A real-time decision-making method for typhoon prevention and emergency evacuation in mountainous villages and towns

Abstract

The present invention discloses a real-time decision-making method for typhoon prevention and emergency evacuation in mountainous villages and towns, including: establishing a probabilistic rainfall forecast: obtaining rainfall forecast samples by sampling method, and inputting the rainfall forecast samples into a physical simulation model for dynamic forecasting of random scenarios; establishing an evacuation decision-making unit: rasterizing the research area in space, and based on the raster, conducting spatial clustering on the research area to divide blocks and obtain the evacuation decision-making unit; conducting risk forecasting on the evacuation decision-making unit: calculating the injury level that residents will face if they are not evacuated, and integrating the injury level that residents will face on the evacuation decision-making unit to obtain the comprehensive risk index forecast of the evacuation decision-making unit, and conducting risk forecasting for all random scenarios respectively; establishing a multi-stage stochastic decision-making model and making decisions based on uncertain risks; rolling decision-making forecast. The present invention makes refined decisions for small-scale evacuations and seeks optimization between over-prevention and untimely evacuation.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency management, and particularly relates to a real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns. Background Art

[0002] Most coastal areas are mountainous and hilly, and typhoons are likely to occur in summer. Different from sudden disasters such as earthquakes, tsunamis, and landslides, a typhoon disaster usually takes several days from formation, approach, landing to decline. During this period, a reasonable and effective method for evacuating emergency personnel can well ensure the safety of people's lives and social stability. However, the existing research on typhoon emergency evacuation decision-making lacks systematic modeling from the disaster end to the decision end, and the uncertainty of weather forecasts and the complexity of mountainous terrain are intertwined and superimposed, exacerbating the difficulty of residents' emergency evacuation decision-making.

[0003] From the characteristics of mountain emergency evacuation command itself: firstly, during the evolution of typhoons, the trajectories and rainfall intensities of uncertain forecasts develop dynamically, with strong forecast randomness, and the necessity of evacuation in each region changes in real time; secondly, under the action of complex terrain, the rainfall field will become fragmented, and the spatial differences in disaster situations are large, which means that the evacuation within mountainous villages and towns is usually limited to within the villages and towns, without the need for a town-wide evacuation in one piece. The spatial differences in risks require a high level of decision-making fineness and are not suitable for following the design idea of large-scale evacuation. Therefore, the specific emergency evacuation command faces two key problems: "uncertain evacuation timing" and "unclear evacuation areas".

[0004] In existing research results, the decision-making optimization for emergency evacuation usually focuses on two key elements: the optimization of evacuation routes and the evacuation start time. Regarding the optimization of evacuation routes, by establishing network flow models, dynamic traffic assignment models, or agent models, etc., the time taken for evacuation clearance is analyzed to optimize the transfer routes of individuals (or families as a unit), aiming to transfer the affected people more efficiently and safely within a limited time. However, this type of work implies the assumption of mandatory evacuation. Therefore, it only focuses on how to effectively clear the dangerous area, but does not discuss whether it is necessary to evacuate a certain area from the perspective of the uncertainty of disaster occurrence. This problem is particularly prominent in the context of disasters with significant spatial differences. For the optimization of the evacuation start time, it can be divided into two types of research methods: one is to make deterministic dynamic decisions. This method establishes a decision model based on preset or re-analyzed historical known information and lacks the ability to flexibly respond to random problems. Therefore, it is usually only used for large-scale (areas above hundreds of square kilometers) evacuation guidance in the initial stage of disasters; the other is to make single-stage or two-stage transfer decisions. Although it considers multiple scenarios, it does not consider the development of randomness therein and cannot make dynamic adjustments to highly variable problems. Typhoon disasters generally have an evolution process of 2 to 3 days. Although the two-stage decision-making method can quickly evaluate the potential risks and benefits under different decision paths, it obviously ignores the change of the necessity of transfer during the disaster process. Summary of the Invention

[0005] In order to effectively consider the uncertainty faced by transfer decisions during the dynamic evolution of typhoons, and to make refined decisions on small-scale (areas of dozens of square kilometers) transfers in the front line for the spatial heterogeneity problem of mountain risks and serve village and town emergency management, the present invention provides a real-time decision-making method for typhoon prevention and emergency transfer for mountain villages and towns. The multi-stage stochastic decision-making model established by this method seeks optimization between over-prevention and untimely transfer, and directly provides real-time, multi-stage, and refined emergency transfer plans for management decision-makers.

[0006] Term Explanation:

[0007] 1. SCS-CN: Soil Conservation Service Curve Number, a runoff coefficient curve method confluence model.

[0008] 2. LisFlood-FP: A semi-open source two-dimensional hydrodynamic model.

[0009] The technical solution adopted by the present invention to overcome its technical problems is:

[0010] A real-time decision-making method for typhoon prevention and emergency transfer for mountain villages and towns, including the following steps:

[0011] S1. Establish probabilistic rainfall forecasts: Obtain rainfall forecast samples through Monte Carlo sampling, and input the rainfall forecast samples into a physical simulation model for dynamic forecasting of random scenarios;

[0012] S2. Establish transfer decision units: Grid the study area spatially, and based on the grids, conduct spatial clustering of the study area to divide it into blocks, obtaining the smallest single entity for transfer decision-making, that is, the transfer decision unit;

[0013] S3. Conduct risk forecasts for transfer decision units: Calculate the injury level that residents will face if they do not transfer, and integrate the injury level that residents will face on the transfer decision units to obtain the comprehensive risk index forecast of the transfer decision units, and conduct risk forecasts for all random scenarios respectively;

[0014] S4. Establish a multi-stage stochastic decision-making model and make decisions based on the uncertain risks obtained in step S3;

[0015] S5. Rolling decision forecasting.

[0016] Furthermore, step S1 includes:

[0017] S11. Establish a stochastic error model: Collect the rainfall forecast data officially released during historical typhoons in the study area and the meteorological station observation data. Using the original forecast value as the independent variable, conduct regression on the mean and variance of the error, then normalize the error under different forecast values based on the mean and variance obtained from the regression, and then fit the processed error distribution to establish a stochastic error model;

[0018] S12. Generate rainfall forecast samples: Based on the deterministic rainfall forecast, extract error samples from the stochastic error model in step S11 through Monte Carlo sampling, and combine the original forecast samples and error samples to obtain several groups of rainfall forecast samples;

[0019] S13. Conduct flash flood forecasts through a physical simulation model: The physical simulation model includes the SCS-CN model and the LisFlood-FP model. Estimate surface runoff based on the SCS-CN model and simulate the dynamic flood process in combination with the LisFlood-FP model. Drive the SCS-CN model and the LisFlood-FP model respectively with the several groups of rainfall forecast samples obtained in step S12 to obtain a set of random scenarios represented by the dynamic water depth of the grids.

[0020] Furthermore, step S2 includes:

[0021] S21. Without considering the village administrative boundaries, rasterize the study area spatially based on a preset resolution. Set traffic control nodes to reflect the surrounding road construction in the study area. Take the nearest road topological node that any raster can reach as the affiliated traffic control node, and divide the study area into several regions according to the traffic control nodes. The regions controlled by different traffic control nodes are studied independently;

[0022] S22. Primary clustering: Divide the study area into two categories, namely valley areas and interfluves, according to the elevation of the raster and the distance from the river;

[0023] S23. Secondary clustering: For the valley area, cluster using four indicators: the slope at the raster location, the roughness at the raster location, the relative height difference from the nearest river channel, and the cross-section slope of the nearest river channel; for the interfluve, cluster using four indicators: the elevation at the raster location, the slope at the raster location, the curvature at the raster location, and the historical landslide density;

[0024] S24. When there is a village administrative boundary passing through the block obtained after secondary clustering, post-processing is performed.

[0025] Furthermore, in step S24, the post-processing specifically refers to: re-segmenting the block after clustering in step S23 according to the village administrative boundary to ensure clear administrative jurisdiction.

[0026] Furthermore, step S3 includes:

[0027] S31. Based on the building function vulnerability curve, calculate the injury level that residents will face if they are not evacuated through the maximum flood depth in the next few hours;

[0028] S32. Integrate the injury level that residents will face obtained in step S31 on the evacuation decision unit to obtain the comprehensive risk index forecast of the evacuation decision unit;

[0029] S33. Conduct risk forecasts respectively based on all the random scenarios obtained in step S1.

[0030] Furthermore, step S3 specifically includes:

[0031] S31. Based on the building function vulnerability curve, calculate the injury level that residents will face if they are not evacuated through the maximum flood depth in the next few hours, which is expressed by the following formula:

[0032] (1)

[0033] (2)

[0034] (3)

[0035] (4)

[0036] In formulas (1)-(4), represents the base of the natural logarithm; represents the current moment; represents the grid water depth at moment represents the grid water depth at moment represents the grid water depth at moment; represents the functional loss rate of the building at different moments; represents the predicted value of the future risk level at moment after considering the command time-consuming in the actual transfer process; represents the building function loss function; represents the maximum immersion depth of the building in the time period predicted by the flood model, represents the preset time interval;

[0037] S32. Integrate the injury levels that residents are about to face obtained in step S31 on the transfer decision unit, that is, weight all the affected buildings inside the transfer decision unit by population to obtain the comprehensive risk index forecast of the transfer decision unit, which is expressed by the following formula:

[0038] (5)

[0039] In formula (5), represents the comprehensive risk index of the th transfer decision unit at moment ; represents the th transfer decision unit, the th building's population; represents the number of flooded buildings inside a certain transfer decision unit; represents the th building's residents' upcoming injury level at moment ; represents the th transfer decision unit's total population;

[0040] S33. Conduct risk forecasts respectively based on all the random scenarios obtained in step S1, which is expressed by wherein, , , represents the sample set of random scenarios.

[0041] Furthermore, step S4 includes:

[0042] S41. Establish a multi-stage stochastic decision-making model:

[0043] (1) Define the decision variables: The multi-stage includes the current stage and several future stages; the decision variables include whether the transfer decision unit transfers in the current stage and several future stages, and the opening status of each shelter corresponding to each stage;

[0044] (2) Design the objective function: Calculate the optimal transfer time of the transfer decision units to be evacuated in the study area from the perspective of maximizing the overall utility, and comprehensively consider the trade-off among the life and property safety that can be protected by the evacuation, the assets abandoned during the evacuation, and the evacuation cost, and select the strategy with the greatest benefit, so as to establish the objective function;

[0045] (3) Design the constraint conditions of the objective function;

[0046] S42. Solve the solution of the multi-stage stochastic decision-making model.

[0047] Furthermore, step S4 specifically includes:

[0048] S41. Establish a multi-stage stochastic decision-making model:

[0049] (1) Define the decision variables: The multi-stage includes the current stage and several future stages; the decision variables include whether the transfer decision unit at the current stage time and several future stages transfers, and the opening status of each shelter corresponding to each stage, which are respectively represented as and two groups of 0-1 binary decision variables, where, , , , represents the number of future stages, represents that the resident stays at home at this time, represents that the resident has left home. For any transfer decision unit , the first time is the immediate transfer time point, represents that the shelter is open, represents that the shelter is not open;

[0050] (2) Design the objective function: Calculate the optimal transfer time of the transfer decision units to be evacuated in the study area from the perspective of maximizing the overall utility, and comprehensively consider the trade-off among the life and property safety that can be protected by the evacuation, the assets abandoned during the evacuation, and the evacuation cost, and select the strategy with the greatest benefit, so as to establish the objective function. The objective function is expressed as follows:

[0051] (6)

[0052] (7)

[0053] (8)

[0054] In formulas (6) - (8), formula (6) represents enabling costs controlled by the number of opened shelters in order to postpone transfer actions to reduce unnecessary production losses and lower resettlement costs; formula (7) represents, in order to reduce the expected cumulative damage within the decision foresight period, represents taking the expected value; represents the loss per person per unit time caused by transfer; represents the fixed cost for enabling a single shelter; represents the expected number of people to be transferred within the decision foresight period; represents the total population living in the

[0055] (3) Design the constraint conditions of the objective function as follows:

[0056] (9)

[0057] (10)

[0058] (11)

[0059] (12)

[0060] (13)

[0061] Formulas (9) and (10) respectively represent the value ranges of the decision variables and ; formulas (11) and (12) mean that once a decision is implemented, the values of the decision variables cannot be reverted to the unimplemented state; in formula (13), represents the maximum capacity of the shelter, and formula (13) means that the number of transferred residents cannot exceed the total capacity of the shelters;

[0062] S42. Solve the solution of the multi - stage stochastic decision - making model:

[0063] (1) The variable is a random variable. Convert the function expectation of in formula (5) into solving and averaging under multiple random samples. At this time, formula (7) is converted into the following expression:

[0064] (14)

[0065] (2) With the minimization of the expected cumulative damage within the decision foresight period as the main objective, the degree of relaxation is controlled based on the ε-constraint algorithm, and ε is adjusted to generate a series of Pareto optimal solutions.

[0066] Furthermore, step S5 includes:

[0067] S51. Based on the current latest rainfall forecast, steps S1, S3, and S4 are sequentially executed. A solution is selected from the Pareto solution set of the multi-stage stochastic decision-making model in step S4 as the transfer plan. This solution includes the locations that need to be immediately transferred in the current stage and the locations that are expected to be transferred in future stages. "Immediate transfer instructions" and "preparatory transfer notices" are issued for the current stage and future stages respectively;

[0068] S52. Step S51 is repeatedly executed based on a preset time interval. During each rolling decision-making process, the previous round of "preparatory transfer notice" is replaced by the current round of "immediate transfer instructions". Each decision-making is a multi-stage rolling decision-making, and each transfer command only corresponds to the current stage in the decision-making solution, and the future stage is a forecast result.

[0069] The beneficial effects of the present invention are:

[0070] 1. By establishing a high-resolution transfer decision-making unit, the present invention can pre-capture the spatial distribution structure of potential hazards, which helps the multi-stage stochastic decision-making model to accurately identify the affected locations; then the flood depths of each building point obtained through physical simulation are integrated into the transfer decision-making unit for risk assessment, so as to balance the fineness and reliability of risk forecasting.

[0071] 2. Focusing on the uncertainty in the dynamic evolution process of disasters, the present invention connects the rolling decision-making through the dual instructions of "immediate transfer" and "preparatory transfer", which helps to promote the dynamic information disclosure between managers and residents and solve practical problems such as difficult command and time-consuming communication.

[0072] 3. The present invention establishes a multi-stage stochastic decision-making model, decomposes the dynamic change process into multiple interrelated decision-making links, thereby decomposing the future uncertainty into each stage, gradually reducing the cumulative effect of decision-making deviation. The deviation in the first stage is small, and the future stage can adjust the strategy in real time according to the updated information. By weighing the retention risk and transfer cost through the multi-stage stochastic decision-making model, the optimal transfer moment of each transfer decision-making unit is optimized, so as to provide a batch-by-batch transfer plan. Description of the Drawings

[0073] Figure 1It is a flowchart of the real-time decision-making method for typhoon prevention and emergency evacuation in mountainous villages and towns according to the embodiments of the present invention.

[0074] Figure 2 It is a curve graph of the first-round probabilistic rainfall forecast according to the embodiments of the present invention.

[0075] Figure 3 It is a curve graph of the second-round probabilistic rainfall forecast according to the embodiments of the present invention.

[0076] Figure 4 It is a curve graph of the third-round probabilistic rainfall forecast according to the embodiments of the present invention.

[0077] Figure 5 It is a curve graph of the fourth-round probabilistic rainfall forecast according to the embodiments of the present invention.

[0078] Figure 6 It is a schematic diagram of the clustering process of the evacuation decision-making unit according to the embodiments of the present invention.

[0079] Figure 7 It is a schematic diagram of the evacuation decision-making according to the embodiments of the present invention.

[0080] Figure 8 It is an effect diagram of the rolling decision-making made during the typhoon evolution process according to the embodiments of the present invention. Detailed implementation manners

[0081] To facilitate better understanding of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following is only exemplary and does not limit the protection scope of the present invention.

[0082] The present invention discloses a real-time decision-making method for typhoon prevention and emergency evacuation in mountainous villages and towns. As Figure 1 shown, it includes the following steps:

[0083] S1. Establish a probabilistic rainfall forecast: Obtain rainfall forecast samples through Monte Carlo sampling, and input the rainfall forecast samples into a physical simulation model for dynamic forecasting of random scenarios;

[0084] S2. Establish an evacuation decision-making unit: Grid the research area spatially, and based on the grids, spatially cluster and divide the research area into blocks to obtain the smallest single entity for evacuation decision-making, that is, the evacuation decision-making unit;

[0085] S3. Conduct risk forecasting for the evacuation decision-making unit: Calculate the injury level that residents will face if they do not evacuate, and integrate the injury level that residents will face on the evacuation decision-making unit to obtain the comprehensive risk index forecast of the evacuation decision-making unit, and conduct risk forecasting for all random scenarios respectively;

[0086] S4. Establish a multi-stage stochastic decision-making model and make decisions based on the uncertainty risks obtained in step S3;

[0087] S5. Rolling decision prediction.

[0088] To better understand the above technical solution, the exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings. These are only the exemplary embodiments of the present invention. However, it should be understood that the present invention can be implemented in various forms and is not limited to the embodiments described herein. These embodiments are for those skilled in the art to understand the present invention more clearly and thoroughly.

[0089] In this embodiment, a certain town (designated as Town B, which has more mountainous terrain) in a certain coastal province (designated as Province A) is taken as an example, and a typhoon (designated as Typhoon C) that occurred in this province in 2019 is used as the disaster background to illustrate a real-time decision-making method for typhoon prevention and emergency transfer for mountainous villages and towns described in the present invention. It should be noted that the maps involved in the drawings of this embodiment are all virtual maps, not real existing areas, and are only for more intuitively illustrating and embodying the real-time decision-making method for typhoon prevention and emergency transfer for mountainous villages and towns described in this embodiment.

[0090] S1. Establish a probabilistic rainfall forecast: Obtain rainfall forecast samples through Monte Carlo sampling, and input the rainfall forecast samples into the physical simulation model for dynamic forecasting of random scenarios.

[0091] In this embodiment, in order to quantify the magnitude of uncertainty faced by the decision-making, it is first necessary to simulate the uncertain disaster field, which specifically includes the following steps:

[0092] S11. Establish a random error model: Collect the rainfall forecast data and meteorological station observation data released by the local meteorological bureau during Typhoon C in the study area (Town B). Using the original forecast value as the independent variable, perform regression on the mean and variance of the error, then normalize the error under different forecast values based on the mean and variance obtained from the regression, and then fit the error distribution obtained after processing to establish a random error model.

[0093] S12. Generate rainfall forecast samples: Based on the deterministic rainfall forecast, extract error samples from the random error model in step S11 through Monte Carlo method, and combine the original forecast samples and error samples to obtain several groups of rainfall forecast samples. In this embodiment, taking obtaining 2000 groups of rainfall forecast samples as an example, as Figures 2 - 5 shown, is the curve graph of the four-round probabilistic rainfall forecast obtained in this embodiment, Figures 2 - 5 respectively representing the curve graphs of the first-round, second-round, third-round, and fourth-round probabilistic rainfall forecasts.

[0094] S13. Conduct flash flood forecasting through physical simulation models: The physical simulation models include the SCS-CN model and the LisFlood-FP model. Estimate surface runoff based on the SCS-CN model and simulate the dynamic flood process in combination with the LisFlood-FP model. Drive the SCS-CN model and the LisFlood-FP model respectively with the 2000 groups of rainfall forecast samples obtained through step S12, so as to obtain a set of random scenarios represented by the dynamic water depth of the grid, where the dynamics here use one minute as the time resolution.

[0095] S2. Establish transfer decision-making units: Grid the B town in the study area spatially, and conduct spatial clustering of the study area based on the grid to divide it into blocks, obtaining the smallest single entity for transfer decision-making, that is, the transfer decision-making unit.

[0096] Due to the significant differences in the distribution of mountain disasters, the transfer decisions made at the administrative village level generally cannot match the actual situation, and the high-spatial-resolution disaster prediction specific to building locations has extremely high uncertainty. In order to find a balance between the uncertainty of forecasting and the refined requirements of transfer, this embodiment selects the key elements that can reflect the disaster-forming environment to establish transfer decision-making units, such as Figure 6 shown as follows:

[0097] S21. On the basis of not considering the village-level administrative boundaries, grid the B town in this embodiment spatially (including topography, landform, roads, buildings, etc.) based on a 30m resolution. Set traffic control nodes to reflect the surrounding road construction situation of the study area. Take the nearest road topological node that any grid can reach as the affiliated traffic control node. Divide the B town into v regions according to the traffic control nodes. The regions controlled by different traffic control nodes are studied independently to ensure the practical feasibility of the transfer decision-making unit in operation command.

[0098] S22. First-level clustering: Divide the B town into two categories, the valley area and the interfluve area, according to the elevation of the grid and the distance from the river. The disaster types in the valley area and the interfluve area are different.

[0099] S23. Second-level clustering: For the valley area, conduct clustering with four indicators: the slope at the grid position, the roughness at the grid position, the relative height difference with the nearest river channel, and the cross-section slope of the nearest river channel; for the interfluve area, conduct clustering with four indicators: the elevation at the grid position, the slope at the grid position, the curvature at the grid position, and the historical landslide density, where the historical landslide density is used to characterize the stability of the affiliated slope unit.

[0100] S24. When there is a village-level administrative boundary passing through the block obtained after the second-level clustering, then divide the block clustered in step S23 according to the village-level administrative boundary to ensure clear administrative jurisdiction.

[0101] S3. Conduct risk forecasting for the transfer decision-making unit: Calculate the injury level that residents will face if they do not transfer, and integrate the injury level that residents will face on the transfer decision-making unit to obtain the comprehensive risk index forecast of the transfer decision-making unit. Conduct risk forecasting for all random scenarios respectively.

[0102] In this embodiment, step S3 includes:

[0103] S31. Based on the building function vulnerability curve , calculate the injury level that residents will face if they do not transfer (i.e., "stay" and carry out normal activities at home) through the maximum flood depth in the next few hours. The purpose is to make a forward-looking assessment of future potential risks, which is specifically expressed by the following formula:

[0104] (1)

[0105] (2)

[0106] (3)

[0107] (4)

[0108] In formulas (1) - (4), represents the base of the natural logarithm; represents the current moment; represents the grid water depth at the moment of represents the grid water depth at the moment of represents the grid water depth at the moment of represents the function loss rate of the building at different moments; represents the forecast value of the future risk level at the moment of considering the command time-consuming in the actual transfer process; represents the building function loss function; represents the maximum immersion depth of the building predicted by the flood model during the time period of represents the preset time interval. The time from when the village cadre receives the superior's instruction to the completion of the transfer usually does not exceed 6 hours. In this embodiment, is taken as the calculation parameter, and this parameter can be corrected according to the actual situation to ensure that the decision-maker can take timely countermeasures.

[0109] S32. Integrate the injury levels that residents are about to face obtained in step S31 on the transfer decision-making units, that is, weight all the affected buildings inside the transfer decision-making units by population to obtain the comprehensive risk index forecast of the transfer decision-making unit. This comprehensive risk index changes over time and is expressed by the following formula:

[0110] (5)

[0111] In formula (5), represents the comprehensive risk index of the th transfer decision-making unit at time; represents the population on the th building in the th transfer decision-making unit; represents the number of flooded buildings inside a certain transfer decision-making unit; represents the degree of injury that the residents on the th building are about to face at time; represents the total population within the th transfer decision-making unit.

[0112] S33. Conduct risk forecasts for all the random scenarios obtained in step S1 respectively, which is represented by , where , , represents the sample set of random scenarios.

[0113] S4. Establish a multi-stage stochastic decision-making model and make decisions based on the uncertain risks obtained in step S3.

[0114] In this embodiment, step S4 specifically includes:

[0115] S41. Establish a multi-stage stochastic decision-making model:

[0116] (1)Define the decision variables: The multi-stage includes the current stage and several future stages. In this embodiment, it is assumed that there are 7 decision-making stages in total; the decision variables include whether the transfer decision-making unit transfers at the current stage time and in several future stages, and the opening status of each refuge corresponding to each stage, which are respectively represented by and two groups of 0-1 binary decision variables, where , , . In this embodiment, the interval between stages is taken as 4 hours, represents that the residents stay at home at this time, Indicates that the residents have left their homes, for any transfer decision unit , the first moment is the immediate transfer time point, Indicates that the shelter is open, Indicates that the shelter is not open.

[0117] (2) Design the objective function: Calculate the optimal transfer time of the transfer decision units to be evacuated in the study area from the perspective of maximizing the overall utility, and comprehensively consider the trade-off between the life and property safety that can be protected by evacuation, the assets abandoned during evacuation, and the evacuation cost, and select the strategy with the greatest benefit, so as to establish the objective function, which is expressed as follows:

[0118] (6)

[0119] (7)

[0120] (8)

[0121] In formulas (6)-(8), formula (6) means to postpone the transfer action to reduce unnecessary production losses and lower the transfer and resettlement costs, and control the activation cost through the number of open shelters; formula (7) means to reduce the expected cumulative damage during the decision foresight period, Indicates the calculation of the expected value; Indicates the loss per person per unit time caused by the transfer; Indicates the fixed cost of activating a single shelter; Indicates the expected number of people to be transferred during the decision foresight period; Indicates the total population living in the

[0122] (3) Design the constraint conditions of the objective function, specifically as follows:

[0123] (9)

[0124] (10)

[0125] (11)

[0126] (12)

[0127] (13)

[0128] Formulas (9) and (10) respectively represent the value ranges of decision variables and ; Formulas (11) and (12) indicate that once a decision is implemented, the value of the decision variable cannot return to the unimplemented state; in Formula (13), represents the maximum capacity of the refuge , and Formula (13) means that the number of transferred residents cannot exceed the total capacity of the refuge.

[0129] S42. Solving the solution of the multi-stage stochastic decision-making model:

[0130] (1) The variable is a random variable. The function expectation in Formula (5) is transformed into solving and averaging under multiple random samples. At this time, Formula (7) is converted into the following expression:

[0131] (14)

[0132] (2) Taking the minimization of the expected cumulative damage within the decision foresight period as the main objective, controlling the relaxation degree based on the ε-constraint algorithm. The ε-constraint algorithm is a multi-objective optimization algorithm used to solve optimization problems with multiple decision variables and multiple objective functions. It is based on the idea of constrained optimization. By introducing a parameter ε to control the weights of the objective functions, an approximate solution set of the optimal solution can be found on the premise of ensuring satisfaction of the constraint conditions, that is, adjusting ε to generate a series of Pareto optimal solutions.

[0133] S5. Rolling decision prediction.

[0134] In this embodiment, step S5 includes:

[0135] S51. Based on the current latest rainfall forecast, successively execute steps S1, S3, and S4, select a solution from the Pareto solution set of the multi-stage stochastic decision-making model in step S4 as the transfer plan. This solution includes the locations that need to be immediately transferred in the current stage and the locations that are expected to be transferred in future stages, and issue an "immediate transfer instruction" and a "preparatory transfer notice" respectively for the current stage and future stages.

[0136] S52. In this embodiment, based on a preset time interval of 4 hours, repeat the execution of step S51. During each rolling decision process, the previous round of "preparatory transfer notice" is replaced by the current round of "immediate transfer instruction". Each decision is a multi-stage rolling decision, and each transfer command only corresponds to the current stage in the decision solution, and the future stage is a forecast result.

[0137] The schematic diagram of the transfer decision is as Figure 7 shown. In this embodiment, with a time interval , taking the transfer actions designed for the next 6 stages in each decision as an example. Of course, it is not limited to a time interval of 4 hours and designing the next 6 stages, and can be adjusted according to the actual situation. The horizontal axis represents the solution obtained by optimizing the current decision. The current stage ( ) is the area where transfer should occur, sending out an "immediate transfer instruction" and immediately commanding the transfer. The future stages are the areas where transfer is expected to occur, sending out a "preparatory transfer notice". The vertical axis represents the actual transfer command process. As the decision-making base point advances, a new optimized solution is generated, and the previous round of "preparatory transfer notice" is replaced by the current round of "immediate transfer instruction". This embodiment shows the rolling decision-making effect during the typhoon evolution process. As Figure 8 shown, a total of seven rounds of evacuation decisions are given, summarizing the early warning release situation in each time period during the emergency process. Figure 8 Each column of Figure 8 represents a transfer decision unit. For example, "P-23" represents the 23rd transfer decision unit.

[0138] Only the basic principles and preferred embodiments of the present invention are described above. Those skilled in the art can make many changes and improvements based on the above description, and these changes and improvements should fall within the protection scope of the present invention.

Claims

1. A real-time decision-making method for typhoon emergency evacuation in mountainous villages and towns, characterized in that: The steps include: S1. Establishing a probabilistic rainfall forecast: obtaining rainfall forecast samples through Monte Carlo sampling, and inputting the rainfall forecast samples into the physical simulation model for dynamic forecasting of random scenarios; S2. Establishing a transfer decision unit: rasterizing the study area spatially, and performing spatial clustering and dividing the study area into blocks based on the raster to obtain the smallest monomer of the transfer decision, i.e., the transfer decision unit; Specifically include: S21. Without considering the village administrative boundaries, the study area is spatially gridded based on a preset resolution, and traffic control nodes are set to reflect the surrounding road construction conditions of the study area. The nearest road topological node that can be reached by any grid is taken as the traffic control node. The study area is divided into several areas according to the traffic control nodes, and the areas controlled by different traffic control nodes are studied independently. S22, first-level clustering: according to the elevation of the grid and the distance from the river, the study area is divided into two categories: river valley area and riverside area; S23, Secondary clustering: For river valleys, clustering is performed based on the slope at the grid position, the roughness at the grid position, the relative height difference with the nearest river channel, and the slope of the nearest river channel section; for inter-river areas, clustering is performed based on the elevation at the grid position, the slope at the grid position, the curvature at the grid position, and the historical landslide density; S3. Risk forecasting for the transfer decision unit: Calculate the level of harm that residents will face if they do not transfer, integrate the level of harm that residents will face on the transfer decision unit to obtain a comprehensive risk index forecast for the transfer decision unit, and conduct risk forecasting for all random scenarios; S4, establish a multi-stage random decision model, and make decisions based on the uncertainty risk obtained in step S3; S5. Rolling decision forecast.

2. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 1 is characterized in that: Step S1 includes: S11. Establish a random error model: collect official rainfall forecast data and meteorological station observation data during historical typhoons in the study area, use the original forecast value as the independent variable, regress the mean and variance of the error, and then normalize the errors under different forecast values ​​based on the mean and variance obtained by regression. Then fit the error distribution obtained after processing to establish a random error model. S12, generating rainfall forecast samples: based on the deterministic rainfall forecast, extracting error samples from the random error model of step S11 by Monte Carlo method, combining the original forecast samples with the error samples to obtain several groups of rainfall forecast samples; S13. Flash flood forecasting through physical simulation models: The physical simulation models include the SCS-CN model and the LisFlood-FP model. The surface runoff is estimated based on the SCS-CN model and the dynamic flood process is simulated in combination with the LisFlood-FP model. The SCS-CN model and the LisFlood-FP model are driven by several groups of rainfall forecast samples obtained in step S12, respectively, to obtain a set of random scenarios represented by the dynamic water depth of the grid.

3. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 1 is characterized in that: Step S2 also includes: S24. When the block obtained after the secondary clustering has a village-level administrative boundary passing through it, post-processing is performed.

4. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 3 is characterized in that: In step S24, post-processing specifically refers to: further dividing the blocks clustered in step S23 according to the village-level administrative boundaries to ensure clear administrative jurisdiction.

5. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 1 is characterized in that: Step S3 includes: S31. Based on the building function vulnerability curve, calculate the level of damage that residents will face if they fail to relocate using the maximum flood depth in the next few hours; S32, integrating the level of harm that the residents will face obtained in step S31 on the transfer decision unit to obtain a comprehensive risk index forecast of the transfer decision unit; S33. Perform risk forecasts for all random scenarios obtained in step S1.

6. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 5 is characterized in that: Step S3 specifically includes: S31. Based on the building function vulnerability curve, the maximum flood depth in the next few hours is used to calculate the level of damage that residents will face if they are not relocated, which is expressed by the following formula: （1） （2） （3） （4） Formula (1)-Formula (4), represents the base of natural logarithms; Indicates the current moment; express The grid water depth at the time, express The grid water depth at the time, express Grid water depth at the moment; Indicates the functional loss rate of the building at different times; Indicates that the actual transfer process takes time to command. The forecast value of future risk level at all times; represents the building function loss function; The flood model predicts The maximum flooding depth of the building during the period, Indicates a preset time interval; S32, integrating the damage level that the residents will face obtained in step S31 on the transfer decision unit, that is, weighting all the affected buildings in the transfer decision unit by population to obtain a comprehensive risk index forecast of the transfer decision unit, which is expressed by the following formula: （5） In formula (5), Indicates The transfer decision unit The comprehensive risk index at the moment; Indicates The transfer decision unit Population of buildings; Indicates the number of flooded buildings within a transfer decision unit; Indicates Residents of the building The degree of harm that is about to happen; Indicates The total population within the transfer decision unit; S33, based on all random scenarios obtained in step S1, risk forecast is performed respectively, through It indicates that, , , Represents a collection of samples of random scenes.

7. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 6 is characterized in that: Step S4 includes: S41. Establish a multi-stage random decision model: (1) Clarify decision variables: Multiple stages include the current stage and several future stages; decision variables include whether the transfer decision unit will transfer in the current stage and several future stages, and the corresponding opening status of each shelter in each stage; (2) Design objective function: Calculate the optimal transfer time of the decision-making units to be evacuated in the study area from the perspective of overall utility maximization, and comprehensively consider the trade-offs between the safety of life and property that can be protected by evacuation, the assets abandoned by evacuation, and the cost of evacuation, and select the strategy with the greatest benefit to establish the objective function; (3) Design the constraints of the objective function; S42. Find the solution to the multi-stage random decision model.

8. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 7 is characterized in that: Step S4 specifically includes: S41. Establish a multi-stage random decision model: (1) Clarify decision variables: multiple stages include the current stage and several future stages; decision variables include transfer decision units At the current stage The time and future stages of whether to transfer, as well as the corresponding opening status of each shelter in each stage, are expressed as and Two sets of 0-1 binary decision variables, where , , , represents the number of future stages, It means that residents are staying at home at this time. Indicates that the resident has left home and any transfer decision unit , the first The moment is the immediate transfer time point, The evacuation shelters are open. It means that the shelter is not open; (2) Design of objective function: Calculate the optimal transfer time of the decision-making units to be evacuated in the study area from the perspective of overall utility maximization, and comprehensively consider the trade-offs between the safety of life and property that can be protected by evacuation, the assets abandoned by evacuation, and the cost of evacuation, and select the strategy with the greatest benefit to establish the objective function. The objective function is expressed as follows: （6） （7） （8） In formula (6)-formula (8), formula (6) indicates that in order to postpone the transfer action to reduce unnecessary production losses and reduce the transfer and resettlement costs, and control the activation costs by opening the number of shelters; formula (7) indicates that in order to reduce the expected cumulative damage within the decision-making forecast period, It means to find the expected value; It indicates the time consumption of a single person unit caused by the transfer; represents the fixed cost of activating a single shelter; It represents the number of people expected to be transferred during the decision forecast period; Indicates The total population living in the transfer decision unit; (3) Design the constraints of the objective function as follows: （9） （10） （11） （12） （13） Formula (9) and Formula (10) represent the decision variables respectively. and The value range of ; Formula (11) and Formula (12) indicate that once a decision is implemented, the value of the decision variable cannot be returned to the unimplemented state; In Formula (13), Indicates a refuge point At the maximum capacity, formula (13) indicates that the number of residents to be transferred cannot exceed the total capacity of the shelter; S42. Solve the solution of the multi-stage random decision model: (1) Variables is a random variable, and the The function expectation is converted into multiple random samples and then averaged. At this time, formula (7) is converted into the following expression: （14） (2) Taking minimizing the expected cumulative damage within the decision forecast period as the main objective, the relaxation degree is controlled based on the ε-constraint algorithm, and ε is adjusted to generate a series of Pareto optimization solutions.

9. The real-time decision-making method for typhoon emergency evacuation for mountainous villages and towns according to claim 8 is characterized in that: Step S5 includes: S51, based on the latest rainfall forecast, sequentially executing steps S1, S3, and S4, selecting a solution from the Pareto solution set of the multi-stage random decision model of step S4 as a transfer plan, the solution including the position that needs to be transferred immediately in the current stage and the position that is expected to be transferred in the future stage, and issuing an "immediate transfer instruction" and a "preparatory transfer notice" for the current stage and the future stage respectively; S52. Repeat step S51 based on a preset time interval. During each rolling decision process, the previous round of "preparatory transfer notification" is replaced by the current round of "immediate transfer instruction". Each decision is a multi-stage rolling decision. Each transfer command only corresponds to the current stage in the execution decision solution, and the future stage is a predictive result.

Images