Intelligent Water Level Control Method and System for Flood Detention Areas Based on Machine Learning

Abstract

This invention relates to the field of intelligent regulation technology, specifically to a machine learning-based intelligent water level regulation method and system for flood storage and detention areas. The method includes the following steps: acquiring remote sensing images and water level records; extracting evolution features; combining change samples and structural maps; identifying response and response numbers; configuring node relationships according to path sequence; adjusting the action connection sequence and number position; and obtaining an intelligent regulation area sequence. In this invention, by combining image number sequences with temporal change features, vegetation, water bodies, and wetland evolution trends are extracted to construct change samples. Spatial contour directions and differences are compared to form a coherent structure. Response paths are generated by combining numbers and action states. Regulation sequences are configured based on connection relationships, enhancing the correspondence between instructions and regional states, optimizing the connection sequence between actions, avoiding path jumps or response omissions, and improving the ability of regulation content to follow the evolution structure and the coherence of number arrangement.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent regulation and control, in particular to an intelligent regulation and control method and system for water level of a flood storage and detention area based on machine learning. BACKGROUND

[0002] The technical field of intelligent regulation and control includes a comprehensive control method for sensing, analyzing, judging and scheduling a complex system in a dynamic environment. Based on data driving, the technical field integrates a sensing network, intelligent algorithms and control models to implement whole-process sensing, fine modeling and efficient regulation and control of a target system. The core content includes spatiotemporal fusion analysis of multi-source heterogeneous data, optimization modeling and calculation oriented to a target state, intelligent generation of control instructions and a feedback mechanism. In the fields of water conservancy, energy, transportation and ecological environment, intelligent regulation and control is widely applied to parameter adjustment and operation strategy formulation in high-dynamic and high-coupling scenarios, thereby improving the scientificity of resource allocation and the stability of system operation. In the management of a flood storage and detention area, intelligent regulation and control can effectively cope with the coordination problem between variable hydrological conditions and ecological requirements, and provide a basic support for the coordinated improvement of flood storage capacity and ecological function.

[0003] The intelligent regulation and control method for water level of a flood storage and detention area based on machine learning refers to modeling and learning of multi-source monitoring data of the flood storage and detention area by using a machine learning model, water level prediction for different rainfall intensities, upstream water flow and historical storage and detention conditions and the like input factors, and generation of precise regulation and control instructions for guiding flood storage and regulation operation. The technical matters targeted by the method include feature extraction of historical flooding information, law modeling of flood evolution process, parameterization expression and dynamic generation of regulation and control rules. The method mainly collects multi-scale spatiotemporal data in a combination mode of remote sensing monitoring and ground sensing, predicts the water level change trend by constructing a long short-term memory network model, improves the data processing accuracy by using a wavelet transform-based noise filtering technology, and applies a genetic algorithm to optimize the regulation and control parameter configuration strategy, so as to realize the prediction and control instruction generation of water level changes at different time periods. The implementation basis of the method includes construction of an ecological hydrological database, training of a dynamic regulation and control rule base and intelligent decision-making program of a regulation and control output scheme.

[0004] The prior art has limitations in the expression of data sample structure. Only the sensor monitoring value is used to construct a prediction model. In the case that regional image information cannot form a time series structure, it is difficult to depict the state evolution track of the numbered region. In the case that the spatial distribution is limited or the regional response is uneven, the model is difficult to output a regulation and control sequence with complete path relationship, which may cause problems such as numbered misplacement, missed action or interruption of local regulation and control instructions. In the actual flood evolution rapid expansion or vegetation degradation acceleration scenario, the system regulation and control response capability decreases, and there are adverse effects such as response node imbalance and regional action lag. SUMMARY

[0005] To solve the technical problems existing in the prior art, the embodiment of the present application provides a machine learning-based intelligent regulation and control method for water level of a flood storage and detention area, comprising the following steps:

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical scheme: a machine learning-based intelligent regulation and control method for water level of a flood storage and detention area, comprising the following steps:

[0007] S1: Obtain a remote sensing image of the flood storage and detention area, a water level monitoring record, and a region number corresponding table, extract vegetation, water body, and wetting change content, align each period of features according to image number order, combine the number corresponding features according to time, and obtain a region change combined sample group;

[0008] S2: Based on the region change combined sample group, compare the boundary expansion direction, wetting continuity, and degradation trend, combine the contour continuous direction consistent change content, compare the spatial difference trend, and obtain an evolution structure diagram group;

[0009] S3: Based on the evolution structure diagram group, compare the diagram group number action state, classify the associated regulation and control information numbers into a response region, and configure them to path nodes according to number order to obtain a regulation and control response associated region;

[0010] S4: Based on the regulation and control response associated region, extract the response number order and spatial connection relationship, embed the to-be-responded number into the response path, adjust the action connection position according to the connection point, and obtain a regulation and control deployment structure chain;

[0011] S5: Based on the regulation and control deployment structure chain, extract the to-be-responded number and access position, map the action node sequence to the path, adjust the number order and corresponding action according to the path structure, and obtain a to-be-executed intelligent regulation and control region sequence.

[0012] As a further scheme of the present application, the region change combined sample group includes vegetation distribution change features, water body range change features, and wetting expansion change features, the evolution structure diagram group includes boundary expansion direction consistency, wetting region continuity, vegetation degradation trend direction consistency, and spatial contour continuity, the regulation and control response associated region includes response region numbers, to-be-responded region numbers, and diagram group arrangement position nodes, the regulation and control deployment structure chain includes action sequence, spatial position relationship, and number front and back connection relationship, and the to-be-executed intelligent regulation and control region sequence includes to-be-responded region numbers, front and back action node positions, action sequence ranking, and output node combination content.

[0013] As a further scheme of the present application, the specific steps of S1 are as follows:

[0014] S101: Obtain the remote sensing image of the flood storage area, the water level monitoring record and the region number corresponding table, correspond to the image layer number at each time period, arrange the change information of the vegetation edge, the water body boundary and the wetting range in time sequence, and obtain the image segment mapping data set;

[0015] S102: Based on the image segment mapping data set, the associated positions between the vegetation expansion zone, the water body change zone and the wetting range in the layer are analyzed in the order of numbering, the continuous change direction and the connected track are identified, the change relationship corresponding to the numbering is analyzed, and a numbering change direction set is obtained.

[0016] S103: Based on the numbering change direction set, the change path corresponding to the same numbering at different times is sequentially connected, the direction corresponding relationship and the structure adjustment process between the element change trend in the numbering region are analyzed, and a regional change combination sample group is obtained.

[0017] As a further scheme of the present application, the specific steps of S2 are:

[0018] S201: Based on the regional change combination sample group, the boundary expansion track, the wetting area water surface extension form and the vegetation index change direction of each sample are called, the sample regions are paired according to the image numbering order, and the consistency of the boundary advancing direction and the outer edge trend of the adjacent region is judged, and a space connection contour sequence is obtained.

[0019] S202: Based on the space connection contour sequence, the included angle amplitude between the vegetation index change angles is calculated, the numbering of the included angle amplitude keeping the same section is screened, the numbering order is arranged according to the numbering adjacency relationship, and a contour convergence path set is obtained.

[0020] S203: Based on the contour convergence path set, it is judged whether the change trend direction of the continuous sample boundary line movement and the vegetation value fluctuation in the region is consistent, the sample numbering arrangement group with uniform change trend direction is selected, and an evolution structure diagram group is obtained.

[0021] As a further scheme of the present application, the calculation formula of the included angle amplitude between the vegetation index change angles is specifically:

[0022] ;

[0023] Wherein, represents the included angle amplitude between the vegetation index change angles of the numbering , represents the number of sampling nodes in the section , represents the vegetation change direction angle of the th node in the th section, represents the segment the node in the the vegetation index observation value of the time layer, represent the segment the node in the the vegetation feature quantity observation value of the time layer, represent the image disturbance error term in the time sequence layer, represent the observation quality value, represent the disturbance intensity of the node in the spatial disturbance dimension, represent the disturbance response value, represent the disturbance bias correction value, represent the upper limit of the total layer number of the image observation time sequence, represent the upper limit of the total number of disturbance response sources.

[0024] As a further scheme of the present application, the specific steps of S3 are:

[0025] S301: Based on the evolution structure diagram set, obtain the graph set number and the corresponding control action state record, identify the number with the control signal identifier, judge whether the signal corresponds to the control item, filter the numbers with the associated actions, and obtain the action response number set;

[0026] S302: Based on the action response number set, extract the spatial boundary point sequence of the corresponding graph set, calculate the angle change rate of the adjacent boundary line segment, filter the numbers with the same angle change trend according to the gradient distribution of the change rate, and obtain the to-be-responded number set;

[0027] S303: Based on the number order in the action response number set and the to-be-responded number set and the path connection order in the graph set, judge the positional relationship of the consecutive numbers in the graph set chain, match the corresponding path segment number, and obtain the control response correlation area.

[0028] As a further scheme of the present application, the calculation formula of the angle change rate of the adjacent boundary line segment is specifically:

[0029] ;

[0030] wherein, represent the angle change rate of the adjacent boundary line segment, represent the angle value between the boundary line segment and the image horizontal reference axis, represent the angle value between the boundary line segment and the image horizontal reference axis, represent the angle value between the Length of the strip boundary line segment, representing the first Length of the strip boundary line segment, representing the first Angle adjustment weight at the first boundary point, representing the first Local gradient value of the first boundary point under the first image feature channel, representing the first Local gradient value of the first boundary point under the first image feature channel, representing the first Normalizing weight value representing the gradient difference influence degree in the first feature channel, representing the first Normalizing weight value representing the gradient difference influence degree in the first feature channel, representing the number of introduced image feature channels.

[0031] As a further scheme of the present application, the specific steps of S4 are:

[0032] S401: Based on the regulation response associated area, identify the number of associated action information and the number of structure characteristics with coincident but without action information, respectively corresponding to the response area and the to-be-responded area, and obtain the number partition set;

[0033] S402: Based on the number partition set, extract the position relationship between adjacent response numbers in order, identify the extension direction of the connection point in the spatial distribution, screen the number combination of the connection position relationship, and obtain the connection sequence sequence group;

[0034] S403: Based on the connection sequence sequence group, insert the to-be-responded number into the connection node between the response number sequence, arrange the number order of each group according to the number before and after, position the number arrangement mode in the continuous segment, and obtain the regulation deployment structure chain.

[0035] As a further scheme of the present application, the specific steps of S5 are:

[0036] S501: Based on the regulation deployment structure chain, call out the number position of each to-be-responded area, retrieve the front and rear action node numbers connected, match the number and the corresponding logical position of the node in the path, and obtain the number connection relationship information;

[0037] S502: Based on the number connection relationship information, identify the path corresponding section between adjacent numbers, adjust the path arrangement order according to the action node distribution state, correct the node sorting rules in the path segment, and obtain the action path sorting information;

[0038] S503: Based on the action path sorting information, organize the action node combination content according to the path order, position the action sequence position of the to-be-responded number in the path structure, and obtain the to-be-executed intelligent regulation area sequence.

[0039] The machine learning-based intelligent regulation system for flood storage and detention basin water level comprises:

[0040] The image recognition module obtains remote sensing image videos, water level monitoring records and a region number corresponding table of the flood storage and detention basin, extracts time variation characteristics of vegetation distribution, water body range and wetting expansion in the numbered region, aligns the variation contents according to the image number sequence, combines the variation characteristics under the same number into a variation content group to obtain a region variation combination sample group;

[0041] The change analysis module compares the boundary expansion direction, the wetting area continuity and the vegetation degradation trend change direction based on the region variation combination sample group, combines the contents with consistent change direction and spatial contour into the same graph group, analyzes the relevance of the graph group through the corresponding relationship between the boundary continuous movement track and the difference direction change in the region, and obtains an evolution structure graph group;

[0042] The response extraction module compares the regulation action state corresponding to the graph group number based on the evolution structure graph group, marks the numbers with associated action information as response regions, marks the numbers with the same structural characteristics but without action information as to-be-responded regions, matches the numbers to the position nodes in the regulation path according to the arrangement of the graph group, and obtains a regulation response associated region;

[0043] The path sorting module selects the response regions in adjacent number sequence based on the regulation response associated region, extracts the action sequence and spatial position relationship, inserts the to-be-responded regions between the response sequences, matches the action between the numbers according to the connection point combination sequence of the upper and lower regions, obtains a regulation deployment structure chain, and delivers the regulation deployment structure chain to the regulation sorting module;

[0044] The regulation sorting module adjusts the number position of each to-be-responded region and the connected front and rear action nodes based on the regulation deployment structure chain, puts the action position into the corresponding section of the adjacent path, adjusts the action sequence ranking of the number in the structure according to the path sequence, combines the contents according to the action path output node, and obtains a to-be-executed intelligent regulation region sequence.

[0045] Compared with the prior art, the advantages and positive effects of the present application are that:

[0046] In the present application, the time sequence variation characteristics are combined through the image number sequence, the vegetation, water body and wetting evolution trend are extracted to construct the change sample, the coherent structure is formed by comparing the spatial contour direction and the difference trend, the response path is generated by combining the number and the action state, the regulation sequence is configured according to the connection relationship, the corresponding relationship between the instruction and the region state is enhanced, the connection sequence between the actions is optimized, the path jump or response omission is avoided, and the following ability of the regulation content to the evolution structure and the coherence of the number arrangement are improved. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative effort.

[0048] Figure 1 The step flowchart of the present application is shown in the figure.

[0049] Figure 2 The system module diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0050] The technical solutions in the present application will be described in detail below with reference to the drawings.

[0051] In the embodiments of the present application, the words such as “example”, “for example” are used to represent as an example, illustration or description. Any embodiment or design scheme described as “example” in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the word “example” is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by “and / or” can be both, or can be one of the two.

[0052] In the embodiments of the present application, “image” and “picture” can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. “Of”, “corresponding” and “relevant” can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0053] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0054] In order to make the technical problems, technical solutions and advantages of the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0055] Please refer to Figure 1 The embodiments of the present application provide a machine learning-based intelligent control method for water level of a flood storage area, which comprises the following steps:

[0056] S1: Obtain remote sensing image of the flood storage area, water level monitoring record and regional number corresponding table, extract the vegetation distribution, water body range and wetting expansion time variation characteristics in the numbered region, align the change content according to the image number order, combine the change characteristics under the same number into a change content group, and obtain a regional change combination sample group;

[0057] S2: Based on the regional change combination sample group, compare the boundary expansion direction, wetting area continuity and vegetation degradation trend change direction, combine the contents with consistent change direction and spatial contour continuity into the same graph group, judge the graph group relevance through the corresponding relationship between the boundary continuous moving track and the difference direction change in the region, and obtain the evolution structure graph group;

[0058] S3: Based on the evolution structure graph group, compare the control action state corresponding to the graph group number, mark the number with associated action information as a response area, mark the number without action information but with the same structure characteristics as a to-be-responded area, match the number to the position node in the control path according to the graph group arrangement, and obtain the control response associated area;

[0059] S4: Based on the control response associated area, select the response area of adjacent number sequence, extract the action sequence and spatial position relationship, insert the to-be-responded area between the response sequence, match the action between the numbers according to the connection point combination sequence of the upper and lower areas, and obtain the control deployment structure chain;

[0060] S5: Based on the control deployment structure chain, call out the number position of each to-be-responded area and the connected front and rear action nodes, include the action position in the corresponding section of the adjacent path, adjust the action sequence ranking of the number in the structure according to the path sequence, combine the contents according to the action path output node, and obtain the to-be-executed intelligent control area sequence.

[0061] The regional change combination sample group includes vegetation distribution change characteristics, water body range change characteristics and wetting expansion change characteristics. The evolution structure graph group includes boundary expansion direction consistency, wetting area continuity, vegetation degradation trend direction consistency and spatial contour continuity. The control response associated area includes response area number, to-be-responded area number and graph group arrangement position node. The control deployment structure chain includes action sequence, spatial position relationship and number front and rear connection relationship. The to-be-executed intelligent control area sequence includes to-be-responded area number, front and rear action node position, action sequence ranking and output node combination content.

[0062] The specific steps of S1 are:

[0063] S101: Obtain the remote sensing image of the flood storage area, the water level monitoring record, and the corresponding table of the region number, correspond to the image layer number at each time period, arrange the change information of the vegetation edge, the water body boundary, and the wetting range in time sequence, and obtain the image segment mapping data set;

[0064] First, the remote sensing image is extracted according to the time sequence order of each stage layer number content, the vegetation edge extraction operation is performed on all numbered regions in each image, the gray value change characteristics in the gray image are used, the pixel combination whose gray difference between continuous pixels is more than 25 is detected as the vegetation edge segment and recorded as the edge unit, the water body boundary extraction is based on the low reflectivity of the near-infrared band, by setting the reflectivity threshold to 0.1, all pixel regions below the value are implemented continuous tracking and given a unique identifier, the continuous labeling of the water body boundary is realized, the identification of the wetting area is based on the sensitive response of the NDWI value, the pixel with NDWI value greater than 0.3 is selected as the wetting response unit, the index coordinate set of the above three types of boundaries is established in the image coordinate, the region number is uniformly labeled, and the layer number of each numbered region is kept consistent at the same time period, to realize the unified standard based on the comparison of cross-period data, then the position data of the regions with the same number at different time points is aligned, the coordinate value set of the vegetation boundary line, the water body boundary line and the wetting boundary line in different layers is extracted, and the point-to-point or segment-to-segment boundary displacement matching is performed in the coordinate index space, for the boundary offset distance in the matching result, the position expansion rate is calculated combined with the corresponding time interval, if the rate is greater than 2 pixel unit length, it is judged that the region has boundary expansion, if the rate is lower than the threshold, it is identified as a boundary contraction region, the boundary change information of all region numbers is arranged in order of layer number, and the region number, data source type (vegetation boundary, water body boundary, wetting area boundary) and threshold value according to which the extraction method is based are marked in turn, and finally the image segment mapping data set is obtained.

[0065] S102: Based on the image segment mapping data set, the associated positions of the vegetation expansion zone, the water body change zone and the wetting range in the layer are analyzed according to the number order, the continuous change direction and the connected track are identified, the change relationship of the corresponding number is analyzed, and the number change direction set is obtained;

[0066] First, according to the numbering order, locate the pixel coordinate points of the vegetation boundary, water body boundary and wetting range in each layer, respectively extract the profile edge point coordinate set of the above three types of boundary lines for the adjacent time points of the continuous layers in the numbered area, call the time sequence comparison of the boundary change trajectory in the same numbered area, calculate the offset vector direction of the end pixel points of each type of boundary line according to the coordinate difference, judge whether the offset direction is within 10 degrees of the angle difference and has continuity, if the numbered area exists vegetation boundary extending to northeast, water body boundary expanding to northeast at the same time, and wetting range also increasing in the same direction, it is judged that the three types of changes in the numbered area have relevance in direction, record the direction as the key value, and mark the direction corresponding quadrant number in the record, for example, the water body boundary in the numbered area A is located at points (103, 205), (106, 208), (109, 211), (112, 214) in the four time slices, the direction offset vector is (3, 3), the direction angle is near 45 degrees, and the vector direction difference in the continuous layers is not more than 5 degrees, which can be classified as northeast direction trajectory, if the vegetation and wetting boundary in the same numbered area also meet the above offset judgment rule, the three directions have consistency, then track the change trajectory of each type of boundary in each numbered area according to the layer sequence, if there is an intersection point between the three types of boundary lines in the adjacent layers, the trajectory is continuous, it is considered that there is a connected path, the intersection point path meets the requirement that the end boundary in the adjacent layer and the starting boundary in the next layer are within the range of two pixel points, it is considered that the path is coherent, for example, the numbered area B, the starting point of the wetting range is (120, 180) in layer n, and (122, 181) in layer n+1, the distance between them is less than 3 pixels, it is judged that the trajectory is connected, finally, the contents of each numbered area that have completed the direction unification and path coherence judgment are unified and arranged, the direction set of each numbered area is output through the numbered matching method, and the numbered change direction set is obtained.

[0067] S103: Based on the numbered change direction set, sequentially connect the change paths corresponding to the same number at different time, analyze the direction corresponding relationship and structure adjustment process between the element change trends in the numbered area, and obtain a region change combination sample group;

[0068] Firstly, the direction sequence corresponding to each number in the set and the timestamp index are called, and the change trajectory image data of the number area in each time layer is arranged in time from early to late order, the spatial transfer path and direction offset of each type of boundary element in the layer are identified, for example, the wetting boundary line end points of number D in four time slices are (110, 203), (113, 206), (117, 209), (120, 212), and its direction trajectory is a coherent sequence towards the northeast direction, the boundary points are connected to form a path segment in sequence using the layer order, each segment of the change path is classified into the change trajectory chain of number D, and it is judged whether the vegetation expansion zone also extends in the same direction, if the boundary points of the vegetation expansion also satisfy the above direction, it is considered that the path directions are consistent, the three element paths are labeled respectively in the number dimension, after connection, the pixel position offset value of each boundary element in the adjacent time slice is called and matched with its direction vector, it is judged whether the angle between the water body vector and the vegetation vector is less than 30 degrees, and the direction deviation is less than 15 degrees, if it is satisfied, it is considered that the change trends of the two are consistent, then it is judged whether the wetting boundary maintains a central trend with an angle less than 20 degrees between the two types of paths, if it is satisfied, it is considered that the three elements constitute a trend equal relationship, the change relationship of the number area in each time slice is recorded as the direction relationship chain of the number, then the intersection and change amplitude of the three types of paths in space are tracked according to the direction relationship chain, the spatial distance change value between each pair of adjacent boundaries is recorded according to the number, for example, the distance between the vegetation boundary and the wetting boundary of number D in layer n is 12 pixels, and in layer n+1 it is 15 pixels, so the change value is +3, the change value is classified as a structure adjustment trajectory according to the time axis, finally, the direction relationship chain and the structure adjustment trajectory under each number are aggregated into a sample unit according to the number as an index, and a regional change combination sample group is obtained.

[0069] The specific steps of S2 are:

[0070] S201: Based on the regional change combination sample group, the boundary expansion trajectory, the wetting area water surface extension form and the vegetation index change direction of each sample are called, the sample regions are paired in sequence according to the image number, and the consistency of the boundary advancing direction and the outer edge trend of the adjacent region is judged, and a spatial connection contour sequence is obtained.

[0071] First, the boundary expansion trajectory information corresponding to each number in the sample group is retrieved. Boundary coordinate data for each number at different time periods are retrieved sequentially from the image sequence, and a trajectory sequence set is established according to the number order. The boundary advancement direction vector in each trajectory is organized, and the water surface edge contour image of the wetted area is extracted accordingly. The morphological change points at different time periods are compared to determine whether the movement trend of the edge coordinates in the coordinate system during the water body extension process is consistent with the boundary expansion direction. Simultaneously, the change direction of the vegetation index NDVI in the same area at different time periods is extracted. Combined with the movement direction of its spatial distribution center point between layers, it is determined whether it is in the same direction as the above two directions. For example, if the center of the high NDVI value area in a certain numbered region moves from coordinates (50, 100) to (58, 108) in images 1 to 4, then its movement direction is... If the water body boundary and the vegetation center move in the northeast direction simultaneously, they are considered to be in the same direction. After the direction vectors of the three elements are sorted, the image numbers are called one by one and the sample pairing task is constructed in chronological order. It is determined whether the boundary advancement direction of the current numbered region in different images has continuity with the outer contour direction of its adjacent regions. The judgment is based on whether the angle between the boundary end point direction vector of the current number and the edge direction vector of its neighboring number is less than 25 degrees. For example, if the boundary advancement direction of number A is east-northeast and the edge direction of number B is also east-northeast with an angle of 15 degrees, they are considered to be in the same direction. The corresponding boundary paths of the regions with the same direction are connected end to end to construct a continuous spatial segment. A continuous boundary sequence is established by extending between the numbers in turn, and finally a spatial continuity contour sequence is obtained.

[0072] S202: Based on the spatial continuity contour sequence, calculate the angle between the angles of vegetation index change, filter the numbering of segments with similar angles, arrange the numbering order according to the numbering relationship, and obtain the contour convergence path set;

[0073] The formula for calculating the angle between the changes in the vegetation index is as follows:

[0074] ;

[0075] in, Representative number is The angle between the changes in the vegetation index, Representative section Number of internal sampling nodes Representing the The first section The direction and angle of vegetation change at each node Indicates the first Section No. Node at the Vegetation index observations over time layers, Indicates the first Section 1 Node 1 in Section 1 Vegetation feature quantity observation value of time layer, Representative image disturbance error term in time sequence layer, Representative observation quality value, Representative disturbance intensity of node 1 in Section 1 in spatial disturbance dimension, Representative disturbance response value, Representative disturbance bias correction value, Representative upper limit of total layer number of image observation time sequence, Representative upper limit of total item number of disturbance response source;

[0076] Hypothesis: (sample section , );

[0077] Node 1 ( ):

[0078] ;

[0079] , , , ;

[0080] , , , ;

[0081] , , ;

[0082] , , ;

[0083] Node 2 ( ):

[0084] ;

[0085] , , , ;

[0086] , , , ;

[0087] , 、 ;

[0088] 、 、 ;

[0089] Node level computation derivation process:

[0090] Node 1:

[0091] Numerator term:

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] Denominator term:

[0097] ;

[0098] ;

[0099] ;

[0100] Node 1 harmonic term:

[0101] ;

[0102] Node 2:

[0103] Numerator term:

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] Denominator term:

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] Node 2 Harmonization Item:

[0114] ;

[0115] Final segment result calculation:

[0116] ;

[0117] The results show that, under the combined effect of dual-time-series remote sensing vegetation index changes and topographic disturbance correction, the average harmonic amplitude of the vegetation change angle in the contour segment numbered 1 is 29.802 degrees. This value can be directly used as an indicator of angle convergence in the selection of map structure paths.

[0118] S203: Based on the contour convergence path set, determine whether the moving direction of the continuous sample boundary line is consistent with the changing trend of the vegetation numerical fluctuation in the region. Select the sample numbering of samples with the same changing trend direction and arrange them into a group of graphs to obtain the evolution structure graph group.

[0119] First, the sample number and its corresponding boundary contour coordinate set recorded in each contour path need to be called. The boundary change data of each sample is organized in ascending order of the numbers. The boundary line segment coordinate pairs under each number are extracted and a boundary line direction vector is constructed. The boundary vector group of consecutively numbered samples is called. For the boundary change direction between two adjacent numbers, the coordinate change trend in two-dimensional space is compared line by line. That is, based on the difference between the x and y values ​​of the boundary line endpoint and the starting point, it is determined whether the horizontal movement is positive (x value increases) or negative (x value decreases), and the vertical movement is upward (y value increases) or downward (y value decreases). Then, the boundary directions of two adjacent samples are compared to see if they are consistent by combining the x and y axis direction trends. If the signs of the x and y axis directions are the same, they are defined as consistent; otherwise, they are inconsistent. Simultaneously, within the same number... Under the specified number, the vegetation value, i.e., NDVI change data, is extracted in the region. Based on the image capture time series, the difference between the mean NDVI values ​​of two consecutive periods is obtained to determine whether the vegetation change direction is increasing (positive difference), decreasing (negative difference), or stable (absolute difference less than 0.01). Then, it is combined with the boundary line direction to determine whether it is consistent. That is, if the boundary expands outward and the NDVI value increases positively, the boundary contracts and the NDVI value increases negatively, or the boundary is stable and the NDVI does not change significantly, it is defined as a consistent direction. Otherwise, it is inconsistent direction. At the same time, the sample numbers of all consistent directions are extracted to form a candidate set, and their regional location map in the two-dimensional image plane is drawn. They are combined into several numbered arrangement map groups, and the numbers in each group are arranged in chronological order to form an image evolution sequence. Finally, the evolution structure map group is obtained.

[0120] The specific steps for S3 are as follows:

[0121] S301: Obtain the group number and the corresponding control action state record based on the evolution structure diagram group, identify the number with the control signal identifier, judge whether the signal corresponds to the control item, filter the numbers with associated actions, and obtain the action response number set;

[0122] First, the control action state record information corresponding to each number in the arrangement order of the group number is extracted one by one, the action label, signal trigger identifier, and action trigger time associated with the number are obtained, and a mapping record table of number and action state is established. It is identified whether there is an explicit control signal identifier in the record, that is, whether the signal trigger item has a valid bit state. If it exists, the number is marked as a control signal identifier number. Then, in the identified control signal number, the matching between the action label and the current control item is judged one by one, that is, the operation category code in the current control item list is matched with the action code in the number action state for consistency. If the two codes are the same or have a superior-inferior mapping relationship in the operation category, it is considered to have an action response relationship. Then, all numbers with the above matching relationship are filtered and collected to form a number set with associated actions. For example, in the action state corresponding to number A101, there is a control signal "TRG1" and the action category is "drainage". If the control item also has a drainage operation task, the number is identified as a response number. At the same time, if number A102 has a signal but the action category is "vegetation maintenance", and the control item does not have this category, it is excluded from the response number set. In this way, all group numbers are screened, and finally a number set that meets the action association condition is obtained, and the action response number set is obtained.

[0123] S302: Based on the action response number set, extract the space boundary point sequence of the corresponding group, calculate the angle change rate of the adjacent boundary line segment, and filter the numbers with the same angle change trend according to the gradient distribution of the change rate to obtain the to-be-responded number set.

[0124] The calculation formula of the angle change rate of the adjacent boundary line segment is as follows:

[0125] ;

[0126] Wherein, represents the angle change rate of the th adjacent boundary line segment, represents the angle value between the th boundary line segment and the image horizontal reference axis, represents the angle value between the th boundary line segment and the image horizontal reference axis, represents the length of the th boundary line segment, represents the length of the th boundary line segment, representing the angle adjustment weight at the representing the local gradient value of the representing the local gradient value of the representing the normalized weight value of the gradient difference influence degree in the representing the number of introduced image feature channels;

[0127] Assume:

[0128] Operation process (assuming m=3):

[0129] radian, radian, then ;

[0130] pixel, pixel, then ;

[0131] Local density: , the average of the whole image ;

[0132] ;

[0133] Gradient value:

[0134] Gray channel: ;

[0135] Sbel edge: ;

[0136] Texture energy: ;

[0137] Weight: , ;

[0138] Gradient weighted sum:

[0139] ;

[0140] ;

[0141] Denominator: ;

[0142] Final calculation:

[0143] ​​​​​​ ;

[0144] The results show that the weighted angle change rate of adjacent boundary segments in structurally complex regions is 4.18, indicating that there is a trend of drastic change in direction at the boundary. This can be used as a basis for trend consistency aggregation screening in subsequent action response numbering and for inclusion in the set of response numbers.

[0145] S303: Based on the numbering order of the action response number set and the response number set and the path connection order in the graph group, determine the positional relationship of consecutive numbers in the graph group chain, match the corresponding path segment number, and obtain the control response associated area;

[0146] First, extract the path position of each number within the graph structure. Combined with the pre-defined path connection table or path node chain in the evolution structure graph group, locate the arrangement level of the number on the path. Then, check the path connection status between any two consecutive numbers to determine if there is a direct connection or an indirect connection via a node jump. If consecutive numbers are sequentially adjacent in the path chain, they are considered to have structural continuity. Next, based on the staggered distribution of response numbers and numbers to be responded to, compare the combination status between consecutive number pairs one by one, and select the path segment numbers composed of response numbers and numbers to be responded to. Then, determine the start and end positions of the numbers involved in this path segment. The intermediate connection status is marked to reflect the coverage of the path segment in the overall path structure. In this process, if there are multiple nodes branching or path interruption in the path segment, the number pair is removed from the comparison set, and only the logically continuous and stable numbered path segments are retained. Then, all the path segment numbers that meet the conditions are indexed and registered, and all regional units within the number range they cover are extracted. Through the mapping relationship between the number and the regional table, the corresponding geographic block name and its control status are obtained to determine its regulation response attributes. Finally, a set of connection relationships between regulation response numbers in the path structure is generated to obtain the regulation response associated region.

[0147] The specific steps of S4 are as follows:

[0148] S401: Based on the control response association region, identify the number of the associated action information and the number of the structural feature-overlapping number but without action information, and assign them to the response region and the region to be responded to, respectively, to obtain the number partition set;

[0149] Extract the set of numbers associated with the response action and mark their location information. Delineate the numbered units with regulatory behavior characteristics in the map group. Combine the image attributes of each number to judge the coverage morphology, water body location and boundary shape in its region. Select numbers with response action trajectory characteristics. For this set of numbers, screen numbers with overlapping structural features but no response action identifiers from the image number structure, geographical coordinates and map group connection path. Compare the similarity of edge band composition, water body outline extension trend and average gradient of vegetation index item by item. Use the average comparison difference value not exceeding 0.2 as the judgment criterion to identify numbers with obvious regional feature overlap but no response action. Align the two types of numbers according to the number structure position in the map group and correspond them to the response area and the area to be responded to, respectively, to obtain the number partition set.

[0150] S402: Based on the numbered partition set, extract the positional relationship between adjacent response numbers in numerical order, identify the extension direction of the connection point in spatial distribution, screen the numbered combinations of the successive positional relationships, and obtain the connection order sequence group;

[0151] First, extract the response number combinations that are adjacent in the numbering sequence, and compare their spatial position difference with the relative position difference between the path number index under geographic coordinate reference. By calculating the direction of the vector connection line of the number position under two-dimensional projection, identify the extension trajectory presented by each number connection segment, and determine whether there is significant directional continuity. If the change value of the direction angle is less than 15 degrees, it is considered to be a consistent extension direction. Then, for all number combinations that meet the conditions, compare whether their boundary direction on the image layer has continuous alignment features, and combine the coverage of the area to which the number belongs at the edge of the wetted zone to determine its boundary connection density. In the example, numbers A3 to A6 have a continuous north-south extension trend and form a linear connection zone in the image group layer. The difference in boundary morphology between the successive positions is less than 3%, thus determining that it can be used as a valid connection sequence, and finally obtain the connection sequence group.

[0152] S403: Based on the connection sequence group, insert the number to be responded to into the connection node between the response number sequence, arrange the number order of each group according to the relationship between the numbers, locate the number arrangement method in the continuous segment, and obtain the control deployment structure chain.

[0153] First, after sorting the existing response numbers according to the map group order, the response numbers without action records are processed by insertion. For segments with positional gaps in the response number combination, the shortest geographical distance between the preceding and following numbers is selected to connect the nodes. Based on the frequency of the response number in the map group and the proximity of its spatial coordinates to the connection point, it is embedded into the corresponding connection segment. In specific execution, the connection priority needs to be determined according to the spatial position point corresponding to the number in the layer image. For example, if there is a gap between numbers B5 and B7, and number B6 is located in the middle of the line connecting B5 and B7 on the image coordinate point, and its wetted boundary extension direction is consistent with B5 and B7, then B6 is inserted into the connection node of that segment. Subsequently, the new sequence formed after insertion is sorted according to the timestamp of the number in the map group, thereby determining the internal numbering order of the connection segment in the sequence. Furthermore, all the adjusted numbering sequences are merged into continuous path segments, and finally the control deployment structure chain is obtained.

[0154] The specific steps of S5 are as follows:

[0155] S501: Based on the control and deployment structure chain, retrieve the number position of each region to be responded to, retrieve the numbers of the preceding and following action nodes of the connection, match the logical positions of the numbers and nodes in the path, and obtain the number connection relationship information.

[0156] First, the number sequence in the deployment chain is retrieved to locate the actual position of each response number in the sequence. Then, adjacent response numbers are extracted from the positions before and after the given number to obtain the start and end numbers of the connection direction. Next, the action node identifiers in the layer images corresponding to each number in the structural chain are called to read the actual records of water expansion, vegetation change, or boundary offset of the node in the image. These records are compared with the labeled data of the response number positions to determine whether a continuous path segment is formed. During the determination process, based on the image number order, it is determined whether the response number is adjacent to the left and right sides of the response number combination. For example, number X9 is located next to number X. If X9 is between X8 and X10, and the direction of the wetted boundary of X9 in the image is the same as that of X8 and X10, then X9 is marked as an internal point of the path as a connection node between X8 and X10. Otherwise, if the direction is opposite or the spatial distance exceeds twice the average spacing of the deployment chain, it is considered an external breakpoint of the path and the connection relationship is not marked for the time being. Then, all number pairs with logical order relationship are listed in the number correspondence table according to the image order. The corresponding predecessor and successor labels are marked between each pair of numbers. For example, (X6→X7) indicates the advancement connection between X6 and X7 in the path. Finally, the logical position pairing of all numbers and path segments is completed to obtain the number connection relationship information.

[0157] S502: Based on the number connection relationship information, identify the corresponding segments of the path between adjacent numbers, adjust the path arrangement order according to the distribution status of action nodes, correct the node sorting rules within the path segment, and obtain action path sorting information.

[0158] First, extract the start and end numbers of the path between each pair of adjacent numbers from the numbering connection table. Use these number pairs as the starting points for the initial path segment division. Next, retrieve the spatial positions indicated by the start and end numbers in the layer image. In terms of spatial layout, represent the actual line form from the start to the end of the path segment using line segments or boundary point sets. Further, use the image number nodes contained within the path segment as the set of action nodes within the segment, and read their positions, action types, and corresponding change direction identifiers on the layer. After constructing each path segment, check whether the position order of its number sequence is consistent with the change trend of the nodes on the layer. If the arrangement direction of the numbers in a certain path segment is reversed in the layer, then according to the distribution of action nodes in the image, rearrange the order of the numbers within the path segment in reverse or partially swap them. For example, if the number sequence is X7→X9→X8, after verification, it is found that X9 is in the layer... If the path is located between X8 and X7, and the corresponding vegetation change direction advances from X7 to X8, then X9 needs to be placed after X7 and adjusted to X7→X8→X9. Furthermore, each adjusted path segment is checked to ensure that nodes within the same path segment have a continuous progression relationship and consistent direction. If a single node is discontinuous, the segment is deemed invalid and deleted. A global numbered sequential linked list is then constructed based on the start and end pairs of numbers. Segments are connected in the linked list according to logical continuity. It is checked for duplicate or misaligned node numbers. If number X10 appears between two segments and its action direction is inconsistent, the segment with consistent direction is selected by comparing the action direction indicators of the preceding and following numbers. The other segment is replaced or discarded. Finally, a set of numbered path sequences without repetition or conflict and arranged in an ordered manner is formed, yielding the action path sorting information.

[0159] S503: Based on the action path sorting information, organize the action node combination content according to the path order, locate the action sequence position of the response number in the path structure, and obtain the intelligent control area sequence to be executed.

[0160] First, extract the corresponding action node numbers from the sorted data sequentially according to the path number, constructing an action node sequence linked list. For each node, read its position in the path and the region identity represented by its number. Establish a correspondence table for the actual control content, execution order, and position of the corresponding number in the sorted chain for each action node. During execution, starting from the first number in the sorted linked list, scan all node numbers in the list sequentially. Compare each node number with the set of numbers to be responded to to determine if it belongs to the target number. If it is a number to be responded to, record its relative position in the sorted chain. For example, if number B15 is a number to be responded to, located at the 4th position in the action path chain, and its preceding numbers B12, B13, and B14 are already responded to, then B15 is marked as the starting point for execution. Simultaneously, continue searching for numbers to be responded to. When multiple numbers to be responded to consecutively (e.g., B15 to B20) appear, they are determined to be... A set of consecutive response numbers is combined with the node sequences between them and the response numbers before and after them to clarify the context of the control path segment where B15 to B20 are located. At the same time, the action types before and after them are extracted to determine the control nature of the path segment in the whole path. Further, the action type of each response number is identified, such as vegetation wetting, soil recharge, water diversion, etc. By comparing the image attribute change trajectory of the area corresponding to the action type in the layer image, it is confirmed whether its response time sequence is consistent with the number path sequence. For example, if number B18 is a vegetation wetting node, and the direction of its path segment is southeastward, and the humidity index of the area indicated by this number in the layer is continuously increasing, then B18 can be included in the action sequence to be executed. Combined with the positional relationship of its preceding and following nodes, its relative position order in the action sequence to be executed is determined. Finally, a sequence of response numbers arranged by path is formed, and the sequence of intelligent control areas to be executed is obtained.

[0161] Please see Figure 2 A machine learning-based intelligent water level control system for flood storage and detention areas includes:

[0162] The image recognition module acquires remote sensing images of the flood storage and detention area, water level monitoring records and a correspondence table of area numbers, extracts the temporal variation characteristics of vegetation distribution, water body range and wetland expansion in the numbered areas, aligns the change content according to the image number order, and combines the change features under the same number into change content groups to obtain regional change combination sample groups.

[0163] The change analysis module is based on a combination of regional change sample groups. It compares the direction of boundary expansion, the continuity of wetland areas and the trend of vegetation degradation. It combines content with consistent change directions and continuous spatial outlines into the same map group. By analyzing the correspondence between the continuous movement trajectory of the boundary and the changes in the direction of differences within the region, the correlation of the map group is obtained, and the evolution structure map group is obtained.

[0164] The response extraction module is based on the evolution structure graph group. It compares the control action status corresponding to the graph group number, marks the number with associated action information as the response area, and marks the number without action information but with the same structural features as the area to be responded to. The number is matched to the position node in the control path according to the graph group arrangement to obtain the control response associated area.

[0165] The path sorting module selects response regions with adjacent numbering based on the control response association region, extracts the sequence of actions and spatial position relationship, inserts the region to be responded to between the response sequences, and performs action matching on the connection relationship between the numbers according to the combination order of the connection points of the upper and lower regions to obtain the control deployment structure chain, which is then passed to the control sorting module.

[0166] The control and sorting module is based on the control deployment structure chain. It retrieves the number position of each region to be responded to and the connected preceding and following action nodes, incorporates the action position into the corresponding segment of the adjacent path, adjusts the action sequence order of the number in the structure according to the path order, and outputs the node combination content according to the action path to obtain the sequence of intelligent control regions to be executed.

[0167] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine learning-based intelligent flood storage basin water level regulation method, characterized in that, The method comprises the following steps: S1: Obtain the remote sensing image of the flood storage area, the water level monitoring record, and the region number corresponding table, extract the vegetation, water body, and wetting change content, align each period feature according to the image number order, combine the numbered corresponding features according to time, and obtain a region change combined sample group; S2: Based on the region change combined sample group, compare the boundary expansion direction, wetting continuity and degradation trend, combine the contour continuous direction consistent change content, compare the spatial difference trend, and obtain an evolution structure diagram group; S3: Based on the evolution structure diagram group, compare the diagram group number action state, classify the numbers with associated control information into the response region, and arrange them to the path node according to the number order to obtain the control response associated region; S4: Based on the control response associated region, extract the response number order and spatial connection relationship, embed the to-be-responded number into the response path, adjust the action connection position according to the connection point, and obtain the control deployment structure chain; S5: Based on the control deployment structure chain, extract the to-be-responded number and access position, map the action node sequence to the path, adjust the number order and corresponding action according to the path structure, and obtain the to-be-executed intelligent control region sequence; The specific steps of S3 are: S301: Based on the evolution structure diagram group, obtain the diagram group number and the corresponding control action state record, identify the numbers with control signal marks, judge whether the signal corresponds to the control item, select the numbers with associated actions, and obtain an action response number set; S302: Based on the action response number set, extract the spatial boundary point sequence of the corresponding diagram group, calculate the angle change rate of the adjacent boundary line segment, select the numbers with the same angle change trend according to the change rate gradient distribution, and obtain a to-be-responded number set; S303: Based on the number order in the action response number set and the to-be-responded number set and the path connection order in the diagram group, judge the position relationship of the continuous numbers in the diagram group chain, match the corresponding path segment number, and obtain the control response associated region; The specific steps of S4 are: S401: Based on the control response associated region, identify the numbers with associated action information and the numbers with structure characteristics but without action information, and correspond to the response region and the to-be-responded region respectively to obtain a number partition set; S402: Based on the number partition set, extract the position relationship between adjacent response numbers according to the number order, identify the extension direction of the connection point in the spatial distribution, screen the number combination with the connection position relationship, and obtain a connection order sequence group; S403: Based on the connection order sequence group, insert the to-be-responded number into the connection node between the response number sequence, arrange the number order of each group according to the number before and after, position the number arrangement mode in the continuous segment, and obtain the control deployment structure chain.

2. The machine learning based intelligent flood storage basin water level regulation method according to claim 1, characterized in that, The region change combination sample set comprises a vegetation distribution change feature, a water body range change feature, and a wetting expansion change feature, the evolution structure diagram set comprises boundary expansion direction consistency, wetting region continuity, vegetation degradation trend direction consistency, and spatial profile continuity, the regulation response associated region comprises a response region number, a region number to be responded to, and a diagram group arrangement position node, the regulation deployment structure chain comprises action sequence, spatial position relationship, and number front and back connection relationship, and the intelligent regulation region sequence to be executed comprises a region number to be responded to, front and back action node positions, action sequence ranking, and output node combination content.

3. The method of claim 1, wherein, The specific steps of S1 are as follows: S101: Obtain a remote sensing image of the flood storage and detention area, a water level monitoring record, and a region number corresponding table, correspond to an image layer number at each time period, arrange change information of a vegetation edge, a water body boundary, and a wetting range in time sequence to obtain an image segment mapping data set; S102: Based on the image segment mapping data set, analyze the associated positions among the vegetation expansion zone, the water body change zone, and the wetting range in the layer according to the number sequence, identify the continuous change direction and the connected track, analyze the change relationship of the corresponding number, and obtain a number change direction set; S103: Based on the number change direction set, sequentially connect the change paths of the same number at different times, analyze the direction corresponding relationship between the element change trends in the number region and the structure adjustment process, and obtain a region change combination sample set.

4. The method of claim 1, wherein, The specific steps of S2 are as follows: S201: Based on the region change combination sample set, call the boundary expansion track, the wetting region water surface extension form, and the vegetation index change direction of each sample, pair the sample regions according to the image number sequence, and judge the consistency of the boundary advancing direction and the trend of the outer edge of the adjacent region to obtain a spatial connection profile sequence; S202: Based on the spatial connection profile sequence, calculate the included angle amplitude between the vegetation index change angles, select the numbers of the same section that maintain the included angle amplitude, arrange the number sequence according to the number adjacency relationship, and obtain a profile convergence path set; S203: Based on the profile convergence path set, judge whether the change trend direction of the boundary line movement direction of the continuous sample and the vegetation value fluctuation in the region is consistent, arrange the sample numbers with uniform change trend direction to obtain an evolution structure diagram set.

5. The machine learning based intelligent flood storage basin water level regulation method according to claim 4, characterized in that, The calculation formula of the included angle amplitude between the vegetation index change angles is as follows: ； in, Representative number is The angle between the changes in the vegetation index, Representative section Number of internal sampling nodes Representing the The first section The direction and angle of vegetation change at each node Indicates the first Section No. Node at the Vegetation index observations over time layers Indicates the first Section No. Node at the Observed vegetation characteristics over time. This represents the image interference error term in the temporal layer. Represents the numerical value of the observation quality. The representative node is at the The intensity of the disturbance in the spatial disturbance dimension. This represents the numerical value of the disturbance response. This represents the perturbation bias correction value. This represents the upper limit of the total number of layers in the image observation time series. This represents the upper limit of the total number of items representing the sources of the disturbance response.

6. The method of claim 1, wherein, The calculation formula of the included angle change rate of the adjacent boundary line segment is as follows: ； wherein, a rate of change of an angle represented by the first boundary line segment, an angle value between the first boundary line segment and the image horizontal reference axis, an angle value between the first boundary line segment and the image horizontal reference axis, a length represented by the first boundary line segment, a length represented by the first boundary line segment, an angle adjustment weight represented by the first boundary point, a local gradient value of the first boundary point under the first image feature channel, a local gradient value of the first boundary point under the first image feature channel, a normalized weight value representing a gradient difference influence degree in the first feature channel, a number of introduced image feature channels.

7. The method of claim 1, wherein, The specific steps of S5 are as follows: S501: Based on the regulation deployment structure chain, call out the number position of each region to be responded to, retrieve the front and back action node numbers connected, match the logical positions of the numbers and nodes in the path, and obtain number connection relationship information; S502: Based on the number connection relationship information, identify the path corresponding section between adjacent numbers, adjust the path arrangement order according to the action node distribution state, correct the node sorting rules in the path segment, and obtain action path sorting information; S503: Based on the action path ordering information, the action node combination content is organized in the path order, the position of the action sequence to be responded in the path structure is located, and the intelligent control region sequence to be executed is obtained.

8. The machine learning-based intelligent flood storage basin water level regulation system, characterized in that, The system is used to implement the machine learning-based intelligent control method of the flood storage and detention area water level, and the system comprises: The image recognition module obtains the remote sensing image of the flood storage and detention area, the water level monitoring record, and the region number corresponding table, extracts the time variation characteristics of the vegetation distribution, water body range, and wetting expansion in the numbered region, aligns the change content according to the image number order, combines the change characteristics under the same number into a change content group, and obtains a region change combination sample group; The change analysis module compares the boundary expansion direction, the wetting area continuity, and the vegetation degradation trend change direction based on the region change combination sample group, combines the content groups with consistent change direction and spatial contour continuity into the same graph group, analyzes the relevance of the graph group through the corresponding relationship between the boundary continuous movement track and the difference direction change in the region, and obtains an evolution structure graph group; The response extraction module compares the control action state corresponding to the graph group number based on the evolution structure graph group, labels the number with associated action information as a response region, labels the number with the same structural characteristics but without action information as a to-be-responded region, matches the number to the position node in the control path according to the arrangement of the graph group, and obtains a control response associated region; The path ordering module selects the response regions in adjacent number order based on the control response associated region, extracts the action sequence and spatial position relationship, inserts the to-be-responded region between the response sequences, matches the action between the numbers according to the connection point combination order of the upper and lower regions, obtains a control deployment structure chain, and delivers it to the control ordering module; The control ordering module obtains the number position of each to-be-responded region and the connected front and rear action nodes based on the control deployment structure chain, includes the action position into the corresponding section of the adjacent path, adjusts the action sequence ranking of the number in the structure according to the path order, outputs the node combination content according to the action path, and obtains the intelligent control region sequence to be executed.

Images