Cooperative treatment method and system for watershed river-lake water area space and city based on artificial intelligence

Abstract

The invention relates to the technical field of watershed space planning, in particular to a watershed river-lake water area space and city collaborative governance method and system based on artificial intelligence. Multi-source data fusion is achieved by constructing a multi-modal feature tensor, and regional styles and features are extracted by adopting a double-track coupling model; problem diagnosis is carried out based on a visual Transform architecture, spatial prior constraints are introduced, a target intention scheme is generated by using a conditional generative adversarial network, and finally treatment measures are matched through a knowledge graph and sorting is optimized. The system adopts a cloud-edge-end architecture to realize distributed computing, and supports continuous optimization of the model. According to the method, the problems of data splitting, experience dependence, scheme homogenization and the like in traditional treatment are solved, and intellectualization, precision and self-adaption of drainage basin city collaborative treatment are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of river basin space planning, and particularly relates to a river basin river and lake space and urban collaborative management method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid advancement of urbanization, the contradiction between river basin river and lake space and its adjacent urban area is increasingly prominent. On the one hand, the encroachment of urban expansion on waterfront areas, the destruction of ecological functions, and the fragmentation of historical context have led to a series of problems such as water space landscape imbalance and ecological system degradation. On the other hand, traditional urban planning and river basin management often belong to different systems, lack effective coordination mechanisms, and make governance strategies have limitations such as fragmentation and single target, making it difficult to cope with the comprehensive challenges of river basin-urban complex systems.

[0003] Currently, in the planning and management of river basin waterfront space, existing technologies rely on manual field research and qualitative analysis, or only use remote sensing and geographic information systems for simple spatial overlay analysis. These methods are not only inefficient, but also heavily dependent on the experience and subjective judgment of planners, making it difficult to deeply mine and analyze massive, diverse spatiotemporal data. Especially in the quantitative analysis of regional landscape characteristics, comprehensive evaluation of spatial quality, and generation of governance schemes with ecological benefits and regional cultural characteristics, existing technologies lack precise, efficient, and dynamically adaptable intelligent support tools.

[0004] Although artificial intelligence technology has made significant progress in image recognition, natural language processing, and other fields, its application in land space planning, especially in river basin management scenarios involving complex coupling of natural ecology and urban society, is still in its infancy. Existing explorations are mostly focused on single-target identification or prediction, and have not formed a complete technical closed loop from data fusion, feature analysis to problem diagnosis, scheme generation, and strategy recommendation. At the same time, most models lack the ability to embed rigid constraints such as ecological protection red lines, and have not effectively integrated regional landscape attributes, resulting in generated schemes often being disconnected from local actual needs and control requirements. SUMMARY

[0005] The purpose of the present application is to provide a basin river and lake water area space and urban collaborative governance method and system based on artificial intelligence, which realizes multi-source data fusion by constructing a multi-modal feature tensor, extracts regional landscape features using a double-track coupled model, conducts problem diagnosis based on a visual Transformer architecture and introduces spatial prior constraints, generates target intention schemes using a conditional generative adversarial network, and finally matches governance measures and optimizes the order through a knowledge graph. The system uses a cloud-edge-end architecture to realize distributed computing and support continuous model optimization. The present application solves the problems of data fragmentation, reliance on experience and scheme homogenization in traditional governance, and realizes the intelligentization, precision and self-adaptation of basin urban collaborative governance.

[0006] To achieve the above technical purposes and effects, the present application is implemented by the following technical solutions: A basin river and lake water area space and urban collaborative governance method based on artificial intelligence, comprising the following steps: Data acquisition and fusion step: acquiring multi-element heterogeneous data of a target basin urban waterfront space, and performing spatio-temporal registration and standardization processing on the data to generate a multi-modal feature tensor F multi ; Attribute modeling step: inputting the multi-modal feature tensor F multi into a double-track coupled model, which performs feature extraction and fusion through parallel ground element analysis tracks and spatial quality evaluation tracks, and outputs a feature landscape attribute vector V style characterizing the common landscape features of the target basin; Analysis and generation step: based on the multi-modal feature tensor F multi and the feature landscape attribute vector V style , parallelly executing landscape problem diagnosis and target intention generation to obtain a problem diagnosis report R issue , a target intention graph I target and a spatial control strategy text T strategy ; Strategy output step: according to the problem diagnosis report R issue , the target intention graph I target and the spatial control strategy text T strategy , matching a recommended planning governance measure set M set from a pre-set governance measure knowledge graph, and outputting after priority sorting.

[0007] Further, in the attribute modeling step, the specific operation mode of the double-track coupled model comprises: In the ground element analysis track, a pre-trained full convolutional network is used to extract F multiThe natural landform index, the original vegetation coverage index, the settlement space texture characteristics and the waterfront interface public accessibility index are resolved and spliced into a geographical element feature vector V local ; The spatial quality evaluation track adopts a multi-layer perception machine to evaluate the global features of F multi , and outputs a spatial quality score vector V including four dimensions of ecological health degree, landscape coordination degree, functional complexity degree and cultural heritage highlighting degree quality ; The V local and V quality are coupled by a feature fusion function to generate the feature landscape attribute vector V style , and the calculation formula is: Wherein, W l and W q are trainable weight matrices for mapping input features to a unified high-dimensional space; and are balance coefficients dynamically adjusted according to different section types in the target basin, including urban core section, rural transition section and natural ecological section.

[0008] Further, the landscape problem diagnosis in the analysis and generation steps is completed by a model based on a visual Transformer architecture, which calculates the abnormal scores of each region in space through a multi-head self-attention mechanism. In the attention weight calculation, a spatial prior mask matrix M is introduced, which is constructed according to the ecological protection red line and the permanent basic farmland range in the geographic information system. A very large negative value is assigned to the position elements corresponding to the above-mentioned restricted areas in the matrix, so as to shield the abnormalities of these areas in the model calculation and focus the diagnosis on the governable urban construction area.

[0009] Further, the target intention generation in the analysis and generation steps is completed by a conditional generative adversarial network. The generator of the network takes random noise and the multi-modal feature tensor F multi as background, and the feature landscape attribute vector V style as condition to learn to generate the target intention map I target . The training loss function L G of the generator is a weighted combination of multi-objective loss, and its expression is: Wherein, L​adv is an adversarial loss, L content is a content-aware loss, L style is a style consistency loss, achieved by calculating the Euclidean distance between the style attribute vector V style re-extracted from the generated image; is a hyper-parameter for controlling the weights of each loss.

[0010] Further, in the strategy output step, the set of planning governance measures M set is prioritized using a ranking function P(m i ) as follows: where m i represents an individual governance measure, U(m i , R issue ) is the implementation urgency calculated based on the problem severity in the problem diagnosis report R issue , B(m i , I target ) is the expected benefit calculated based on the contribution of the measure m i to achieving the goal intention graph I target , and are adjustable coefficients for balancing the weights of urgency and benefit.

[0011] In another aspect, the present application proposes an artificial intelligence-based watershed river and lake water area space and urban collaborative governance system for implementing the above method, which comprises: a data collection and fusion module for performing the data collection and fusion step; a dual-track coupled attribute modeling module for performing the attribute modeling step; an intention generation and problem diagnosis module for performing the analysis and generation step; a strategy output and feedback module for performing the strategy output step.

[0012] Further, the system further comprises a model updating and management module for receiving user's correction feedback on the output result, and based on the correction feedback, incrementally learning the style problem diagnosis model and the conditional generative adversarial network to achieve continuous optimization of model parameters.

[0013] Further, the system is deployed on a cloud-edge-end collaborative computing architecture, wherein: the cloud center is responsible for centralized training of heavy AI models and maintenance of global knowledge graph; The edge server is arranged in a city-level management unit, and is responsible for local model inference and real-time data processing. The terminal device provides a graphical user interface for the planner to issue tasks, interact with the scheme and review the results.

[0014] The beneficial effects of the present application are as follows: The present application realizes the deep fusion of multi-source data by constructing a multi-modal feature tensor, and innovatively adopts a double-track coupled modeling mechanism to solve the problem of single feature representation. In traditional governance methods, remote sensing images, geographic information system data and field survey data are often managed separately, making it difficult to form a unified analysis basis. The present application integrates heterogeneous data such as spectral features, terrain elevation, land use types and environmental monitoring indicators into a unified multi-modal feature tensor through spatial and temporal registration and standardization, providing a complete data base for subsequent analysis. The double-track coupled model performs feature extraction from two dimensions of micro-morphology and macro-quality through parallel ground element analysis track and spatial quality evaluation track: the full convolution network accurately quantifies specific element features such as natural landform morphology, original vegetation coverage and settlement space texture to form a ground element feature vector; the multilayer perceptron comprehensively evaluates functional quality dimensions such as ecological health degree and landscape coordination degree to form a spatial quality score vector. The two are coupled through a dynamic weighted feature fusion function, and the balance coefficient can be adaptively adjusted according to different section types, so that the generated feature landscape attribute vector can accurately reflect regional characteristic details and reflect spatial overall quality, providing a feature representation with both geographic precision and semantic depth for subsequent analysis.

[0015] The present application realizes accurate positioning of governance problems and regional adaptability generation of governance schemes by introducing a spatial prior constraint visual Transformer architecture and a conditional generative adversarial network. Traditional methods rely on manual experience to identify spatial problems, which often have strong subjectivity and incomplete coverage. The present application adopts a visual Transformer-based landscape problem diagnosis model to calculate the abnormal scores of each region in space through a multi-head self-attention mechanism, and constructs a spatial prior mask matrix based on rigid control boundaries such as ecological protection red lines and permanent basic farmland, which effectively shields the restricted areas in attention weight calculation to ensure that the diagnosis focus is concentrated on the governable urban construction area. In the target intention generation link, the conditional generative adversarial network embeds the feature landscape attribute vector as a style condition into the generation process, and through the optimization of the multi-objective loss function, the generator always maintains high consistency with the regional landscape features while solving specific diagnostic problems.

[0016] The present application realizes intelligent mapping and accurate recommendation from problems to strategies by constructing a governance measure knowledge graph and designing a multi-factor priority ranking function. Traditional governance measure recommendation is often based on experience judgment and lacks systematic argumentation. Based on the pre-set governance measure knowledge graph, the present application intelligently matches and diagnoses a governance measure set related to the problem and target intention through graph query and semantic similarity calculation. On this basis, the system uses a mathematical ranking function to evaluate the priority of the measures, which comprehensively considers the implementation urgency based on the severity of the problem and the expected benefit based on the contribution of the target intention, and balances the weights of the two through adjustable coefficients.

[0017] Of course, implementing any product of the present application does not necessarily require all the advantages described above. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed for the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1 The present application is a whole process schematic diagram; Figure 2 The present application is an embodiment planning process schematic diagram. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] Embodiment 1 The method for basin river and lake water space and urban collaborative governance based on artificial intelligence described in this embodiment is for the Lijiang River basin, which includes the following steps: S1: Multi-source heterogeneous data acquisition and spatio-temporal alignment: Through the space-air-ground integrated monitoring network deployed on satellites, unmanned aerial vehicles and ground sensors, multi-source heterogeneous data of urban waterfront space in the Lijiang River basin is collected; the multi-source heterogeneous data specifically includes: Remote sensing image data set: including panchromatic and multispectral images, and synthetic aperture radar images for extracting terrain and building height; Geographic Information System Datasets: These include digital elevation models, land use / cover classification maps based on national land survey results, and vector boundary data including ecological protection red lines and urban development boundaries; Field survey dataset: 360-degree panoramic images acquired by GPS-equipped data acquisition devices, vector boundary and attribute data of landscape elements (including but not limited to vegetation, walkways, revetments, and buildings) acquired by surveying and mapping methods, and water quality, air quality, and noise data from fixed monitoring stations; Planning text dataset: Digitalized urban master plans, regulatory detailed plans, waterfront space landscape control guidelines, and lists of historical and cultural heritage protection obtained from government functional departments at all levels; The aforementioned diverse and heterogeneous data are input into the data fusion module, where a spatiotemporal registration algorithm is executed to unify all data to the same geographic coordinate system and pixel size. Non-image data is then rasterized or vectorized, ultimately outputting a multimodal feature tensor that includes spatial, attribute, and semantic information. Where H and W represent the height and width of the spatial grid after the study area is divided, respectively, and C represents the total number of all feature channels after fusion. This channel is composed of spectral bands, elevation values, land use type codes, and quantitative values ​​of environmental monitoring indicators. S2: Construction of regional adaptive landscape attribute vectors based on dual-track coupling mechanism: Construct a dual-track coupling model, using the multimodal feature tensor F multi As input, deep features are extracted through two parallel analysis tracks and coupled to generate a feature landscape attribute vector that highly summarizes the regional characteristics of the Lijiang River basin. D s The preset dimension of the attribute vector; the specific operation mode of the dual-track coupling model includes: First Track – Quantitative Analysis of Local Elements: Employing a fully convolutional network model with an attention mechanism, F multi Perform pixel-level semantic segmentation and feature extraction, and quantitatively output the feature vector of local elements. The vector V local It is composed of the following normalized exponents: Karst landform index I landform The topographic relief and surface roughness were calculated by analyzing DEM data. Native vegetation cover index I vegetation Normalized vegetation index and leaf area index obtained by interpreting multispectral images; Traditional settlement spatial texture characteristics I settlement This is obtained by analyzing the contour, density, and orientation distribution entropy of building vector data; Waterfront Accessibility Index I accessibility It is calculated using a cost-distance algorithm based on road network data and entry point locations; Second Track – Multi-dimensional Spatial Quality Intelligent Assessment Track: A multilayer perceptron model is used, whose input is F passed through a global pooling layer. multi The compressed global feature vector is used to output a multi-dimensional quality score vector. The four dimensions of this vector correspond to: Ecological Health S eco The calculation is based on a comprehensive assessment of indicators such as water quality, vegetation coverage, and connectivity of biological habitats. Landscape Harmony S scene The calculation is based on a comprehensive assessment of factors such as visual transparency, harmony between architectural color and form, and the beauty of the skyline. Functional Complexity S function The calculation is based on the combination and convenience of functions such as leisure, transportation, commerce, and ecology; Cultural Heritage Prominence S culture Calculations based on the preservation status of historical elements, the intensity of cultural activities, and their visibility; Dynamically weighted feature coupling mechanism: Through feature fusion function The feature vectors from the two tracks are deeply fused to generate the final V. style Its mathematical expression is: in, and This is the weight matrix learned during the model training phase, used to map features from different sources to the same high-dimensional space; and This is a dynamic equilibrium coefficient, and its value is not fixed. Instead, it is automatically generated by a small neural network based on the specific section of the Lijiang River basin corresponding to the input data (pre-divided into urban core section, rural transition section, and natural ecological section). This ensures that the emphasis varies in different areas; for example, in the urban core section... The weight of (corresponding space quality) will be relatively higher; S3: Precise diagnosis of landscape issues and generation of target intentions by integrating regional attributes: This step executes two core tasks in parallel: S3.1: Detailed Diagnosis of Landscape Issues: The multimodal feature tensor F multiInput a landscape problem diagnosis model based on the visual Transformer architecture; through its core multi-head self-attention mechanism, calculate the correlation between any two positional features in space, thereby identifying abnormal regions that are inconsistent with the overall landscape; the j-th attention head is calculated as follows: Among them, Q j K j V j They are respectively composed of F multi The query, key, and value matrix obtained by linear transformation, d k The dimension of the key vector is used for scaling. It is a spatial prior mask matrix; this matrix M is not all zeros, and its element values ​​are set according to the restrictive boundaries in the GIS data, such as ecological protection red lines, basic farmland, and historical and cultural relic protection areas, and a very large negative value (e.g., -10) is assigned at the corresponding positions. 9 This forces the model to ignore the "anomalies" in these undevelopable areas during softmax calculations, focusing its attention on manageable urban built-up areas. The final output is a problem diagnosis report (R) accurate to the specific grid, including problem type (e.g., "abrupt architectural style," "lack of greenery"), geographical location, and severity level. issue ; S3.2: Generation of target intentions under conditional constraints: The characteristic landscape attribute vector V style As a style condition, it is consistent with the problem diagnosis report R. issue A co-input conditional generative adversarial network; the generator G of this network uses random noise z and F. multi As input, with V style As a condition, the learning process generates methods that not only conform to the characteristics of the Lijiang River basin but also specifically address the R... issue Target intention diagram I for the problems listed target To ensure generation quality, the generator's training loss function L... G It is a weighted combination of multi-target losses: Among them, L adv It is to combat loss and encourage the generation of images. It appears real to discriminator D; L content It is a content-aware loss, which calculates the generated image and compares it with an ideal reference image drawn by an expert. ref Differences in feature space; L style It is a landscape consistency loss, achieved through a pre-trained encoder f extractRe-extract the landscape attributes from the generated image and require them to match the input condition V. style To be as close as possible to ensure a faithful representation of the regional style; These are hyperparameters that control the weights of various losses; simultaneously, the system also outputs the corresponding spatial control strategy text T based on the generation logic. strategy Examples include "restricting the building height on the first waterfront surface" and "increasing the local karst vegetation". S4: Knowledge Graph-Driven Governance Measure Matching and System Adaptive Feedback: S4.1: Matching and prioritizing intelligent measures: The strategy reasoning module receives the target intention map I. target Space control strategy text T strategy And the problem diagnosis report R issue This module connects to a pre-built knowledge graph of governance measures, which stores a large number of "problem-measure-effect" entities and their relationships in the form of triples. Through graph query and semantic similarity calculation, the module retrieves and matches the most relevant set of specific planning governance measures. Subsequently, the system employs a priority sorting function P(m) i ) for M set The measures in the document are sorted in order: Wherein, U(m) i R issue ) is the urgency function, which is based on the measure m i The problems that can be solved in R issue The severity of the disease is calculated; B(m) i I target ) is the expected benefit function, which evaluates measure m i To achieve I target The degree of contribution of the described objective; and It is a weighting coefficient used to balance urgency and benefits, which can be adjusted by the user according to the actual needs of the project; S4.2: Human-Machine Collaborative Feedback and Model Iterative Optimization The target intent map I is fully presented to urban planners through a graphical user interface. target Problem Diagnosis Report R issue and the sorted set of governance measures M set Planners can adjust plans, select measures, or propose modifications through the interface; these actions are recorded as correction feedback. feedback The model update and management module within the system utilizes F feedbackThe system also uses new data collected after implementation to fine-tune the parameters of the landscape problem diagnosis model and the conditional generative adversarial network through incremental learning algorithms, enabling the system to continuously adapt to the dynamic changes in the Lijiang River basin and continuously improve the accuracy of recommendations.

[0022] In this embodiment, the fully convolutional network model used in the first track in step S2 is a variant of U-Net with attention gate. The attention gate automatically learns to focus on key spatial regions related to the landform and settlement features of the Lijiang River Basin during the decoding process by calculating the importance weights of different positions in the feature map, thereby suppressing the interference of irrelevant background information.

[0023] In this embodiment, the construction of the spatial prior mask matrix M in step S3.1 is specifically as follows: for positions i and j in space, if position i or j is located within the ecological protection red line, permanent basic farmland, or first-level historical and cultural relic protection area, then the diagonal and neighboring elements of all rows and columns related to position i in matrix M are set to a very large negative value. This almost completely shields the connection between these restricted areas and other areas in the attention weight calculation.

[0024] On the other hand, as described in this embodiment, an artificial intelligence-based collaborative governance system for river and lake water space and city in the Lijiang River Basin, used to implement the above method, comprises the following mutually communicating modules: The data acquisition and fusion module includes a hardware configuration of a network interface for receiving air-space-ground data and a high-speed memory, and software logic for executing step S1 to complete the spatiotemporal registration of multi-source heterogeneous data and the multimodal feature tensor F. multi The generation of; The dual-track coupling attribute modeling module encapsulates the dual-track coupling model defined in claim 1, including the fully convolutional network, the multilayer perceptron, and the feature fusion function. The implementation code is used to execute step S2 and output the feature landscape attribute vector V. style ; The intention generation and problem diagnosis module integrates the aforementioned visual Transformer-based landscape problem diagnosis model and the aforementioned conditional generative adversarial network, and is used to concurrently execute steps S3.1 and S3.2, outputting a problem diagnosis report R. issue And Target Intention Map I target and text T strategy ; The strategy output and feedback module includes a graph database storing the knowledge graph of the governance measures, an inference engine for executing the sorting function P(·), and an interactive unit that provides a graphical user interface for performing the feedback receiving function of steps S4.1 and S4.2. The model update and management module, including a version control system and an incremental learning trainer, is used to update the model based on feedback data F. feedback Manage and iteratively update the artificial intelligence models in the system.

[0025] In this embodiment, the system is deployed on a cloud-edge-device collaborative computing architecture: The cloud center, equipped with a high-performance GPU cluster, is responsible for storing massive amounts of historical data, centralized training of initial heavy AI models, and maintenance of the global knowledge graph. Edge servers are deployed in the computer rooms of the Natural Resources and Planning Bureau in Guilin City and its subordinate districts and counties. They carry lightweight inference models and are responsible for real-time data processing and solution generation in their respective jurisdictions, reducing network latency and protecting data privacy. The terminal device is a desktop or mobile workstation used by the planner. It accesses the graphical user interface through a browser or client software to assign tasks, interact with solutions, and review results.

[0026] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for collaborative governance of watershed river and lake water space and urban areas based on artificial intelligence, characterized in that, Includes the following steps: Data acquisition and fusion steps: Acquire multi-dimensional heterogeneous data of urban waterfront spaces in the target watershed, and perform spatiotemporal registration and standardization on the data to generate a multimodal feature tensor F including spatial, spectral, and attribute features. multi ; Attribute modeling steps: Convert the multimodal feature tensor F... multi The input is fed into a dual-track coupled model, which extracts and fuses features through parallel local feature analysis and spatial quality assessment tracks, and outputs a feature landscape attribute vector V representing the common landscape characteristics of the target watershed. style ; Analysis and generation steps: Based on the multimodal feature tensor F multi and the feature landscape attribute vector V style Parallel execution of landscape problem diagnosis and target intention generation yields a problem diagnosis report R. issue Target Intention Map I target and space control strategy text T strategy ; Strategy output steps: Based on the problem diagnosis report R issue The target intention diagram I target and the space control strategy text T strategy The recommended set of planning and governance measures M is matched from the pre-set knowledge graph of governance measures. set Then, sort the results by priority and output them.

2. The method for collaborative governance of watershed river and lake water space and city based on artificial intelligence as described in claim 1, characterized in that: In the attribute modeling step, the specific operation mode of the dual-track coupling model includes: The local feature parsing trajectory uses a pre-trained fully convolutional network, from F... multi The natural landform morphology index, native vegetation cover index, settlement spatial texture characteristics, and waterfront public accessibility index were extracted and concatenated into a local feature vector V. local ; The space quality assessment track employs a multilayer perceptron to assess F. multi The overall characteristics are evaluated, and the spatial quality score vector V is output, which includes four dimensions: ecological health, landscape harmony, functional integration, and cultural heritage prominence. quality ; Through feature fusion function V local and V quality Coupled to generate the feature landscape attribute vector V style The formula for its calculation is: Among them, W l and W q It is a trainable weight matrix used to map input features to a unified high-dimensional space; and The balance coefficient is dynamically adjusted according to different segment types within the target watershed, including urban core segments, rural transition segments, and natural ecological segments.

3. The method for collaborative governance of watershed river and lake water space and city based on artificial intelligence as described in claim 1, characterized in that: The landscape problem diagnosis in the analysis and generation steps is completed by a model based on the visual Transformer architecture, which calculates the anomaly score of each region in the space through a multi-head self-attention mechanism. In the calculation of attention weights, a spatial prior mask matrix M is introduced. This matrix is ​​constructed based on the ecological protection red line and the scope of permanent basic farmland in the geographic information system. The location elements in the matrix corresponding to the above-mentioned restricted areas are assigned a very large negative value, thereby shielding the anomalies in these areas in the model calculation and focusing the diagnostic focus on the manageable urban construction area.

4. The method for collaborative governance of watershed river and lake water space and city based on artificial intelligence as described in claim 1, characterized in that: The generation of target intent in the analysis and generation steps is accomplished by a conditional generative adversarial network. The generator of this network uses random noise and the multimodal feature tensor F multi Against this background, the feature landscape attribute vector V style As a condition, the target intention map I is generated through learning. target ; The training loss function L of the generator G The weighted combination of multi-objective losses is expressed as follows: Among them, L adv It is to combat losses, L content It is content-perceived loss, L style It is the landscape consistency loss, calculated by the conditional vector V. style This is achieved by using the Euclidean distance between the topographical attribute vectors and the topographical attribute vectors re-extracted from the generated image; Hyperparameters are used to control the weights of various losses.

5. The method for collaborative governance of watershed river and lake water space and city based on artificial intelligence as described in claim 1, characterized in that: In the strategy output step, the set of planning and governance measures M is... set The sorting function P(m) used for priority sorting i )for: Where, m i Representing a single governance measure, U(m) i R issue (Based on the aforementioned problem diagnosis report R) issue The urgency of implementation is calculated based on the severity of the problem, B(m) i I target ) for evaluation measures m i To achieve the target intention diagram I target The expected benefits are calculated based on the contribution of the individual. and This is an adjustable coefficient used to balance the weights of urgency and benefit.

6. An artificial intelligence-based watershed river and lake water space and urban collaborative governance system for implementing the method as described in any one of claims 1-5, characterized in that: The system includes: The data acquisition and fusion module is used to perform data acquisition and fusion steps; The dual-track coupled property modeling module is used to perform property modeling steps; The intent generation and problem diagnosis module is used to perform the analysis and generation steps; The strategy output and feedback module is used to execute the strategy output steps.

7. The watershed river and lake water space and urban collaborative governance system based on artificial intelligence as described in claim 6, characterized in that: The system also includes a model update and management module, which is used to receive user feedback on the output results and perform incremental learning on the landscape problem diagnosis model and the conditional generative adversarial network based on the feedback, so as to achieve continuous optimization of model parameters.

8. The watershed river and lake water space and urban collaborative governance system based on artificial intelligence as described in claim 6, characterized in that: The system is deployed on a cloud-edge-device collaborative computing architecture, wherein: The cloud center is responsible for the centralized training of heavy-duty AI models and the maintenance of the global knowledge graph; Edge servers are deployed at city-level management units and are responsible for inference and real-time data processing of localized models; The terminal device provides planners with a graphical user interface for task assignment, solution interaction, and result review.

Images