Artificial intelligence-based big data human traffic analysis and prediction method and system
By applying artificial intelligence technology in the flow analysis, a prediction model with spatial and temporal characteristics is constructed, and the problems of insufficient consideration of data feature extraction and spatial and temporal characteristics in the existing methods are solved, and efficient and accurate prediction of traffic is achieved.
Patent Information
- Application Number
- CN202411575809.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2025-02-25
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing traffic analysis and prediction methods are difficult to accurately extract features from massive data, and fail to fully consider the spatiotemporal characteristics of traffic data, resulting in a general prediction result.
Using a big data analysis method based on artificial intelligence, collect flow data through cameras and mobile phone signals, build regional flow matrix and time series attributes, combine external attributes such as meteorological data and holiday data, use the LSTM model to perform feature extraction and prediction model construction, and The model parameters are optimized by the stochastic gradient descent method.
It realizes efficient feature extraction in massive traffic data, builds a prediction model that takes into account the spatial and temporal characteristics, improves the accuracy and efficiency of prediction, reduces the investment in manpower and material resources, and is promoted and applied to other big data prediction fields.
Smart Images
Figure CN119514778A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of big data analysis, and more specifically relates to a method and system for analyzing and predicting human traffic based on big data using artificial intelligence. Background Art
[0002] The analysis and prediction of pedestrian flow spans a wide range of fields, including urban planning, commercial operations, traffic management, and public safety. With the acceleration of urbanization, population mobility is increasingly impacting urban operations, and obtaining pedestrian flow data has become easier. Using this data to help city managers make effective decisions has become a crucial issue.
[0003] Traditional methods for analyzing and predicting foot traffic flow primarily rely on social surveys and statistical analysis, requiring significant human and material resources and incurring high error rates. In recent years, with the advancement of information technology and artificial intelligence (AI), AI methods such as deep learning and machine learning have been introduced to foot traffic analysis. While these methods can process large amounts of data and achieve high prediction accuracy, they still face numerous challenges.
[0004] First, big data itself is characterized by large volume, multiple dimensions, strong timeliness, and variable quality. Accurately and quickly processing this data is the primary challenge in big data analysis. This requires building an effective data processing model to perform operations such as feature extraction, feature selection, and feature encoding on the data.
[0005] Secondly, pedestrian flow data inherently exhibits specific temporal and spatial characteristics. For example, traffic distribution differs between weekdays and weekends, and between peak and off-peak hours. Incorporating this temporal and spatial distribution information into prediction models presents a challenge.
[0006] Thirdly, most of the existing crowd flow prediction models are built based on global samples, without taking spatial diversity and spatiotemporal dependencies into account, which is likely to lead to overly general prediction results. Summary of the Invention
[0007] The present invention proposes a method for analyzing and predicting pedestrian flow based on big data using artificial intelligence. The main technical problem solved by the present invention is how to accurately and efficiently extract features from massive amounts of pedestrian flow data, construct a prediction model that takes into account the temporal and spatial characteristics of pedestrian flow data, and effectively train and verify the model, so as to achieve the purpose of accurately predicting pedestrian flow and provide scientific and accurate decision-making support for related fields such as urban planning and commercial operations.
[0008] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the method comprises:
[0009] Data collection: collects a large amount of crowd flow data through cameras and mobile phone signals, including people's location information, movement paths, and time information;
[0010] Data feature extraction, including construction of regional traffic matrix, time series attribute division, and external attribute splicing;
[0011] Model construction, building crowd flow analysis and prediction models;
[0012] Model training: training of crowd flow analysis and prediction models based on a large amount of historical data and corresponding crowd flow information;
[0013] Model validation uses a validation set or test set to evaluate the model's performance, including its accuracy and recall rate. The trained model is applied to actual scenarios to analyze and predict pedestrian traffic in real time and provide decision support.
[0014] In one embodiment, the data collection includes:
[0015] Data is collected through cameras and mobile phone signals, and the collected models are divided into training set, validation set and test set; the data set is divided according to the ratio of training set: validation set: test set = 60%: 20%: 20%.
[0016] In one solution, the regional traffic matrix is constructed by dividing the region into 32*32 grid areas and counting the inflow and outflow of each area every half hour. The resulting data format is [2,32,32]. When the time dimension is added, the format of the data set is [numbers_time_slots,2,32,32], which serves as the input of the pedestrian flow analysis and prediction model.
[0017] In one embodiment, the time series attributes are divided; the nearest, near and far key frames are selected to predict the time interval t, which are defined as The specific expressions are as follows:
[0018]
[0019] Where p and q are the spans of the daily and weekly cycles respectively; l c 、l p and l q They are the lengths of the three time attributes, which can be set as three hyperparameters during initialization.
[0020] In one solution, the external attribute splicing includes: a meteorological data set and a holiday data set, wherein the meteorological data set includes four attribute lists of time, temperature, wind speed, and weather, which are represented by time_vector, holiday_vector, and meteorol_vector, and finally the three parts are spliced into a vector meta_vector.
[0021] In one solution, the pedestrian flow analysis and prediction model inputs the construction of the regional flow matrix, the time series attribute division, and the external attribute splicing vector into LSTM as a feature parameter, so that the model combines the spatiotemporal correlation of the pedestrian flow data.
[0022] In one solution, the LSTM layer directly performs time series prediction on the input data. The input data has 32-dimensional features, which include information about the day of the prediction. The algorithm removes the information about the day when it is actually input.
[0023] A two-layer LSTM is used, with the output of the first layer as the input of the second layer. The step size is 3, and two fully connected layers are used to increase the model depth and enhance the model fitting ability. Finally, the dimensionality is reduced to output the predicted flow of people.
[0024] In one embodiment, the model training includes:
[0025] The model training phase includes:
[0026] (1) Use the stochastic gradient descent (SGD) optimization algorithm to find the optimal solution for the model parameters. During each training process, a portion of data is randomly selected and input into the model. Based on the difference between the model's predicted results and the actual results, the error is back-propagated to update the model parameters.
[0027] (2) Define the loss function and design the RMSE loss function as follows:
[0028]
[0029] Among them, y i is the actual traffic value, y' i is the predicted flow value, and n is the number of samples.
[0030] (3) Perform model training, set the number of training batches (Epoch), the amount of data corresponding to each training batch (Batchsize), and the step size of each parameter update (Learningrate); repeatedly iterate the model training until the model converges, that is, the loss function value almost no longer decreases, and the training ends.
[0031] In one embodiment, the model validation includes:
[0032] Use the validation set to validate the model and evaluate the performance of the model on unknown data to prevent overfitting;
[0033] During the training process, an early stopping strategy is set, that is, monitoring a set indicator. If the loss function value does not decrease for N consecutive epochs, the training is stopped to prevent overfitting.
[0034] In another aspect, a pedestrian flow analysis and prediction system based on big data of artificial intelligence is provided, wherein the system is applicable to the method described above, and the system comprises:
[0035] Data collection module: used to collect a large amount of pedestrian flow data through cameras, mobile phone signals, etc., including people's location information, movement path, and time information;
[0036] Data feature extraction module: processes and analyzes the collected pedestrian flow data, including constructing a regional flow matrix, dividing time series attributes, and merging with external attribute data;
[0037] Model building module: builds a crowd flow analysis and prediction model based on the extracted data features, and inputs the extracted feature parameters into the model;
[0038] Model training module: trains the crowd flow analysis and prediction model, and optimizes the model parameters through repeated iterations based on a large amount of historical data and corresponding crowd flow information;
[0039] Model verification module: Use the verification set or test set to evaluate the performance of the model, including the accuracy and recall rate of the model, and provide the verification results to the decision maker according to the needs. Beneficial effects of the invention:
[0040] The beneficial effects of the present invention are mainly reflected in the following aspects: First, through the intelligent analysis and processing of a large amount of pedestrian flow data, it is possible to accurately predict pedestrian flow and further provide decision support, thereby providing great convenience for urban management, commercial operations, traffic planning, etc. Secondly, the method takes into account the spatiotemporal characteristics of pedestrian flow data, making the prediction more accurate and targeted. Furthermore, the method improves the accuracy and efficiency of the prediction by constructing a training model and conducting effective verification, avoiding the large amount of manpower and material resources invested in traditional methods. In addition, the model and training method in this method are highly versatile. In addition to pedestrian flow analysis, it can also be applied to other fields that require big data prediction. In general, this method of pedestrian flow analysis and prediction based on big data of artificial intelligence can assist urban planning, traffic management, commercial operations, etc. to achieve precise management and scientific decision-making, and promote the intelligent development of related industries. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1Flow chart of the method of the present invention;
[0042] Figure 2 This is a model block diagram of the method of the present invention;
[0043] Figure 3 This is a system block diagram of the present invention. DETAILED DESCRIPTION
[0044] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate exemplary embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.
[0045] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.
[0046] Example 1
[0047] like Figure 1 As shown, a method for analyzing and predicting passenger flow based on big data using artificial intelligence includes:
[0048] S1. Data collection: A large amount of crowd flow data is collected through cameras and mobile phone signals, including people's location information, movement paths, and time information.
[0049] To collect data from cameras, visual processing algorithms are required. For example, object detection algorithms can detect people in an image, while object tracking algorithms can track their movements. This allows us to obtain information such as the number of people, their location, and their movement speed.
[0050] One way to collect data through mobile phone signals is to obtain location data through the phone's GPS signal. Alternatively, the phone's location information can be obtained through wireless networks, such as cellular and Wi-Fi networks. This can be used to track a person's movements and where they stay.
[0051] For example, a shopping mall can use infrared sensors installed at its entrances and exits to collect data on the number of people entering and leaving. Its surveillance cameras can also capture data on the number of people and their movement paths within the mall. Furthermore, the mall's Wi-Fi network can be used to obtain mobile phone location information. This allows for more accurate analysis and prediction of mall foot traffic. The collected data is divided into training, validation, and test sets; the dataset is split 60%:20%:20% for training:validation:test.
[0052] S2. Data feature extraction, including the construction of regional traffic matrix, time series attribute division, and external attribute splicing.
[0053] The construction of the regional traffic matrix divides the area into 32*32 grid areas and counts the inflow and outflow of each area every half hour. The resulting data format is [2,32,32]. When the time dimension is added, the format of the data set is [numbers_time_slots,2,32,32], which serves as the input of the pedestrian flow analysis and prediction model.
[0054] The time series attribute division mentioned above; With the input form of each time interval mentioned above, the input data of each LSTM layer will be divided below. The entire grid will not take the entire time series as input. Secondly, different key frames are selected along the time series. The nearest, near and far key frames are selected to predict the time interval t, which are defined as The specific expressions are as follows:
[0055]
[0056] Where p and q are the spans of the daily and weekly cycles respectively; l c 、l p and l q They are the lengths of the three time attributes, which can be set as three hyperparameters during initialization.
[0057] The external attribute splicing includes: meteorological data set and holiday data set, wherein the meteorological data set includes four attribute lists of time, temperature, wind speed, and weather, which are represented by time_vector, holiday_vector, and meteorol_vector. Finally, the three parts are spliced into a vector meta_vector.
[0058] For each time interval, time_vector has eight dimensions, the first seven of which are one-hot encoded values, and the last dimension indicates whether the day is a weekday (1 for a weekday, 0 for a non-workday). The vector holiday_vector has one dimension, indicating whether it is a holiday (0 for a weekday, 1 for a holiday). The vector meteorol_vector has a total of nineteen dimensions. The first seventeen dimensions use one-hot encoding to represent a type of weather, such as sunny, blizzard, rainstorm, sandstorm, etc. The last two dimensions represent wind speed and temperature, respectively. Finally, the three vectors are concatenated into a twenty-eight-dimensional metadata vector meta_vector, which is the external feature.
[0059] S3. Model construction, building a pedestrian flow analysis and prediction model.
[0060] The pedestrian flow analysis and prediction model constructs a regional flow matrix, divides time series attributes, and inputs external attribute splicing vectors into LSTM as a feature parameter, so that the model combines the spatiotemporal correlation of pedestrian flow data.
[0061] The LSTM layer directly performs time series prediction on the input data. The input data has 32-dimensional features, which include information about the day of prediction. The algorithm removes the information about the day of prediction when it is actually input. Assuming x is the input, the input dimension is: x(3,28). Figure 2 As shown in the LSTM model design diagram: the algorithm adopts a two-layer LSTM, with the output of the first layer serving as the input of the second layer. The algorithm adopts a step size of 3 and uses two fully connected layers to increase the model depth and enhance the model fitting ability. Finally, the dimensionality is reduced to output the predicted pedestrian flow.
[0062] Assume that sij represents the weight from neuron i to neuron j in the model. w represents input, a represents output. p and q represent input and output gates, respectively. c represents cells, n represents the number of cells in the input model, g represents the cell output activation function, G represents the number of cells in the LSTM hidden layer, and r represents the cell state of the model cell. δ represents the activation function. The input of the LSTM input gate is then expressed as:
[0063]
[0064] The input to the output gate is represented as:
[0065]
[0066] The output of the output gate is expressed as:
[0067]
[0068] The final prediction result is expressed as:
[0069]
[0070] S4. Model training: training the crowd flow analysis and prediction model based on a large amount of historical data and corresponding crowd flow information.
[0071] The model training phase includes:
[0072] (1) Use the stochastic gradient descent (SGD) optimization algorithm to find the optimal solution for the model parameters. During each training process, a portion of data is randomly selected and input into the model. Based on the difference between the model's predicted results and the actual results, the error is back-propagated to update the model parameters.
[0073] (2) Define the loss function and design the RMSE loss function as follows:
[0074]
[0075] Among them, y i is the actual traffic value, y' i is the predicted flow value, and n is the number of samples.
[0076] (3) Perform model training, set the number of training batches (Epoch), the amount of data corresponding to each training batch (Batchsize), and the step size of each parameter update (Learningrate); repeatedly iterate the model training until the model converges, that is, the loss function value almost no longer decreases, and the training ends.
[0077] LSTM model parameter settings
[0078] The LSTM model divides historical foot traffic into training and validation sets in a ratio of 9 to 1. The model uses two LSTM layers, one for input and one for output. The number of neurons in the first LSTM hidden layer is set to 128, and the number of neurons in the second layer is set to 64. The fully connected parameter of the first layer is 32, and the fully connected parameter of the second layer is set to 1. To ensure optimal training results, the epochs are set between 100 and 1000. The number of samples used for training is 32 to 128.
[0079] S5. Model verification: Use the validation set or test set to evaluate the performance of the model, including the model's accuracy and recall rate; apply the trained model to actual scenarios, analyze and predict pedestrian flow in real time, and provide decision support.
[0080] The model validation includes:
[0081] Use the validation set to validate the model and evaluate the performance of the model on unknown data to prevent overfitting;
[0082] During the training process, an early stopping strategy is set, that is, monitoring a set indicator. If the loss function value does not decrease for N consecutive epochs, the training is stopped to prevent overfitting.
[0083] For Early Stopping:
[0084] After each epoch, the loss function value on the validation set is calculated. If the validation set loss value of the current epoch is smaller than the validation set loss value of all previous epochs, the current model parameters are stored.
[0085] Set an integer N as the early stopping threshold. That is, when the loss value of the validation set does not decrease in N consecutive epochs, stop training.
[0086] Let minValidLoss be the previous minimum validation set loss value. If the validation set loss value validLoss of the current Epoch is greater than minValidLoss, and this situation persists for N Epochs, early stopping is triggered.
[0087] Use the trained model to predict the samples in the test set and get the prediction result y' i .
[0088] Compare the prediction results y' i and the true output y i Compare and calculate the performance indicators of the model.
[0089] For example, the accuracy calculation formula is as follows: Acc=∑(y′ i ==y i ) / n, where n is the total sample size.
[0090] Recall and precision are often used to evaluate binary classification problems. The calculation formulas are as follows: Precision = TP / (TP+FP) and Recall = TP / (TP+FN), where TP is the true positive sample, which represents the number of samples that the model predicts to be positive and are actually positive; FP is the false positive sample, which represents the number of samples that the model predicts to be positive but are actually negative; and FN is the false negative sample, which represents the number of samples that the model predicts to be negative but are actually positive.
[0091] Example 2
[0092] like Figure 3 As shown, a pedestrian flow analysis and prediction system based on big data of artificial intelligence is applicable to the method described above, and the system includes:
[0093] The data collection module collects large amounts of pedestrian flow data, including location, movement paths, and time, through cameras and mobile phone signals. Its core function is to gather data related to pedestrian flow. Data sources include CCTV, mobile phone signals, and Wi-Fi connection records. This data helps us understand pedestrian movement paths, dwell time, and the number of people at each point in time. The acquisition of large amounts of high-precision, real-time pedestrian flow data helps improve the accuracy of pedestrian flow analysis and forecasting.
[0094] The Data Feature Extraction Module processes and analyzes collected pedestrian flow data, including constructing a regional flow matrix, partitioning time series attributes, and merging them with external attribute data. Key features useful for pedestrian flow analysis and prediction, such as time, location, and volume, are extracted. A flow matrix is then constructed to represent pedestrian flow at a given time and location. These feature parameters help improve the accuracy of model predictions.
[0095] Model building module: Build a crowd flow analysis and prediction model based on the extracted data features, and input the extracted feature parameters into the model.
[0096] Model training module: Trains the crowd flow analysis and prediction model, optimizes the model parameters through repeated iterations based on a large amount of historical data and corresponding crowd flow information.
[0097] Model validation module: Use validation sets or test sets to evaluate the performance of the model, including the accuracy and recall rate of the model, and provide the validation results to decision makers as needed.
[0098] Example 3
[0099] Data Collection: We collected pedestrian traffic data for one quarter (approximately three months), including each person's location information (latitude and longitude), movement path, and time. We also collected daily meteorological data, such as temperature, wind speed and direction, humidity, weather conditions (sunny, rainy, snowy, foggy, etc.), and air pressure, as well as holiday data and special event data (such as concerts and art festivals).
[0100] Data feature extraction: The commercial district is divided into 32x32 grid areas, and the number of people entering and leaving each area is counted every half hour. For example, the first grid has 50 people entering and exiting during the half-hour period from 8:00 to 8:30 a.m., with 30 entering and 20 leaving. From 8:30 to 9:00 a.m., the number of people entering and exiting is 60, with 35 entering and 25 leaving, and so on. Furthermore, pedestrian flow data from the last three hours (near distance), the last 12 hours (medium distance), and the last 24 hours (long distance) is used to predict the time interval t. Furthermore, meteorological data, holiday data, and information on special events are used as feature parameters.
[0101] Model construction: Build a multi-layer LSTM model, taking the above-mentioned pedestrian flow, time series, weather information, holidays, and special event information as input features.
[0102] Model training: The first two months of collected quarterly data were used to train the model, and the last month's data was used for validation. Backpropagation training was performed using stochastic gradient descent, so that the model weights minimized the RMSE loss function between the predicted and actual foot traffic.
[0103] Model validation: The model is validated using the data from the last month to evaluate various indicators such as accuracy and recall. If the model performs well on the validation dataset, then the model can be used to predict future pedestrian traffic in commercial districts.
[0104] After implementing the prediction method of this invention, the flow of people in commercial blocks can be predicted more accurately, and then reasonable flow of people can be managed, thereby improving business results and user satisfaction.
[0105] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
[0106] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for analyzing and predicting human traffic based on big data using artificial intelligence, characterized in that: The method includes: Data collection: collects a large amount of human flow data through cameras and mobile phone signals, including people's location information, movement paths, and time information; Data feature extraction, including construction of regional traffic matrix, time series attribute division, and external attribute splicing; Model construction, building a model for analyzing and predicting human traffic; Model training: training the crowd flow analysis and prediction model based on a large amount of historical data and corresponding crowd flow information; Model verification: Use a validation set or a test set to evaluate the performance of the model, including the accuracy and recall of the model; apply the trained model to actual scenarios, analyze and predict pedestrian flow in real time, and provide decision support.
2. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The data collection includes: Data is collected through cameras and mobile phone signals, and the collected models are divided into training set, verification set and test set; the data set is divided according to training set: verification set: test set = 60%: 20%: 20%.
3. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The construction of the regional traffic matrix divides the area into 32*32 grid areas, and counts the inflow and outflow of each area every half an hour. The data format is [2,32,32]. When the time dimension is added, the format of the data set is [numbers-time-slots,2,32,32], which serves as the input of the pedestrian flow analysis and prediction model.
4. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The time series attribute division described above; selecting the nearest, near and far key frames to predict the time interval t is defined as The specific expressions are as follows: Where p and q are the spans of the daily and weekly cycles respectively; l c , l p and l q They are the lengths of the three time attributes, which can be set as three hyperparameters during initialization.
5. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The external attribute splicing includes: a meteorological data set and a holiday data set, wherein the meteorological data set includes four attribute lists of time, temperature, wind speed, and weather, which are represented by time_vector, holiday_vector, and meteorol_vector, and finally the three parts are spliced into a vector meta_vector.
6. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The pedestrian flow analysis and prediction model inputs the construction of the regional flow matrix, the time series attribute division, and the external attribute splicing vector into the LSTM as a feature parameter, so that the model combines the spatiotemporal correlation of the pedestrian flow data at the same time.
7. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 6, characterized in that: The LSTM layer directly performs time series prediction on the input data. The input data has 32-dimensional features, which include information about the day of prediction. The information about the day is removed when the algorithm is actually input. A double-layer LSTM is used, with the output of the first layer as the input of the second layer. The step size is 3, and two fully connected layers are used to increase the model depth and enhance the model fitting ability. Finally, the dimensionality is reduced to output the predicted flow of people.
8. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 6, characterized in that: The model training includes: The model training phase includes: (1) Use the stochastic gradient descent (SGD) optimization algorithm to find the optimal solution for the model parameters. During each training process, a portion of data is randomly selected to input into the model. Based on the gap between the model's predicted results and the actual results, the error is back-propagated to update the model parameters. (2) Define the loss function and design the RMSE loss function as follows: Among them, y i is the actual flow value, y' i is the predicted flow value, n is the number of samples; (3) Perform model training, set the number of training batches (Epoch), the amount of data corresponding to each training batch (Batchsize), and the step size of each parameter update (Learningrate); repeatedly iterate the model training until the model converges, that is, the loss function value almost stops decreasing, and the training ends.
9. The method for analyzing and predicting human traffic based on big data using artificial intelligence according to claim 1, characterized in that: The model validation includes: Use the validation set to validate the model and evaluate the performance of the model on unknown data to prevent overfitting; During the training process, an early stopping strategy is set, that is, a set indicator is monitored. If the loss function value does not decrease for N consecutive Epochs, the training is stopped to prevent overfitting.
10. A big data passenger flow analysis and prediction system based on artificial intelligence, the system being applicable to the method as claimed in any one of claims 1 to 9, characterized in that: The system comprises: Data collection module: used to collect a large amount of human traffic data through cameras, mobile phone signals, etc., including people's location information, movement path, and time information; Data feature extraction module: processes and analyzes the collected traffic data, including building a regional traffic matrix, dividing the time series attributes, and merging with external attribute data; Model building module: build a crowd flow analysis and prediction model based on the extracted data features, and input the extracted feature parameters into the model; Model training module: trains the crowd flow analysis and prediction model, and optimizes the model parameters through repeated iterations based on a large amount of historical data and corresponding crowd flow information; Model validation module: Use validation sets or test sets to evaluate the performance of the model, including the accuracy and recall rate of the model, and provide the validation results to decision makers as needed.