs.

[0058] The value network of the agent decision-making model is based on the state input. Estimate the corresponding state value and based on the cumulative reward. Update the value network parameters; The policy network of the agent decision-making model is based on the evaluation results of the value network, and aims to maximize the cumulative reward. Update the strategy network parameters and optimize the control strategy for drainage facilities.

[0059] Through continuous iteration of the training process described above, the agent can accumulate experience under various rainfall scenarios and urban flooding evolution conditions, gradually forming a comprehensive optimization strategy that takes into account waterlogging control effectiveness, risk prevention, and operating costs. This method not only improves the adaptive capability of decision-making but also enables differentiated regulation based on the characteristics of different urban areas, demonstrating strong practical application value.

[0060] Step S9: Implement urban flooding decision-making based on the trained agent decision-making model, specifically including: In practice, under actual or predicted rainfall scenarios, the water accumulation status information of each spatial grid unit within the current urban area is first obtained. Combined with real-time or predicted rainfall data, the state of the urban flooding digital baseline model is updated to construct the current state input S(t). This state input reflects the spatial distribution characteristics and development trend of current urban flooding, providing a basis for subsequent decision-making.

[0061] The state input S(t) is fed into the trained policy network, which directly outputs the control actions for each drainage facility, including the on / off state and opening parameters. Since the policy network has been trained using extensive historical data and extended scenarios, it can quickly generate near-optimal control strategies in complex environments, thus avoiding the time delay caused by iterative calculations in traditional optimization methods and meeting real-time decision-making requirements.

[0062] The generated control actions are then applied to the actual drainage facilities or dispatching system to regulate the operating status of pumping stations, drainage outlets, and related facilities. For example, the start and stop of drainage facilities can be controlled based on the output switch status, and the drainage flow rate or valve opening can be adjusted based on the opening parameters, thereby achieving dynamic allocation of drainage capacity. In this way, drainage resources can be rationally allocated among different areas of the city, prioritizing the alleviation of waterlogging problems in high-risk or high-value areas.

[0063] Example 2: This embodiment provides an urban flooding decision-making system based on agent simulation and evolutionary inference, such as... Figure 2 As shown, it includes an urban flooding digital infrastructure module, a drainage facility agent module, an intelligent agent decision-making module, a training data processing module, and a control execution module.

[0064] The Urban Flood Digital Base Module divides the city into spatial grid units, setting static attributes (topographic elevation, land use type, drainage capacity) and dynamic state variables (water depth) for each unit to simulate the evolution of water accumulation under rainfall.

[0065] The drainage facility agent module abstractly represents facilities such as urban pumping stations and drainage outlets. It has on / off status, adjustable opening degree and physical capability parameters. It can calculate the actual drainage flow based on the control actions generated by the intelligent agent and feed it back to the digital base module to regulate water accumulation.

[0066] The agent decision-making module includes a policy network and a value network. The policy network generates drainage control actions based on the dynamic state of the grid cells, while the value network evaluates the cumulative rewards and guides policy optimization, thereby achieving global dynamic intelligent regulation.

[0067] The training data processing module is based on historical rainfall and water accumulation data, and generates various rainfall scenarios through perturbation expansion to construct the intelligent agent training dataset.

[0068] The control execution module issues control actions to the intelligent agent and receives feedback to form a closed-loop control, enabling real-time or near-real-time regulation of urban flooding.

[0069] The system can comprehensively consider the economic losses of different land uses, drainage facility capacity and operating costs, to achieve intelligent prevention and optimized scheduling of urban flooding.

[0070] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0071] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for urban flooding decision-making based on intelligent agent simulation and evolutionary inference, characterized in that, Includes the following steps: A digital baseline model for urban flooding is constructed, dividing the target urban area into multiple spatial grid units, and setting static attributes and dynamic state variables for each spatial grid unit; the dynamic state variable is the water depth. A proxy model for drainage facilities is established within the aforementioned digital infrastructure model for urban flooding. Construct an agent decision-making model; A training dataset for an intelligent agent decision-making model is constructed based on historical flooding data of the target city; the training dataset includes initial flooding status data and corresponding time-series rainfall data. Based on the initial water accumulation state data in the training dataset, the dynamic state variables of each spatial grid cell are initialized; The intelligent agent decision model uses the dynamic state variables of each spatial grid cell as state input, generates control actions for the drainage facility based on the state input, and applies the control actions to the drainage facility agent model to adjust the operating state of the drainage facility. Based on the operating status of the drainage facility proxy model, the static attributes of each spatial grid unit and the time-series rainfall data, the dynamic state variables in the urban waterlogging digital base model are updated to simulate the evolution process of urban waterlogging under the control action. The reward value is calculated based on the updated dynamic state variables, and the agent decision-making model is trained based on the reward value; Urban flooding decision-making is achieved based on a trained agent decision-making model.

2. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The static attributes include topographic elevation, land use type, and drainage capacity; wherein, the topographic elevation is used to represent the relative terrain height of each spatial grid cell, and determines the natural flow direction of water and the susceptibility to water accumulation and flooding. The land use types include rooftops, roads, green spaces, and water bodies; The drainage capacity is used to represent the maximum drainage volume of the drainage facilities covered by the spatial grid cell per unit time.

3. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The construction of the urban flooding digital infrastructure model involves dividing the target urban area into multiple spatial grid units and setting static attributes and dynamic state variables for each spatial grid unit, specifically including: Acquire topographic data, land use data, and drainage facility data for the target urban area. Then, based on a preset spatial resolution, perform grid-based discretization on the target urban area, generating a set of spatial grid cells, each denoted as . The terrain data includes digital elevation model data of the target urban area; the land use data includes spatial distribution data of roofs, roads, green spaces, water bodies and other surface cover types in the target urban area; the drainage facility data includes drainage network data, pump station location data, drainage outlet location data and corresponding drainage facility design drainage capacity parameters. Set corresponding static properties for each spatial grid cell, including the following steps: The terrain data is mapped to each spatial grid cell, and the terrain elevation of the corresponding spatial grid cell is determined by calculating the average value or the center point elevation value of the terrain elevation data within the spatial grid cell. ; The land use data is spatially overlaid with the spatial grid cells. Based on the area proportion of each land use type in the spatial grid cells, the land use type with the largest area proportion is selected as the land use type of the spatial grid cell. ; The service area of ​​the drainage facilities corresponding to each spatial grid unit is determined based on the drainage facility data, and the drainage capacity of the corresponding spatial grid unit is determined based on the design drainage capacity of the drainage facilities. ; Set dynamic state variables for each spatial grid cell Dynamic state variables Indicates time Spatiotemporal grid cells The depth of surface water accumulation was used to obtain a digital baseline model of urban flooding.

4. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The drainage facility proxy model is used to represent the drainage facilities corresponding to each spatial grid unit of the target city. Each drainage facility has an adjustable operating status and physical capabilities.

5. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The establishment of a drainage facility proxy model in the urban flooding digital base model specifically includes: The drainage facilities in the target city are abstracted as facility proxy points set in spatial grid cells, and each drainage facility is denoted as . ;in, Indicates the drainage facility number; For each drainage facility Record the corresponding connected spatial grid cells, including the entrance spatial grid cell. and export space grid unit The inlet space grid cell is used to represent the pumping location of the drainage facility, and the outlet space grid cell is used to represent the drainage location of the drainage facility. For each drainage facility The operating status is set, which includes a switch status and an opening degree. The switch status indicates whether the drainage facility is in operation, and the opening degree indicates the degree of adjustment of the drainage facility in operation. The opening degree is denoted as... ,and ; For each drainage facility Set physical capacity parameters, including maximum design drainage flow. Water level at facility inlet and facility outlet water level ; When the drainage facility When the switch is in the "on" state, the drainage facility At any moment Actual drainage flow The calculation formula is: in, Indicates drainage facilities Maximum design drainage flow rate; Indicates the drainage facilities at any time The degree of opening; Indicates the entrance space grid cell The water level of the accumulated water; Represents the exit space grid cell water level; The head influence coefficient is obtained by fitting the characteristic curve of the drainage facility. When the drainage facility When the switch is in the off state, the actual drainage flow rate of the drainage facility .

6. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The training dataset for constructing the agent decision-making model based on historical flooding data of the target city specifically includes: Historical flooding data of the target city is obtained, including historical rainfall data and water accumulation status data corresponding to the rainfall data; the water accumulation status data is acquired through sensors. The historical rainfall data and water accumulation status data are time-aligned and spatially mapped to map the historical rainfall data to each spatial grid cell, and the water accumulation status data is converted into the initial water accumulation status data of the corresponding spatial grid cell. Time-series rainfall data is constructed based on the time-aligned historical rainfall data, and corresponding initial water accumulation status data is constructed based on the water accumulation status data. Based on the historical flooding data, multiple sets of extended data for different rainfall scenarios are generated by perturbating and expanding the rainfall intensity, duration, and spatial distribution. The initial waterlogging status data, the time-series rainfall data, and the extended data are combined to construct a training dataset containing various urban flooding scenarios.

7. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The agent decision-making model includes a policy network and a value network; The state input of the agent decision-making model is represented as follows: in, Indicates time Status input; Indicates time Spatiotemporal grid cells The depth of surface water accumulation; Indicates the total number of spatial grid cells; The control actions of the agent decision-making model are represented as follows: in, Indicates time A set of control actions; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The opening degree is used to characterize the degree of adjustment of the drainage facility when it is in the open state; Indicates the total number of drainage facilities; The policy network is used to adjust based on state input. Generate control actions The value network is used to evaluate the cumulative reward of the control action in the current state and guide the training of the agent's decision-making model.

8. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The process of updating the dynamic state variables in the urban flooding digital infrastructure model based on the operational status of the drainage facility proxy model, the static attributes of each spatial grid unit, and time-series rainfall data specifically includes: At any moment to Within the time step, for each spatial grid unit in the digital baseline model of urban flooding dynamic state variables The update is performed using the following formula: in, Represents the updated spatial grid cell Dynamic state variables; Indicates the time step; Indicates time Acting on spatial grid cells The rainfall intensity was obtained based on time-series rainfall data; Represents spatial grid cells The amount of infiltration or retention loss; Representation and spatial grid cells Adjacent grid cell sets; Indicates from spatial grid cells Flow to adjacent grid cells Surface runoff; Indicates spatial grid cells The drainage flux discharged through drainage facilities; This indicates that water flows into the spatial grid cell through the drainage system. The drainage flux; The amount of infiltration or retention loss Based on the land use type in the static attributes of each spatial grid unit Confirmed, indicated as: in, Indicates land use type The corresponding infiltration coefficient is obtained by calibration based on historical observation data or empirical parameters; The surface runoff Determined based on the terrain elevation difference between spatial grid cells, expressed as: in, , Representing spatial grid cells and Terrain elevation in static attributes; The water exchange coefficient between spatial grid cells is determined by a combination of the distance between spatial grid cells, the connection relationship between them, and the surface roughness parameters. The drainage flux is based on the connection to the spatial grid unit. The set of drainage facilities is defined as follows: in, Indicates drainage facilities The entrance space grid unit; Indicates drainage facilities The exit space grid cell; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The actual drainage flow rate.

9. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 1, characterized in that, The training process for the intelligent agent decision-making model includes: At every moment The reward value is calculated based on the updated dynamic state variables of each spatial grid cell. ; Based on the reward value Calculate the cumulative return from the current moment. , represented as: in, Indicates from time Initial cumulative returns; Indicates the discount factor. Indicates the termination time of the simulation process; The value network of the agent decision-making model is based on the state input. Estimate the corresponding state value, and based on the cumulative reward. Update the value network parameters; The policy network of the agent decision-making model, based on the evaluation results of the value network, maximizes the cumulative reward. Update the strategy network parameters and optimize the control strategy for drainage facilities.

10. The urban flooding decision-making method based on agent simulation and evolutionary inference according to claim 9, characterized in that, The reward value The formula is: in, Indicates time The reward value; These are the weighting coefficients; Indicates time Spatial grid unit The dynamic state variable, namely the water depth; Represents spatial grid cells At any moment An indicator function for whether water accumulation has exceeded a threshold, when The value is 1 if the condition is met, and 0 otherwise. Represents spatial grid cells The preset water depth threshold; Represents spatial grid cells Land use types; Indicates land use type The corresponding economic loss weighting coefficient is used to characterize the potential economic loss of different land use types when waterlogging occurs; Indicates drainage facilities At any moment The on / off state; Indicates drainage facilities At any moment The degree of opening; Indicates drainage facilities At any moment The actual drainage flow rate; Indicates the total number of drainage facilities; This indicates the total number of spatial grid cells.

Images