Urban territorial space dynamic monitoring and intelligent analysis system and method

Abstract

The invention discloses an urban territorial space dynamic monitoring and intelligent analysis system and method, and relates to the technical field of territorial space dynamic monitoring. The system comprises an initialization module used for constructing a lightweight learning framework, a dynamic modeling module used for establishing a dynamic network of geographical environment factors and a monitoring target based on parameters and a characteristic mode set output by the initialization module, and an optimization feedback module used for establishing a monitoring result feedback channel, and comparing the actual monitoring data set with a model prediction data set output based on the dynamic modeling module to calculate a prediction error. Through cooperative work of the three, the problem of weak model generalization ability caused by spatial heterogeneity of geographic phenomena is solved, and efficient and accurate monitoring and analysis of urban territorial spaces in different regions are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic monitoring of land space, and specifically to a system and method for dynamic monitoring and intelligent analysis of urban land space. Background Art

[0002] In the field of dynamic monitoring of urban land space, traditional geological models face serious model generalization problems when dealing with geographic spatial heterogeneity. Geographic environmental factors such as topography, climate conditions, and economic development levels vary greatly in different regions, making it difficult for traditional models trained based on a single structure and fixed parameters to maintain monitoring accuracy in new areas. For example, when migrating from plains to mountainous areas, or from economically developed cities to economically underdeveloped towns, the model needs to re-collect and process massive amounts of data for retraining, which is a cumbersome and inefficient process. The traditional model structure cannot dynamically adapt to the complex and changing relationship between the geographical environment and the monitoring target. Faced with real-time changing geographic information, it is difficult to quickly and accurately adjust the monitoring strategy, resulting in delayed monitoring results and large errors. It cannot meet the urgent needs of real-time, efficient, and cross-regional monitoring of urban land space. Summary of the Invention

[0003] The purpose of the present invention is to provide a system and method for dynamic monitoring and intelligent analysis of urban land space to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a system for dynamic monitoring and intelligent analysis of urban land space, comprising the following steps: The initialization module is used to build a lightweight learning framework. When the system enters a new area, it uses a preset number of sample data to trigger model adaptation. The learning algorithm extracts the characteristic pattern set of the new area data and initializes and adjusts the pre-trained model parameter set based on the characteristic pattern. A dynamic modeling module is used to establish a dynamic network of geographical environmental factors and monitoring targets based on the parameters and feature pattern set output by the initialization module, calculate the edge weights between network nodes through the influence degree calculation function, and introduce a graph neural network to dynamically update the network structure; The optimization feedback module is used to establish a monitoring result feedback channel, compare the actual monitoring data set with the model prediction data set based on the output of the dynamic modeling module to calculate the prediction error, introduce a metric learning algorithm to measure the similarity between samples, and update the metric matrix and the parameters of the initialization module and the dynamic modeling module through the reinforcement learning algorithm based on the prediction error.

[0005] Preferably, the initialization module includes a feature extraction unit and a parameter adjustment unit; The feature extraction unit is used to extract a feature pattern set of new area data from the preset number of sample data using the learning algorithm; The parameter adjustment unit is used to initialize and adjust the pre-trained model parameter set according to the feature pattern set.

[0006] Preferably, the feature extraction unit uses a neural network combined with an attention mechanism to extract features from the sample data.

[0007] Preferably, the dynamic modeling module includes a network construction unit, a weight calculation unit and a structure update unit; The network construction unit is configured to construct a dynamic network with geographical environmental factors and monitoring targets as nodes based on the parameters and characteristic pattern set output by the initialization module; The weight calculation unit is used to calculate the edge weights between nodes in the dynamic network using the influence degree calculation function; The structure updating unit is used to dynamically update the dynamic network structure using the graph neural network.

[0008] Preferably, the impact degree calculation function is obtained based on historical data and machine learning algorithm training, and is used to quantify the impact degree of geographical environmental factors on the monitoring target.

[0009] Preferably, the optimization feedback module includes an error calculation unit, a metric learning unit and a parameter updating unit; The error calculation unit is used to compare the actual monitoring data set with the model prediction data set to calculate the prediction error; The metric learning unit is used to measure the similarity between samples using a metric function and generate a parameter update strategy using the reinforcement learning algorithm based on the prediction error; The parameter updating unit is used to update the measurement matrix and the parameters of the initialization module and the dynamic modeling module according to the parameter updating strategy.

[0010] Preferably, the metric learning unit adopts an optimization method based on gradient descent to iteratively update the metric matrix.

[0011] The present invention also provides a method for dynamic monitoring and intelligent analysis of urban land space, comprising: Build a lightweight learning framework that triggers model adaptation using a preset number of sample data when entering a new area. The learning algorithm extracts a set of characteristic patterns from the new area data and initializes and adjusts the pre-trained model parameter set based on the characteristic patterns. Based on the initialized and adjusted parameters and the characteristic pattern set, a dynamic network of geographical environmental factors and monitoring targets is established, the edge weights between network nodes are calculated through an influence degree calculation function, and a graph neural network is introduced to dynamically update the network structure; A monitoring result feedback channel is established, the actual monitoring data set is compared with the model prediction data set obtained based on the dynamically updated network structure to calculate the prediction error, a metric learning algorithm is introduced to measure the similarity between samples, and the metric matrix and the pre-trained model parameters and dynamic network structure parameters are updated through a reinforcement learning algorithm based on the prediction error.

[0012] The present invention further provides an electronic device, wherein the electronic device is a physical device and comprises: a processor and a memory, wherein the memory is communicatively connected to the processor; The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the above-mentioned urban land space dynamic monitoring and intelligent analysis method.

[0013] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned urban land space dynamic monitoring and intelligent analysis method.

[0014] Compared with the prior art, the present invention has the following beneficial effects: Through the initialization module, a lightweight learning framework is constructed, and a small amount of sample data is used to quickly complete the adaptation preparation of the model in the new area, realizing the rapid adaptation of the model in different regions and improving the adaptation speed. Through the dynamic modeling module, a dynamic relationship model between geographical environmental factors and monitoring targets is established, and the complex relationship between the two is captured in real time, which improves the accuracy of monitoring. Through the optimization feedback module, the model is continuously optimized according to the actual monitoring results, and the model parameters are continuously adjusted, which effectively improves the generalization ability and adaptability of the model. The three work together to solve the problem of weak model generalization ability caused by the spatial heterogeneity of geographical phenomena, and realize efficient and accurate monitoring and analysis of urban land space in different regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the structure of a system for dynamic monitoring and intelligent analysis of urban land space provided by an embodiment of the present invention; Figure 2 A flow chart of a method for dynamic monitoring and intelligent analysis of urban land space provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] The execution subject of the method of this embodiment is a terminal, which can be a mobile phone, tablet computer, PDA, notebook or desktop computer, etc. Of course, it can also be other devices with similar functions, and this embodiment is not limited.

[0018] See also Figure 1 The present invention provides a system for dynamic monitoring and intelligent analysis of urban land space, comprising: The initialization module is used to build a lightweight learning framework. When the system enters a new area, it uses a preset number of sample data to trigger model adaptation. The learning algorithm extracts the characteristic pattern set of the new area data and initializes and adjusts the pre-trained model parameter set based on the characteristic pattern. A dynamic modeling module is used to establish a dynamic network of geographical environmental factors and monitoring targets based on the parameters and feature pattern set output by the initialization module, calculate the edge weights between network nodes through the influence degree calculation function, and introduce a graph neural network to dynamically update the network structure; The optimization feedback module is used to establish a monitoring result feedback channel, compare the actual monitoring data set with the model prediction data set based on the output of the dynamic modeling module to calculate the prediction error, introduce a metric learning algorithm to measure the similarity between samples, and update the metric matrix and the parameters of the initialization module and the dynamic modeling module through the reinforcement learning algorithm based on the prediction error.

[0019] In an optional embodiment, the initialization module aims to build a lightweight learning framework, using a small amount of sample data to quickly complete the adaptation preparation of the model in the new area, thereby improving the adaptation speed of the model in the new area; the dynamic modeling module is used to establish a dynamic relationship model between geographical environmental factors and monitoring targets, capturing the complex relationship between the geographical environment of different regions and monitoring targets in real time, and improving monitoring accuracy; the optimization feedback module is responsible for continuously optimizing the model, continuously adjusting the model parameters according to actual monitoring results, and improving the generalization ability and adaptability of the model.

[0020] For example, in a specific implementation, when the system enters a new urban area for land space monitoring, the three modules function in turn. The initialization module first uses a small amount of sample data to prepare for model adaptation. The dynamic modeling module constructs a dynamic relationship model based on the adapted parameters. The optimization feedback module optimizes the model based on actual monitoring and model prediction conditions to complete the monitoring and analysis of the area.

[0021] The quantitative indicators of the lightweight learning framework include: Sample size threshold: The preset sample data volume that triggers model adaptation is 50-100 groups (a single group of data contains multi-dimensional features of geographical environmental factors and monitoring targets, and the data volume is ≤10MB), which is much lower than the thousands of samples required by traditional models; Model parameter scale: pre-trained model parameter set The total number of parameters is ≤ 1 million, and the number of convolutional neural network layers of the feature extraction unit is ≤ 5, ensuring that the model can run in real time on terminal devices (such as edge computing nodes); Adaptation time: On a standard server (Intel Core i7 CPU), the total time required to complete new region feature extraction and parameter adjustment is ≤ 5 minutes, meeting the requirements for rapid deployment.

[0022] In an optional embodiment, the initialization module includes a feature extraction unit and a parameter adjustment unit; The feature extraction unit is used to extract a feature pattern set of new area data from the preset number of sample data using the learning algorithm; The parameter adjustment unit is used to initialize and adjust the pre-trained model parameter set according to the feature pattern set.

[0023] It should be noted that the learning algorithm adopts a hybrid architecture that integrates meta-learning and convolutional neural networks. The specific implementation process is as follows: Meta-learning Rapid Adaptation Phase: The model-agnostic meta-learning (MAML) algorithm is used to perform migration and adaptation training on sample data from new regions. Using 50-100 small sample data sets, the algorithm quickly captures the distribution patterns of regional features and outputs initial feature extraction parameters suitable for the target region. This process effectively addresses the generalization challenges of traditional algorithms in data-scarce scenarios. Convolutional neural network feature extraction stage: Based on the initial parameters generated by meta-learning, a convolutional neural network (CNN) model with an attention mechanism is constructed. The specific implementation process is as follows: Convolutional layer: uses a 3×3 convolution kernel to extract local features of the input data to capture spatial details such as land use changes and building forms; Pooling layer: Reduces data dimensions through maximum pooling operations, reducing computational complexity while retaining key features; Attention layer: Introducing the self-attention module, the feature importance weight is calculated through the multi-layer perceptron (MLP). The formula is: , where Indicates the This mechanism can effectively highlight key features such as land use change trends and building density distribution, and ultimately output a set of spatial feature patterns. .

[0024] Among them, sample data refers to the original data about the new area obtained through sensor collection, data collection, etc., including land use, building distribution, population density and other information. Learning algorithm It plays a core role in the feature extraction unit. It can analyze and process sample data and mine representative feature patterns.

[0025] Among them, the feature pattern set It is a high-level summary and abstraction of the data characteristics of the new region. ( ) represents a unique characteristic pattern. For example, It may indicate the evolution pattern of land use types in the temporal dimension, such as the trend of conversion from cultivated land to construction land; These patterns may represent the relationship between building distribution and transportation networks, with higher building density near major roads. These characteristic patterns encompass multiple dimensions, including spatial distribution characteristics, temporal trends, and inter-element relationships, characterizing the new region's national space from various perspectives.

[0026] In one possible implementation, the feature extraction unit uses a convolutional neural network combined with an attention mechanism. Convolutional neural networks automatically extract local features from data and, through multiple layers of convolution and pooling, gradually extract higher-level feature representations. The attention mechanism enables the network to focus on key features in the data and suppress unimportant information, thereby highlighting key feature patterns and improving the accuracy and efficiency of feature extraction.

[0027] Furthermore, the parameter adjustment unit is used to adjust the pre-trained model parameter set according to the feature pattern set extracted by the feature extraction unit. Perform initialization adjustment. The pre-trained model parameter set is the model parameters that have been pre-trained on other related data and may not be applicable in the new area. The parameter adjustment unit uses the formula , combined with the feature pattern set, the pre-training model parameters are optimized, where is the adjusted parameter, is the learning rate adjustment coefficient, by adjusting The value of is used to control the step size of parameter adjustment, ensuring that the model can adapt quickly while avoiding over-adjustment, so that the model can better adapt to the data characteristics of the new area. is the parameter update function of the meta-learning algorithm (specifically the 'inner loop update function' in the MAML algorithm), represents the parameter adjustment calculation function based on the new regional feature pattern and pre-trained parameters, is a feature pattern set The characteristic patterns (such as land use change trends, building distribution associations, etc.), is a set of pre-trained model parameters The parameters.

[0028] Among them, the parameter adjustment calculation function The inner loop update mechanism based on meta-learning is implemented. The specific process includes: Input: New regional feature model (such as the “mountainous terrain-low building density” association feature), pre-training parameters (e.g. weight parameters related to building density prediction); Similarity calculation: using the cosine similarity formula , the matching degree between the quantized feature pattern and the pre-trained parameters (the value range is [−1, 1], the closer the value is to 1, the higher the matching degree); Adjustment amount generation: If (low matching degree), the adjustment amount is , ( is the adaptation coefficient, with a value of 0.05∼0.1), and is corrected by back propagation ;like (high matching degree), then the adjustment amount is (fine-tuning); Output: Adjustment amount ,Right now ; This process dynamically generates adjustments based on feature-parameter matching to ensure that pre-trained parameters quickly adapt to new regional features.

[0029] For example, in a specific embodiment, after the system obtains 70 sets of sample data of a new area, the feature extraction unit uses a convolutional neural network combined with an attention mechanism to extract a set of feature patterns such as land use change trends, building density distribution, and the relationship between population mobility and land use from the sample data. The parameter adjustment unit adjusts the pre-trained land space monitoring model parameter set based on these feature patterns. Perform initial adjustments.

[0030] In an optional embodiment, the feature extraction unit uses a neural network combined with an attention mechanism to extract features from the sample data.

[0031] Specifically, a neural network refers to a computational model that simulates the structure and function of biological neural networks. It is composed of multiple neurons and learns the characteristics and patterns of data through training. In this embodiment, a convolutional neural network is used, which includes a structure such as a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer performs a convolution operation on the input data by convolution kernels to extract local features of the data; the pooling layer downsamples the output of the convolutional layer to reduce the amount of data while retaining the main features; the fully connected layer integrates the output of the pooling layer to obtain the final feature representation.

[0032] The attention mechanism enhances key feature extraction within the feature extraction unit. It calculates the importance weights of different features, allowing the network to focus more on key components of the data. In one possible implementation, a self-attention mechanism is employed. This mechanism calculates attention weights between different positions within the same input sequence, capturing long-range dependencies in the data. For example, when processing land use data, the self-attention mechanism can focus on the associations between land use types in different regions, thereby more accurately extracting characteristic patterns of land use change.

[0033] For example, in a specific embodiment, when extracting features from the land space data of a new area of ​​a city, the convolutional neural network in the feature extraction unit first performs a convolution operation on the sample data to extract the preliminary features of the land use type. Then, the self-attention mechanism calculates the importance weight of each feature, assigning higher weights to key features related to land use change trends and lower weights to secondary features. Finally, through operations such as weighted summation, the key feature patterns are highlighted, and an accurate feature pattern set is obtained. , providing high-quality input data for subsequent parameter adjustment units. At the same time, the characteristic pattern set It will also serve as an important basis for building dynamic networks in the dynamic modeling module, providing support for determining node characteristics and connection relationships.

[0034] In an optional embodiment, the dynamic modeling module includes a network construction unit, a weight calculation unit and a structure update unit; The network construction unit is configured to construct a dynamic network with geographical environmental factors and monitoring targets as nodes based on the parameters and characteristic pattern set output by the initialization module; The weight calculation unit is used to calculate the edge weights between nodes in the dynamic network using the influence degree calculation function; The structure updating unit is used to dynamically update the dynamic network structure using the graph neural network.

[0035] Specifically, the network construction unit adopts a construction method based on knowledge graph. Knowledge graph is a semantic network that displays the relationship between entities in a graphical way. By taking geographical environmental factors and monitoring targets as entities and the influence relationship between them as edges, a dynamic network is constructed. This method can make full use of existing knowledge and experience, making the constructed dynamic network more in line with the actual situation. Feature pattern set The characteristic information in the data will be integrated into the attribute description of the network nodes to further enrich the node characteristics and enable the dynamic network to more accurately reflect the characteristics of the regional land space.

[0036] Furthermore, the weight calculation unit is used to calculate the function by the degree of influence Achieve accurate calculation of edge weights between dynamic network nodes. This function is based on massive historical data and uses machine learning algorithms for deep training and optimization, which can accurately quantify a set of geographical environmental factors. Any element in , for the monitoring target set Target In the actual calculation process, the weight calculation unit inputs the multi-dimensional correlation data of geographical environment factors and monitoring targets into the trained Function model. After model calculation, the model outputs the weight value corresponding to each edge. The larger the weight value, the stronger the impact of the corresponding geographical environmental factor on the monitoring target, providing a key quantitative basis for subsequent spatial dynamic analysis.

[0037] Furthermore, the structure update unit is used to dynamically update the dynamic network structure using a graph convolutional neural network (GNN). The graph convolutional neural network can perform convolution operations on graph structure data and update the feature representation of nodes. The structure update unit uses the formula Update the dynamic network structure. Represents the number of network layers. As the number of layers increases, the network can extract higher-level features. It is an activation function used to introduce nonlinear factors and enhance the expression ability of the network. Using the ReLU function ( ), used to introduce nonlinear features and solve the gradient vanishing problem, and is suitable for nonlinear relationship modeling of geographic features; , For the unit matrix, add self-connection so that each node can integrate its own information; is the degree matrix, which is based on the adjacency matrix The constructed diagonal matrix has the following elements on the diagonal: Representation node The degree of the node The number of connected edges is calculated as ; Through Perform the corresponding operations (such as ), the adjacency matrix Perform normalization processing to ensure balanced information transmission between different nodes, and avoid some nodes dominating the information transmission due to large differences in node degrees, which affects the accuracy of network structure updates; For the The weight matrix of the layer learns the feature transformation relationship between nodes; is the node feature matrix, which includes the initial features of the node and the features after network update.

[0038] Among them, the node feature matrix for A matrix of dimension, is the total number of nodes of geographical environmental factors and monitoring targets in the network, The node feature matrix is ​​the characteristic dimension of each node (such as terrain nodes including slope, aspect and other indicators), include: Initial features: refers to the basic feature values ​​assigned to nodes at the initial stage of network construction, such as terrain slope measurements directly extracted from sample data, and the proportion of land use types obtained from satellite remote sensing interpretation; Updated features: refers to the feature values ​​obtained after iterative updating of the graph convolutional neural network, that is, through the formula The calculated new feature matrix, where For the The feature matrix of the layer, For the sutra The updated feature matrix of the layer network contains the association information between nodes (such as the impact of terrain on land use).

[0039] Among them, the weight matrix It is used to implement linear transformation of node features. Its construction process and variable meanings are as follows: Dimension definition: for , a matrix where Indicates the The input feature dimension of the layer (that is, the node feature dimension of the previous layer output), Indicates the The output feature dimension of the layer (set by the model design based on the complexity of the task, such as 64, 128, etc.); Initialization method: Use He initialization method to meet ,in, Represents a normal distribution, the first parameter "0" is the mean of the distribution, the second parameter " " is the variance of the distribution. Initializing in this way can keep the network stable in the early stage of training. Update mechanism: During the model training process, Iterative optimization through back propagation algorithm to predict error Build a loss function based on ,in is the regularization coefficient (the range is ), used to suppress overfitting; updated by Adam optimizer , the learning rate of the optimizer is set to , first-order moment estimate index , the second moment estimate index , until the loss function convergence.

[0040] Among them, the prediction error (generated by the error calculation unit) is passed to the structure update unit through the optimization feedback module as Updated loss signal: When >0.2 (large error), trigger Forced update; when When the error is less than or equal to 0.2 (acceptable), only fine-tune The timing relationship between the two is: the error calculation unit output →Optimize feedback module to transmit signal→Structure update unit starts renew; At the same time, in order to make the update mechanism of the Adam optimizer clear and definite, its update method is now explained in detail. The specific steps include: Step s1, initialize the first-order moment estimate , second-order moment estimation , number of iterations ; Step s2, each iteration: ; Calculating gradients ( is the loss function ); Update the first moment: ( ); Update the second moment: ( ); Bias correction: , ; renew : ( ) is the learning rate, A constant to prevent division by zero); Step s3, repeat step s2 until (Loss converges), stop updating.

[0041] In step s2, is the deviation correction value of the first-order moment estimate, which is the first-order moment in the iterative process. The correction eliminates the deviation of the initial iteration moment estimation and makes the parameter update more accurate; ; indicates the The first-order moment estimate (momentum term) at iteration , which is used to accumulate the exponential moving average of the gradient; is the deviation correction value of the second-order moment estimation, which is the second-order moment in the iterative process. Correction to ensure the accuracy of the second-order moment estimation; Indicates the The second-order moment estimate at iteration , which is used as the exponential moving average of the accumulated gradient squared; To prevent division by zero for extremely small constants, avoid Leading to calculation errors; Specifically represents the learning rate of the Adam optimizer, the value , used to control the step size of parameter updates. The smaller the learning rate, the finer the parameter adjustment but the slower the convergence speed. The larger the learning rate, the faster the convergence speed but the more likely it is to cause oscillation. The value here is the optimal choice after comprehensively considering the model stability and convergence efficiency. Indicates the The loss function at the iteration About the weight matrix The gradient reflects the loss function Follow The trend of change is the core basis for parameter updating, and the negative gradient direction is the direction in which the loss function decreases fastest.

[0042] Furthermore, through the formula The specific steps for updating the dynamic network structure include: Data input: Import the current node feature matrix , which integrates multi-dimensional information such as geographical environment parameters and monitoring target feature vectors; at the same time, the adjacency matrix is ​​input , where the elements Representation Node and The initial connection weight of Matrix normalization: Augment the adjacency matrix: ( is the unit matrix, used to enhance the self-connection characteristics); calculate the degree matrix The diagonal elements of : ; Generate inverse square root degree matrix , complete normalization preprocessing; Feature iterative update: through matrix multiplication operation , combined with the ReLU activation function to achieve nonlinear transformation and output a new node feature matrix ; Dynamic adjustment of edge weights: based on the updated node feature matrix , using the cosine similarity formula Quantify the feature correlation between nodes; then calculate the similarity result with the influence degree function The output value of is weighted, that is, , complete the adaptive update of edge weights and realize dynamic optimization of network structure.

[0043] For example, in a specific embodiment, when the system monitors the land space of a certain urban area, the network construction unit uses a knowledge graph-based method based on the parameters and feature pattern set output by the initialization module to construct a dynamic network containing geographical environmental factor nodes such as terrain slope and climate type and monitoring target nodes such as land use change and infrastructure operation status. The characteristics of building distribution and traffic in the feature pattern set are integrated into the node attributes to make the network more realistic. The weight calculation unit calculates the influence degree through the function The calculated edge weight for terrain slope on land use change is 0.8, indicating that terrain slope has a significant impact on land use change. As monitoring data is updated, the structure update unit uses a graph convolutional neural network to update the dynamic network structure according to the above formula, adjusting node features and edge weights to ensure that the network better reflects the relationship between the current geographical environment and the monitoring target.

[0044] In an optional embodiment, the impact degree calculation function is obtained based on historical data and machine learning algorithm training, and is used to quantify the impact degree of geographical environmental factors on the monitoring target.

[0045] Specifically, the influence degree calculation function , which is based on historical data and machine learning algorithms to accurately quantify geographical environmental factors For monitoring targets Historical data refers to the geographical environment factor data and monitoring target data collected in similar geographical environments or related areas over a period of time. These data include various changes in geographical environment factors and the responses of corresponding monitoring targets.

[0046] Furthermore, machine learning algorithms can be of various types. In one possible implementation, a random forest algorithm is used for training. The random forest algorithm is an ensemble learning algorithm based on decision trees. It constructs multiple decision trees and integrates the results of these decision trees to obtain the final prediction result. When training the impact calculation function, the geographical environmental factor data in the historical data is used as the input feature, and the monitoring target data is used as the output label. The mapping relationship between the input feature and the output label is learned by the random forest algorithm, thereby obtaining the impact calculation function. .

[0047] Among them, the influence degree calculation function The random forest algorithm includes the following steps: Data preprocessing: normalize historical geographic environmental factor data (such as terrain slope and climate type) to the [0,1] interval, and convert monitoring target data (such as land use change rate) into continuous numerical labels; Model construction: The number of decision trees is set to 100, the maximum depth is 10, and the out-of-bag data (OOB) is used to evaluate the model performance; Training process: Using geographic environmental factor data as input features and monitoring target data as output labels, a training set is generated through bootstrap sampling. Each decision tree divides features based on the Gini coefficient, and ultimately outputs a predicted value through a voting mechanism. Function solidification: solidify the trained random forest model parameters (such as the split threshold of each decision tree and the output value of the leaf node) into a function that can be directly called , input geographical factors The monitoring target can be output The degree of influence (the value range is [0,1], the larger the value, the stronger the influence).

[0048] For example, in one embodiment, to calculate the terrain slope Land use change To determine the degree of impact, we collected the terrain slope data (such as slope value, slope change range, etc.) and land use change data (such as land use type conversion, land use area change, etc.) of the city and surrounding areas over the past five years as historical data. These data were input into the training model based on the random forest algorithm, and the impact degree calculation function was obtained after training. When there is new terrain slope data and land use change monitoring needs, the terrain slope data is substituted into the function The influence of terrain slope on land use change can be calculated, which provides a basis for calculating the edge weights between nodes in the dynamic network.

[0049] In an optional embodiment, the optimization feedback module includes an error calculation unit, a metric learning unit, and a parameter updating unit; The error calculation unit is used to compare the actual monitoring data set with the model prediction data set to calculate the prediction error; The metric learning unit is used to measure the similarity between samples using a metric function and generate a parameter update strategy using the reinforcement learning algorithm based on the prediction error; The parameter updating unit is used to update the measurement matrix and the parameters of the initialization module and the dynamic modeling module according to the parameter updating strategy.

[0050] Specifically, the error calculation unit is used to collect the actual monitoring data and model prediction data set Compare and calculate the prediction error. The actual monitoring data set is the real data about the urban land space collected by various monitoring equipment (such as sensors, satellite remote sensing, etc.), reflecting the actual geographical space status and changes. The model prediction data set is the prediction result of the future or current land space status based on the output of the dynamic modeling module in the system. The error calculation unit uses the formula Calculate the prediction error, where represents the number of data samples, Indicates the actual monitoring data data, Indicates the first This formula measures the accuracy of the model prediction by calculating the average of the sum of squares of the differences between the actual data and the predicted data. The smaller the prediction error, the more accurate the model prediction.

[0051] When calculating prediction errors, the error calculation unit can perform data preprocessing based on different data types and requirements. In one possible implementation, for numerical data, normalization is first performed, mapping the data to a specific interval such as [0, 1] or [−1, 1] to prevent the impact of data dimension differences on error calculation. For non-numeric data (such as categorical data such as land use types), it is converted to numerical form using methods such as one-hot encoding before error calculation.

[0052] Furthermore, the metric learning unit is used to pass the metric function Measure the similarity between samples and use reinforcement learning algorithm based on prediction error Generate parameter update strategy. Sample feature vector Contains feature information extracted from actual monitoring data and model prediction data, such as land use change trend characteristics, building density distribution characteristics, etc. These feature information and the feature pattern set extracted by the feature extraction unit in the initialization module Correlation is the specific embodiment of the characteristic pattern in actual data. Is a learnable matrix that determines the distance measurement method in the sample feature space. The element value of can change the weight of different feature dimensions when calculating sample similarity, so that the measurement function can better reflect the true similarity relationship between samples.

[0053] For example, when is the identity matrix, the metric function Equivalent to Euclidean distance; when By learning a specific non-unit matrix structure, we can adaptively adjust the distance metric in the feature space to highlight the key feature differences related to the urban land space monitoring task.

[0054] To ensure that the measurement function The distance metric validity, metric matrix It needs to satisfy the symmetric positive definite property, which is achieved through Cholesky decomposition: The measurement matrix Decompose into In the form of is a lower triangular matrix with all diagonal elements being positive; During model training, only the lower triangular matrix To ensure that the measurement matrix The positive definite property of , avoids the non-real number result of distance calculation, for the lower triangular matrix The diagonal elements of are constrained to be no less than ; During iterative updates, The initial value of is set to the unit lower triangular matrix. At the same time, to ensure the convergence stability, its learning rate is set to be consistent with the Adam optimizer and is taken as .

[0055] Reinforcement Learning Algorithms It is an algorithm that learns based on environmental feedback. Its core idea is to let the agent continuously perform actions in the environment and adjust the strategy according to the reward signal to maximize the long-term cumulative reward. In this system, the agent is the strategy set used to adjust the model parameters, the environment is the current model state (including dynamic network structure, model parameters, etc.) and prediction error, and the action is the measurement matrix As well as adjusting the parameters of the initialization module and the dynamic modeling module, the reward signal is designed based on the change in prediction error. Specifically, when the prediction error decreases after adjusting the parameters, it means that the action is effective, and a positive reward is given to encourage the adjustment direction; conversely, if the prediction error increases, a negative reward is given to encourage the strategy to avoid such adjustments. Through continuous trial and error and learning, we explore the optimal parameter update strategy to maximize model performance improvement.

[0056] The parameter update unit is used to update the metric matrix according to the parameter update strategy generated by the metric learning unit And the parameters of the initialization module and the dynamic modeling module. During the update process, the parameter update unit strictly follows the parameter update strategy to adjust the relevant parameters. For example, for the pre-trained model parameter set in the initialization module , adjust some of its parameter values ​​according to the strategy to make the model better adapt to new data; for the weight matrix of the graph neural network in the dynamic modeling module Parameters such as [1] and [2] are also updated accordingly to optimize the dynamic network structure.

[0057] For example, in a specific embodiment, when monitoring land use changes in a certain urban area, the error calculation unit collects the actual land use change data set collected by the sensor. Land use change data set with model predictions For comparison, according to the formula The prediction error is calculated to be 0.15. The metric learning unit uses the metric function based on the sample feature vector (such as the land use type distribution feature vector in different time periods) Calculate the similarity between samples and combine it with the prediction error through reinforcement learning algorithm Generate parameter update strategies, such as adjusting the metric matrix The parameter update unit adjusts the values ​​of some elements of the metric matrix according to the strategy. And update the relevant module parameters to complete the optimization of the model.

[0058] In an optional embodiment, the metric learning unit uses an optimization method based on gradient descent to iteratively update the metric matrix.

[0059] Gradient descent is a commonly used optimization algorithm. Its basic idea is to find the optimal solution by calculating the gradient of the objective function with respect to the parameters and updating the parameters in the opposite direction of the gradient to gradually reduce the value of the objective function. In this embodiment, the objective function is constructed based on the prediction error and sample similarity. The purpose is to adjust the metric matrix , making the model prediction results closer to the actual monitoring data, that is, reducing the prediction error.

[0060] Specifically, different variations of gradient descent optimization can be employed. In one possible implementation, stochastic gradient descent (SGD) is employed. SGD uses only one sample per iteration to calculate gradients and update parameters. This approach is computationally efficient, can process large amounts of data more quickly, and, to a certain extent, helps escape local optimal solutions. Another possible implementation is mini-batch gradient descent, which uses a small number of samples (i.e., a mini-batch) to calculate gradients. This combines the advantages of batch gradient descent and stochastic gradient descent, reducing the variance of parameter updates and making the training process more stable while also improving computational efficiency.

[0061] For example, in a specific embodiment, during the optimization process of the land space monitoring model of a certain urban area, the metric learning unit constructs the objective function based on the prediction error and the sample feature vector. , using the mini-batch gradient descent method, 32 samples are selected each time to form a mini-batch data. By calculating the objective function About the Metric Matrix Gradient , according to the formula Perform iterative updates, where Indicates the The metric matrix at the iteration, is the learning rate, which is used to control the step size of each parameter update. After multiple iterations, the metric matrix is ​​continuously adjusted , making the similarity measurement between samples more accurate, thereby optimizing the model performance.

[0062] In this embodiment, a lightweight learning framework is constructed through the initialization module, and a small amount of sample data is used to quickly complete the adaptation preparation of the model in the new area, thereby realizing the rapid adaptation of the model in different areas and improving the adaptation speed. Through the dynamic modeling module, a dynamic relationship model between geographical environmental factors and monitoring targets is established, and the complex relationship between the two is captured in real time, thereby improving the accuracy of monitoring. Through the optimization feedback module, the model is continuously optimized according to the actual monitoring results, and the model parameters are continuously adjusted, which effectively improves the generalization ability and adaptability of the model. The three work together to solve the problem of weak model generalization ability caused by the spatial heterogeneity of geographical phenomena, and realize efficient and accurate monitoring and analysis of urban land space in different regions.

[0063] Based on the above embodiments, the present invention further provides a specific embodiment of a system for dynamic monitoring and intelligent analysis of urban land space. The background of the specific embodiment is the entire process of urban land space monitoring in a mountainous city. The specific embodiment at least includes: Initialize the module: Step 111: Input: 50 sets of sample data (including terrain slope, land use type, etc.) for the mountainous city; Step 112, processing: Rapidly adapt through the meta-learning algorithm (MAML) and output the initial parameters for feature extraction , generating a set of feature patterns (“Slope>25°-Proportion of cultivated land<10%”), "Altitude > 800m - Construction land ratio < 5%")... ; Dynamic Modeling Module: Step 113, network construction: construct a dynamic network with terrain and climate as geographical environment nodes, and land use and infrastructure as monitoring target nodes; Step 114, weight calculation: through the influence function Calculate the slope Proportion of cultivated land ” has a weight of 0.75; Step 115, structure update: based on graph convolutional neural network formula , after 3 iterations, output the updated node feature matrix (Integrating the correlation characteristics of slope and cultivated land); Optimize feedback module: Step 116, error calculation: actual monitoring data With forecast data Error ; Step 117, parameter update: pass Calculate the adjustment amount, correct ; Updated by Adam optimizer ,make down to 0.09; Step 118, output: final monitoring results (priority areas for farmland protection, infrastructure optimization recommendations).

[0064] In this embodiment, mountainous cities are taken as monitoring objects, which fully demonstrates the targeted improvement of the present invention on the defects of traditional geological models: the initialization module completes the adaptation of the mountain environment with only 50 groups of samples, which greatly improves the efficiency compared with the traditional model that requires massive data retraining; the dynamic modeling module calculates the edge weight of the slope to the cultivated land ratio of 0.75, combines the graph convolutional neural network to iteratively update the feature matrix, accurately captures the relationship between the unique geographical environment of the mountains and the monitoring target, and overcomes the problem of insufficient accuracy caused by the fixed structure of the traditional model; the optimization feedback module reduces the prediction error from 0.18 to 0.09, realizes dynamic adjustment of parameters, and solves the problem of monitoring lag and large error after the traditional model is migrated to the new area, and finally efficiently outputs accurate results such as the priority area for cultivated land protection, meeting the needs of cross-regional and real-time land space monitoring.

[0065] Based on the above embodiments, Figure 2 As shown, the present invention also provides a method for dynamic monitoring and intelligent analysis of urban land space, which is used to support the system for dynamic monitoring and intelligent analysis of urban land space in the above embodiment. The method for dynamic monitoring and intelligent analysis of urban land space includes: Step 210: construct a lightweight learning framework. When entering a new area, use a preset number of sample data to trigger model adaptation, extract a set of characteristic patterns of the new area data through a learning algorithm, and initialize and adjust the pre-trained model parameter set based on the characteristic patterns.

[0066] Specifically, step 210 further includes: Step 211, extracting a feature pattern set of new region data from the preset number of sample data using the learning algorithm; Step 212: Initialize and adjust the pre-trained model parameter set according to the feature pattern set.

[0067] Among them, step 211 uses a neural network combined with an attention mechanism to extract features from the sample data.

[0068] Step 220: Based on the initialized and adjusted parameters and the characteristic pattern set, a dynamic network of geographical environmental factors and monitoring targets is established, the edge weights between network nodes are calculated through the influence degree calculation function, and a graph neural network is introduced to dynamically update the network structure.

[0069] Specifically, step 220 further includes: Step 221: constructing a dynamic network with geographical environmental factors and monitoring targets as nodes based on the parameters and characteristic pattern set output by the initialization module; Step 222: Calculate the edge weights between nodes in the dynamic network using the influence calculation function. Step 223: Use the graph neural network to dynamically update the dynamic network structure.

[0070] Among them, the impact degree calculation function is trained based on historical data and machine learning algorithms, and is used to quantify the impact of geographical environmental factors on monitoring targets.

[0071] Step 230: establish a monitoring result feedback channel, compare the actual monitoring data set with the model prediction data set obtained based on the dynamically updated network structure to calculate the prediction error, introduce a metric learning algorithm to measure the similarity between samples, and update the metric matrix and the pre-trained model parameters and dynamic network structure parameters through a reinforcement learning algorithm based on the prediction error.

[0072] Specifically, step 230 further includes: Step 231, comparing the actual monitoring data set with the model prediction data set to calculate the prediction error; Step 232: generating a parameter update strategy using the reinforcement learning algorithm based on the similarity between the samples measured by the metric function and the prediction error; Step 233: Update the metric matrix, the pre-trained model parameters, and the dynamic network structure parameters according to the parameter update strategy.

[0073] Wherein, step 232 adopts the optimization method based on gradient descent to iteratively update the metric matrix.

[0074] In this embodiment, by constructing a lightweight learning framework and using sample data to trigger model adaptation, a set of characteristic patterns of the new area is extracted to initialize and adjust the pre-trained model parameters, thereby achieving rapid adaptation of the model in the new area and shortening the preparation time. By establishing a dynamic network of geographical environmental factors and monitoring targets, calculating the node edge weights and dynamically updating the network structure, the complex relationship between the two is accurately captured, thereby improving the accuracy of monitoring. By comparing the calculation errors of actual and predicted data, the model parameters are updated using metric learning and reinforcement learning algorithms, and the model is continuously optimized, thereby enhancing the generalization ability and adaptability of the model. These steps work together to effectively solve the model adaptation problem caused by geographic spatial heterogeneity, and achieve efficient and accurate monitoring and intelligent analysis of urban land space.

[0075] Based on the above embodiments, Figure 3 As shown, the present invention also provides an electronic device, the electronic device comprising: At least one processor 22, at least one memory 21, a communication interface 23 and a communication bus 24, wherein the processor 22 is communicatively connected to the memory 21; In this embodiment, the memory 21 can be implemented in any appropriate manner, for example, the memory 21 can be a read-only memory, a mechanical hard disk, a solid-state drive, or a USB flash drive, etc. The memory 21 is used to store at least one executable instruction executed by the processor; In this embodiment, the processor 22 can be implemented in any appropriate manner. For example, the processor 22 can take the form of a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASIC), programmable logic controllers, and embedded microcontrollers, etc.; the processor is used to execute the executable instructions to implement the above-mentioned urban land space dynamic monitoring and intelligent analysis method.

[0076] Based on the above embodiments, the present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method for dynamic monitoring and intelligent analysis of urban land space as described above is implemented.

[0077] Those skilled in the art will appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0078] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, equipment and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or units can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or equipment, which can be electrical, mechanical or other forms.

[0080] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0081] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0082] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program instructions, such as a USB flash drive, a mobile hard disk, a read-only storage server, a random access storage server, a magnetic disk, or an optical disk.

[0083] In addition, it should be noted that the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0084] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many similar variations are possible. All variations directly derived from or associating with the present invention by those skilled in the art are intended to fall within the scope of protection of the present invention.

[0085] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A system for dynamic monitoring and intelligent analysis of urban land space, characterized by: include: The initialization module is used to build a lightweight learning framework. When the system enters a new area, it uses a preset number of sample data to trigger model adaptation. The learning algorithm extracts the characteristic pattern set of the new area data and initializes and adjusts the pre-trained model parameter set based on the characteristic pattern. A dynamic modeling module is used to establish a dynamic network of geographical environmental factors and monitoring targets based on the parameters and feature pattern set output by the initialization module, calculate the edge weights between network nodes through the influence degree calculation function, and introduce a graph neural network to dynamically update the network structure; The optimization feedback module is used to establish a monitoring result feedback channel, compare the actual monitoring data set with the model prediction data set based on the output of the dynamic modeling module to calculate the prediction error, introduce a metric learning algorithm to measure the similarity between samples, and update the metric matrix and the parameters of the initialization module and the dynamic modeling module through the reinforcement learning algorithm based on the prediction error.

2. The urban land space dynamic monitoring and intelligent analysis system according to claim 1 is characterized in that: The initialization module includes a feature extraction unit and a parameter adjustment unit; The feature extraction unit is used to extract a feature pattern set of new area data from the preset number of sample data using the learning algorithm; The parameter adjustment unit is used to initialize and adjust the pre-trained model parameter set according to the feature pattern set.

3. The urban land space dynamic monitoring and intelligent analysis system according to claim 2 is characterized in that: The feature extraction unit uses a neural network combined with an attention mechanism to extract features from the sample data.

4. The urban land space dynamic monitoring and intelligent analysis system according to claim 1 is characterized in that: The dynamic modeling module includes a network construction unit, a weight calculation unit and a structure update unit; The network construction unit is configured to construct a dynamic network with geographical environmental factors and monitoring targets as nodes based on the parameters and characteristic pattern set output by the initialization module; The weight calculation unit is used to calculate the edge weights between nodes in the dynamic network using the influence degree calculation function; The structure updating unit is used to dynamically update the dynamic network structure using the graph neural network.

5. The urban land space dynamic monitoring and intelligent analysis system according to claim 4 is characterized in that: The impact degree calculation function is obtained based on historical data and machine learning algorithm training, and is used to quantify the impact of geographical environmental factors on monitoring targets.

6. The urban land space dynamic monitoring and intelligent analysis system according to claim 1 is characterized in that: The optimization feedback module includes an error calculation unit, a metric learning unit and a parameter updating unit; The error calculation unit is used to compare the actual monitoring data set with the model prediction data set to calculate the prediction error; The metric learning unit is used to measure the similarity between samples using a metric function and generate a parameter update strategy using the reinforcement learning algorithm based on the prediction error; The parameter updating unit is used to update the measurement matrix and the parameters of the initialization module and the dynamic modeling module according to the parameter updating strategy.

7. The urban land space dynamic monitoring and intelligent analysis system according to claim 6 is characterized in that: The metric learning unit adopts an optimization method based on gradient descent to iteratively update the metric matrix.

8. A method for dynamic monitoring and intelligent analysis of urban land space, characterized in that: The following steps are involved: Build a lightweight learning framework that triggers model adaptation using a preset number of sample data when entering a new area. The learning algorithm extracts a set of characteristic patterns from the new area data and initializes and adjusts the pre-trained model parameter set based on the characteristic patterns. Based on the initialized and adjusted parameters and the characteristic pattern set, a dynamic network of geographical environmental factors and monitoring targets is established, the edge weights between network nodes are calculated through an influence degree calculation function, and a graph neural network is introduced to dynamically update the network structure; A monitoring result feedback channel is established, the actual monitoring data set is compared with the model prediction data set obtained based on the dynamically updated network structure to calculate the prediction error, a metric learning algorithm is introduced to measure the similarity between samples, and the metric matrix and the pre-trained model parameters and dynamic network structure parameters are updated through a reinforcement learning algorithm based on the prediction error.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory, wherein the memory is communicatively connected to the processor; The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the urban land space dynamic monitoring and intelligent analysis method as described in claim 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the urban land space dynamic monitoring and intelligent analysis method as described in claim 8.

Images