Urban population dynamic monitoring and trajectory analysis method and system based on big data

Abstract

The invention provides an urban population dynamic monitoring and trajectory analysis method and system based on big data. The method comprises the steps of obtaining a multi-modal data set in a city preset range, performing preprocessing and gridding mapping to obtain a multi-modal optimization data set corresponding to a city grid, then performing processing through a preset cross-mechanism federated learning model in combination with a dynamic weight distribution mechanism to generate a cross-mechanism fusion feature matrix, and finally obtaining a multi-modal fusion feature matrix. Coding processing is carried out through a preset LSTM model, processing is carried out through a preset DPC clustering algorithm, and a high-frequency path and a parking high-density area are obtained for visual output; according to the method, the data richness is improved through multi-modal data fusion, the data privacy is protected by utilizing cross-mechanism federated learning, and the real-time performance and accuracy of the data are improved by combining the LSTM model and the DPC clustering algorithm, so that high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis are realized.

Description

Technical Field

[0001] The present application relates to the field of urban intelligent management technology, and specifically to a method and system for urban population dynamic monitoring and trajectory analysis based on big data. Background Art

[0002] The rapid development and increasing intelligence of cities are placing higher demands on urban population monitoring and analysis. Traditional demographic methods suffer from problems such as a single data source, untimely data collection and updates, and insufficient analysis granularity, making them unable to meet the data analysis needs of smart cities. Existing technologies such as CN115665677A, while involving population monitoring, primarily analyze signaling data and fail to achieve the coordinated optimization of multimodal data integration, data privacy protection, and real-time analysis. Therefore, there is an urgent need for a method for urban population dynamics monitoring and trajectory analysis that balances data multi-source, data privacy protection, and analysis timeliness.

[0003] In response to the above problems, effective technical solutions are urgently needed. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for urban population dynamic monitoring and trajectory analysis based on big data, which can improve data richness through multimodal data fusion, protect data privacy through cross-institutional federated learning, and improve data real-time and accuracy by combining LSTM model and DPC clustering algorithm, thereby achieving high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis.

[0005] First, this application provides a method for urban population dynamic monitoring and trajectory analysis based on big data, including the following steps: Obtain a multimodal dataset within a preset urban area, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the urban grid; Processing the multimodal optimization dataset through a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; According to the cross-institutional fusion feature matrix, encoding processing is performed through a preset LSTM model, and processing is performed through a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; The high-frequency path and the high-density dwelling area are visualized and output.

[0006] Optionally, in the method for urban population dynamic monitoring and trajectory analysis based on big data described in the present application, obtaining a multimodal dataset within a preset range of the city, and performing preprocessing and grid mapping to obtain a multimodal optimized dataset corresponding to the urban grid includes: Acquire multimodal datasets within a preset city area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; Divide the preset range of the city into grids according to the preset grid size to obtain multiple city grids; Pre-cleaning, differential privacy protection, and standardization of the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapping them to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

[0007] Optionally, in the method for urban population dynamic monitoring and trajectory analysis based on big data described in the present application, the multimodal optimization dataset is processed by a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix, including: Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; Each institution conducts local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset in combination with preset initial model parameters and the principle of minimizing the joint loss function to obtain the corresponding institutional model parameters and parameter gradients; Obtaining data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; The data integrity, data timeliness, data relevance and data consistency are weighted and summed to obtain the corresponding data quality score, and the dynamic weight value corresponding to each institution model is calculated according to the data quality score; Performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; Performing weighted summation processing on the aggregated parameter gradients to obtain a global gradient, and performing global model parameter update processing to obtain global model optimization parameters; Generate a global shared feature vector based on the global model optimization parameters, and generate a private feature vector for each institution based on the institution model parameters; The global shared feature vector and the private feature vector of each institution are concatenated and fused to obtain a cross-institutional fusion feature matrix.

[0008] Optionally, in the method for dynamic monitoring and trajectory analysis of urban population based on big data described in the present application, encoding processing is performed using a preset LSTM model according to the cross-institutional fusion feature matrix, and processing is performed using a preset DPC clustering algorithm to obtain high-frequency paths and high-density areas of stay, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

[0009] Optionally, the method for urban population dynamic monitoring and trajectory analysis based on big data described in this application further includes: Inputting the multimodal feature dataset and the low-dimensional trajectory feature vector into a preset lightweight deep learning model for processing to obtain the population density data of the city grid at a first preset time; Comparing the population density data with a preset population density warning threshold; If it is greater than the preset population density warning threshold, then count the number of times it is continuously greater than the preset population density warning threshold within the second preset time period; A grid population density heat map is generated based on the personnel density data and the corresponding frequency values.

[0010] Optionally, the method for urban population dynamic monitoring and trajectory analysis based on big data described in this application further includes: Obtain the real-time population density and historical mean population density of the city grid; Obtain the average data of personnel flow between city grids within a preset time period; Extracting data based on the multimodal feature data set to obtain grid node attribute data, including static attribute data, dynamic attribute data, and timeliness data; According to the real-time population density and the historical population density mean combined with the static attribute data, dynamic attribute data and timeliness data, a personnel flow graph is constructed with the mean data of personnel flow as the node edge weight value to obtain a grid node personnel flow graph.

[0011] In a second aspect, the present application provides a system for monitoring and analysing the dynamics of an urban population based on big data, the system comprising: a memory and a processor, the memory comprising a program for a method for monitoring and analysing the dynamics of an urban population based on big data, the program for monitoring and analysing the dynamics of an urban population based on big data, when executed by the processor, implementing the following steps: Obtain a multimodal dataset within a preset urban area, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the urban grid; Processing the multimodal optimization dataset through a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; According to the cross-institutional fusion feature matrix, encoding processing is performed through a preset LSTM model, and processing is performed through a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; The high-frequency path and the high-density dwelling area are visualized and output.

[0012] Optionally, in the big data-based urban population dynamic monitoring and trajectory analysis system described in the present application, the step of obtaining a multimodal dataset within a preset range of the city, performing preprocessing and grid mapping, and obtaining a multimodal optimized dataset corresponding to the urban grid includes: Acquire multimodal datasets within a preset city area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; Divide the preset range of the city into grids according to the preset grid size to obtain multiple city grids; Pre-cleaning, differential privacy protection, and standardization of the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapping them to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

[0013] Optionally, in the big data-based urban population dynamic monitoring and trajectory analysis system described in the present application, the multimodal optimization dataset is processed by a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix, including: Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; Each institution conducts local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset, combined with preset initial model parameters and the principle of minimizing the joint loss function, to obtain the corresponding institutional model parameters and parameter gradients; Obtaining data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; The data integrity, data timeliness, data relevance and data consistency are weighted and summed to obtain the corresponding data quality score, and the dynamic weight value corresponding to each institution model is calculated according to the data quality score; Performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; Performing weighted summation processing on the aggregated parameter gradients to obtain a global gradient, and performing global model parameter update processing to obtain global model optimization parameters; Generate a global shared feature vector based on the global model optimization parameters, and generate a private feature vector for each institution based on the institution model parameters; The global shared feature vector and the private feature vector of each institution are concatenated and fused to obtain a cross-institutional fusion feature matrix.

[0014] Optionally, in the urban population dynamic monitoring and trajectory analysis system based on big data described in the present application, the cross-institutional fusion feature matrix is ​​encoded by a preset LSTM model and processed by a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

[0015] From the above, it can be seen that the urban population dynamic monitoring and trajectory analysis method and system based on big data provided by this application improves data richness through multimodal data fusion, protects data privacy through cross-institutional federated learning, and combines the LSTM model with the DPC clustering algorithm to improve data real-time and accuracy, thereby realizing high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis.

[0016] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or understood by practicing the embodiments of the present application. The objectives and other advantages of the present application can be achieved and obtained through the structures particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of a method for urban population dynamic monitoring and trajectory analysis based on big data provided in an embodiment of the present application; Figure 2 A flowchart of obtaining a multimodal optimized data set corresponding to an urban grid in a method for urban population dynamic monitoring and trajectory analysis based on big data provided in an embodiment of the present application; Figure 3 A flowchart of generating a cross-institutional fusion feature matrix for the method for urban population dynamic monitoring and trajectory analysis based on big data provided in an embodiment of the present application; Figure 4 A high-level flowchart of methods according to various embodiments of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0020] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0021] Please refer to Figure 1, Figure 1 This is a flowchart of a method for monitoring and analysing the dynamics of urban populations based on big data in some embodiments of the present application. This method for monitoring and analysing the dynamics of urban populations based on big data is used in terminal devices such as computers and mobile phones. This method for monitoring and analysing the dynamics of urban populations based on big data includes the following steps: S11. Obtain a multimodal dataset within a preset range of the city, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the city grid; S12. Processing the multimodal optimization dataset using a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; S13, encoding the cross-institutional fusion feature matrix using a preset LSTM model, and processing it using a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; S14: Visually output the high-frequency path and the high-density parking area.

[0022] It should be noted that multimodal data sets are collected at edge nodes by multiple different institutions in the city, and the preset range of the city is divided into multiple grids of preset sizes. The collected multimodal data sets are then mapped to each grid. In order to improve data security and processing accuracy, a cross-institutional federated learning model is adopted and a federated learning framework is introduced to ensure that the collected data does not leave the local institution. Joint model training is carried out and weights are dynamically allocated according to the data evaluation of each institution to ensure the accuracy of the model analysis of each local institution while ensuring the privacy of the data. The generated cross-institutional fusion feature matrix is ​​then encoded and clustered to determine high-frequency paths and high-density areas, which are then displayed through a visualization module.

[0023] Please refer to Figure 2 , Figure 2 This is a flowchart of obtaining a multimodal optimized dataset corresponding to a city grid in a method for urban population dynamic monitoring and trajectory analysis based on big data in some embodiments of the present application. According to an embodiment of the present invention, obtaining a multimodal dataset within a preset range of a city, performing preprocessing and grid mapping, and obtaining a multimodal optimized dataset corresponding to the city grid includes: S21. Obtain multimodal datasets within a preset urban area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; S22, dividing the preset city range into grids according to a preset grid size to obtain multiple city grids; S23. Pre-cleaning, differential privacy protection, and standardization are performed on the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapped to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; S24. The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

[0024] It should be noted that mobile phone signaling data, such as base station ID, signal access time, signal departure time, and signal strength, is collected by the three major operators at a dynamic collection frequency (e.g., 1-minute intervals during peak hours and 5-minute intervals at other times). Remote sensing image data is collected daily via high-resolution satellites to obtain building density maps and nighttime light radiation values. IoT data, including smart meter data and shared bicycle operation records, is collected from IoT companies. Consumption record data within a preset city range is collected through mobile payment platforms. The specific data collected will be dynamically adjusted by those skilled in the art based on the specific requirements of population dynamic monitoring and trajectory analysis. The preset city range is divided into multiple city grids according to a preset grid size (e.g., 500mx500m). Simultaneously, the collected multimodal datasets undergo data pre-cleaning, including outlier detection and missing value repair, differential privacy processing, including location blurring and data aggregation desensitization. After undergoing temporal and spatial normalization, the datasets are mapped to the divided city grids. This generates optimized mobile phone signaling data, remote sensing imagery data, IoT data, and consumption record data for each city grid for subsequent cross-federated model training and data analysis.

[0025] Please refer to Figure 3 , Figure 3 This is a flowchart of generating a cross-institutional fusion feature matrix for a method for urban population dynamic monitoring and trajectory analysis based on big data in some embodiments of the present application. According to an embodiment of the present invention, the cross-institutional fusion feature matrix is ​​generated by processing the multimodal optimization dataset using a preset cross-institutional federated learning model combined with a dynamic weight allocation mechanism, including: S31. Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; S32. Each institution performs local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset, combined with preset initial model parameters and the principle of minimizing the joint loss function, to obtain the corresponding institution model parameters and parameter gradients; S33. Obtain data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; S34. Perform a weighted summation of the data integrity, data timeliness, data relevance, and data consistency to obtain a corresponding data quality score, and calculate a dynamic weight value corresponding to each institution model based on the data quality score; S35, performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; S36. Perform weighted summation processing according to the aggregated parameter gradient to obtain a global gradient, and perform global model parameter update processing to obtain global model optimization parameters; S37, generating a global shared feature vector based on the global model optimization parameters, and generating a private feature vector for each institution based on the institution model parameters; S38. Perform splicing and fusion processing based on the global shared feature vector and the private feature vector of each institution to obtain a cross-institutional fusion feature matrix.

[0026] It should be noted that, first, the urban population monitoring and analysis management department constructs a global feature dictionary including basic spatiotemporal features (such as urban grid ID serial number, timestamp), population flow characteristics (such as urban grid inflow, urban grid outflow, length of stay, inter-grid flow speed) and behavioral attribute characteristics (such as POI visit times, travel mode proportion), and performs data feature engineering extraction processing on the multimodal optimization dataset according to the preset dimensions. Feature engineering extraction processing is to convert the original multi-source heterogeneous data into structured features that can be efficiently learned by the model, such as marking the signal situation according to mobile phone signaling data, base station switching frequency and signal strength fluctuation data to generate "signal stability" features, and generating "consumption activity" features according to consumption conditions, and determining the mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset or consumption record feature dataset corresponding to each institution. Then, each institution uses the Adam optimizer to perform local model training based on its own multimodal feature dataset and the principle of minimizing the joint loss function, and outputs the corresponding institutional model parameters and parameter gradients. At the same time, technical personnel in this field evaluate the quality of the data to obtain data integrity and data timeliness. , data correlation and data consistency, and then perform weighted averaging to obtain the dynamic weight value corresponding to each institution's model. Then, the parameter gradients of institutions of the same type (such as the three major operators) are weighted averaged to obtain the aggregated parameter gradients of institutions of the same type. The parameter gradients of single institutions without institutions of the same type (such as institutions holding remote sensing image data) are the aggregated parameter gradients. The obtained aggregated parameter gradients are combined with the preset weight values ​​for weighted summation to obtain the global gradient of the preset cross-institutional federated learning model, and the global model parameters are optimized and updated to obtain the global model optimization parameters, which are encrypted and sent to each participating institution. Finally, the preset cross-institutional federated learning model generates a global shared feature vector (such as a 16-dimensional feature vector including the standardized value of grid population density, morning rush hour inflow probability, and commercial activity activity). The local model of each institution generates each institution's private feature vector based on the institutional model parameters (such as an 8-dimensional feature vector including the operator's user residence characteristics and the transportation department's road network congestion correlation characteristics). Then, the splicing is performed according to the urban grid and time distribution to obtain a cross-institutional fusion feature matrix, realizing hierarchical aggregation, privacy protection and dynamic weight collaborative processing.

[0027] According to an embodiment of the present invention, encoding processing is performed using a preset LSTM model based on the cross-institutional fusion feature matrix, and processing is performed using a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

[0028] It should be noted that the user's mobile record data within a preset time period (such as 1 day) is analyzed to obtain each user's trajectory sequence, such as user A [(06:00-08:00) grid 1, (08:00-11:00) grid 2, (11:00-14:00) grid 1, (14:00-18:00) grid 2, (18:00-06:00) grid 1], combined with the obtained cross-institutional fusion feature matrix, it is encoded through the LSTM-Autoencoder model to obtain a fixed-length low-dimensional trajectory feature vector. These vectors compress the dimension of the original trajectory data while retaining the key spatiotemporal and behavioral features, which can be used for subsequent clustering analysis and pattern mining. Among them, the preset LSTM model is obtained by training the cross-institutional fusion feature matrix and grid trajectory sequence and the corresponding low-dimensional trajectory feature vector of a large number of historical samples; Based on the low-dimensional trajectory feature vector, the technicians calculated the Euclidean distance between trajectory vectors to determine the local density of each trajectory vector, which is used to indicate the number of similar trajectory vectors around a certain trajectory vector. Then, they calculated the minimum distance from each trajectory vector to the vector with the highest local density, which is used to reflect the distance between the trajectory vector and the high-density area, and measure the relative position of the trajectory vector in the cluster. Finally, they selected the point with the largest local density and the largest minimum distance as the cluster center. Based on the cluster center, they divided the other trajectory vectors into corresponding clusters according to the distance and density relationship with the cluster center, thereby dividing the low-dimensional trajectory embedding vector set into different trajectory categories. According to the clustering results, they counted the starting and ending positions of the trajectories in each trajectory category, as well as the grid areas where they stayed, to determine high-frequency paths (such as from a residential area to an urban office building, from an industrial park to a transportation hub) and high-density areas (such as a shopping mall or a subway station).

[0029] According to an embodiment of the present invention, the further embodiment includes: Inputting the multimodal feature dataset and the low-dimensional trajectory feature vector into a preset lightweight deep learning model for processing to obtain the population density data of the city grid at a first preset time; Comparing the population density data with a preset population density warning threshold; If it is greater than the preset population density warning threshold, then count the number of times it is continuously greater than the preset population density warning threshold within the second preset time period; A grid population density heat map is generated based on the personnel density data and the corresponding frequency values.

[0030] It should be noted that the MobileNetV3 lightweight deep learning model is used to take the city grid as the unit. Based on the obtained multimodal feature dataset and low-dimensional trajectory feature vector, the convolution layer, pooling layer and other structures in the network are used to extract the population density data of the city grid in the first preset time period (such as 15 minutes later). If it is greater than the preset population density warning threshold, the number of times that it is continuously greater than the preset population density warning threshold in the second preset time period (such as within 1 hour) is further cyclically counted. Finally, a grid population density heat map is rendered based on the population density data and the corresponding number values. For example, red represents grids with a population density greater than the preset population density warning threshold and a number value greater than the preset number threshold, yellow represents grids with a population density greater than the preset population density warning threshold and a number value not greater than the preset number threshold, and green represents grids with a population density not greater than the preset population density warning threshold. The preset lightweight deep learning model is trained by obtaining a large number of historical samples of multimodal feature datasets and low-dimensional trajectory feature vectors and corresponding population density data.

[0031] According to an embodiment of the present invention, the further embodiment includes: Obtain the real-time population density and historical mean population density of the city grid; Obtain the average data of personnel flow between city grids within a preset time period; Extracting data based on the multimodal feature data set to obtain grid node attribute data, including static attribute data, dynamic attribute data, and timeliness data; According to the real-time population density and the historical population density mean combined with the static attribute data, dynamic attribute data and timeliness data, a personnel flow graph is constructed with the mean data of personnel flow as the node edge weight value to obtain a grid node personnel flow graph.

[0032] It should be noted that the obtained multimodal feature data set is subjected to data extraction to obtain grid node attribute data, including static attribute data such as grid area data, density of shopping malls within the grid, number of bus stops within the grid, dynamic attribute data such as the inflow and outflow of people during the morning rush hour within the grid, and timeliness data such as real-time weather data and holiday label data. The specific grid node attribute data is dynamically adjusted by technical personnel in this field according to the requirements of specific personnel dynamic monitoring and trajectory analysis; the real-time population density and the mean of historical population density, as well as the static attribute data, dynamic attribute data and timeliness data are standardized into numerical vectors as input features of urban grid nodes, and the mean data of personnel flow is used as the node edge weight value, which can be dynamically adjusted. A personnel flow map is then constructed in units of time slices (such as 15 minutes) to obtain a grid node personnel flow map, which is stored in real time in the form of a graph database or tensor.

[0033] Please refer to Figure 4, Figure 4 It is a high-level flow chart of the methods of various embodiments of the present application, which can be used for the method of dynamic monitoring and trajectory analysis of urban population based on big data. According to the embodiment of the present invention, the preset range of the city to be analyzed is divided into several urban grids, and the obtained multimodal data set is pre-processed and mapped to the urban grid. Local training, gradient upload and global update are realized through the cross-institutional federated learning model to obtain a cross-institutional fusion feature matrix, which is encoded and processed by the preset LSTM model and processed by the preset DPC clustering algorithm to obtain high-frequency paths and high-density areas of parking, which are displayed through visual output; at the same time, based on the obtained multimodal feature data set combined with the low-dimensional trajectory feature vector encoded and processed by the preset LSTM model, the preset lightweight deep learning model is used for processing to obtain urban grid personnel density data, and further generate a grid population density heat map; based on the obtained real-time population density and historical population density average combined with the average data of personnel flow between grid nodes, a personnel flow map is constructed to obtain a grid node personnel flow map, thereby realizing high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis based on big data.

[0034] It is worth mentioning that according to an embodiment of the present invention, the present invention further includes: Obtaining a personnel flow graph of grid nodes within a third preset time period, and performing splicing processing based on time nodes to obtain a personnel flow time sequence graph sequence; Input the personnel flow time sequence diagram into a preset spatiotemporal graph neural network model for processing to obtain the predicted number of personnel at the city grid node corresponding to the preset time node; Processing the predicted number of people at the city grid nodes to obtain a population growth rate within a fourth preset time period; If the predicted number of personnel at the city grid node is greater than the preset personnel quantity warning threshold, and the personnel growth rate is greater than the preset personnel growth rate monitoring threshold, an early warning response is output.

[0035] It should be noted that by obtaining the grid node personnel flow map within the third preset time period, such as the 12 grid node personnel flow maps from 7 to 10 am, they are spliced ​​in sequence from morning to night, and then input into the preset ST-GNN spatiotemporal graph neural network model for analysis and processing, and the predicted number of city grid node personnel corresponding to the preset time node is predicted. For example, it is predicted that the number of personnel at a city grid node at 11:00 am is 1,000 people, and 11:10 is 1,500 people, then (1,500-1,000) / 1,000=0.5 is the personnel growth rate. The personnel carrying capacity of the grid node is determined by threshold comparison. If both exceed the threshold, an early warning response is output, otherwise, monitoring is continued. The preset spatiotemporal graph neural network model is trained by obtaining a large number of historical sample personnel flow time series diagram sequences and the corresponding city grid node personnel prediction quantities.

[0036] The present invention also discloses a system for monitoring and analysing the dynamics of urban populations based on big data, comprising a memory and a processor. The memory comprises a program for monitoring and analysing the dynamics of urban populations based on big data. When the program is executed by the processor, the following steps are implemented: Obtain a multimodal dataset within a preset urban area, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the urban grid; Processing the multimodal optimization dataset through a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; According to the cross-institutional fusion feature matrix, encoding processing is performed through a preset LSTM model, and processing is performed through a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; The high-frequency path and the high-density dwelling area are visualized and output.

[0037] It should be noted that multimodal data sets are collected at edge nodes by multiple different institutions in the city, and the preset range of the city is divided into multiple grids of preset sizes. The collected multimodal data sets are then mapped to each grid. In order to improve data security and processing accuracy, a cross-institutional federated learning model is adopted and a federated learning framework is introduced to ensure that the collected data does not leave the local institution. Joint model training is carried out and weights are dynamically allocated according to the data evaluation of each institution to ensure the accuracy of the model analysis of each local institution while ensuring the privacy of the data. The generated cross-institutional fusion feature matrix is ​​then encoded and clustered to determine high-frequency paths and high-density areas, which are then displayed through a visualization module.

[0038] According to an embodiment of the present invention, the step of obtaining a multimodal dataset within a preset range of a city, performing preprocessing and grid mapping, and obtaining a multimodal optimized dataset corresponding to the city grid includes: Acquire multimodal datasets within a preset city area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; Divide the preset range of the city into grids according to the preset grid size to obtain multiple city grids; Pre-cleaning, differential privacy protection, and standardization of the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapping them to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

[0039] It should be noted that mobile phone signaling data, such as base station ID, signal access time, signal departure time, and signal strength, is collected by the three major operators at a dynamic collection frequency (e.g., 1-minute intervals during peak hours and 5-minute intervals at other times). Remote sensing image data is collected daily via high-resolution satellites to obtain building density maps and nighttime light radiation values. IoT data, including smart meter data and shared bicycle operation records, is collected from IoT companies. Consumption record data within a preset city range is collected through mobile payment platforms. The specific data collected will be dynamically adjusted by those skilled in the art based on the specific requirements of population dynamic monitoring and trajectory analysis. The preset city range is divided into multiple city grids according to a preset grid size (e.g., 500mx500m). Simultaneously, the collected multimodal datasets undergo data pre-cleaning, including outlier detection and missing value repair, differential privacy processing, including location blurring and data aggregation desensitization. After undergoing temporal and spatial normalization, the datasets are mapped to the divided city grids. This generates optimized mobile phone signaling data, remote sensing imagery data, IoT data, and consumption record data for each city grid for subsequent cross-federated model training and data analysis.

[0040] According to an embodiment of the present invention, the multimodal optimization dataset is processed by a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix, including: Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; Each institution conducts local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset, combined with preset initial model parameters and the principle of minimizing the joint loss function, to obtain the corresponding institutional model parameters and parameter gradients; Obtaining data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; The data integrity, data timeliness, data relevance and data consistency are weighted and summed to obtain the corresponding data quality score, and the dynamic weight value corresponding to each institution model is calculated according to the data quality score; Performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; Performing weighted summation processing on the aggregated parameter gradients to obtain a global gradient, and performing global model parameter update processing to obtain global model optimization parameters; Generate a global shared feature vector based on the global model optimization parameters, and generate a private feature vector for each institution based on the institution model parameters; The global shared feature vector and the private feature vector of each institution are concatenated and fused to obtain a cross-institutional fusion feature matrix.

[0041] It should be noted that, first, the urban population monitoring and analysis management department constructs a global feature dictionary including basic spatiotemporal features (such as urban grid ID serial number, timestamp), population flow characteristics (such as urban grid inflow, urban grid outflow, length of stay, inter-grid flow speed) and behavioral attribute characteristics (such as POI visit times, travel mode proportion), and performs data feature engineering extraction processing on the multimodal optimization dataset according to the preset dimensions. Feature engineering extraction processing is to convert the original multi-source heterogeneous data into structured features that can be efficiently learned by the model, such as marking the signal situation according to mobile phone signaling data, base station switching frequency and signal strength fluctuation data to generate "signal stability" features, and generating "consumption activity" features according to consumption conditions, and determining the mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset or consumption record feature dataset corresponding to each institution. Then, each institution uses the Adam optimizer to perform local model training based on its own multimodal feature dataset and the principle of minimizing the joint loss function, and outputs the corresponding institutional model parameters and parameter gradients. At the same time, technical personnel in this field evaluate the quality of the data to obtain data integrity and data timeliness. , data correlation and data consistency, and then perform weighted averaging to obtain the dynamic weight value corresponding to each institution's model. Then, the parameter gradients of institutions of the same type (such as the three major operators) are weighted averaged to obtain the aggregated parameter gradients of institutions of the same type. The parameter gradients of single institutions without institutions of the same type (such as institutions holding remote sensing image data) are the aggregated parameter gradients. The obtained aggregated parameter gradients are combined with the preset weight values ​​for weighted summation to obtain the global gradient of the preset cross-institutional federated learning model, and the global model parameters are optimized and updated to obtain the global model optimization parameters, which are encrypted and sent to each participating institution. Finally, the preset cross-institutional federated learning model generates a global shared feature vector (such as a 16-dimensional feature vector including the standardized value of grid population density, morning rush hour inflow probability, and commercial activity activity). The local model of each institution generates each institution's private feature vector based on the institutional model parameters (such as an 8-dimensional feature vector including the operator's user residence characteristics and the transportation department's road network congestion correlation characteristics). Then, the splicing is performed according to the urban grid and time distribution to obtain a cross-institutional fusion feature matrix, realizing hierarchical aggregation, privacy protection and dynamic weight collaborative processing.

[0042] According to an embodiment of the present invention, encoding processing is performed using a preset LSTM model based on the cross-institutional fusion feature matrix, and processing is performed using a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

[0043] It should be noted that the user's mobile record data within a preset time period (such as 1 day) is analyzed to obtain each user's trajectory sequence, such as user A [(06:00-08:00) grid 1, (08:00-11:00) grid 2, (11:00-14:00) grid 1, (14:00-18:00) grid 2, (18:00-06:00) grid 1], combined with the obtained cross-institutional fusion feature matrix, it is encoded through the LSTM-Autoencoder model to obtain a fixed-length low-dimensional trajectory feature vector. These vectors compress the dimension of the original trajectory data while retaining the key spatiotemporal and behavioral features, which can be used for subsequent clustering analysis and pattern mining. Among them, the preset LSTM model is obtained by training the cross-institutional fusion feature matrix and grid trajectory sequence and the corresponding low-dimensional trajectory feature vector of a large number of historical samples; Based on the low-dimensional trajectory feature vector, the technicians calculated the Euclidean distance between trajectory vectors to determine the local density of each trajectory vector, which is used to indicate the number of similar trajectory vectors around a certain trajectory vector. Then, they calculated the minimum distance from each trajectory vector to the vector with the highest local density, which is used to reflect the distance between the trajectory vector and the high-density area, and measure the relative position of the trajectory vector in the cluster. Finally, they selected the point with the largest local density and the largest minimum distance as the cluster center. Based on the cluster center, they divided the other trajectory vectors into corresponding clusters according to the distance and density relationship with the cluster center, thereby dividing the low-dimensional trajectory embedding vector set into different trajectory categories. According to the clustering results, they counted the starting and ending positions of the trajectories in each trajectory category, as well as the grid areas where they stayed, to determine high-frequency paths (such as from a residential area to an urban office building, from an industrial park to a transportation hub) and high-density areas (such as a shopping mall or a subway station).

[0044] According to an embodiment of the present invention, the further embodiment includes: Inputting the multimodal feature dataset and the low-dimensional trajectory feature vector into a preset lightweight deep learning model for processing to obtain the population density data of the city grid at a first preset time; Comparing the population density data with a preset population density warning threshold; If it is greater than the preset population density warning threshold, then count the number of times it is continuously greater than the preset population density warning threshold within the second preset time period; A grid population density heat map is generated based on the personnel density data and the corresponding frequency values.

[0045] It should be noted that the MobileNetV3 lightweight deep learning model is used to take the city grid as the unit. Based on the obtained multimodal feature dataset and low-dimensional trajectory feature vector, the convolution layer, pooling layer and other structures in the network are used to extract the population density data of the city grid in the first preset time period (such as 15 minutes later). If it is greater than the preset population density warning threshold, the number of times that it is continuously greater than the preset population density warning threshold in the second preset time period (such as within 1 hour) is further cyclically counted. Finally, a grid population density heat map is rendered based on the population density data and the corresponding number values. For example, red represents grids with a population density greater than the preset population density warning threshold and a number value greater than the preset number threshold, yellow represents grids with a population density greater than the preset population density warning threshold and a number value not greater than the preset number threshold, and green represents grids with a population density not greater than the preset population density warning threshold. The preset lightweight deep learning model is trained by obtaining a large number of historical samples of multimodal feature datasets and low-dimensional trajectory feature vectors and corresponding population density data.

[0046] According to an embodiment of the present invention, the further embodiment includes: Obtain the real-time population density and historical mean population density of the city grid; Obtain the average data of personnel flow between city grids within a preset time period; Extracting data based on the multimodal feature data set to obtain grid node attribute data, including static attribute data, dynamic attribute data, and timeliness data; According to the real-time population density and the historical population density mean combined with the static attribute data, dynamic attribute data and timeliness data, the personnel flow mean data is used as the node edge weight value to construct a personnel flow graph to obtain a grid node personnel flow graph.

[0047] It should be noted that the obtained multimodal feature data set is subjected to data extraction to obtain grid node attribute data, including static attribute data such as grid area data, shopping mall density within the grid, and the number of bus stops within the grid; dynamic attribute data such as the inflow and outflow of people during the morning rush hour within the grid; and timeliness data such as real-time weather data and holiday label data. The specific grid node attribute data are dynamically adjusted by technical personnel in this field according to the requirements of specific personnel dynamic monitoring and trajectory analysis; the real-time population density and the mean historical population density as well as the static attribute data, dynamic attribute data and timeliness data are standardized into numerical vectors as input features of urban grid nodes, and the mean data of personnel flow is used as the node edge weight value, which can be dynamically adjusted. A personnel flow map is then constructed in units of time slices (such as 15 minutes) to obtain a grid node personnel flow map, which is stored in real time in the form of a graph database or tensor.

[0048] According to an embodiment of the present invention, a preset range of a city to be analyzed is divided into several city grids, which are mapped to the city grids after preprocessing based on the obtained multimodal data set. Local training, gradient upload and global update are realized through a cross-institutional federated learning model to obtain a cross-institutional fusion feature matrix, which is encoded and processed by a preset LSTM model and processed by a preset DPC clustering algorithm to obtain high-frequency paths and high-density areas of residence, which are displayed through visual output. At the same time, based on the obtained multimodal feature data set combined with the low-dimensional trajectory feature vector encoded and processed by the preset LSTM model, the preset lightweight deep learning model is used to obtain urban grid personnel density data, and further generate a grid population density heat map. A personnel flow map is constructed based on the obtained real-time population density and historical population density mean values ​​combined with the mean personnel flow data between grid nodes to obtain a grid node personnel flow map, thereby realizing high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis based on big data.

[0049] It is worth mentioning that according to an embodiment of the present invention, the present invention further includes: Obtaining a personnel flow graph of grid nodes within a third preset time period, and performing splicing processing based on time nodes to obtain a personnel flow time sequence graph sequence; Input the personnel flow time sequence diagram into a preset spatiotemporal graph neural network model for processing to obtain the predicted number of personnel at the city grid node corresponding to the preset time node; Processing the predicted number of people at the city grid nodes to obtain a population growth rate within a fourth preset time period; If the predicted number of personnel at the city grid node is greater than the preset personnel quantity warning threshold, and the personnel growth rate is greater than the preset personnel growth rate monitoring threshold, an early warning response is output.

[0050] It should be noted that by obtaining the grid node personnel flow map within the third preset time period, such as the 12 grid node personnel flow maps from 7 to 10 am, they are spliced ​​in sequence from morning to night, and then input into the preset ST-GNN spatiotemporal graph neural network model for analysis and processing, and the predicted number of city grid node personnel corresponding to the preset time node is predicted. For example, it is predicted that the number of personnel at a city grid node at 11:00 am is 1,000 people, and 11:10 is 1,500 people, then (1,500-1,000) / 1,000=0.5 is the personnel growth rate. The personnel carrying capacity of the grid node is determined by threshold comparison. If both exceed the threshold, an early warning response is output, otherwise, monitoring is continued. The preset spatiotemporal graph neural network model is trained by obtaining a large number of historical sample personnel flow time series diagram sequences and the corresponding city grid node personnel prediction quantities.

[0051] The method and system for urban population dynamic monitoring and trajectory analysis based on big data disclosed in the present invention improve data richness through multimodal data fusion, protect data privacy through cross-institutional federated learning, and improve data real-time and accuracy by combining the LSTM model with the DPC clustering algorithm, thereby achieving high-precision, high-accuracy and high-timeliness urban population dynamic monitoring and trajectory analysis.

[0052] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0053] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0054] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0055] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware related to program instructions, and the aforementioned program may be stored in a readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0056] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as standalone products, they can also be stored on a readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A method for monitoring and trajectories analysis of urban population dynamics based on big data, characterized by: The following steps are involved: Obtain a multimodal dataset within a preset urban area, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the urban grid; Processing the multimodal optimization dataset through a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; According to the cross-institutional fusion feature matrix, encoding processing is performed through a preset LSTM model, and processing is performed through a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; The high-frequency path and the high-density dwelling area are visualized and output.

2. The method for urban population dynamic monitoring and trajectory analysis based on big data according to claim 1 is characterized in that: The method of obtaining a multimodal dataset within a preset range of a city, performing preprocessing and grid mapping to obtain a multimodal optimized dataset corresponding to the city grid includes: Acquire multimodal datasets within a preset city area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; Divide the preset range of the city into grids according to the preset grid size to obtain multiple city grids; Pre-cleaning, differential privacy protection, and standardization of the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapping them to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

3. The method for urban population dynamic monitoring and trajectory analysis based on big data according to claim 2 is characterized in that: The multimodal optimization dataset is processed by a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix, including: Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; Each institution conducts local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset, combined with preset initial model parameters and the principle of minimizing the joint loss function, to obtain the corresponding institutional model parameters and parameter gradients; Obtaining data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; The data integrity, data timeliness, data relevance and data consistency are weighted and summed to obtain the corresponding data quality score, and the dynamic weight value corresponding to each institution model is calculated according to the data quality score; Performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; Performing weighted summation processing on the aggregated parameter gradients to obtain a global gradient, and performing global model parameter update processing to obtain global model optimization parameters; Generate a global shared feature vector based on the global model optimization parameters, and generate a private feature vector for each institution based on the institution model parameters; The global shared feature vector and the private feature vector of each institution are concatenated and fused to obtain a cross-institutional fusion feature matrix.

4. The method for urban population dynamic monitoring and trajectory analysis based on big data according to claim 3 is characterized in that: The encoding process is performed by a preset LSTM model according to the cross-institutional fusion feature matrix, and processed by a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

5. The method for urban population dynamic monitoring and trajectory analysis based on big data according to claim 4 is characterized in that: Also includes: Inputting the multimodal feature dataset and the low-dimensional trajectory feature vector into a preset lightweight deep learning model for processing to obtain the population density data of the city grid at a first preset time; Comparing the population density data with a preset population density warning threshold; If it is greater than the preset population density warning threshold, then count the number of times it is continuously greater than the preset population density warning threshold within the second preset time period; A grid population density heat map is generated based on the personnel density data and the corresponding frequency values.

6. The method for urban population dynamic monitoring and trajectory analysis based on big data according to claim 5 is characterized in that: Also includes: Obtain the real-time population density and historical mean population density of the city grid; Obtain the average data of personnel flow between city grids within a preset time period; Extracting data based on the multimodal feature data set to obtain grid node attribute data, including static attribute data, dynamic attribute data, and timeliness data; According to the real-time population density and the historical population density mean combined with the static attribute data, dynamic attribute data and timeliness data, a personnel flow graph is constructed with the mean data of personnel flow as the node edge weight value to obtain a grid node personnel flow graph.

7. Urban population dynamic monitoring and trajectory analysis system based on big data, characterized by: The system comprises a memory and a processor, wherein the memory comprises a program for a method for monitoring and analysing the dynamics of urban population and trajectories based on big data, and when the program is executed by the processor, the following steps are implemented: Obtain a multimodal dataset within a preset urban area, perform preprocessing and grid mapping, and obtain a multimodal optimized dataset corresponding to the urban grid; Processing the multimodal optimization dataset through a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix; According to the cross-institutional fusion feature matrix, encoding processing is performed through a preset LSTM model, and processing is performed through a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas; The high-frequency path and the high-density dwelling area are visualized and output.

8. The urban population dynamic monitoring and trajectory analysis system based on big data according to claim 7 is characterized in that: The method of obtaining a multimodal dataset within a preset range of a city, performing preprocessing and grid mapping to obtain a multimodal optimized dataset corresponding to the city grid includes: Acquire multimodal datasets within a preset city area, including mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data; Divide the preset range of the city into grids according to the preset grid size to obtain multiple city grids; Pre-cleaning, differential privacy protection, and standardization of the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, and mapping them to a city grid to obtain a multimodal optimized dataset corresponding to the city grid; The multimodal optimization data set includes mobile phone signaling optimization data, remote sensing image optimization data, Internet of Things optimization data and consumption record optimization data.

9. The urban population dynamic monitoring and trajectory analysis system based on big data according to claim 8 is characterized in that: The multimodal optimization dataset is processed by a preset cross-institutional federated learning model combined with a dynamic weight distribution mechanism to generate a cross-institutional fusion feature matrix, including: Performing data feature engineering processing on the multimodal optimized dataset to obtain a multimodal feature dataset, including a mobile phone signaling feature dataset, a remote sensing image feature dataset, an Internet of Things feature dataset, and a consumption record feature dataset; Each institution conducts local model training based on the corresponding mobile phone signaling feature dataset, remote sensing image feature dataset, Internet of Things feature dataset, or consumption record feature dataset, combined with preset initial model parameters and the principle of minimizing the joint loss function, to obtain the corresponding institutional model parameters and parameter gradients; Obtaining data quality assessment parameters corresponding to the mobile phone signaling data, remote sensing image data, Internet of Things data, and consumption record data, including data integrity, data timeliness, data relevance, and data consistency; The data integrity, data timeliness, data relevance and data consistency are weighted and summed to obtain the corresponding data quality score, and the dynamic weight value corresponding to each institution model is calculated according to the data quality score; Performing weighted average aggregation processing on the parameter gradients of the same type of institutions according to the dynamic weight value to obtain the aggregated parameter gradients corresponding to the same type of institutions; Performing weighted summation processing on the aggregated parameter gradients to obtain a global gradient, and performing global model parameter update processing to obtain global model optimization parameters; Generate a global shared feature vector based on the global model optimization parameters, and generate a private feature vector for each institution based on the institution model parameters; The global shared feature vector and the private feature vector of each institution are concatenated and fused to obtain a cross-institutional fusion feature matrix.

10. The urban population dynamic monitoring and trajectory analysis system based on big data according to claim 9 is characterized in that: The encoding process is performed by a preset LSTM model according to the cross-institutional fusion feature matrix, and processed by a preset DPC clustering algorithm to obtain high-frequency paths and high-density parking areas, including: Obtaining movement record data of a preset user within a preset time period, and performing splicing processing based on the movement record data and the city grid according to time to obtain a grid trajectory sequence within the preset time period; According to the cross-institutional fusion feature matrix and the grid trajectory sequence, encoding processing is performed through a preset LSTM model to obtain a low-dimensional trajectory feature vector; The low-dimensional trajectory feature vector is processed by a preset DPC clustering algorithm to obtain a high-frequency path and a high-density parking area.

Images