Pedestrian flow short-term prediction processing method and system based on spatio-temporal data, and medium

Abstract

The invention provides a short-term people flow prediction processing method and system based on spatio-temporal data and a medium, and belongs to the technical field of big data analysis, and the method comprises the steps: getting through signaling data of operator communication network users, obtaining data of different service calibers of signaling, and building a sample data set; analyzing the sample data set, introducing various feature data such as weather features, time features and event features, and completing feature engineering construction; a short-term people flow prediction model is constructed by adopting a hybrid architecture learning algorithm, the defect that a single prediction model cannot reversely learn and discover time sequence periodic change is overcome, training is performed through data samples under different time and space, and the prediction precision is improved; and calculating sample data by using the short-term people flow prediction model, and outputting a short-term prediction data result of the urban space-time people flow. According to the method, the influence of meteorological data, event information and the like on the human traffic is comprehensively calculated by constructing the human traffic prediction processing model, and human traffic prediction is carried out for different application scenes.

Description

Technical Field

[0001] This invention belongs to the field of big data analysis technology, specifically relating to a method, system, and medium for short-term prediction and processing of pedestrian flow based on spatiotemporal data. Background Technology

[0002] Urban pedestrian flow data processing involves massive, complex, and dynamic data. Existing methods for processing and predicting urban pedestrian flow data have several significant drawbacks and challenges.

[0003] First, data acquisition and processing are often fragmented, lacking a unified data integration and preprocessing workflow. The diverse signaling data formats provided by various telecommunications operators, coupled with the complexity of urban geographic information data, make data integration and cleaning a challenging task. This not only increases data processing time and costs but may also affect the accuracy of subsequent analyses due to data inconsistencies. Second, existing data processing methods have limitations in urban functional area division and population location. Specifically, traditional area division methods often rely on static geographic information, making it difficult to accurately reflect dynamic population changes. For example, on weekdays and during working hours, the population is mainly concentrated in the city center and along major transportation routes; while on weekends and during rest periods, the population may be more dispersed. Simultaneously, population location technology is limited by data accuracy and timeliness, leading to potentially significant errors in location results. Furthermore, existing population prediction models often lack sufficient information on influencing factors, resulting in a failure to fully consider the impact of geographical location, temporal changes, and changes in the surrounding environment on population size when processing time-series data, thus reducing the accuracy and reliability of prediction results. Summary of the Invention

[0004] In a first aspect, embodiments of this application provide a method for short-term prediction and processing of pedestrian flow based on spatiotemporal data, comprising the following steps: S1. Establish communication with the operator's network to obtain user signaling data, including device object data, base station object data, and user roaming data; simultaneously collect meteorological data, event data, and urban geographic information data related to the impact of pedestrian traffic; clean and fuse the signaling data, specifically including: parsing base station location information and associating it with geographic area boundaries; generating time point labels for user entry / exit from the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; finally, establish a spatiotemporal pedestrian traffic sample dataset; S2. Analyze the sample dataset and construct a hybrid feature set including geographic block attribute features, user behavior attribute features, weather features and event features based on the spatiotemporal tags, meteorological data and event data; S3. A prediction model is constructed using a hybrid architecture learning algorithm combining LightGBM and LSTM, wherein: the hybrid feature set is aggregated according to unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, the LightGBM model learns the feature influence weights, and the LSTM model learns the temporal change pattern, thus obtaining a trained short-term pedestrian flow prediction model based on spatiotemporal data; S4. Calculate the real-time sample data using the trained short-term pedestrian flow prediction model and output the short-term prediction data of urban spatiotemporal pedestrian flow.

[0005] Furthermore, the specific steps of step S1 are as follows: S11. Obtain signaling data from network users of various communication operators, and select the time update granularity of the signaling data according to the required prediction time period; the signaling data includes device object data, base station object data, user object data and user roaming data, and the trajectory data is merged through the multi-device IMSI association rule of the same user; S12. Collect meteorological data, spatiotemporal characteristic data, and event data to prepare urban area division and geographical location information data that characterize urban spatial planning; S13. According to the required prediction time range and spatial range, establish sample set data by associating each data table with base station ID and area code and with timestamp and interval; S14. Perform useless field deletion, text feature processing, missing value processing, and standardization on the sample set data. Then, obtain a standard dataset by associating various labels.

[0006] Furthermore, the specific steps of step S13 are as follows: S131. Obtain geographic location information data and base station object data, parse latitude and longitude information location and connect it with base station location, construct the association between regional boundary information and base station location, and construct geographic region labels; S132. Combine geographic region labels with urban area division information data to add geographic attributes to different geographic regions and form geographic block attribute labels; S133. Construct the regional boundary based on geographic location information data, determine the signaling trajectory data based on equipment object data and base station object data, and associate the signaling trajectory data with the regional boundary to obtain the population flow trajectory; S134. Add time point labels to different entry and exit locations based on the location update time period and base station location in the base station information data to obtain the entry time point and exit time point; S135. Calculate the user’s dwell time in each area based on the entry and exit times, then add a daily location tag list to the dwell time statistics, summarize the daily location tag list to form a monthly area tag list, and associate it with the actual scene to form spatiotemporal tags. S136. Add population residence duration attribute tags based on residence duration, and then add work / residence attribute tags and regional residence attribute tags according to the time period to which the residence duration belongs, to obtain the population attribute tags of the geographic block; S137. Based on geographic block attribute tags, geographic block population attribute tags, and spatiotemporal tags, track the activity trajectory of each user at different times and geographic spaces to form a spatiotemporal population flow feature dataset; S138. Perform statistical analysis on the crowd trajectory of each number in the spatiotemporal pedestrian flow characteristic dataset. Combine the business survey results in different scenarios such as scenic spots, business districts, and stations to construct a sample dataset that conforms to the actual business scenarios of working or living and passing by people. By removing the two groups in the region, a sample dataset is obtained.

[0007] Furthermore, the specific steps of step S14 are as follows: S141. Obtain geographic location information data and base station object data, parse latitude and longitude location information and base station location information, construct a dynamic spatial association between regional boundary information and base station location, and construct geographic region features; S142. Combine geographical region characteristics with urban area division information data to add geographical attributes to different geographical regions and form geographical block attribute tags; S143. Construct regional boundaries based on geographic location information data, determine signaling trajectory data based on device object data and base station object data, and merge the trajectory of the same user through multi-device IMSI association, and associate the merged signaling trajectory data with the regional boundaries to obtain the population flow trajectory; S144. Add time point labels to different entry and exit locations based on the location update time period and base station location in the base station information data to obtain the entry time point and exit time point; S145. Based on the entry time and exit time, perform statistical calculations according to the preset time interval to obtain the user's dwell time in each area. Then, perform daily statistics on the dwell time and add a daily location tag list. Summarize the daily location tag list to form a monthly area tag list and associate it with the actual scene to form a spatiotemporal tag. S146. Add population residence duration attribute tags based on residence duration, and then add work / residence attribute tags and regional residence attribute tags according to the time period to which the residence duration belongs, to obtain the population attribute tags of the geographic block; S147. Based on geographic block attribute tags, spatiotemporal tags, and geographic block population attribute tags, track the activity trajectory of each user at different times and geographic spaces to form a spatiotemporal population flow feature dataset; S148. Perform statistical analysis on the trajectory of each mobile phone number in the spatiotemporal pedestrian flow feature dataset, identify duplicate trajectories of multiple device IDs under the same IMSI and duplicate signaling trajectories of the same base station within the same time period, and perform deduplication to ensure that the redundancy rate of the dataset after deduplication is less than the preset redundancy rate threshold, thereby obtaining the spatiotemporal pedestrian flow sample dataset.

[0008] Furthermore, the specific steps of step S2 are as follows: S21. Based on weather and meteorological data, feature classification is performed, which is divided into text-based and numerical-based related features, and a range of machine-recognizable numerical features is assigned to text-based content. S22. Based on the time cycle data of pedestrian flow, construct a time feature engineering for identifying time feature ranges, including hourly, daily, and weekly time dimension divisions; S23. Define major events and construct event feature engineering based on major event data, including event scale weight and event type weight, and calibrate the weights by the deviation rate between actual and predicted pedestrian flow in the last N similar events. S24. Construct historical data feature engineering based on the traffic flow forecast cycle, and select traffic flow data, meteorological data, and event data from the same period in the past M months as historical features, and the time granularity of the historical data is consistent with the forecast cycle.

[0009] Furthermore, the specific steps of step S3 are as follows: S31. Aggregate the sample data based on the added spatiotemporal feature label set, summarize the data within a unit statistical time period, add an index to the summarized data based on the unit statistical time period, and then adjust the dimensions according to the requirements of the LSTM model to form the current dataset. S32. Construct an LSTM model, set one input of the LSTM model to the current dataset, and set the output of the LSTM model to population prediction data of the same dimension as the current dataset, and select the activation functions required for the hidden layers and fully connected layers of the LSTM model; S33. Construct a LightGBM model and set the input of the LightGBM model to the current dataset. Combine the LightGBM model and the LSTM model by setting the output of the LightGBM model as another input of the LSTM model to form a hybrid architecture pedestrian flow prediction model. S34. Divide the current dataset into training, validation, and test sets, and set the loss function and optimizer; S35. The traffic flow prediction model is trained using the training set. During the training process, the learning rate, number of iterations, number of trees, and number of leaf nodes are continuously adjusted to continuously optimize the LightGBM model. The influence of time feature values ​​on time series data is learned. The number of iterations of the LSTM model is controlled according to the loss function and iteration threshold. The LSTM model is continuously optimized through the optimizer. At the same time, the performance of the LSTM model during the iteration process is monitored through the validation set. S36. Use the test set to evaluate the performance of the trained pedestrian flow prediction model, and input the current dataset into the evaluated pedestrian flow prediction model to predict short-term pedestrian flow data.

[0010] Furthermore, the specific steps of step S31 are as follows: S311. Based on the spatiotemporal label feature set, filter out data related to pedestrian flow; S312. Integrate the geographic block attribute labels, geographic block population attribute labels, and spatiotemporal labels from the spatiotemporal label feature set into the dataset to ensure that each data entry contains complete spatiotemporal information; S313. Aggregate the data according to the unit statistical time, merge the data records within the same time period, and summarize the traffic flow in each time period; S314. Add time period indexes to the aggregated data, and adjust the dimensions and format of the data according to the requirements of the LSTM model to obtain the current dataset.

[0011] Furthermore, the specific steps of step S32 are as follows: S321. Determine the number of layers and the number of neurons in each layer, and initialize an LSTM model; S322. Encode the data related to pedestrian traffic and the spatiotemporal feature label set in the current dataset into vector form and set it as the input of the LSTM model; S323. Set the output of the LSTM model to pedestrian flow data of the same dimension as the current dataset; S324. Set the number of LSTM layers as needed, and set the number of neurons and activation function for each LSTM layer.

[0012] Furthermore, the specific steps of step S36 are as follows: S361. Use the test set to evaluate the performance of the trained pedestrian flow prediction model; If the evaluation is successful, proceed to step S362; If the evaluation fails, return to step S35; S362. Select a dataset from the current dataset that is less than a set threshold away from the current time point, and input the selected dataset into the pedestrian flow prediction model to obtain the predicted pedestrian flow data.

[0013] Secondly, embodiments of this application also provide a short-term pedestrian flow prediction and processing system based on spatiotemporal data, comprising: The sample dataset construction module is used to connect to the operator's communication network and obtain user signaling data, including device object data, base station object data, and user roaming data. Simultaneously, it collects meteorological data, event data, and urban geographic information data related to the impact of pedestrian traffic. The signaling data is then cleaned and fused, specifically including: parsing base station location information and associating it with geographic area boundaries; generating time point labels for user entry / exit from the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; and finally establishing a spatiotemporal pedestrian traffic sample dataset. The feature engineering construction module is used to analyze the sample dataset and construct a hybrid feature set including geographic block attribute features, user behavior attribute features, weather features and event features based on the spatiotemporal tags, meteorological data and event data. The pedestrian flow prediction model construction module is used to construct a prediction model using a hybrid architecture learning algorithm combining LightGBM and LSTM. Specifically: the hybrid feature set is aggregated by unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, learning the feature influence weights through the LightGBM model and learning the temporal variation patterns through the LSTM model, resulting in a trained short-term pedestrian flow prediction model based on spatiotemporal data. The pedestrian flow prediction module is used to calculate real-time sample data using the trained short-term pedestrian flow prediction model and output short-term prediction data results of urban spatiotemporal pedestrian flow.

[0014] Thirdly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the short-term pedestrian flow prediction processing method based on spatiotemporal data as described in the first aspect.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: The method, system, and medium for short-term pedestrian flow prediction based on spatiotemporal data provided in this application improve the efficiency and accuracy of urban pedestrian flow data processing by constructing a spatiotemporal feature label set and using machine learning algorithms to build a pedestrian flow prediction model. It can effectively solve the problems of fragmented data, lack of unified integration process, limitations in urban functional area division and population positioning, and lack of spatiotemporal features and meteorological information in existing urban pedestrian flow data processing methods. At the same time, through multi-time series model hybrid architecture processing and parameter adjustment of multiple features, it forms a prediction method, system, and medium service capability for different fields such as scenic spots, transportation, and urban areas. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the short-term pedestrian flow prediction processing method based on spatiotemporal data according to the present invention.

[0018] Figure 2 This is a schematic diagram of the short-term pedestrian flow prediction and processing system based on spatiotemporal data according to the present invention. Detailed Implementation

[0019] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the short-term pedestrian flow prediction processing method based on spatiotemporal data. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0020] For example, when processing urban pedestrian flow data, we face massive, complex, and dynamic datasets, and current processing methods reveal some significant shortcomings and challenges. First, the data acquisition and processing are quite fragmented, lacking a unified and systematic data integration and preprocessing framework. The varying signaling data formats provided by different telecommunications operators, coupled with the inherent complexity and variability of urban geographic information, make data integration and cleaning exceptionally arduous. This not only significantly increases the time and economic costs of data processing but may also introduce inconsistencies that could compromise the accuracy of subsequent data analysis. Second, existing data processing methods have significant limitations in delineating urban functional areas and locating populations. Traditional regional division methods rely primarily on static geographic information, making it difficult to accurately capture the dynamic characteristics of population movement. For instance, on weekdays and during working hours, the population is mainly concentrated in the city center and along major transportation routes; while on weekends and rest days, population movement may be more widespread and dispersed. Furthermore, due to limitations in data accuracy and timeliness, population location technology often introduces errors, further affecting the accuracy and reliability of the data. Furthermore, current population flow prediction models are insufficient in incorporating spatiotemporal and event-related characteristics of population flow impacts. When processing time-series data, these models fail to fully consider the profound impact of geographical location and temporal changes, weather variations, and major events on population flow patterns, thus reducing the accuracy and reliability of the prediction results.

[0021] To address the aforementioned issues, this embodiment provides a short-term pedestrian flow prediction processing method based on spatiotemporal data. By integrating multi-source data, a dataset of historical pedestrian flow, spatiotemporal features, historical meteorological features, and event features is constructed. The LightGBM and LSTM hybrid architecture learning algorithm is then used for prediction, which can more accurately reflect the dynamic changes in urban population flow. This enables short-term pedestrian flow prediction at different granularities in different urban areas, while also reducing problems such as the inability of a single prediction model to learn in reverse and discover temporal periodic changes.

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 The diagram shows a flowchart of a short-term pedestrian flow prediction method based on spatiotemporal data in a specific embodiment. The method includes the following steps: S1. Establish signaling data for users on the operator's communication network, acquiring user signaling data including device object data, base station object data, and user roaming data; simultaneously collect meteorological data, event data, and urban geographic information data related to the impact of pedestrian traffic; clean and fuse the signaling data, specifically including: parsing base station location information and associating it with geographic area boundaries; generating time point labels for user entry / exit from the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; finally, establish a spatiotemporal pedestrian traffic sample dataset; It should be noted that the signaling data of operators' communication network users were connected, and multi-source data were integrated, providing a rich data foundation for subsequent predictions. Data related to the impact of pedestrian traffic were collected and organized, and a sample dataset was established. S2. Analyze the sample dataset and, based on the spatiotemporal labels, meteorological data, and event data, construct a hybrid feature set including geographic block (e.g., business district) attribute features, user behavior (e.g., working population, tourists) attribute features, weather features, and event features; It should be noted that a spatiotemporal feature set was constructed, which provides important spatiotemporal information support for the division of urban functional areas, population positioning and prediction, and provides a basis for subsequent analysis; a weather feature set and an event feature set were constructed, which provide information support for learning to identify the changes in pedestrian flow under different spatiotemporal conditions caused by weather and event activities, and for mining the importance of feature sets to pedestrian flow, and provide important input parameters for subsequent model prediction; S3. A prediction model is constructed using a hybrid architecture learning algorithm combining LightGBM and LSTM, wherein: the hybrid feature set is aggregated according to unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, the LightGBM model learns the feature influence weights, and the LSTM model learns the temporal change pattern, thus obtaining a trained short-term pedestrian flow prediction model based on spatiotemporal data; It should be noted that, based on the spatiotemporal characteristic data patterns, calculations and training are conducted for different times in the city, such as weekdays, rest days, and holidays, and for different spatiotemporal areas, such as administrative districts, scenic spots, business districts, and transportation hubs, to construct optimal adaptation parameters and build an optimal short-term population flow prediction model method based on spatiotemporal data that is adapted to different regional granularities. A machine hybrid architecture learning algorithm is used to construct a population flow prediction model based on spatiotemporal data, and sample data is used for training to obtain accurate prediction results. Through hybrid learning and iterative optimization and performance evaluation, the learning and recognition of the impact characteristics of population flow under different time series are deepened, ensuring the stability and reliability of the prediction model. S4. Calculate the real-time sample data using the trained short-term pedestrian flow prediction model and output the short-term prediction data of urban spatiotemporal pedestrian flow.

[0024] This embodiment integrates multi-source data, constructs a feature set, and uses a hybrid architecture learning algorithm for prediction, which can more accurately reflect the dynamic changes in urban pedestrian traffic.

[0025] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, a method for short-term prediction and processing of pedestrian flow based on spatiotemporal data is provided. This method includes the following steps: S1. Establish communication with the operator's network to obtain user signaling data, including device object data, base station object data, and user roaming data; simultaneously collect meteorological data, event data, and urban geographic information data related to the impact of pedestrian traffic; clean and fuse the signaling data, specifically including: parsing base station location information and associating it with geographic area boundaries; generating time point labels for user entry / exit from the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; finally, establish a spatiotemporal pedestrian traffic sample dataset; The specific steps of step S1 are as follows: S11. Obtain signaling data from network users of various communication operators, and select the time update granularity (15 seconds / time) to distinguish the signaling data according to the required prediction time range (e.g., 1 hour / 3 hours); the signaling data includes device object data (e.g., mobile phone IMSI), base station object data (e.g., base station ID, latitude and longitude), user object data (e.g., IMSI) and user roaming data, and merge trajectory data through the multi-device IMSI association rule of the same user; For example, the required forecast time range can be selected with granularity such as week, day, or hour; S12. Collect meteorological data (such as temperature, rainfall, and wind), spatiotemporal characteristic data (such as road network density), and event data (scale, type, and duration) to prepare urban area division and geographical location information data (boundary vectors and latitude and longitude range) that characterize urban spatial planning. For example, meteorological data includes data such as maximum temperature, minimum temperature, wind force, wind direction, and air quality; urban area division and geographic location information data are geographic layer information data obtained from the government information disclosure of location A, which are used to construct geographic information data such as location A, districts, scenic spots, business districts, and transportation hubs. S13. According to the required prediction time range (e.g., 1 hour / 3 hours) and spatial range (e.g., a single business district or multiple community clusters), link the data tables through "base station ID-region code" and "time stamp-statistical interval" to establish sample set data; S14. Perform useless field removal (such as user privacy fields unrelated to location), text feature processing, missing value processing (such as filling with the mean of the three most recent data from the same period) and standardization on the sample set data, and obtain a standard dataset through various label processing associations; It should be noted that by going through the steps of data acquisition and preprocessing, including the acquisition and analysis of signaling data, the acquisition of urban geographic information data, the acquisition of historical weather information data, the acquisition of information data on major urban events, the construction and association of data tables, and the cleaning and standardization of data, we can ensure the integrity and consistency of data, provide a high-quality data foundation for subsequent processing, reduce the time and cost of data processing, and improve the efficiency of data processing. S2. Analyze the sample dataset. Based on the spatiotemporal labels, meteorological data, and event data, construct a hybrid feature set including geographic block (e.g., business district) attribute features, user behavior (e.g., working population, tourists) attribute features, weather features, and event features. The specific steps of step S2 are as follows: S21. Based on weather and meteorological data, feature classification is performed, dividing it into text-based features (such as "heavy rain" and "cloudy") and numerical features (such as temperature 25℃ and wind force level 3). The text-based content is assigned a machine-recognizable numerical feature range ("heavy rain" = 1.0, "heavy rain" = 0.8, "moderate rain" = 0.5, "light rain" = 0.3, "cloudy" = 0.1, "sunny" = 0.0). For example, the text class can be wind direction, such as northeast wind, southwest wind, southeast wind, etc., to construct the range of numerical features [1-8]; S22. Based on the time cycle data of pedestrian flow, construct a time feature engineering for identifying time feature ranges, including time dimension divisions of "hourly level (6:00-9:00 morning peak / 12:00-14:00 noon / 18:00-21:00 evening peak)", "daily level (weekdays / weekends / holidays)" and "weekly level (Monday to Friday / Saturday to Sunday)"; For example, year: 2024, month: December, day: 1st, Spring Festival, Dragon Boat Festival, May Day, etc.; S23. Define the event (e.g., "Large-scale event: ≥10,000 people", "Medium-scale event: 5,000-10,000 people", "Small-scale event: <5,000 people"), and construct event feature engineering based on the event data, including event size weight (large-scale = 1.0, medium-scale = 0.6, small-scale = 0.3) and event type weight (concert = 0.8, exhibition = 0.6, market = 0.4), and calibrate the weights by the deviation rate (deviation rate <10%) between the actual and predicted pedestrian flow of the last 3 similar events; For example, the event time characteristic engineering may include the number of large-scale performance events, the number of large-scale sporting events, the number of large-scale exhibitions, etc. S24. Construct historical data feature engineering based on the traffic flow prediction cycle (1 hour / 3 hours), select traffic flow data, meteorological data, and event data from the same period in the past 6 months (such as the same day last week or the same day last month) as historical features, and the time granularity of the historical data is consistent with the prediction cycle (1 hour prediction corresponds to 1 hour historical granularity, and 3 hours prediction corresponds to 3 hours historical granularity). For example, historical data feature engineering may include the average of the last 7 days, the average of the last 3 days, the average of the last 1 day, etc. It should be noted that constructing weather and meteorological features, time features, major event features, and historical data features can effectively classify the features affecting pedestrian flow and provide feature datasets for LightGBM feature selection. S3. A prediction model is constructed using a hybrid architecture learning algorithm combining LightGBM and LSTM, wherein: the hybrid feature set is aggregated according to unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the dimension-adjusted current dataset itself; the hybrid model is trained using the sample dataset, the LightGBM model learns the feature influence weights, and the LSTM model learns the temporal variation patterns, resulting in a trained short-term pedestrian flow prediction model based on spatiotemporal data; the specific steps of step S3 are as follows: S31. Aggregate the sample data based on the added spatiotemporal feature label set, summarize the data within a unit statistical time period, add an index to the summarized data based on the unit statistical time period (e.g., "2024-05-01 08:00-08:15"), and adjust the dimensions according to the requirements of the LSTM model (time step = 12, corresponding to a 3-hour time series window) as the current dataset; S32. Construct an LSTM model, set one input of the LSTM model to the current dataset, set the output of the LSTM model to population prediction data of the same dimension as the current dataset, and select the activation functions required for the hidden layer (2 layers) and fully connected layer (1 layer) of the LSTM model (tanh for the LSTM layer and ReLU for the fully connected layer). S33. Construct a LightGBM model and set the input of the LightGBM model to the current dataset. Combine the LightGBM model and the LSTM model by setting the output of the LightGBM model as another input of the LSTM model to form a hybrid architecture pedestrian flow prediction model. S34. Divide the current dataset into training, validation, and test sets, and set the loss function and optimizer; S35. The traffic flow prediction model is trained using the training set. During the training process, the learning rate, number of iterations, number of trees, and number of leaf nodes are continuously adjusted to continuously optimize the LightGBM model. The influence of time feature values ​​on time series data is learned. The number of iterations of the LSTM model is controlled according to the loss function and iteration threshold. The LSTM model is continuously optimized through the optimizer. At the same time, the performance of the LSTM model during the iteration process is monitored through the validation set. S36. Use the test set to evaluate the performance of the trained pedestrian flow prediction model, and input the current dataset into the evaluated pedestrian flow prediction model to predict short-term pedestrian flow data; It should be noted that the process of building a spatiotemporal data-based population flow prediction model using a hybrid learning prediction model includes data aggregation and indexing, model construction and training, and model performance evaluation. The LSTM model can fully consider the characteristics of time series data, while the LightGBM model effectively filters the features that affect population flow as supplementary parameters to the LSTM model, thereby improving the accuracy and reliability of population flow prediction. S4. Calculate the real-time sample data using the trained short-term pedestrian flow prediction model and output the short-term prediction data of urban spatiotemporal pedestrian flow.

[0026] In one embodiment of the present invention, based on step S13, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0027] The specific steps of step S13 are as follows: S131. Obtain geographic location information data and base station object data, parse latitude and longitude information location and connect it with base station location, construct the association between regional boundary information and base station location, and construct geographic region labels; S132. Combine geographic region labels with urban area division information data to add geographic attributes to different geographic regions and form geographic block attribute labels; S133. Construct the regional boundary based on geographic location information data, determine the signaling trajectory data based on equipment object data and base station object data, and associate the signaling trajectory data with the regional boundary to obtain the population flow trajectory; S134. Add time point labels to different entry and exit locations based on the location update time period and base station location in the base station information data to obtain the entry time point and exit time point; S135. Calculate the user’s dwell time in each area based on the entry and exit times, then add a daily location tag list to the dwell time statistics, summarize the daily location tag list to form a monthly area tag list, and associate it with the actual scene to form spatiotemporal tags. S136. Add population residence duration attribute tags based on residence duration, and then add work / residence attribute tags and regional residence attribute tags according to the time period to which the residence duration belongs, to obtain the geographic block population attribute tags; such as working population, residential population, etc. For example, the geographic block population attribute tags may include visitor tags, work tags, and residence tags; S137. Based on geographic block attribute tags, geographic block population attribute tags, and spatiotemporal tags, track the activity trajectory of each user at different times and geographic spaces to form a spatiotemporal population flow feature dataset; S138. Statistical analysis of the crowd trajectory of each number in the spatiotemporal pedestrian flow characteristic dataset is performed. Combined with the business survey results of different scenarios such as scenic spots, business districts, and stations, a sample dataset is constructed that conforms to the actual business scenario of working or living and passing by people. By removing the two groups in the region, the sample dataset is obtained. It should be noted that by constructing a dataset that associates location signaling with urban geospatial data, the impact of geographical location and temporal changes on population flow can be fully considered. By parsing geographical location information, constructing geographical area labels, and population flow trajectories, the spatiotemporal characteristics of population flow can be accurately captured.

[0028] In one embodiment of the present invention, based on steps S132, S133, S134 and S138, a possible embodiment will be given below, and its specific implementation will be described in a non-limiting manner.

[0029] The specific steps of step S132 are as follows: S1321. Based on the GIS information of the map, regional boundary information, and POI points within the region, make an initial determination of the attributes of urban functional areas; If there is a mismatch in the urban functional area attributes, proceed to step S1322; If the city's functional area attributes are correct, proceed to step S1323; S1322. Notify the city functional area to be revised, return to step S1; S1323. Based on the roads and residential areas in the GIS map, and combined with the trajectories of the population, analyze and verify the rationality of urban functional areas; If reasonable, proceed to step S1324; If this is not reasonable, return to step S1322; S1324. Determine if there are areas that cannot be obtained through POI points and GIO information; If so, proceed to step S1325; If not, return to step S1322; S1325. Based on the known behavioral preferences and relationship networks of communities within the community, identify and store the data in residential areas, set new urban functional area markers, and return to step S1322; The specific steps of step S133 are as follows: S1331. Add an internal ID with displacement encoding to the signaling data, and count the incoming data in each cell based on the internal ID and count the local users based on the user IMSI to complete the preprocessing of the signaling data; S1332. Based on the preprocessed signaling data, statistically analyze the cell-level user location and historical trajectory, and manage them with the base station latitude and longitude to obtain the user's latitude and longitude; S1333. Based on the mapping of communities and regions, locate the population area and determine whether the population area location does not match the population activity trajectory; If not, proceed to step S134; If so, proceed to step S1334; For example, the area mapping can be a residential area, business district, scenic spot, billboard location, etc. S1334. Return to step S1 if the incorrect population area location is incorrect. The specific steps of step S134 are as follows: S1341. Set time periods based on users' daily routines and collect information on users' activity areas within each time period; S1342. By setting the number of days the user stays in the community information, obtain the user area data, and accumulate and summarize the data according to the time period to obtain the daily and monthly area population size data; S1343. Compare daily regional population size data with daily location tag list, and compare monthly regional population size data with monthly regional tag list; If they match, proceed to step S135; If there is a discrepancy, proceed to step S1344; S1344. Notification of spacetime label error, return to step S1; The specific steps of step S138 are as follows: S1381. Construct an encrypted chain from the user's signaling data trajectory, with the work location grid as the head of the encrypted chain and the residential location grid as the tail of the encrypted chain, and set the effective dwell point, entry time and exit time in the main body of the encrypted chain to generate the encrypted chain; The cryptographic chain is represented in the following format: Head: Work location grid (GeoHash) Tail: Residence grid (GeoHash) Chain body: Effective dwell point {GEOHASHCODE, entry time, exit time}; For example, User A: {Workplace grid WX4ER}, {Residence grid WX4G3}, {wtw37m, entry time, departure time}, {wtw37n, entry time, departure time}… S1382. Select two users from each geographic region and determine a certain month as the statistical period; S1383. Calculate the dwell time of two users in the same grid within the same time period during the statistical period, and sum them to obtain the first total duration T1; S1384. Calculate the dwell time of two users in different networks within the same time period during the statistical period, and sum them to obtain the second total duration T2. S1384. Calculate the effective dwell time of two users simultaneously existing within the statistical time period, and sum them to obtain the third time sum T3; S1385. T1 / T3 is denoted as the trajectory overlap degree, and T2 / T3 is denoted as the trajectory non-overlap degree; S1386. Determine whether the trajectory overlap is greater than a preset overlap threshold, while the trajectory non-overlap is less than a preset non-overlap threshold. If so, proceed to step S1387; If not, proceed to step S138; S1387. Determine that the two selected users are the same person, notify the population statistics deduplication process, and return to step S1; For example, the overlap threshold can be set to 90%, and the non-overlap threshold can be set to 10%. It should be noted that, in order to reduce the amount of calculation, this calculation only applies to numbers with the same work and residence locations, so a preliminary user screening is required beforehand.

[0030] In one embodiment of the present invention, based on step S14, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0031] The specific steps of step S14 are as follows: S141. Obtain geographic location information data (boundary vector, latitude and longitude range) and base station object data (base station ID, latitude and longitude), parse latitude and longitude location information and base station location information, construct the dynamic spatial association between regional boundary information and base station location, and construct geographic region features; S142. Combine geographical region characteristics with urban area division information data to add geographical attributes to different geographical regions and form geographical block attribute tags; S143. Construct the area boundary (vector polygon) based on the geographic location information data, determine the signaling trajectory data based on the device object data (multi-device ID) and the base station object data (base station latitude and longitude), and merge the trajectory of the same user by associating the multi-device IMSI, and associate the merged signaling trajectory data with the area boundary to obtain the population flow trajectory; S144. Add time point labels to different entry and exit locations based on the location update time period and base station location in the base station information data to obtain the entry time point and exit time point; S145. Calculate the user's dwell time in each area by statistically analyzing the entry and exit times in 30-minute intervals. Then, perform daily statistics on the dwell time (statistics from 6:00 to 24:00 every day). Add a daily location tag list containing "dwelling during working hours (e.g., 9:00-18:00) / dwelling during non-working hours". Summarize the daily location tag list by calendar month to form a monthly area tag list containing "monthly dwell days ≥ 20 days / 10-19 days / < 10 days". Associate this with the actual scene (work area, residential area, leisure area) to form spatiotemporal tags. S146. Based on the length of stay, add population stay duration attribute tags of "stay ≥ 4 hours (weight 0.8) / 1-4 hours (weight 0.5) / < 1 hour (weight 0.2)". Then, according to the time period to which the stay duration belongs, add regional stay attribute tags of "work attribute (9:00-18:00 stay ≥ 4 hours) / residence attribute (20:00-next day 8:00 stay ≥ 6 hours)" and "high frequency stay (monthly stay ≥ 20 days) / medium frequency stay (10-19 days) / low frequency stay (< 10 days)" to obtain the geographic block population attribute tags; S147. Based on geographic block attribute tags, spatiotemporal tags, and geographic block population attribute tags, track the activity trajectory of each user at different times and geographic spaces to form a spatiotemporal population flow feature dataset; S148. Perform statistical analysis on the trajectory of each mobile phone number in the spatiotemporal pedestrian flow feature dataset, identify "duplicate trajectories of multiple device IDs under the same IMSI" and "duplicate signaling trajectories of the same base station within the same time period" and perform deduplication to ensure that the redundancy rate of the dataset after deduplication is ≤5%, and obtain the spatiotemporal pedestrian flow sample dataset.

[0032] In one embodiment of the present invention, based on steps S31, S32, S34 and S35, a possible embodiment will be given below, and its specific implementation will be described in a non-limiting manner.

[0033] The specific steps of step S31 are as follows: S311. Based on the spatiotemporal label feature set, filter out data related to pedestrian flow; S312. Integrate the geographic block attribute labels, geographic block population attribute labels, and spatiotemporal labels from the spatiotemporal label feature set into the dataset to ensure that each data entry contains complete spatiotemporal information; S313. Aggregate the data according to the unit statistical time, merge the data records within the same time period, and summarize the traffic flow in each time period; S314. Add time period indexes to the aggregated data, and adjust the dimensions and format of the data according to the requirements of the LSTM model to obtain the current dataset; It should be noted that by aggregating and indexing data, including filtering data related to pedestrian flow, integrating spatiotemporal label feature sets, and aggregating data according to unit statistical time, the integrity and consistency of the data can be ensured, providing an accurate data foundation for subsequent pedestrian flow prediction. Table 1 shows the calculation logic of population flow monitoring indicators based on spatiotemporal data in location A: Table 1

[0034] The specific steps of step S32 are as follows: S321. Determine the number of layers and the number of neurons in each layer, and initialize an LSTM model; S322. Encode the data related to pedestrian traffic and the spatiotemporal feature label set in the current dataset into vector form and set it as the input of the LSTM model; S323. Set the output of the LSTM model to pedestrian flow data of the same dimension as the current dataset; S324. Set the number of LSTM layers as needed, and set the number of neurons and activation function for each LSTM layer; The specific steps of step S34 are as follows: S341. Extract features that may affect pedestrian flow from weather feature sets, spatiotemporal pedestrian flow datasets, and major event datasets, and scale the numerical features to ensure the stability of the values; S342. Construct LightGBM and set model parameters, including learning rate, number of iterations, number of trees, and number of leaf nodes; For example, the learning rate is usually set between 0.01 and 0.3, the number of iterations is initially set to 100, the maximum depth of the tree is set between 3 and 20, and the minimum number of leaf node samples is set to 20 by default. The specific settings need to be adjusted according to the data. S343. Use the training set data to perform LightGBM, and adjust the iterative model parameters by monitoring the performance of the validation set data. S344. Analyze the degree of influence of important feature values ​​on the model's predictive performance, evaluate the importance of different features on the prediction of spatiotemporal pedestrian flow, and form data results on the important influence of pedestrian flow prediction; S345. LightGBM is used to filter features and generate a highly generalized intermediate representation. That is, the feature data and the time series feature relationship data group are used as the LSTM model parameter input for learning, so as to capture the activity pattern of people flow under the influence of different events, weather and other features in different time series and form prediction results. The specific steps of step S35 are as follows: S351. The performance of the trained LSTM model is evaluated using a test set. Root mean square error (RMSE) and mean absolute error (MAE) are selected as two metrics to measure the prediction accuracy of the pedestrian flow prediction model. The calculation formulas are as follows:

[0035]

[0036] In the formula: m is the number of samples; This represents the actual passenger flow. To predict passenger flow values ​​for the model, This represents the absolute error between the actual passenger flow and the model-predicted passenger flow. If the evaluation is successful, proceed to step S352; If the evaluation fails, return to step S34; S352. Select a dataset from the current dataset that is less than a set threshold from the current time point, and input the selected dataset into the LSTM model to obtain the predicted traffic flow data; The short-term prediction model constructed in this invention is based on LSTM. The model framework is mainly divided into two parts: one is the preparation of the input layer dataset, and the other is the LSTM module. (1) Specifically, the model dataset structure is as follows:

[0037] After normalizing the dataset, the dataset is then divided into training and training sets. and test set The training set serves as the input to the model, and the test set is used to evaluate the model's performance.

[0038] (2) Specifically, the LSTM model method includes three gates: forget, input, and output. The forget gate uses the sigmoid function to filter the remembered information and the information newly input from the previous moment. When the gate value is 1, all information is retained; when the gate value is 0, all information is discarded. The input gate is used to control how much network input enters the unit at the current moment; the output gate is used to control how much control state enters the output of the current unit. The specific operation process of the LSTM model is as follows: Forget Gate: Deletes information from the cell state.

[0039] Input gate: Adds information to the cell state

[0040]

[0041] Update cell status:

[0042] Output gate: Selects useful information from the cell state and transmits it as output.

[0043]

[0044]

[0045] In the above formula: For the current time step input, , , These are the weight matrices for the forget gate, input gate, and output gate, respectively. , , These are the bias terms for each gating unit.

[0046] After repeated testing with multiple sets of data, the main LSTM parameters were determined as follows: 100 neurons, 4 features per sequence element (including date, pedestrian traffic, weather, and event), tanh activation function, and weight initialization method. , , The truncated normal distribution; It should be noted that the construction and training of the LSTM model are described in detail, including model initialization, input and output settings, determination of the number of LSTM layers and neurons, and selection of activation functions. This ensures the stability and accuracy of the model and improves the effectiveness of pedestrian flow prediction.

[0047] For example, predicting pedestrian flow using an LSTM model can be achieved with the following code: Create a time series model: model = Sequential() Add an LSTM network with 100 neurons model.add(LSTM(100, activation='relu', input_shape=(n_timesteps, 1))) Add a fully connected layer model.add(Dense(1)) Using the Adam optimizer, mean_squared_error is used as the loss function. model.compile(optimizer='adam', loss='mean_squared_error') The model was trained in 500 epochs. model.fit(X_train, y_train, epochs=500, batch_size=32, verbose=0) Input data for a future period is fed into a trained LSTM model to obtain predictions for the future.

[0048] predictions = model.predict(data).

[0049] This application constructs a pedestrian flow prediction and processing model based on spatiotemporal data, effectively grasping the impact of meteorological data and event information on pedestrian flow. It conducts pedestrian flow predictions for key pedestrian gathering areas such as administrative districts, scenic spots, commercial areas, and transportation hubs during different time periods, including rest days and holidays. By applying the prediction algorithm to different granularities within the city and continuously training and validating it, it obtains adaptation parameters for different regional ranges, thereby obtaining short-term pedestrian flow prediction results based on spatiotemporal data that are adapted to different granularities. This effectively improves the ability to analyze the patterns of urban pedestrian flow prediction and processing, as well as the assessment of holiday pedestrian flow gatherings in different granular areas, empowering urban population management and commercial management services.

[0050] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0051] like Figure 2 As shown, the following are embodiments of the short-term pedestrian flow prediction processing system based on spatiotemporal data provided in this disclosure. This system and the short-term pedestrian flow prediction processing method based on spatiotemporal data in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the short-term pedestrian flow prediction processing system based on spatiotemporal data, please refer to the embodiments of the short-term pedestrian flow prediction processing method based on spatiotemporal data described above.

[0052] The system includes: The sample dataset construction module is used to connect the signaling data of operators' communication network users, obtain data from different business calibers of signaling, and collect and organize meteorological data, spatiotemporal characteristic data and event data related to the impact of traffic flow to establish a sample dataset. The feature engineering construction module is used to analyze the sample dataset and construct a hybrid feature set including geographic block attribute features, user behavior attribute features, weather features and event features based on the spatiotemporal tags, meteorological data and event data. The pedestrian flow prediction model construction module is used to construct a prediction model using a hybrid architecture learning algorithm combining LightGBM and LSTM. Specifically: the hybrid feature set is aggregated by unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, learning the feature influence weights through the LightGBM model and learning the temporal variation patterns through the LSTM model, resulting in a trained short-term pedestrian flow prediction model based on spatiotemporal data. The pedestrian flow prediction module is used to calculate real-time sample data using the trained short-term pedestrian flow prediction model and output short-term prediction data results of urban spatiotemporal pedestrian flow.

[0053] This embodiment implements a short-term pedestrian flow prediction system based on spatiotemporal data by constructing a dataset, building a feature engineering model, training a prediction model, and implementing a pedestrian flow prediction module.

[0054] The storage medium provided in this application stores a program product capable of implementing a short-term pedestrian flow prediction and processing method based on spatiotemporal data.

[0055] The short-term pedestrian flow prediction processing method based on spatiotemporal data includes: connecting to the operator's communication network to obtain user signaling data, which includes device object data, base station object data, and user roaming data; simultaneously collecting meteorological data, event data, and urban geographic information data related to the impact of pedestrian flow; cleaning and fusing the signaling data, specifically including: parsing base station location information and associating it with geographic area boundaries; generating time point tags for users entering / leaving the area based on the signaling update period; calculating user dwell time based on the time point tags and generating spatiotemporal tags characterizing user spatiotemporal behavior patterns; finally establishing a spatiotemporal pedestrian flow sample dataset; analyzing the sample dataset, and constructing a dataset including geographic block attribute features, user behavior attribute features, weather features, and event features based on the spatiotemporal tags, meteorological data, and event data. A mixed feature set is obtained; a prediction model is constructed using a hybrid architecture learning algorithm combining LightGBM and LSTM, wherein: the mixed feature set is aggregated by unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, the LightGBM model learns the feature influence weights, and the LSTM model learns the temporal change patterns, resulting in a trained short-term pedestrian flow prediction model based on spatiotemporal data; the trained short-term pedestrian flow prediction model is used to calculate real-time sample data, and the short-term prediction data results of urban spatiotemporal pedestrian flow are output.

[0056] In some possible implementations, the short-term pedestrian flow prediction processing method based on spatiotemporal data disclosed herein can be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0057] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0058] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for short-term prediction and processing of pedestrian flow based on spatiotemporal data, characterized in that, Includes the following steps: S1. Connect to the operator's communication network and obtain the user's signaling data, which includes device object data, base station object data, and user roaming data; at the same time, collect meteorological data, event data, and urban geographic information data related to the impact of pedestrian traffic. The signaling data is cleaned and fused, specifically including: parsing base station location information and associating it with geographical area boundaries; generating time point labels for users entering / leaving the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; and finally establishing a spatiotemporal pedestrian flow sample dataset. S2. Analyze the sample dataset and construct a hybrid feature set including geographic block attribute features, user behavior attribute features, weather features and event features based on the spatiotemporal tags, meteorological data and event data; S3. A prediction model is constructed using a hybrid architecture learning algorithm combining LightGBM and LSTM, wherein: the hybrid feature set is aggregated according to unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, the LightGBM model learns the feature influence weights, and the LSTM model learns the temporal change pattern, thus obtaining a trained short-term pedestrian flow prediction model based on spatiotemporal data; S4. Calculate the real-time sample data using the trained short-term pedestrian flow prediction model and output the short-term prediction data of urban spatiotemporal pedestrian flow.

2. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Obtain signaling data from network users of various communication operators, and select the time update granularity of the signaling data according to the required prediction time range; the signaling data includes device object data, base station object data, user object data and user roaming data, and the trajectory data is merged through the multi-device IMSI association rule of the same user; S12. Collect meteorological data, spatiotemporal characteristic data, and event data to prepare urban area division and geographical location information data that characterize urban spatial planning; S13. According to the required prediction time range and spatial range, establish sample set data by associating each data table with base station ID and area code and with timestamp and interval; S14. Perform useless field deletion, text feature processing, missing value processing, and standardization on the sample set data. Then, obtain a standard dataset by associating various labels.

3. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 2, characterized in that, The specific steps of step S14 are as follows: S141. Obtain geographic location information data and base station object data, parse latitude and longitude location information and base station location information, construct a dynamic spatial association between regional boundary information and base station location, and construct geographic region features; S142. Combine geographical region characteristics with urban area division information data to add geographical attributes to different geographical regions and form geographical block attribute tags; S143. Construct regional boundaries based on geographic location information data, determine signaling trajectory data based on device object data and base station object data, and merge the trajectory of the same user through multi-device IMSI association, and associate the merged signaling trajectory data with the regional boundaries to obtain the population flow trajectory; S144. Add time point labels to different entry and exit locations based on the location update time period and base station location in the base station information data to obtain the entry time point and exit time point; S145. Based on the entry time and exit time, perform statistical calculations according to the preset time interval to obtain the user's dwell time in each area. Then, perform daily statistics on the dwell time and add a daily location tag list. Summarize the daily location tag list by natural month to form a monthly area tag list, and associate it with the actual scene to form a spatiotemporal tag. S146. Based on the length of stay, add population length of stay attribute tags, and then add work / residence attribute tags and regional length of stay attribute tags according to the time period to which the length of stay belongs, to obtain the population attribute tags of the geographic block; S147. Based on geographic block attribute tags, spatiotemporal tags, and geographic block population attribute tags, track the activity trajectory of each user at different times and geographic spaces to form a spatiotemporal population flow feature dataset; S148. Perform statistical analysis on the trajectory of each mobile phone number in the spatiotemporal pedestrian flow feature dataset, identify duplicate trajectories of multiple device IDs under the same IMSI and duplicate signaling trajectories of the same base station within the same time period, and perform deduplication to ensure that the redundancy rate of the dataset after deduplication is less than the preset redundancy rate threshold, thereby obtaining the spatiotemporal pedestrian flow sample dataset.

4. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 2, characterized in that, The specific steps of step S2 are as follows: S21. Based on meteorological data, feature classification is performed, dividing it into text-based and numerical-based related features, and assigning machine-recognizable numerical feature ranges to text-based content; S22. Based on the time cycle data of pedestrian flow, construct a time feature engineering for identifying time feature ranges, including hourly, daily, and weekly time dimension divisions; S23. Define the event and construct the event feature engineering based on the event data, including event scale weight and event type weight, and calibrate the weights by the deviation rate between the actual flow and the predicted flow of the last N similar events. S24. Construct historical data feature engineering based on the traffic flow forecast cycle, and select traffic flow data, meteorological data, and event data from the same period in the past M months as historical features, and the time granularity of the historical data is consistent with the forecast cycle.

5. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 4, characterized in that, The specific steps of step S3 are as follows: S31. Aggregate the sample data based on the added spatiotemporal feature label set, summarize the data within a unit statistical time period, add an index to the summarized data based on the unit statistical time period, and adjust the dimensions according to the requirements of the LSTM model to form the current dataset; S32. Construct an LSTM model, set one input of the LSTM model to the current dataset, set the output of the LSTM model to population prediction data of the same dimension as the current dataset, and select the activation functions required for the hidden layers and fully connected layers of the LSTM model; S33. Construct a LightGBM model, set the input of the LightGBM model to the current dataset, and combine the LightGBM model and the LSTM model by setting the output of the LightGBM model as another input of the LSTM model to form a hybrid architecture pedestrian flow prediction model; S34. Divide the current dataset into training, validation, and test sets, and set the loss function and optimizer; S35. The traffic flow prediction model is trained using the training set. During the training process, the learning rate, number of iterations, number of trees, and number of leaf nodes are continuously adjusted to continuously optimize the LightGBM model. The influence of time feature values ​​on time series data is learned. The number of iterations of the LSTM model is controlled according to the loss function and iteration threshold. The LSTM model is continuously optimized through the optimizer. At the same time, the performance of the LSTM model during the iteration process is monitored through the validation set. S36. Use the test set to evaluate the performance of the trained pedestrian flow prediction model, and input the current dataset into the evaluated pedestrian flow prediction model to predict short-term pedestrian flow data.

6. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 5, characterized in that, The specific steps of step S31 are as follows: S311. Based on the spatiotemporal label feature set, filter out data related to pedestrian flow; S312. Integrate the geographic block attribute labels, geographic block population attribute labels, and spatiotemporal labels from the spatiotemporal label feature set into the dataset to ensure that each data entry contains complete spatiotemporal information; S313. Aggregate the data according to the unit statistical time, merge the data records within the same time period, and summarize the traffic flow in each time period; S314. Add time period indexes to the aggregated data, and adjust the dimensions and format of the data according to the requirements of the LSTM model to obtain the current dataset.

7. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 6, characterized in that, The specific steps of step S32 are as follows: S321. Determine the number of layers and the number of neurons in each layer, and initialize an LSTM model; S322. Encode the data related to pedestrian traffic and the spatiotemporal feature label set in the current dataset into vector form and set it as the input of the LSTM model; S323. Set the output of the LSTM model to pedestrian flow data of the same dimension as the current dataset; S324. Set the number of LSTM layers as needed, and set the number of neurons and activation function for each LSTM layer.

8. The method for short-term prediction and processing of pedestrian flow based on spatiotemporal data according to claim 5, characterized in that, The specific steps of step S36 are as follows: S361. Use the test set to evaluate the performance of the trained pedestrian flow prediction model; If the evaluation is successful, proceed to step S362; If the evaluation fails, return to step S35; S362. Select a dataset from the current dataset that is less than a set threshold away from the current time point, and input the selected dataset into the pedestrian flow prediction model to obtain the predicted pedestrian flow data.

9. A short-term pedestrian flow prediction and processing system based on spatiotemporal data, characterized in that, include: The sample dataset construction module is used to connect to the operator's communication network and obtain user signaling data, which includes device object data, base station object data and user roaming data; at the same time, it collects meteorological data, event data and urban geographic information data related to the impact of traffic flow. The signaling data is cleaned and fused, specifically including: parsing base station location information and associating it with geographical area boundaries; generating time point labels for users entering / leaving the area based on the signaling update period; calculating user dwell time based on the time point labels and generating spatiotemporal labels representing user spatiotemporal behavior patterns; and finally establishing a spatiotemporal pedestrian flow sample dataset. The feature engineering construction module is used to analyze the sample dataset and construct a hybrid feature set including geographic block attribute features, user behavior attribute features, weather features and event features based on the spatiotemporal tags, meteorological data and event data. The pedestrian flow prediction model construction module is used to construct a prediction model using a hybrid architecture learning algorithm combining LightGBM and LSTM. Specifically: the hybrid feature set is aggregated by unit statistical time and a time index is added to form the current dataset; the input of the LightGBM model is the current dataset, and its output is used as one input of the LSTM model; the other input of the LSTM model is the current dataset itself after dimensionality adjustment; the hybrid model is trained using the spatiotemporal pedestrian flow sample dataset, learning the feature influence weights through the LightGBM model and learning the temporal variation patterns through the LSTM model, resulting in a trained short-term pedestrian flow prediction model based on spatiotemporal data. The pedestrian flow prediction module is used to calculate real-time sample data using the trained short-term pedestrian flow prediction model and output short-term prediction data results of urban spatiotemporal pedestrian flow.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the short-term pedestrian flow prediction processing method based on spatiotemporal data as described in any one of claims 1 to 8.

Images