Flood path prediction method based on particle swarm adaptive grouping

Abstract

The application discloses a flood path prediction method based on particle swarm adaptive grouping, S1. Collecting multi-source heterogeneous data, and preprocessing the multi-source heterogeneous data; S2. Constructing a graph neural network model based on the multi-source heterogeneous data; S3. Using the graph neural network model to predict the key propagation path of the target area; S4. According to the initial reference information, the target area is divided into sub-regions; S5. For each sub-region, initialize the particle swarm and set the flood propagation path parameters of the particles; S6. Iterative optimization of particle swarm based on adaptive grouping strategy; S7. The new multi-source heterogeneous data is input into the graph neural network model to update the prediction result of the key propagation path; S8. Forming the overall flood path prediction result of the target area. The dynamic particle swarm optimization of the sub-region level also provides higher reliability for accurately identifying waterlogging points and endangered areas, and improves the accuracy of flood control and drainage scheme planning and early warning information release.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flood, in particular to a flood path prediction method based on particle swarm adaptive grouping. BACKGROUND

[0002] With the frequent occurrence of extreme weather events and the accelerating process of urbanization, flood disasters have become an important factor threatening people's life and property safety in many regions. In order to effectively carry out flood prevention warning and emergency disposal, flood path prediction technology has received widespread attention in recent years. On the one hand, accurately predicting the propagation trajectory of flood in a river basin or urban area is crucial for formulating timely and effective defense measures and personnel evacuation strategies. On the other hand, with the rapid growth of multi-source data such as remote sensing monitoring, hydrological observation and meteorological forecasting, how to fully utilize and integrate these data has become a hot research topic.

[0003] In the prior art, common flood path prediction methods are mainly divided into two categories: one is the numerical simulation method based on physical mechanism, which uses fluid dynamics model to calculate the flow rate, flow direction and inundation range of flood in the region; the other is to rely on machine learning or data-driven model, which trains historical hydrological and meteorological data to predict the propagation process of flood. The existing methods have improved the accuracy and efficiency of flood path prediction to some extent, but there are still some problems that cannot be ignored. First, the physical simulation method has very strict requirements for boundary conditions and initial conditions and often has large computational overhead. In the scene of complex terrain and multi-source heterogeneous data coupling, it is difficult to realize real-time or near real-time prediction. Second, data-driven models often rely on a large amount of high-quality training data, and have insufficient adaptability to sudden abnormal conditions (floods caused by extreme rainfall or geological disasters) and are easily affected by regional differences and data missing in terms of prediction accuracy.

[0004] In addition, the existing technology generally lacks dynamic coupling ability of multi-source data and cannot fully explore the comprehensive influence of terrain undulation, vegetation distribution and meteorological element factors on flood propagation path. Most researches only update data in offline mode, and the prediction results often lag behind the actual flood evolution speed, which is difficult to meet the real-time demand of emergency response. On the other hand, although some flood path prediction methods based on traditional optimization algorithms have certain optimization ability, they are easy to fall into local optimal solution in small range or complex terrain, and lack adaptive ability for multi-objective or multi-regional differentiated prediction.

[0005] In summary, the prior art has deficiencies in real-time prediction at high spatiotemporal resolution, dynamic fusion of multi-source heterogeneous data, and adaptive optimization for complex terrain, and it is difficult to quickly and accurately identify the key path of flood propagation and carry out fine early warning, thereby bringing challenges to current flood control and emergency management. Therefore, how to update in real time while utilizing multi-source data, improve the prediction accuracy in a small area, and adapt to changing hydrological conditions has become a technical problem to be solved in the field of flood path prediction. SUMMARY

[0006] One object of the present application is to provide a flood path prediction method based on adaptive grouping of particle swarm, and the dynamic particle swarm optimization at the sub-region level of the present application also provides higher reliability for accurately identifying flood-prone points and endangered areas, and improves the accuracy of flood control and drainage scheme planning and early warning information release.

[0007] According to the flood path prediction method based on adaptive grouping of particle swarm, the method comprises the following steps:

[0008] S1. Collecting multi-source heterogeneous data and pre-processing the multi-source heterogeneous data;

[0009] S2. Constructing a graph neural network model based on the multi-source heterogeneous data, taking the regional features in the historical flood data as nodes, and taking the interaction relationship formed by the terrain features and the meteorological data as edges;

[0010] S3. Using the graph neural network model to predict the key propagation path of the target region, and outputting the initial reference information of the key propagation path;

[0011] S4. Dividing the target region into sub-regions according to the initial reference information, and determining the boundary range of each sub-region;

[0012] S5. For each sub-region, initializing a particle swarm and setting the flood propagation path parameters of the particles, so that each particle represents the flow velocity, flow direction or residence time parameters related to the flood propagation in the sub-region;

[0013] S6. Iterative optimization of the particle swarm based on the adaptive grouping strategy, calculating the fitness value and dynamically adjusting the particle swarm size and algorithm parameters in the iteration process, and using the local data in the sub-region to real-time correct the search direction and speed of the particles;

[0014] S7. When receiving new multi-source heterogeneous data, inputting the new multi-source heterogeneous data into the graph neural network model to update the prediction result of the key propagation path, and making corresponding adjustments to the sub-region division and particle swarm parameters;

[0015] S8. Integrating the particle swarm optimization results of each sub-region to form the overall flood path prediction results of the target region.

[0016] Optionally, S1 comprises the following steps:

[0017] S11. Collecting historical flood data D h , including the flood propagation path, peak flow, and duration of the i-th historical flood event;

[0018] S12. Obtaining terrain feature data D t , including the slope, aspect, and surface roughness of the i-th geographical unit;

[0019] S13. Collecting meteorological data D m , including the rainfall intensity, wind speed, and temperature at the i-th time point in the target region;

[0020] S14. Obtaining water level data D w of hydrological stations, including water depth, flow velocity, and flow rate of the i-th hydrological station;

[0021] S15. Obtaining satellite remote sensing image data D r , containing water body distribution, vegetation coverage, and inundation range in the target region of the i-th satellite image;

[0022] S16. Standardizing the multi-source heterogeneous data obtained in S11-S15 to construct a unified multi-source heterogeneous data set D:

[0023]

[0024] wherein, represents the set of the i-th flood event in the historical flood data, d hi contains information about the i-th flood event, totaling n flood events, represents the set of terrain feature data, g j is the terrain feature of the j-th geographical unit, totaling m geographical units, represents the set of meteorological data, r k is the meteorological data at the k-th time point, totaling p time points, represents the set of water level data of hydrological stations, s l is the real-time water level data of the l-th hydrological station, totaling q hydrological stations, represents the set of satellite remote sensing image data, v o is the o-th satellite remote sensing image data, totaling k images, and ∪ represents the set union operator for the collection of individual elements in each type of data;

[0025] S17. Map all data in the multi-source heterogeneous dataset D to the interval [0, 1] by linear normalization, and use time interpolation method to complete the missing data at time points, and use spatial interpolation method to supplement the missing spatial data in the region, to obtain the preprocessed multi-source heterogeneous dataset D c .

[0026] Optionally, the S2 comprises the following steps:

[0027] S21. From the preprocessed multi-source heterogeneous dataset D c , select the historical flood data D in the target area and the hydrological station water level data as the node feature source, and construct the node set N = {n1, n2, …, nu} u}, wherein n i ,i∈{1,2,...,u} represents the historical flood event features and corresponding hydrological station water level features contained in the i-th node;

[0028] S22. Combine the terrain feature data, meteorological data and satellite remote sensing image data in the multi-source heterogeneous dataset D c , construct a multi-relation adjacency matrix wherein R represents the number of different relationship types, the connection of the nodes under the r-th relationship type, and construct the edge set E = {e ij};

[0029] S23. Define the node feature matrix wherein u represents the number of nodes, f represents the joint feature dimension contained in each node, and the feature of each node is composed of the historical flood event feature, the hydrological station water level data feature and the local geographical and meteorological attributes corresponding to the node;

[0030] S24. Construct a graph neural network model GNN, introduce a multi-head attention mechanism to weight and aggregate different relationship types based on the multi-relation adjacency matrix and the node feature matrix X, and define the node feature update as:

[0031]

[0032] wherein H (l) represents the node feature matrix of the l-th layer of the graph neural network, H (0) =X initially, W (l) is the trainable weight matrix of the l-th layer, a (m,r) is the weight coefficient of the m-th attention head under the r-th relationship type, M represents the number of attention heads, s represents the activation function, and || represents the splicing operation.

[0033] Optionally, the construction of the edge set is based on the following multi-source relationships:

[0034] The spatial adjacency between nodes is identified by using the slope and slope direction in the terrain feature data;

[0035] The time correlation between nodes is identified by using the rainfall intensity or wind speed in the meteorological data;

[0036] The flood influence correlation degree between nodes is identified by using the regional inundation range in the satellite remote sensing image data;

[0037] The hydrodynamic coupling strength between nodes is calculated by comprehensively using the water depth and flow information in the hydrological station water level data.

[0038] Optionally, the S3 comprises the following steps:

[0039] S31. The node feature matrix H is calculated according to the graph neural network model (L) , wherein L represents the total number of layers of the graph neural network, and the node feature matrix H (L) represents the importance and dynamic characteristics of each node in the flood propagation network under the comprehensive conditions of multi-source heterogeneous data;

[0040] S32. The node importance score I(n (L) ) of each node n i in the target region is calculated based on the node feature matrix H i , which is used to evaluate the criticality of the node in the flood propagation process:

[0041]

[0042] , wherein represents the flood influence weight of the node n i on the connected node n j under the rth type of relationship, represents the flood propagation feature of the node n i in the Lth layer, R represents the number of multi-source relationship types, and u is the total number of nodes;

[0043] S33. The K nodes with the highest scores are selected to form a key node set N k in descending order of the node importance score I(n i );

[0044] S34. For each node in the key node set N k , the node feature and the connection relationship in the multi-relationship adjacency matrix are extracted, the flood propagation path between the key nodes is identified by traversing the adjacent nodes, and a preliminary propagation path set is generated:

[0045] P={p1,p2,…,p M};

[0046] Among them, p M represents the Mth propagation path, including the node sequence and the flow or water level changes between nodes in the path;

[0047] S35. For each path p in the preliminary propagation path set P j Calculate the path weight W(p j ), the path weight is used to quantify the priority of the path in flood propagation:

[0048]

[0049] Among them, I(n ki ) is the path p j The importance score of the node reflects the influence of the node in flood propagation, w ij is the edge e in the path ij The weight of represents the intensity of flood propagation between nodes;

[0050] S36. According to the path weight W(p j ) Sort the preliminary propagation path set P, select the first M' paths with the highest weight to form the key propagation path set, the key path set P k Indicates the most likely flood propagation path within the target area:

[0051] P k ={p k1 ,p k2 ,…,p kM′}.

[0052] Optionally, the S4 includes the following steps:

[0053] S41. According to the key propagation path set P k Extract the node set N on each path k and the connection relationship set E k , where the node set N k Represents the flood propagation area nodes on the key propagation path, and the connection relationship set E k Represents the flooding association between nodes;

[0054] S42. Based on the key propagation path set P k Calculate the node cluster center C in the target area = {c1, c2, ..., c q}, where q represents the number of sub-regions initially divided, and each cluster center c q Determined by:

[0055]

[0056] Among them, w jimportance weight of node n j , reflecting the key degree of node in flood propagation, p j represents the spatial position vector of node n j ;

[0057] S43. Spatially divide the node set N k in the target area, and assign each node n j to the nearest cluster center c i , complete the sub-area division, and the sub-area set is represented as R:

[0058]

[0059] wherein R i represents the i-th sub-area, containing all nodes closest to the cluster center c i , and ||p j -c i || represents the Euclidean distance between node n j and cluster center c i ;

[0060] S44. According to the sub-area set R, further optimize the boundary range of each sub-area R k in combination with the connection relationship set E i between nodes, and calculate the boundary function B i of each sub-area:

[0061]

[0062] wherein B i represents the boundary of the i-th sub-area, containing all edges e jk connected to other sub-areas.

[0063] Optionally, the S5 comprises the following steps:

[0064] S51. For each sub-area R i , set the particle swarm size P i for each sub-area, and respectively construct a particle swarm set in the sub-area R i :

[0065]

[0066] S52. In the sub-area R i , define the parameter vector of the particle according to the search space determined by the flow velocity, flow direction and residence time parameters related to flood propagation: represents the initial position of the j-th particle:

[0067]

[0068] wherein, denotes the initial flow velocity of the jth particle in the sub-region R i , reflecting the magnitude of the flood flow velocity, denotes the initial flow direction angle, representing the propagation direction of the flood, denotes the initial residence time, embodying the residence or stay characteristics of the flood in the sub-region;

[0069] S53. Randomly initializing the parameter vector of each particle, making the particle swarm uniformly distributed in the feasible range, and the random initialization is:

[0070]

[0071] wherein, d e {1, 2, 3} respectively correspond to the parameters of flow velocity, flow direction and residence time of three dimensions, and respectively denote the minimum and maximum feasible values of the dth parameter, and rand(0, 1) represents a random number in the interval [0, 1];

[0072] S54. Setting the initial velocity vector for each particle , making the initial velocity component of each dimension follow a random or zero value distribution, and storing the initialized particle swarm set G i in the particle swarm database of the sub-region R i ;

[0073] S55. Outputting the initial position and initial velocity of the particle swarm in each sub-region R i , used to represent the flood propagation path parameters of the sub-region, and completing the particle swarm initialization of each sub-region.

[0074] Optionally, the S6 comprises the following steps:

[0075] S61. On the basis of the initialized particle swarm set G i , defining the upper limit of the iteration number for each sub-region R i , and executing the particle swarm optimization process based on the adaptive grouping strategy;

[0076] S62. At each iteration, calculating the fitness value of the particle , which reflects the performance of the particle in terms of the accuracy and stability of the predicted flood propagation path, and defining the fitness function:

[0077]

[0078] wherein, D iSub-region R i The local data in the sub-region R i , including terrain features, meteorological data and hydrological station water level information, are used to measure the fitting degree of the current particle in the flood propagation process in the sub-region R;

[0079] S63. According to the fitness value of each particle, the individual optimal position of the particle is updated, and the position with the highest fitness value is saved as the individual optimal position of the particle and the global optimal position gbest of the sub-region R; i

[0080] S64. The velocity and position of the particle are updated, and the velocity is updated as follows:

[0081]

[0082] wherein ω is an inertia weight, balancing the exploration and exploitation capabilities, c1 and c2 are learning factors, and rand1 (0, 1) and rand2 (0, 1) represent random numbers in the interval [0, 1];

[0083] S65. In the iteration process, the change of the local data in the sub-region Ris detected in real time, and if the key monitoring indicators change significantly, the particle swarm size and algorithm parameters ω, c1 and c2 are dynamically adjusted according to the adaptive grouping strategy, and the search direction and velocity of the particle are corrected, and part of the particle swarm is reinitialized or merged to meet the dynamic propagation conditions of the flood;

[0084] S66. The iteration steps S61-S65 are repeated until the upper limit of the iteration number is reached or the convergence condition is met, and the optimal position set {gbest i} of the particle swarm in the sub-region Ris output, representing the particle solution set with the highest fitness in the flood propagation path prediction in the sub-region R . i

[0085] The beneficial effects of the present application are:

[0086] (1) The present application can independently iterate and dynamically adjust the algorithm parameters for each sub-region according to the local characteristics of the sub-region by fusing terrain feature data, hydrological station water level data, meteorological data and satellite remote sensing image data in the particle swarm initialization and iteration optimization process, and differentiating the sub-region division based on the key propagation path extraction. In the case of different hydrological conditions or sudden increase of upstream inflow, the adaptive grouping mechanism can quickly locate the affected areas and optimize the particle swarm search, so that the real-time flood path prediction can both consider the whole and achieve higher accuracy in local areas.

[0087] (2) The application introduces an improved graph neural network on the basis of traditional physical models and empirical models to deeply mine the mutual correlation between historical flood data, topography and meteorological conditions, accurately captures the main path of flood propagation through key node extraction and multi-relation adjacency matrix construction, and at the same time, the adaptive grouping strategy focuses on the change of key monitoring indicators in the sub-region in the particle swarm iteration process to dynamically adjust the particle swarm size and parameters, avoiding the deficiency that a single particle swarm is easy to fall into local optimum or lack of adaptability when facing different hydrological scenes, and realizing more flexible and efficient prediction of the flood propagation process with high spatio-temporal dynamics.

[0088] (3) After refining and screening the key propagation path, the application performs independent particle swarm iteration search on each local area based on sub-region division, and merges the optimal solutions of each sub-region at the global level, thereby realizing comprehensive control of the overall flood propagation situation, and continuously correcting the flood flow rate, flow direction and residence time prediction indicators under the real-time update of multi-source data, so that the flood control command department has more initiative and pertinence when responding to sudden disasters, and at the same time, the dynamic particle swarm optimization refined to the sub-region level also provides higher reliability for accurately identifying waterlogging points and endangered areas, and improves the accuracy of flood control and drainage scheme planning and early warning information release. BRIEF DESCRIPTION OF DRAWINGS

[0089] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, together with the embodiments of the application, to explain the application, and do not constitute a limitation on the application. In the drawings:

[0090] Figure 1 A flowchart of a flood path prediction method based on particle swarm adaptive grouping according to the application. DETAILED DESCRIPTION

[0091] The application will now be described in further detail with reference to the drawings. These drawings are simplified schematic diagrams, and only schematically show the basic structure of the application.

[0092] REFERENCE Figure 1 A flood path prediction method based on particle swarm adaptive grouping, comprising the following steps:

[0093] S1. Collecting multi-source heterogeneous data and pre-processing the multi-source heterogeneous data;

[0094] S2. Building a graph neural network model based on the multi-source heterogeneous data, taking the regional characteristics in the historical flood data as nodes, and taking the interaction relationship formed by the topographic characteristics and meteorological data as edges;

[0095] S3. Predict the key propagation path of the target area using a graph neural network model, and output the initial reference information of the key propagation path;

[0096] S4. Divide the target area into sub-regions according to the initial reference information, and determine the boundary range of each sub-region;

[0097] S5. For each sub-region, initialize the particle swarm and set the flood propagation path parameters of the particles, so that each particle represents the flow velocity, flow direction or residence time parameters related to the flood propagation in the sub-region;

[0098] S6. Perform iterative optimization of the particle swarm based on the adaptive grouping strategy, calculate the fitness value, and dynamically adjust the particle swarm size and algorithm parameters during the iteration process, and use the local data in the sub-region to real-time correct the search direction and speed of the particles;

[0099] S7. When receiving new multi-source heterogeneous data, input the new multi-source heterogeneous data into the graph neural network model to update the prediction result of the key propagation path, and make corresponding adjustments to the sub-region division and particle swarm parameters;

[0100] S8. Integrate the particle swarm optimization results of each sub-region to form the flood path prediction result of the whole target area.

[0101] In this embodiment, S1 includes the following steps:

[0102] S11. Collect historical flood data D h , including the flood propagation path, peak flow and duration of the i-th historical flood event;

[0103] S12. Obtain terrain feature data D t , including the slope, slope direction and surface roughness of the i-th geographic unit;

[0104] S13. Collect meteorological data D m , including the rainfall intensity, wind speed and temperature at the i-th time in the target area;

[0105] S14. Obtain water level data D w of hydrological station, including water depth, flow velocity and flow of the i-th hydrological station;

[0106] S15. Obtain satellite remote sensing image data D r , containing water body distribution, vegetation coverage and inundation range in the target area of the i-th satellite image;

[0107] S16. Standardize the multi-source heterogeneous data obtained by S11-S15 to construct a unified multi-source heterogeneous data set D:

[0108]

[0109] in, represents the set of the i-th flood events in historical flood data, d hi Contains information about the i-th flood event, with a total of n flood events, Represents a collection of terrain feature data, g j is the topographical feature of the jth geographical unit, with a total of m geographical units, represents a collection of meteorological data, r k is the meteorological data at the kth time point, with a total of p time points, Represents the collection of water level data of hydrological stations, s l is the real-time water level data of the lth hydrological station, with a total of q hydrological stations. Represents a collection of satellite remote sensing image data, v o is the oth satellite remote sensing image data, with a total of k images. ∪ represents the set union operator, which is used to aggregate the individual elements in each type of data.

[0110] S17. Map all data in the multi-source heterogeneous dataset D to the interval [0,1] through linear normalization, and use the time interpolation method to fill in the missing time point data, and the spatial interpolation method to fill in the missing spatial data in the region, to obtain the preprocessed multi-source heterogeneous dataset D c .

[0111] In this embodiment, S2 includes the following steps:

[0112] S21. From the preprocessed multi-source heterogeneous dataset D c In the process, the historical flood data D in the target area and the water level data of the hydrological station are selected as the node feature sources, and the node set N = {n1, n2, ..., n u}, where n i ,i∈{1,2,...,u} represents the historical flood event characteristics and the corresponding hydrological station water level characteristics contained in the i-th node;

[0113] S22. Combining Multi-source Heterogeneous Datasets D c The terrain feature data, meteorological data and satellite remote sensing image data in the data are used to construct a multi-relationship adjacency matrix Where R represents the number of different relationship types, Show the node connectivity under the rth relationship type, and construct the edge set E = {e ij};

[0114] S23. Define node feature matrix wherein u represents the number of nodes, and f represents the joint feature dimension contained by each node, and the feature of each node is composed of the historical flood event feature, the hydrological station water level data feature, and the local geographical and meteorological attributes corresponding to the node;

[0115] S24. A graph neural network model GNN is constructed, and a multi-head attention mechanism is introduced on the basis of the multi-relation adjacency matrix and the node feature matrix X to perform weighted aggregation on different relation types, and the node feature update is defined as:

[0116]

[0117] wherein H (l) represents the node feature matrix of the lth layer of the graph neural network, and initially H (0) =X, W (l) is the trainable weight matrix of the lth layer, a (m,r) is the weight coefficient of the mth attention head under the rth relation type, M represents the number of attention heads, σ represents an activation function, and || represents a splicing operation.

[0118] In the embodiment, the construction of the edge set is based on the following multi-source relations:

[0119] The spatial adjacency between nodes is identified by using the slope and slope direction in the terrain feature data;

[0120] The temporal correlation between nodes is identified by using the rainfall intensity or wind speed in the meteorological data;

[0121] The flood influence correlation degree between nodes is identified by using the regional inundation range in the satellite remote sensing image data;

[0122] The hydrodynamic coupling strength between nodes is calculated by comprehensively using the water depth and flow information in the hydrological station water level data.

[0123] In the embodiment, S3 includes the following steps:

[0124] S31. The node feature matrix H (L) is calculated according to the graph neural network model, wherein L represents the total number of layers of the graph neural network, and the node feature matrix H (L) represents the importance and dynamic characteristics of each node in the flood propagation network under the comprehensive condition of multi-source heterogeneous data;

[0125] S32. The node importance score I(n i ) of each node n i in the target region is calculated based on the node feature matrix H (L) , which is used to evaluate the criticality of the node in the flood propagation process:

[0126]

[0127] wherein, denotes the node n i under the rth type of relationship, j the flood impact weight of the node n denotes the flood propagation feature of the node n i in the Lth layer, R represents the number of multi-source relationship types, and u is the total number of nodes;

[0128] S33. According to the descending order of the node importance score I(n i ), the top K nodes with the highest scores are selected to form a key node set N k ;

[0129] S34. For each node in the key node set N k , its node feature and the connection relationship in the multi-relationship adjacency matrix are extracted, the flood propagation path between the key nodes is identified by traversing the adjacent nodes, and a preliminary propagation path set P is generated:

[0130] P={p1,p2,…,p M};

[0131] wherein, p M represents the Mth propagation path, which contains the node sequence in the path and the flow or water level change between the nodes;

[0132] S35. For each path p j in the preliminary propagation path set P, the path weight W(p j ) is calculated, which is used to quantify the priority of the path in the flood propagation:

[0133]

[0134] wherein, I(n ki ) is the importance score of the node on the path p j , reflecting the influence degree of the node in the flood propagation, w ij is the weight of the edge e ij in the path, representing the propagation strength of the flood between the nodes;

[0135] S36. According to the path weight W(p j ), the preliminary propagation path set P is sorted, and the top M' paths with the highest weights are selected to form a key propagation path set P k , which represents the most likely flood propagation path in the target area:

[0136] P k ={p k1 ,p k2 ,…,pkM′}.

[0137] In this embodiment, S4 includes the following steps:

[0138] S41. According to the key propagation path set P k Extract the node set N k and the connection relationship set E k on each path, where the node set N k represents the flood propagation area node on the key propagation path, and the connection relationship set E k represents the flood propagation association between nodes.

[0139] S42. Based on the key propagation path set P k Calculate the node clustering center C = {c1, c2, …, c q} in the target area, where q represents the number of initial partition sub-regions, and each clustering center c q is determined by:

[0140]

[0141] where w j is the importance weight of node n j , reflecting the key degree of the node in flood propagation, and p j represents the spatial position vector of node n j .

[0142] S43. Spatially divide the node set N k in the target area, and assign each node n j to the nearest clustering center c i , complete the sub-region division, and the sub-region set is represented as R:

[0143]

[0144] where R i represents the i-th sub-region, containing all nodes closest to the clustering center c i , and ||p j -c i || represents the Euclidean distance between node n j and clustering center c i .

[0145] S44. According to the sub-region set R, in combination with the connection relationship set E k between nodes, further optimize the boundary range of each sub-region R i , and calculate the boundary function B i of each sub-region:

[0146]

[0147] where B i denotes the boundary of the i-th sub-region, containing all edges e jk connecting to other sub-regions.

[0148] In this embodiment, S5 comprises the following steps:

[0149] S51. For each sub-region R i , set the population size P i for each sub-region. S52. Within each sub-region R i , construct a particle swarm set G

[0150]

[0151] S52. Within each sub-region R i , define the parameter vector of each particle according to the search space determined by the flow velocity, flow direction and residence time parameters related to flood propagation. denotes the initial position of the j-th particle:

[0152]

[0153] where denotes the initial flow velocity of the j-th particle within the sub-region R i , reflecting the magnitude of flood flow velocity, denotes the initial flow direction angle, representing the propagation direction of the flood, denotes the initial residence time, embodying the retention or stay characteristics of the flood within the sub-region;

[0154] S53. Randomly initialize the parameter vector of each particle , so that the particle swarm is uniformly distributed within the feasible range, and the random initialization is:

[0155]

[0156] where d e {1, 2, 3} respectively correspond to the parameters of flow velocity, flow direction and residence time in three dimensions, and respectively denote the minimum and maximum feasible values of the d-th parameter, and rand(0, 1) denotes a random number in the interval [0, 1];

[0157] S54. Set the initial velocity vector for each particle , so that the initial velocity components in each dimension are subject to random or zero value distribution, and store the initialized particle swarm set G i in the sub-region R iin the particle swarm database;

[0158] S55. Output each sub-region R i The initial position and initial velocity of the inner particle swarm are used to represent the flood propagation path parameters of the sub-region and complete the initialization of the particle swarm in each sub-region.

[0159] In this embodiment, S6 includes the following steps:

[0160] S61. In the initialized particle swarm set G i Based on the i Define the upper limit of the number of iterations and execute the particle swarm optimization process based on the adaptive swarming strategy;

[0161] S62. At each iteration, calculate the particles The fitness value of The fitness value reflects the performance of the particle in predicting the accuracy and stability of flood propagation paths. The fitness function is defined as:

[0162]

[0163] Among them, D i Represents sub-region R i The local data within the sub-region, including topographic features, meteorological data, and water level information of hydrological stations, are used to measure the degree of fit of the current particle in the flood propagation process in the sub-region;

[0164] S63. According to the fitness value Particles The individual optimal position pbest pij and the global optimal position gbest in the sub-region i Update and save the position with the highest fitness value as the individual optimal position of the particle and the global optimal position of the sub-region;

[0165] S64. Particles Speed Along with the position update, the velocity update is:

[0166]

[0167] Among them, ω is the inertia weight, balancing exploration and development capabilities, c1 and c2 are learning factors, rand1(0,1) and rand2(0,1) represent random numbers in the interval [0,1];

[0168] S65. In the iteration process, real-time detection of local data changes in the sub-region, if the key monitoring indicators change significantly, the particle swarm size and algorithm parameters ω, c1, c2 are dynamically adjusted according to the adaptive grouping strategy, and the search direction and speed of the particle are corrected, and part of the particle swarm is reinitialized or merged to meet the dynamic propagation conditions of the flood;

[0169] S66. Repeat the iteration steps S61-S65 until the upper limit of the number of iterations is reached or the convergence condition is met, and output the optimal position set {gbest i} of the particle swarm in the sub-region R i , representing the particle solution set with the highest fitness in the prediction of the flood propagation path in the sub-region.

[0170] Example 1:

[0171] Example: Application of the flood path prediction method based on deep learning assisted particle swarm adaptive grouping in a mountain flood disaster on June 15, 2024

[0172] At 15:30 on June 15, 2024, the hydrological monitoring station located in a mountainous area issued a heavy rain warning signal. The regional rainfall increased sharply to 120 mm in a short time, and was expected to continue until 20:00 that night. At this time, the flood control department of the low-lying village along the river received the warning information and quickly started the flood path prediction system of the invention to monitor the real-time dynamics of flood propagation and guide the evacuation of residents.

[0173] The target area includes five small villages, two river-crossing bridges, and part of the farmland irrigation facilities. The terrain is complex and the slope is large. The heavy rain triggered the rapid confluence of floods from the two main catchment points upstream, and the flood propagated along the low-lying area to the downstream. To cope with this complex situation, the flood control department immediately collected multi-source data including real-time rainfall, river level, flow rate, and satellite remote sensing image data after receiving the heavy rain warning. The data covered an area of about 150 square kilometers in the target area.

[0174] At 15:35, through the deployment of multi-source monitoring equipment, the first batch of real-time data collected includes:

[0175] The upstream hydrological station water level monitoring point (number: WS-01) shows that the water level rises sharply to 4.2 meters, and the flow rate reaches 2.8 meters / second;

[0176] The rainfall monitoring point (number: RS-02) measures the rainfall intensity of 40 mm per hour during this period;

[0177] Satellite remote sensing images show that 1.8 square kilometers of the low-lying area upstream has been flooded, and the water flow path is spreading in the southeast direction.

[0178] The system first normalizes the collected data, inputs the terrain, rainfall intensity, and water level information into the constructed graph neural network model, and analyzes the current conditions based on historical data to generate preliminary reference information for the key propagation path:

[0179] Path 1: Propagates downstream from WS-01 along the river, expected to reach Village A in 30 minutes;

[0180] Path 2: Low-lying areas converge, possibly threatening Bridge B in 45 minutes;

[0181] According to the above key path information, the system automatically divides the target area into 6 sub-regions (R1 to R6), each corresponding to different hydrological characteristics and terrain conditions. In sub-region R2 (near Village A), the system simulates flood propagation through a particle swarm algorithm, with the following parameters at particle initialization:

[0182] Initial flow rate: 3.2 meters / second;

[0183] Initial flow direction: South 15° East;

[0184] Residence time estimate: 12 minutes.

[0185] At 15:40, the particle swarm algorithm completes the 10th iteration optimization, obtaining the global optimal solution of the particle swarm. The prediction result shows that the flood will submerge the low-lying area of Village A at 16:10, with an estimated submergence area of 2.5 square kilometers and water depth exceeding 1.2 meters.

[0186] At 16:00, rainfall monitoring point RS-03 sends an abnormal signal, with rainfall intensity suddenly increasing to 55 mm / hour. The system detects abnormal water level fluctuations in sub-region R4 (downstream of the dam) through an adaptive grouping strategy, automatically adjusts the size and algorithm parameters of the particle swarm in R4, and re-simulates the propagation path of the flood in the low-lying area. The result shows that the flood flow rate increases from 3.5 meters / second to 4.1 meters / second, and is expected to threaten the northern area of Village B before 16:20.

[0187] Based on the prediction path and affected area map generated by the system, the flood control department immediately activates the emergency evacuation plan for Village A and Village B, and mobilizes resources to reinforce Bridge B.

[0188] During the entire flood propagation process, the invention is compared and analyzed with the traditional HEC-RAS numerical model, and the following Table 1 is the comparison of key indicators:

[0189] Table 1 Comparison of key indicators in flood path prediction between the method of the invention and the traditional method

[0190]

[0191]

[0192] It can be seen from the comparison that the application not only significantly shortens the processing time and simulation time, but also performs excellently in the accuracy of the propagation path and the accuracy of the flooded area prediction, and the prediction map finally generated by the system at 16:30 is basically consistent with the distribution of the actual flooded area, which wins valuable response time for the flood control department, successfully avoids the threat of thousands of people due to floods, and reduces the facility loss in the region by about 30%.

[0193] The embodiment verifies the superior performance of the application under complex dynamic flood propagation conditions, and has significant advantages in real-time performance, refinement and decision support.

[0194] The application fuses topographic feature data, hydrological station water level data, meteorological data and satellite remote sensing image data of multiple sources and heterogeneous information in the particle swarm initialization and iterative optimization process, and differentiates the target region into sub-regions based on the extraction of key propagation paths, so that independent iteration and dynamic adjustment of algorithm parameters can be performed for the local characteristics of each sub-region, and in the case of different hydrological conditions or sudden increase of upstream inflow, the adaptive grouping mechanism can quickly locate the affected areas and optimize the particle swarm search, so that the real-time flood path prediction can not only consider the whole, but also achieve higher accuracy in local areas.

[0195] The application introduces an improved graph neural network on the basis of traditional physical models and empirical models to deeply mine the mutual relationship between historical flood data, topography and meteorological conditions, accurately captures the main path of flood propagation through key node extraction and multi-relation adjacency matrix construction, and at the same time, the adaptive grouping strategy dynamically adjusts the particle swarm size and parameters in real time according to the changes of key monitoring indicators in the sub-regions during the particle swarm iteration process, avoiding the shortcomings of single particle swarm in different hydrological scenarios, such as being easily trapped in local optimum or lacking adaptability, and realizing more flexible and efficient prediction of the flood propagation process with high spatio-temporal dynamics.

[0196] After refining and screening the key propagation paths, the application performs independent particle swarm iterative search on each local area based on sub-region division, and merges the optimal solutions of each sub-region on a global level, so as to realize comprehensive control of the overall flood propagation situation, and continuously correct the flood flow rate, flow direction and residence time prediction index under the real-time update of multi-source data, so that the flood control command department has more initiative and pertinence when dealing with sudden disasters, and at the same time, the dynamic particle swarm optimization at the sub-region level also provides higher reliability for accurately identifying flood-prone points and endangered areas, and improves the accuracy of flood control and drainage scheme planning and early warning information release. <000065

[0197] The above description is only the preferred embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can make equivalent substitutions or changes according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A flood path prediction method based on particle swarm adaptive clustering, characterized in that, The method comprises the following steps: S1. Collecting multi-source heterogeneous data and preprocessing the multi-source heterogeneous data; S2. Constructing a graph neural network model based on the multi-source heterogeneous data, taking regional features in historical flood data as nodes, and taking interaction relationships formed by topographic features and meteorological data as edges; S3. Using the graph neural network model to predict the key propagation path of the target region and output initial reference information of the key propagation path; S4. Dividing the target region into sub-regions according to the initial reference information and determining the boundary range of each sub-region; S5. For each sub-region, initializing a particle swarm and setting the flood propagation path parameters of the particles, so that each particle represents the flow velocity, flow direction or residence time parameters related to the flood propagation in the sub-region; S6. Iteratively optimizing the particle swarm based on an adaptive grouping strategy, calculating the fitness value and dynamically adjusting the particle swarm size and algorithm parameters in the iteration process, and using the local data in the sub-region to real-time correct the search direction and speed of the particles; S7. When receiving new multi-source heterogeneous data, inputting the new multi-source heterogeneous data into the graph neural network model to update the prediction result of the key propagation path, and adjusting the sub-region division and particle swarm parameters accordingly; S8. Integrating the particle swarm optimization results of each sub-region to form the flood path prediction result of the target region as a whole.

2. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, wherein, The S1 comprises the following steps: S11. Collect historical flood data D h including flood propagation path, peak flow and duration of the i-th historical flood event; S12. Obtain terrain feature data D t including the slope, aspect and surface roughness of the i-th geographical unit; S13. Collecting meteorological data D m including the rainfall intensity, wind speed and temperature at the i-th time in the target area; S14. Obtain hydrological station water level data D w , including the water depth, flow rate and flow of the i-th hydrological station; S15. Obtain satellite remote sensing image data D r , containing the water body distribution, vegetation coverage and inundation range in the target area of the i-th satellite image; S16. Standardizing the multi-source heterogeneous data obtained in S11-S15 to construct a unified multi-source heterogeneous data set D: wherein, denotes the set of i-th flood event in historical flood data, d hi contains information of i-th flood event, totaling n flood events, denotes the set of terrain feature data, g j is the terrain feature of j-th geographical unit, totaling m geographical units, denotes the set of meteorological data, r k is the meteorological data of k-th time point, totaling p time points, denotes the set of hydrological station water level data, s l is the real-time water level data of l-th hydrological station, totaling q hydrological stations, denotes the set of satellite remote sensing image data, v o is the o-th satellite remote sensing image data, totaling k images, and ∪ denotes the union operator of sets, used to collect individual elements in each type of data. S17. Map all data in the multi-source heterogeneous dataset D to the interval [0, 1] by linear normalization, and use time interpolation method to complete the missing data at time points, and use spatial interpolation method to supplement the missing spatial data in the region, to obtain the preprocessed multi-source heterogeneous dataset D c .

3. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, characterized in that, The S2 comprises the following steps: S21. From the pre-processed multi-source heterogeneous dataset D c Among them, the historical flood data D and the hydrological station water level data in the target area are selected as the node feature sources together to construct the node set N = {n1, n2, …, n u}, wherein n i ,i∈{1,2,...,u} represents the historical flood event features and the corresponding hydrological station water level features contained in the i th node. S22. Combine the terrain feature data, weather data, and satellite remote sensing image data in the multi-source heterogeneous data set D c to construct a multi-relation adjacency matrix where R represents the number of different relation types, illustrating the node connectivity under the rth relation type, to construct an edge set E = {e ij}; S23. Define the node feature matrix where u represents the number of nodes, and f represents the joint feature dimension contained by each node. The features of each node are composed of the historical flood event features, the hydrological station water level data features, and the local geographical and meteorological attributes corresponding to the node. S24. Constructing a graph neural network model GNN, introducing a multi-head attention mechanism to weight and aggregate different relationship types based on the multi-relationship adjacency matrix and the node feature matrix X, and defining the node feature update as: where H (l) denotes the node feature matrix of the l-th layer of the graph neural network, initially H (0) = X, W (l) is the trainable weight matrix of the l-th layer, and a (m,r) is the weight coefficient of the m-th attention head under the r-th relation, M denotes the number of attention heads, σ denotes the activation function, and ∥ denotes the concatenation operation.

4. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 3, characterized in that, The construction of the edge set is based on the following multi-source relationships: Using the slope and slope direction in the topographic feature data to identify the spatial adjacency between nodes; Using the rainfall intensity or wind speed in the meteorological data to identify the time correlation between nodes; Using the regional inundation range in the satellite remote sensing image data to identify the flood influence correlation degree between nodes; Comprehensive water level data in the hydrological station water depth and flow information to calculate the hydrodynamic coupling strength between nodes.

5. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, wherein, The S3 comprises the following steps: S31. Calculate the node feature matrix H according to the graph neural network model (L) wherein L represents the total number of layers of the graph neural network, the node feature matrix H (L) represents the importance and dynamic characteristics of each node in the flood propagation network under the comprehensive conditions of multi-source heterogeneous data; S32. Based on the node feature matrix H (L) For each node n in the target region i Calculate its node importance score I(n i ), to assess the criticality of the node in the flood propagation process: wherein, denotes the node n i under the rth type of relationship, j the flood influence weight of the node n denotes the flood propagation feature of the node n i in the Lth layer, R represents the number of multi-source relationship types, and u is the total number of nodes. S33. According to the node importance score I(n i ) in descending order, select the top K nodes with the highest scores to form a set of key nodes N k ; S34. Extract the node features of each node in the key node set N k and the connection relationship in the multi-relationship adjacency matrix, identify the flood propagation path between the key nodes by traversing the adjacent nodes, and generate a preliminary propagation path set:​ P = {p1, p2,..., pn} ; and M} ; where p M represents the Mth propagation path, containing the sequence of nodes and the flow or water level change between nodes in the path; S35. For each path p in the preliminary set of propagation paths P j The path weight W(p j ) is computed, which quantifies the priority of the path in the propagation of the flood: Among them, I(n ki ) is the path p j The importance score of the node reflects the influence of the node in flood propagation, w ij is the edge e in the path ij The weight of represents the intensity of flood propagation between nodes; S36. Sort the preliminary propagation path set P according to the path weight W(p j ), select the top M' paths with the highest weights to form the key propagation path set P k , which represents the most likely flood propagation paths in the target area. P k = {p k1 , p k2 ,..., p kM′}.

6. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, wherein, The S4 comprises the following steps: S41. According to the key propagation path set P k Extract the node set N on each path k And the connection relationship set E k Where the node set N k Represents the flood propagation area node on the key propagation path, and the connection relationship set E k Represents the flood propagation association between nodes; S42. Determine the key propagation path set P based on the initial partitioning of the target region k Compute the node cluster centers C = {c1, c2, …, c q} within the target region, where q denotes the number of sub-regions of the initial partitioning, and each cluster center c q is determined by wherein w j is the importance weight of node n j , reflecting the key degree of the node in the flood propagation, p j represents the spatial position vector of node n j ; S43. A set of nodes N in the target region is selected k The space is divided, and each node n j is assigned to the nearest cluster center c i The sub-region division is completed, and the set of sub-regions is denoted as R: wherein R i represents the i-th sub-region, containing all nodes closest to the cluster center c i ; p j represents the i-th node in the sub-region; and i represents the Euclidean distance between node n j and cluster center c i . S44. According to the sub-region set R, combine the connection relationship set E between nodes k Further optimize the boundary range of each sub-region R i Calculate the boundary function B of each sub-region i : where B i represents the boundary of the i-th sub-region, containing all edges e jk connected to other sub-regions.

7. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, wherein, The S5 comprises the following steps: S51. For each sub-region R i Set the particle swarm size P for each sub-region i Within each sub-region R i Construct a particle swarm set S52. In the sub-region R i define the parameter vector of the particle according to the search space determined by the flow velocity, flow direction and residence time parameters related to the flood propagation represent the initial position of the jth particle: wherein, represents the initial flow velocity of the jth particle in the sub-region R i reflecting the size of the flood flow speed, represents the initial flow direction angle, representing the propagation direction of the flood, represents the initial residence time, embodying the residence or stay characteristics of the flood in the sub-region; S53. Parameter vector for each particle Random initialization is performed to uniformly distribute the particle swarm in the feasible range, and the random initialization is as follows: where d e {1, 2, 3} corresponds to the flow rate, flow direction and residence time respectively, with min d and max d represent the minimum and maximum feasible values of the dth parameter respectively, and rand(0, 1) represents a random number in the interval [0, 1]. S54. For each particle Setting initial velocity vectors Let the initial velocity components in each dimension be subject to a random or zero value distribution, and initialize the particle swarm set G i Stored in the particle swarm database of the sub-region R i ; S55. output each sub-region R i The initial position and initial velocity of the inner particle group are used to represent the flood propagation path parameters of the sub-region, and the particle group initialization of each sub-region is completed.

8. The flood routing method based on adaptive swarm grouping of particle swarm according to claim 1, wherein, The S6 comprises the following steps: S61. On the basis of the initialized particle swarm set G i , define the upper limit of iteration number for each sub-region R i , and execute the particle swarm optimization process based on the adaptive swarm strategy. S62. At each iteration, the fitness value of the particle is calculated S63. The particle with the highest fitness value is selected as the best particle The fitness value reflects the performance of the particle in terms of accuracy and stability in predicting the flood propagation path, and the fitness function is defined as: where D i represents the local data within the sub-region R i , including topographic features, meteorological data, and water level information of hydrological stations, for measuring the fitting degree of the current particle in the flood propagation process in the sub-region. S63. The individual optimal position of the particle is updated according to the fitness value The individual optimal position of the particle is updated according to the fitness value The individual optimal position of the particle is updated according to the fitness value i The individual optimal position of the particle is updated according to the fitness value S64. The velocity of the particle is updated with the position, the velocity update being: ​ Wherein, ω is the inertia weight, balancing the exploration and exploitation capabilities, c1 and c2 are learning factors, rand1(0,1) and rand2(0,1) represent random numbers in the interval [0,1]. S65. In the iterative process, local data changes in the sub-region are detected in real time. If significant changes in key monitoring indicators occur, the particle swarm size and algorithm parameters ω, c1, c2 are dynamically adjusted according to the adaptive grouping strategy, and the search direction and speed of the particle are corrected, and part of the particle swarm is reinitialized or merged to meet the dynamic propagation conditions of the flood. S66. Repeat the iteration steps S61-S65 until reaching an upper limit of iteration number or satisfying a convergence condition, output the sub-region R i a set of optimal positions of inner particles {gbest i}, representing a set of particle solutions with the highest fitness in the flood propagation path prediction in the sub-region.

Images