Smart city management system integrating artificial intelligence and big data

Abstract

The invention discloses a smart city management system fusing artificial intelligence and big data, and relates to the technical field of smart city management, and the system comprises a data collection and feature extraction module which is used for obtaining and processing multi-source city data, extracting road network topological features, functional area distribution features and people flow dynamic features through a graph neural network, and obtaining a data collection and feature extraction module; generating and storing a structured feature vector; the cross-domain knowledge migration model construction module is used for constructing a city simulator, generating virtual data through a generative adversarial network and reinforcement learning, bridging a source domain and a target domain based on the virtual data, and realizing migration of source domain knowledge to the target domain by utilizing graph matching and an optimal transmission theory; and the model optimization and alignment module is used for aligning feature distribution of a source domain and a target domain through confrontation training of a domain discriminator and a gradient inversion layer, dynamically adjusting data weight, realizing seamless migration from virtual distribution to real distribution of the target domain, and realizing seamless knowledge migration from a mature urban area to a new urban area.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart city management, in particular to a smart city management system integrating artificial intelligence and big data. BACKGROUND

[0002] With the rapid development of smart city construction, artificial intelligence and big data technology are increasingly widely used in city management. The existing technology usually trains AI models based on historical data of mature urban areas, analyzes multiple source data such as road network topology, POI distribution, and human flow dynamics, and realizes intelligent management of tasks such as traffic flow prediction and functional area planning. For example, traditional methods use graph neural networks to extract urban spatial features and combine time series models to capture dynamic patterns, which perform well in mature urban areas with sufficient data.

[0003] However, the existing technology has significant shortcomings. First, new urban areas lack historical data accumulation, resulting in missing spatiotemporal features, making it difficult for traditional AI models to accurately capture their dynamic evolution patterns, and the prediction results have large deviations. Second, when models trained in mature urban areas are directly transferred to new urban areas, the model performance significantly decreases due to domain shift problems, forming a small sample learning dilemma. For example, the road network structure and human flow distribution of new urban areas differ greatly from those of mature urban areas, and existing models cannot effectively adapt to such distribution differences. In addition, the existing technology has insufficient coverage of long-tail scenarios. New urban areas may have sparse or unevenly distributed data in the early stages of development, and traditional methods have difficulty learning generalizable features from limited data, leading to prediction failures for rare scenarios such as sudden human flow peaks. Although some research attempts to alleviate this problem through data augmentation or transfer learning, it lacks systematic integration of virtual data generation and cross-domain knowledge transfer, making it difficult to achieve seamless knowledge transfer from mature urban areas to new urban areas. SUMMARY

[0004] (I) Technical problems solved To address the shortcomings of the existing technology, the present application provides a smart city management system integrating artificial intelligence and big data, which solves the problem of model performance degradation due to data scarcity in new urban areas by constructing a cross-domain knowledge transfer model and a virtual data generation mechanism, and achieves seamless knowledge transfer from mature urban areas to new urban areas, significantly improving the prediction accuracy and adaptability of the smart city management system in data-scarce scenarios.

[0005] (II) Technical solutions To achieve the above purpose, the present application is implemented by the following technical solutions: a smart city management system integrating artificial intelligence and big data, comprising: The data acquisition and feature extraction module is configured to acquire and process multi-source urban data, including road network topology data, POI data, and mobile signaling data, extract road network topology features, functional area distribution features, and human flow dynamic features through a graph neural network, generate a structured feature vector, and store the structured feature vector; The cross-domain knowledge transfer model construction module is configured to construct a city simulator, generate virtual data through a generative adversarial network and reinforcement learning, bridge the source domain and the target domain based on the virtual data, and realize the transfer of the source domain knowledge to the target domain by using graph matching and optimal transport theory. The model optimization and alignment module is configured to align the feature distribution of the source domain and the target domain through adversarial training of a domain discriminator and a gradient reversal layer, and dynamically adjust the data weight to realize seamless migration from a virtual distribution to a real distribution of the target domain.

[0006] Further, in the data acquisition and feature extraction module, the road network topology data includes road node coordinates, road edge geometry information, road grades, and connectivity; the POI data includes POI names, geographic coordinates, categories, and operating status; and the mobile signaling data includes device unique identifiers, timestamps, and base station locations.

[0007] Further, in the data acquisition and feature extraction module, the graph neural network includes a node embedding layer, a multi-scale aggregation layer, and a feature output layer; the node embedding layer converts road nodes, POIs, and signaling base stations into multi-dimensional vectors; the multi-scale aggregation layer extracts road network topology features, functional area distribution features, and human flow dynamic features; and the feature output layer integrates the extracted features into a structured feature vector, which is stored in a database and indexed with a spatial grid ID and a timestamp as joint primary keys.

[0008] Further, in the cross-domain knowledge transfer model construction module, the city simulator includes a data input layer, a generation module, a reinforcement learning module, and an output layer; the generation module uses a Wasserstein generative adversarial network with gradient penalty to generate virtual data, including virtual road networks, virtual POI distributions, and virtual human flow dynamics; and the reinforcement learning module uses the distribution similarity of the virtual data and the real data as a reward signal, and uses a deep deterministic policy gradient algorithm to train the generator parameters.

[0009] Further, in the cross-domain knowledge transfer model construction module, the source domain is a mature urban area, and the target domain is a newly built urban area; graph structures of the source domain, the target domain, and the virtual data are constructed, with spatial grids as basic nodes, node features including road network grade average, POI category proportion, and peak human flow, and edge weights being the feature similarity of adjacent grids; the Louvain community discovery algorithm is used to aggregate adjacent nodes to form superpoints, and the superpoint features are the average of the features of the included basic nodes.

[0010] Further, in the cross-domain knowledge transfer model construction module, the source domain and the target domain are bridged through a virtual graph, the matching degree of super points is calculated, and the corresponding relationship with a matching degree not lower than a matching degree threshold is reserved; a domain adaptation algorithm based on graph matching is used to calculate the initial correspondence of the super points of the source domain and the target domain, and the feature mapping is optimized through the optimal transport theory, the super point features of the source domain are weighted and mapped to the target domain, and knowledge transfer is realized.

[0011] Further, in the model optimization and alignment module, the domain discriminator is a fully connected neural network, the source domain super point features and the target domain enhanced features are input, and the probability of the features belonging to the domain is output; a gradient reversal layer is inserted between the multi-scale aggregation layer and the domain discriminator, the gradient is multiplied by a fixed coefficient -1 during back propagation, and the multi-scale aggregation layer is forced to learn domain-independent general features.

[0012] Further, in the model optimization and alignment module, the target of the adversarial training is to minimize the feature extraction loss and the domain adversarial loss; the weight is adjusted according to the amount of real data of the target domain; the features of different sources are fused through the weighted average method, and the features input into the model are the weighted sum of the source domain features, the virtual features and the real features of the target domain; the training target is to minimize the prediction error until the accuracy of the model on the target domain validation set converges, and seamless knowledge transfer is realized.

[0013] An electronic device includes a memory, a processor, and a computer program stored on the memory and running on the processor, and the processor implements a smart city management system integrating artificial intelligence and big data when executing the computer program.

[0014] A computer readable storage medium has a computer program stored thereon, and the computer program implements the above-mentioned smart city management system integrating artificial intelligence and big data when executed.

[0015] (Three) beneficial effects The present application provides a smart city management system integrating artificial intelligence and big data, which has the following beneficial effects: (1) By integrating multi-source city data, using a graph neural network to extract a structured feature vector, the standardization and efficient processing of city data are realized, the data quality is improved, the feature expression capability is enhanced, and a unified, indexable high-dimensional feature basis is provided for subsequent cross-domain knowledge transfer, thereby significantly improving the analysis accuracy and decision efficiency of the smart city management system.

[0016] (2) By generating virtual data and constructing a city simulator through reinforcement learning, combining graph matching and optimal transport theory, effective transfer of mature city knowledge to newly built cities is realized, the problem of data scarcity in newly built cities is solved, the influence of domain shift is reduced, and the model generalization capability is improved, thereby significantly improving the planning accuracy and adaptability of the smart city management system in data scarce scenarios.

[0017] (3) Through the domain discriminator and the gradient reversal layer, the feature distribution of the source domain and the target domain is effectively aligned, the domain shift influence is reduced, the data weight is dynamically adjusted, the model is ensured to smoothly transition from the virtual distribution to the real distribution, the domain-independent general features are generated, the model generalization is enhanced, the multi-source data is fused through the weighted fusion, and the prediction accuracy and the migration stability of the smart city management system in the data scarce scene are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Fig. 1 It is a structure schematic diagram of the smart city management system of the present application fusing artificial intelligence and big data; Fig. 2 It is a process schematic diagram of the model optimization and alignment module. DETAILED DESCRIPTION

[0019] 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 part of the embodiments of the present application, rather than all the 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 protection scope of the present application.

[0020] Please refer to Figs. 1-2 The present application provides a smart city management system fusing artificial intelligence and big data, comprising: a data acquisition and feature extraction module, a cross-domain knowledge migration model construction module and a model optimization and alignment module, wherein: The data acquisition and feature extraction module is used for acquiring and processing multi-source city data, including road network topology data, POI data and mobile signaling data, extracting road network topology features, functional area distribution features and human flow dynamic features through a graph neural network, and generating a structured feature vector; Specifically, the road network topology data of the target city is acquired through the road network data API interface provided by the map open platform, such as the Gaode map open API interface, including road node coordinates, road edge geometry information, road grades, road lengths and connectivity attributes, etc., wherein the road grades include expressways, main roads, secondary roads and branch roads, the collection frequency is set to once a day, and the temporary construction or road diversion information is dynamically updated; POI data reflecting the distribution of urban functional areas are collected through the city geographic information public service platform and commercial POI (Point of Interest) databases such as Meituan and Dianping POI open data. The POI categories include residential, commercial, public service and transportation hub, among which the residential category includes residential buildings and communities, the commercial category includes shopping malls and catering, the public service category includes schools and hospitals, and the transportation hub category includes subway stations and bus stations. The data fields include POI name, geographic coordinates, category and opening status. Effective POIs are selected during collection, and records that have been cancelled or closed are excluded. Anonymized mobile phone signaling data are obtained in cooperation with communication operators, from which user identity identifiers have been removed, leaving only device unique identifiers, timestamps and base station locations, which are used to reflect human flow dynamics. The collection range is explicitly the administrative area of the target city, and the time granularity is every 15 minutes, including the spatiotemporal location sequence of the user. For road network data, the standard deviation method is used to identify outliers, calculate the mean μ and standard deviation σ of road length, and exclude abnormal road records with length greater than μ+3σ or less than μ-3σ. Isolated nodes, i.e. nodes without connected roads, are excluded through topological verification. For POI data, POIs outside the administrative boundaries of the target city are excluded through geographic coordinate verification. Duplicate records are excluded through repeated name + coordinate matching, such as a POI being marked multiple times. For mobile phone signaling data, abnormal positions caused by signal drift are excluded, such as unreasonable movement records with a position span exceeding 50 kilometers within 15 minutes, and null records caused by device failure. Standardization processing is performed on the multi-source city data to convert them into the WGS84 coordinate system, ensuring the consistency of spatial positions. A graph neural network is constructed, which includes a node embedding layer, a multi-scale aggregation layer and a feature output layer. The node embedding layer converts entities such as road nodes, POIs and signaling base stations in multi-source city data into multi-dimensional vectors. The multi-scale aggregation layer extracts road network topology features, functional area distribution features and human flow dynamic features. The road network topology features include road grade features and connectivity features. The road grade enumeration values in the road network data are embedded through the graph neural network to output road grade vectors as road grade features. Based on the road network adjacency matrix, the connectivity probability between nodes is calculated through the graph neural network to generate a connectivity heat map. The standardized road grade enumeration values are converted into one-hot encoding vectors. For example, road grade enumeration values: 1=highway, 2=main road, 3=secondary road, 4=branch road, and one-hot encoding vectors: highway corresponds to [1, 0, 0, 0], main road corresponds to [0, 1, 0, 0], and so on, forming an initial grade feature vector. In the node embedding layer of the multi-scale graph neural network, the embedding dimension is set, the embedding layer weight matrix is a trainable parameter matrix, and the initial grade feature vector is mapped to a multi-dimensional embedding vector through matrix multiplication, with the formula: embedding vector = initial grade feature vector x weight matrix + bias term. The road intersection is taken as a node to construct a road network adjacency matrix, and the value of the road network adjacency matrix element is determined according to the following rules: if the nodes are directly connected, that is, there is a physical road, the value is the grade weight of the road, highway = 4, main road = 3, secondary road = 2, branch = 1; if not directly connected, it is 0; the graph convolution network (GCN) is used to calculate the connectivity probability, the GCN includes 2 layers of hidden layers, the output dimension of the first layer is 64, the output dimension of the second layer is 32, and the activation function is ReLU, the input is the adjacency matrix, and for any two nodes, the connectivity probability is the dot product of the second layer output vector after normalization by the softmax function, ranging from 0 to 1; The functional area distribution characteristics include a POI heat map and a format ratio, and at a 1km*1km grid scale, the POI quantity in the grid is aggregated by a graph neural network to generate a spatial heat value matrix of each POI category, which is used as the POI heat map; the format ratio vector is output by calculating the proportion of the number of each POI in each grid, such as the number of commercial POIs / total POI number in the grid, which is used as the format ratio; The human flow dynamic characteristics include an OD matrix and a space-time trajectory, and based on mobile phone signaling data, the number of personnel flow between grids within 15 minutes is counted by a graph neural network to generate an OD matrix, row = starting grid, column = ending grid, and value = number of flow; the space-time trajectory is obtained by time series modeling of the signaling position sequence of a single device to extract the stopping point and moving path of the trajectory, and the stopping point is a position with a duration of more than or equal to 30 minutes; The feature output layer integrates the extracted road network topology characteristics, functional area distribution characteristics and human flow dynamic characteristics into a structured feature vector, which is stored in a database, an index is established with the spatial grid ID and the time stamp as the joint primary key, and the urban structure basic data is formed.

[0021] By integrating multi-source urban data and using a graph neural network to extract a structured feature vector, the standardization and efficient processing of urban data are realized, the data quality is improved, the feature expression capability is enhanced, and a unified and indexable high-dimensional feature basis is provided for subsequent cross-domain knowledge transfer, thereby significantly improving the analysis accuracy and decision-making efficiency of the smart city management system.

[0022] The cross-domain knowledge transfer model construction module is used to construct an urban simulator, generate virtual data through a generative adversarial network and reinforcement learning, and bridge the source domain and the target domain based on the virtual data, and realize the transfer of the knowledge of the source domain to the target domain by using graph matching and optimal transport theory; Specifically, the urban simulator including a data input layer, a generation module, a reinforcement learning module and an output layer is constructed, wherein the input layer receives the urban structure data; the generation module and the reinforcement learning module work cooperatively to output virtual data; The generation module adopts a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) as the generation network core to avoid mode collapse, wherein the generator inputs a random noise vector and a structured feature vector such as a format ratio vector, and outputs virtual data including a virtual road network, a virtual POI distribution, and virtual human flow dynamics; the discriminator inputs real data and virtual data, and the real data is standardized multi-source urban data, and the continuity of the discriminator is constrained by a gradient penalty term, and the loss function of the discriminator is composed of three parts: the expected discrimination result of the real data, the expected discrimination result of the virtual data, and the gradient penalty term, and the penalty coefficient is set to 10; The reinforcement learning module takes the distribution similarity of the virtual data and the real data as the reward signal, the state space is the feature distribution of the currently generated virtual data, the action space is the parameter adjustment amount of the generator, the reward value is measured by the JS divergence of the difference between the real data distribution and the virtual data distribution, and the negative exponential result of the difference is taken, a deep deterministic policy gradient algorithm is used for training, the Actor network outputs the parameter adjustment amount, the Critic network evaluates the action value, and the iteration number is set to 5000 times, and the sample amount of each iteration is 1024; When the loss of the discriminator of the generative adversarial network converges and the reward value of the reinforcement learning is not less than 0.8, the loss converges, that is, the loss fluctuation of 50 consecutive iterations does not exceed 0.01, and the virtual data is output, which contains structured features homologous to the real data, such as virtual road network topology, virtual POI heat map, and virtual OD matrix, to form a virtual city copy that can match the source domain and the target domain; The source domain is defined as a mature urban area, that is, the built-up time is not less than 10 years, and the target domain is a newly built urban area, that is, the built-up time is not more than 3 years; the source domain data and the target domain initial data are split from the urban structure basic data, and are associated with the generated virtual data based on the spatial grid ID; The graph structure is constructed for the source domain and the target domain, respectively, with a 500m x 500m spatial grid as the basic node, and the node features include the average road network level, the POI category proportion, and the 15-minute peak human flow in the grid, which are extracted from the source domain data and the target domain data; the weight of the edge is the feature similarity of two adjacent grids, which is calculated by the cosine similarity; if the physical distance between two grids exceeds 1km, the edge weight is set to 0, which is considered as non-adjacent; similarly, the graph structure is constructed based on the virtual data to generate a virtual graph; In the source domain and the target domain, adjacent nodes are aggregated to form superpoints, representing urban functional units such as business districts and residential areas, and the Louvain community discovery algorithm is used for aggregation, with the maximum modularity as the target, and the modularity threshold is set to 0.5; each superpoint contains not less than 5 basic nodes, and the superpoint feature is the average feature of the contained basic nodes; finally, the source domain and the target domain form a double-layer structure of basic nodes and superpoints, and the number of superpoints is 1 / 10~1 / 5 of the number of basic nodes; The association between the source domain and the target domain is established by taking the virtual graph as an intermediate bridge, the super node matching degree of the source domain and the virtual graph and the super node matching degree of the virtual graph and the target domain are calculated, the super node matching degree is calculated based on the cosine similarity of the super node feature vector, the corresponding relationship of the super nodes with a matching degree not lower than a matching degree threshold is reserved, a double-mode graph structure of source domain→virtual graph→target domain is formed, the matching degree threshold is set based on the actual situation, and the threshold value that makes the prediction accuracy highest and the domain adversarial loss smallest and stable, such as 0.7, is selected; A domain adaptation algorithm based on graph matching is used to calculate the node-level corresponding relationship of the super nodes in the source domain and the target domain, the super node adjacency matrix of the source domain and the target domain is extracted, and then the Laplacian matrix of each is calculated, the Laplacian matrix is the difference value of the degree matrix and the adjacency matrix, the feature vectors corresponding to the first 20 largest eigenvalues are obtained through eigenvalue decomposition, and the feature matrix of the source domain and the target domain is constructed; The value of an element in the super node adjacency matrix is calculated in the following manner: all the basic nodes contained in the ith super node are extracted and denoted as set P, and all the basic nodes contained in the jth super node are denoted as set Q, the edge weight of all basic node pairs in sets P and Q, i.e., the feature similarity, is calculated, the value range is 0~1, and the average value of these edge weights is taken as the value of the element; if the two super nodes are not connected, i.e., the edge weight of all basic node pairs is 0, the value of the element is 0; The feature vector similarity is taken as a measure to construct a matching matrix, each element in the matrix is calculated through the Euclidean distance of the super node feature vectors of the source domain and the target domain, specifically the negative exponential result of the Euclidean distance, and a larger value indicates that the two super node features are more similar; the Hungarian algorithm is used to solve the maximum weight matching of the matching matrix to obtain the initial corresponding relationship of the source domain super nodes and the target domain super nodes; Based on the initial corresponding relationship, the optimal transport theory is used to optimize the feature mapping from the source domain to the target domain, the source domain feature distribution (super node feature distribution) and the target domain feature distribution are defined, each element in the transport cost matrix is the Euclidean distance of the source domain super node feature and the target domain super node feature, i.e., the square root of the sum of the squares of the corresponding element differences of the two feature vectors; the Sinkhorn algorithm is used to solve the optimal transport matrix, the iteration number is 100, and the regularization parameter is 0.01; the optimal transport matrix satisfies that the row sum is equal to the source domain distribution, the column sum is equal to the target domain distribution, and the total transport cost is minimized; According to the optimal transport matrix, the source domain super node features are mapped to the target domain to generate enhanced features of the target domain, i.e., each target domain super node feature is a weighted sum of the source domain super node features, and the weight is determined by the transport matrix, realizing the preliminary migration of the source domain knowledge to the target domain.

[0023] The city simulator is constructed by virtual data generation and reinforcement learning, and the mature city knowledge is effectively migrated to the newly built city by combining graph matching and optimal transmission theory, solving the problem of data scarcity in newly built cities, reducing the influence of domain bias, and improving the model generalization ability, thereby significantly improving the planning accuracy and adaptability of the smart city management system in data scarce scenarios.

[0024] The model optimization and alignment module is used to align the feature distribution of the source domain and the target domain through adversarial training of the domain discriminator and the gradient reversal layer, and dynamically adjust the data weight to realize seamless migration from the virtual distribution to the real distribution of the target domain. A domain discriminator is constructed to distinguish the source of the features, which inputs the source domain superpoint features and the target domain enhanced features, both of which are structured features at the superpoint level, and outputs the judgment result of the domain to which the features belong, i.e., the source domain or the target domain. The domain discriminator adopts a 3-layer fully connected neural network structure, with node numbers of 256, 128, and 1 respectively, and activation functions of ReLU, ReLU, and Sigmoid respectively, for outputting the probability of the features belonging to the source domain, which is between 0 and 1. A gradient reversal layer is inserted between the multi-scale aggregation layer and the domain discriminator. The gradient reversal layer does not change the feature data during forward propagation, but multiplies the gradient by a fixed coefficient -1 during backward propagation, so that the multi-scale aggregation layer receives the gradient signal opposite to the domain discriminator. An adversarial loss function is defined, which includes a feature extraction loss and a domain adversarial loss. The feature extraction loss adopts a cross-entropy loss, which trains the multi-scale aggregation layer based on the labels of the source domain data, such as known traffic flow and functional area attributes, to ensure that it can accurately extract task-related features. The domain adversarial loss adopts a binary cross-entropy loss, which takes the difference between the output of the domain discriminator and the real domain label as the optimization target. The domain label is 1 for the source domain and 0 for the target domain. Through the gradient reversal layer, the multi-scale aggregation layer is forced to learn domain-independent general features, i.e., the domain discriminator cannot distinguish the source of the features. The multi-scale aggregation layer and the domain discriminator are jointly trained, with a batch size of 64 and a learning rate of 0.001. The Adam optimizer is used, and when the accuracy of the domain discriminator approaches 50%, it is determined that the feature distribution of the source domain and the target domain is aligned, with a fluctuation of 48%-52% in the accuracy of the last 30 iterations. The virtual data is defined as an intermediate domain, which is divided into an intermediate domain training set and a validation set in a 7:3 ratio, and has the same data format as the source domain and the target domain, and contains road network, POI, and human flow dynamic characteristics. The source domain data (mature urban area data) and the intermediate domain training set are used as inputs to train the multi-scale aggregation layer and the domain adaptation algorithm. The training target is to minimize the mapping error between the source domain features and the intermediate domain features. The mean square error is used as the measurement, and the number of iterations is set to 300. When the validation set error decreases to less than 30% of the initial value, the training is stopped, so that the cross-domain knowledge transfer model is preliminarily adapted to the virtual distribution; The intermediate domain validation set and the initial data of the target domain (new urban area data) are used to continue training the model. The mean square error is used as the measurement to optimize the mapping precision of the target domain features. During the training process, the proportion of intermediate domain data is gradually reduced from 70% to 30%, and the proportion of target domain data is gradually increased from 30% to 70%. The number of iterations is set to 200, and the smooth transition from virtual distribution to real target domain distribution is realized. The KL divergence is used to calculate the feature distribution difference between the source domain and the intermediate domain, and the difference between the intermediate domain and the target domain. If the ratio of the direct difference between the source domain and the target domain to the sum of the differences is greater than or equal to 1.5, the path is determined to be effective, i.e., the domain difference of the gradual transfer is reduced by more than 50% compared with the direct transfer. In the initial stage of model training, the weight of the source domain data is set to 0.5, the weight of the virtual data (intermediate domain) is set to 0.4, and the weight of the real data of the target domain is set to 0.1. The total weight is 1. When the amount of real data of the target domain reaches 50% of the initial data amount, the first weight adjustment is started. Every 30% of the subsequent data amount, i.e., relative to the total amount at the last adjustment, triggers an adjustment, until the real data amount reaches 80% of the source domain data amount. In each adjustment, the real data weight of the target domain = current weight + 0.2, and the upper limit is 0.8. The source domain data weight = current weight - 0.1, and the lower limit is 0.1. The virtual data weight = 1 - real data weight - source domain data weight, and the lower limit is 0.1. The weighted average method is used to fuse the data features from different sources, i.e., the final input feature of the model = (source domain feature x source domain weight) + (virtual feature x virtual data weight) + (real target domain feature x real data weight). The fused data is used to update the feature extraction module and the cross-domain mapping model. The training target is to minimize the prediction error, such as traffic flow prediction error and planning evaluation deviation. The model accuracy on the target domain validation set converges, such as a continuous 20-time iteration fluctuation ≤2%, to realize seamless knowledge transfer from the source domain to the target domain.

[0025] The source domain and the target domain are effectively aligned in feature distribution through the adversarial training of the domain discriminator and the gradient reversal layer, the influence of domain shift is reduced, the data weight is dynamically adjusted, the model is ensured to smoothly transition from the virtual distribution to the real distribution, the domain-independent general features are generated, the model generalization is enhanced, the multi-source data is fused through weighting, and the prediction accuracy and migration stability of the smart city management system in the data scarce scene are significantly improved.

[0026] The application further provides an electronic device, including a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the smart city management system integrating artificial intelligence and big data when executing the computer program.

[0027] The application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program implements the smart city management system integrating artificial intelligence and big data when being executed.

[0028] In the application, the several formulas involved are all dimensionless numerical calculations, the formula is a formula obtained by software simulation of a large amount of collected data to reflect the most real situation, and the coefficients in the formula are set by a person skilled in the art according to the actual situation.

[0029] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software and a combination of electronic hardware and computer software. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions.

[0030] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, and can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0031] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the application, which should be covered within the protection scope of the application.

Claims

1. A smart city management system fusing artificial intelligence and big data, characterized in that: Comprise: a data acquisition and feature extraction module for acquiring and processing multi-source urban data, including road network topology data, POI data and mobile signaling data, extracting road network topology features, functional area distribution features and human flow dynamic features through graph neural networks, generating structured feature vectors and storing them; a cross-domain knowledge transfer model construction module for constructing a city simulator, generating virtual data through generative adversarial networks and reinforcement learning, bridging the source domain and the target domain based on the virtual data, and realizing the transfer of source domain knowledge to the target domain using graph matching and optimal transport theory; a model optimization and alignment module for aligning the feature distribution of the source domain and the target domain through adversarial training of the domain discriminator and the gradient reversal layer, and dynamically adjusting the data weight to realize seamless migration from the virtual distribution to the real distribution of the target domain.

2. The smart city management system integrating artificial intelligence and big data according to claim 1, characterized in that: in the data acquisition and feature extraction module, the road network topology data includes road node coordinates, road edge geometry information, road grades and connectivity; the POI data includes POI names, geographic coordinates, categories and operating status; and the mobile signaling data includes device unique identifiers, timestamps and base station locations. 3.The smart city management system of fusing artificial intelligence with big data according to claim 2, characterized in that: in the data acquisition and feature extraction module, the graph neural network includes a node embedding layer, a multi-scale aggregation layer and a feature output layer; the node embedding layer converts road nodes, POIs and signaling base stations into multi-dimensional vectors; the multi-scale aggregation layer extracts road network topology features, functional area distribution features and human flow dynamic features; the feature output layer integrates the extracted features into structured feature vectors, which are stored in a database and indexed with spatial grid ID and timestamp as the joint primary key. 4.The smart city management system fusing artificial intelligence and big data according to claim 1, characterized in that: in the cross-domain knowledge transfer model construction module, the city simulator includes a data input layer, a generation module, a reinforcement learning module and an output layer; the generation module uses a Wasserstein generative adversarial network with gradient penalty to generate virtual data, including virtual road networks, virtual POI distributions and virtual human flow dynamics; the reinforcement learning module uses the distribution similarity of virtual data and real data as a reward signal and uses a deep deterministic policy gradient algorithm to train the generator parameters. 5.The smart city management system fusing artificial intelligence and big data according to claim 4, characterized in that: in the cross-domain knowledge transfer model construction module, the source domain is a mature urban area and the target domain is a newly built urban area; graph structures of the source domain, the target domain and the virtual data are constructed, with spatial grids as basic nodes and node features including road network grade average, POI category proportion and peak human flow; edge weights are the feature similarity of adjacent grids; the Louvain community discovery algorithm is used to aggregate adjacent nodes to form superpoints, and the superpoint features are the average of the features of the included basic nodes. 6.The smart city management system fusing artificial intelligence and big data according to claim 5, characterized in that: in the cross-domain knowledge transfer model construction module, the source domain and the target domain are bridged through a virtual graph, the superpoint matching degree is calculated and the corresponding relationship with a matching degree not lower than the matching degree threshold is retained; a domain adaptation algorithm based on graph matching is used to calculate the initial correspondence of the source domain and the target domain superpoints, and the optimal transport theory is used to optimize the feature mapping, realizing the weighted mapping of the source domain superpoint features to the target domain and achieving knowledge transfer. 7.The smart city management system fusing artificial intelligence and big data according to claim 1, characterized in that: In the model optimization and alignment module, the domain discriminator is a fully connected neural network, which inputs the source domain superpoint feature and the target domain enhanced feature, and outputs the probability of the feature belonging to the domain; a gradient reversal layer is inserted between the multi-scale aggregation layer and the domain discriminator, and when back propagation, the gradient is multiplied by a fixed coefficient-1, so that the multi-scale aggregation layer receives the gradient signal opposite to the domain discriminator. 8.The smart city management system fusing artificial intelligence and big data according to claim 7, characterized in that: In the model optimization and alignment module, the target of the adversarial training is to minimize the feature extraction loss and the domain adversarial loss; the weight is adjusted according to the amount of real data in the target domain; the data features of different sources are fused by the weighted average method, and the input feature of the model is the weighted sum of the source domain feature, the virtual feature and the real feature of the target domain; The training target is to minimize the prediction error until the accuracy of the model on the target domain validation set converges, realizing seamless knowledge transfer.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to realize the steps performed by the smart city management system integrating artificial intelligence and big data in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed to realize the steps performed by the smart city management system integrating artificial intelligence and big data in any one of claims 1-8.

Images