An AI-based multi-modal traffic flow prediction system

Abstract

The application relates to the technical field of traffic flow prediction, in particular to a multi-modal traffic flow prediction system based on AI, which comprises a data collection module, a data preprocessing module, a data analysis and feature extraction module, an AI prediction model module, a real-time event response module, a prediction result analysis module and a traffic control module; wherein the data collection module is used for collecting multi-modal traffic data, the multi-modal traffic data comprising vehicle flow, pedestrian flow, bus flow, traffic signal state and weather conditions; the data preprocessing module receives original data from the data collection module, and carries out cleaning, standardization and integration, generates a data quality report and transmits the data quality report to the real-time event response module. According to the application, the AI-based prediction system effectively improves the urban traffic efficiency, enhances the travel experience of residents and helps to reduce environmental pollution by analyzing the multi-modal traffic data in real time and adaptively adjusting the control strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic flow prediction, and particularly relates to an AI-based multi-modal traffic flow prediction system. BACKGROUND

[0002] With the acceleration of urbanization, urban traffic flow management has become an increasingly significant challenge. Traditional traffic management methods mainly rely on experience and static traffic models, which often cannot respond to sudden traffic incidents and daily traffic fluctuations in a timely manner. Therefore, traffic congestion, travel delays, and environmental pollution have become increasingly severe, causing great inconvenience to urban residents.

[0003] In order to more accurately and efficiently manage urban traffic flow, researchers and engineers seek to use more advanced technical methods. Among them, data-based methods, especially the use of big data and artificial intelligence technology, provide new possibilities for traffic flow prediction and management. By collecting multi-modal traffic data such as vehicle flow, pedestrian flow, bus flow, traffic signal status, and weather conditions, more comprehensive information can be provided for traffic management. However, how to effectively process, analyze, and transform these data into practical traffic control strategies remains a key problem to be solved.

[0004] In addition, the quality and integrity of the data are crucial to the accuracy of traffic prediction. In practical applications, data loss or anomalies may occur due to sensor failure, data transmission interruption, or other reasons. How to effectively identify and handle these data problems to ensure the accuracy and robustness of the prediction model is also a hot topic of current research.

[0005] In summary, developing an AI-based multi-modal traffic flow prediction system that can collect, process, and analyze traffic data in real time and automatically adjust traffic control strategies based on prediction results is an urgent need and research trend for current urban traffic management. SUMMARY

[0006] Based on the above purpose, the present application provides an AI-based multi-modal traffic flow prediction system.

[0007] An AI-based multi-modal traffic flow prediction system includes a data collection module, a data preprocessing module, a data analysis and feature extraction module, an AI prediction model module, a real-time event response module, a prediction result analysis module, and a traffic control module. Wherein,

[0008] The data collection module is used to collect multi-modal traffic data, including vehicle flow, pedestrian flow, bus flow, traffic signal status, and weather conditions.

[0009] Data Preprocessing Module: Receives raw data from the Data Collection Module, cleans, standardizes, and integrates the data, and generates a data quality report, which is passed to the Real-Time Event Response Module;

[0010] Data Analysis and Feature Extraction Module: Receives processed data from the Data Preprocessing Module and performs data analysis to extract key features;

[0011] AI Prediction Model Module: Receives key features from the Data Analysis and Feature Extraction Module and performs traffic flow prediction based on pre-trained AI models;

[0012] Real-Time Event Response Module: Receives data quality reports and key features to identify factors that can affect prediction accuracy in real-time, adjusting the parameters of the AI Prediction Model Module;

[0013] Prediction Result Analysis Module: Receives prediction results from the AI Prediction Model Module and performs result interpretation and optimization;

[0014] Traffic Control Module: Receives prediction results from the Prediction Result Analysis Module and adjusts traffic control measures accordingly.

[0015] Further, the Data Collection Module includes a vehicle flow data collection unit, a pedestrian flow data collection unit, a public transportation flow data collection unit, a traffic signal state data collection unit, and a weather condition data collection unit;

[0016] Vehicle Flow Data Collection Unit: Equipped with vehicle counters to monitor vehicle flow at each intersection and major road in real time, and mark the collected vehicle flow data with timestamps;

[0017] Pedestrian Flow Data Collection Unit: Uses ground and elevated human sensors or CCTV cameras to monitor pedestrian flow on sidewalks and intersections in real time, and associates these data with time and location information;

[0018] Public Transportation Flow Data Collection Unit: Connects with the API interface of the public transportation system or the GPS tracking system to collect the position, speed, direction, and passenger capacity of each bus;

[0019] Traffic Signal State Data Collection Unit: Uses sensors or interfaces with the traffic signal control system to obtain the state of traffic lights in real time;

[0020] Weather Condition Data Collection Unit: Connects with weather stations or online weather services to collect real-time weather information such as temperature, humidity, wind speed, and precipitation.

[0021] Further, the data preprocessing module includes a data cleaning unit, a data standardization unit, a data integration unit, and a data quality report generation unit; wherein,

[0022] The data cleaning unit can identify and delete outliers, duplicate values and irrelevant information in the data, and handle missing or incomplete data entries to ensure the accuracy of subsequent analysis;

[0023] The data standardization unit converts various data into a unified standard format through a preset algorithm, and converts the traffic and passenger flow data into hourly flow to ensure data consistency in subsequent analysis;

[0024] The data integration unit is used to automatically identify and integrate data from different data collection units, and merge them into a structured data set. Specifically, the traffic, passenger flow, bus flow, traffic signal status and weather data are integrated by time and location;

[0025] The data quality report generation unit is used to generate a data quality report based on the original data and processed data, which lists the details of data cleaning, data standardization standards, and data integration structure description.

[0026] Further, the data analysis and feature extraction module specifically includes:

[0027] The time series analysis unit performs time series analysis on the received traffic, passenger flow and bus flow data to identify the periodicity, trend and potential seasonal impact of traffic flow;

[0028] The correlation analysis unit analyzes the relationship between traffic signals, traffic, passenger flow and bus flow;

[0029] The environmental factor analysis unit analyzes the relationship between weather data and traffic flow, specifically the impact of rain or snow on traffic and passenger flow;

[0030] The feature extraction unit automatically extracts key feature vectors, specifically traffic peak time, main traffic bottleneck section, and weather impact index on traffic. The feature extraction unit can also automatically select and optimize features based on analysis results for AI prediction model use.

[0031] Further, the AI prediction model module includes a model training unit, a model optimization unit, and a real-time prediction unit; wherein,

[0032] Model training unit: using historical traffic data as the training set, combined with the extracted key features of traffic peak time, main traffic bottleneck section and the influence index of weather on traffic, the model is trained. The training process is based on a deep learning regression model, and the specific algorithm is as follows:

[0033]

[0034] Wherein, L(θ) is the loss function, representing the average square difference between the predicted value and the true value; N is the number of training samples; y i is the true value of the i-th sample; x i is the input feature of the i-th sample; f is the prediction function, which is associated with the model parameter θ;

[0035] Model optimization unit: the random gradient descent algorithm is used to optimize the model parameters. For the random gradient descent, the specific update rule is as follows:

[0036]

[0037] Wherein, θ t is the model parameter at time t, η is the learning rate, is the gradient of the loss function L at time t;

[0038] Real-time prediction unit: receiving the key features from the data analysis and feature extraction module, inputting the key features into the pre-trained AI model to obtain the real-time prediction result of traffic flow.

[0039] Further, the real-time event response module includes a data quality monitoring unit, a feature deviation identification unit, an influence factor analysis unit and a parameter adjustment recommendation unit; wherein,

[0040] Data quality monitoring unit: for receiving and reading the data quality report from the data preprocessing module, monitoring by setting the threshold that the missing data is not more than 2% and the abnormal data is not more than 1%, to ensure the data integrity and consistency;

[0041] Feature deviation identification unit: comparing the received key features with the historical average feature value, when any key feature deviates from the historical average value by more than 10%, a deviation alarm is triggered;

[0042] Influence factor analysis unit: combining the results of the data quality monitoring unit and the feature deviation identification unit, using the preset logical rule, specifically, when the data missing is more than 2% and the feature deviation is more than 10%, it is judged as high risk;

[0043] Parameter adjustment recommendation unit: when it is identified as high risk, the parameter adjustment recommendation unit will automatically adjust the model weight.

[0044] Further, the parameter adjustment recommendation unit automatically adjusts the formula of the model weight as follows:

[0045]

[0046] where w new is the new weight, w old is the original weight, a is a predefined learning rate, which is set to 0.01, is the gradient of the loss function J with respect to the weight.

[0047] Further, the prediction result analysis module specifically includes:

[0048] The result analysis unit is configured to receive the prediction result from the Al prediction model module and to structure the prediction result, specifically, to comprehensively analyze the quality and accuracy of the prediction result by using explicit index standards including the average error rate, the maximum error, and the error distribution.

[0049] The prediction error calculation unit is configured to calculate the error between the prediction and the actual data, specifically, the formula is as follows:

[0050]

[0051] where y i,pred is the prediction value of the i-th sample, y i,true is the true value of the i-th sample, and N is the number of samples.

[0052] The optimization suggestion unit is configured to generate specific optimization suggestions when the error exceeds a preset threshold based on the result of the prediction error calculation unit, the optimization suggestions including re-adjusting the parameters of the Al model, increasing the amount of training data, or adding feature factors.

[0053] Further, the traffic control module specifically includes:

[0054] The result receiving unit is configured to receive the prediction result, the prediction result including the future traffic volume, the future passenger flow, the future bus flow, and the traffic signal state of each road segment.

[0055] The control strategy generation unit is configured to calculate the traffic pressure value of each road segment based on the received prediction result, specifically, the formula is as follows:

[0056] P i = a · V car,i + b · V ped,i + g · V bus,i ,

[0057] where P i is the traffic pressure value of the i-th road segment, V car,i , V ped,i , and Vbus,i are the traffic volume, passenger volume and bus volume prediction values of the ith road segment respectively, and α, β and γ are the weights of each traffic type, which are preset as fixed values;

[0058] Traffic signal adjustment unit: when the P i When the pressure value exceeds the preset upper limit value of 1000 units, the traffic signal adjustment unit will automatically adjust the traffic signal of the road segment.

[0059] The beneficial effects of the present application are:

[0060] The present application, by introducing an AI-based multi-modal traffic flow prediction system, can comprehensively and real-time collect multi-modal traffic data of the city. This system uses advanced data processing and analysis methods to ensure the quality and integrity of the data. Compared with traditional traffic management methods, this system can more accurately capture the subtle changes and potential trends of traffic flow when predicting traffic flow, thereby greatly improving the accuracy and real-time performance of the prediction.

[0061] The present application can adjust the traffic control strategy in real time according to the prediction results. When it is predicted that traffic pressure in a certain road segment or area may increase, the system can automatically adjust the traffic signal duration or take other control measures to ensure smooth traffic and reduce the possibility of traffic congestion. This adaptive control strategy makes the city traffic management more intelligent and efficient. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the present application, and other drawings can also be obtained by those skilled in the art without creative labor based on these drawings.

[0063] Figure 1 The traffic flow prediction system of the present application is shown in the schematic diagram. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present application more clear and explicit, the following will further describe the present application in combination with specific embodiments.

[0065] It should be noted that unless otherwise defined, technical terms or scientific terms used in the present application shall have the common meaning to those skilled in the art of the present application. The "first", "second" and similar words used in the present application do not represent any order, number or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationship, which may change accordingly when the absolute position of the described object changes.

[0066] As shown in Figure 1 An AI-based multi-modal traffic flow prediction system includes a data collection module, a data preprocessing module, a data analysis and feature extraction module, an AI prediction model module, a real-time event response module, a prediction result analysis module, and a traffic control module. Among them,

[0067] Data collection module: used to collect multi-modal traffic data, including traffic flow, pedestrian flow, bus flow, traffic signal status and weather conditions;

[0068] Data preprocessing module: receives raw data from data collection module, and performs cleaning, standardization and integration, generates a data quality report, and transmits it to real-time event response module;

[0069] Data analysis and feature extraction module: receives processed data from data preprocessing module, and performs data analysis to extract key features;

[0070] AI prediction model module: receives key features from data analysis and feature extraction module, and performs traffic flow prediction based on pre-trained AI model;

[0071] Real-time event response module: receives data quality report and key features to identify factors that can affect prediction accuracy in real time, and adjusts the parameters of AI prediction model module;

[0072] Prediction result analysis module: receives the prediction results of AI prediction model module, and performs result analysis and optimization;

[0073] Traffic control module: receives the prediction results of prediction result analysis module, and adjusts traffic control measures accordingly.

[0074] The data collection module includes a vehicle flow data collection unit, a pedestrian flow data collection unit, a bus flow data collection unit, a traffic signal state data collection unit, and a weather condition data collection unit; wherein,

[0075] The vehicle flow data collection unit is equipped with a vehicle counter for real-time monitoring of vehicle flow at each intersection and major road, and the collected vehicle flow data is marked in the form of a timestamp;

[0076] The pedestrian flow data collection unit uses ground and elevated human body sensors or CCTV cameras to monitor pedestrian flow at crosswalks and intersections in real time, and associates these data with time and location information;

[0077] The bus flow data collection unit connects with the API interface of the bus system or the GPS tracking system to collect the position, speed, direction, and passenger capacity of each bus;

[0078] The traffic signal state data collection unit uses sensors or interfaces with the traffic signal control system to obtain the state of traffic lights (red and green light duration, cycle, etc.) in real time;

[0079] The weather condition data collection unit connects with the meteorological station or online weather service interface to collect real-time weather information such as local temperature, humidity, wind speed, and precipitation.

[0080] The data preprocessing module includes a data cleaning unit, a data standardization unit, a data integration unit, and a data quality report generation unit; wherein,

[0081] The data cleaning unit can identify and delete abnormal values, repeated values, and irrelevant information in the data, and handle missing or incomplete data entries to ensure the accuracy of subsequent analysis;

[0082] The data standardization unit converts various data into a unified standard format through a pre-set algorithm, and converts vehicle flow and pedestrian flow data into hourly flow to ensure data consistency in subsequent analysis;

[0083] The data integration unit is used to automatically identify and integrate data from different data collection units into a structured data set, specifically, the data of vehicle flow, pedestrian flow, bus flow, traffic signal state, and weather condition are integrated by time and location;

[0084] The data quality report generation unit is used to generate a data quality report based on raw data and processed data, which lists the details of data cleaning, the standards of data standardization, and the structure description of data integration.

[0085] The data analysis and feature extraction module specifically includes:

[0086] a time series analysis unit that performs time series analysis on the received traffic volume, passenger flow, and bus flow data to identify periodicity, trend, and potential seasonal effects of traffic flow;

[0087] a correlation analysis unit that analyzes the interrelationships between traffic signal status, traffic volume, passenger flow, and bus flow;

[0088] an environmental factor analysis unit that analyzes the relationship between weather data and traffic flow, specifically the impact of rain or snow on traffic volume and passenger flow;

[0089] a feature extraction unit that automatically extracts key feature vectors, specifically traffic peak time, major traffic bottleneck sections, and weather impact index on traffic, and can also automatically select and optimize features based on analysis results for use by the AI prediction model.

[0090] The AI prediction model module includes a model training unit, a model optimization unit, and a real-time prediction unit; wherein,

[0091] the model training unit uses historical traffic data as a training set, combined with the extracted key features of traffic peak time, major traffic bottleneck sections, and weather impact index on traffic, to train the model, with the training process based on a deep learning regression model whose specific algorithm is:

[0092]

[0093] where L(θ) is the loss function representing the average squared difference between the predicted value and the true value; N is the number of training samples; y i is the true value of the i-th sample; x i is the input feature of the i-th sample; f is the prediction function associated with the model parameter θ;

[0094] the model optimization unit uses the stochastic gradient descent (SGD) algorithm to optimize the model parameters, with the specific update rule for stochastic gradient descent being:

[0095]

[0096] where θ t is the model parameter at time t, η is the learning rate, is the gradient of the loss function L at time t;

[0097] the real-time prediction unit receives the key features from the data analysis and feature extraction module and inputs them into the pre-trained AI model to obtain real-time prediction results of traffic flow.

[0098] The real-time event response module comprises a data quality monitoring unit, a feature deviation identification unit, an influencing factor analysis unit and a parameter adjustment recommendation unit; wherein,

[0099] The data quality monitoring unit is configured to receive and read the data quality report from the data preprocessing module, monitor the data quality by setting thresholds of less than 2% missing data and less than 1% abnormal data, and ensure data integrity and consistency.

[0100] The feature deviation identification unit is configured to compare the received key features with the historical average feature values, and trigger a deviation alarm when any key feature deviates from the historical average value by more than 10%.

[0101] The influencing factor analysis unit is configured to combine the results of the data quality monitoring unit and the feature deviation identification unit, and use a pre-set logical rule, specifically, when the data missing is more than 2% and the feature deviation is more than 10%, it is judged as high risk.

[0102] The parameter adjustment recommendation unit is configured to automatically adjust the model weight when it is identified as high risk.

[0103] The formula for the parameter adjustment recommendation unit to automatically adjust the model weight is:

[0104]

[0105] wherein w new is the new weight, w old is the original weight, a is a pre-defined learning rate set to 0.01, is the gradient of the loss function J with respect to the weight.

[0106] The prediction result analysis module specifically comprises:

[0107] The result analysis unit is configured to receive the prediction results from the A1 prediction model module and structure the prediction results, specifically, use clear index standards including average error rate, maximum error and error distribution to comprehensively analyze the quality and accuracy of the prediction results.

[0108] The prediction error calculation unit is configured to calculate the error between the prediction and the actual data, and the specific formula is:

[0109]

[0110] wherein y i,pred is the prediction value of the i-th sample, y i,true is the true value of the i-th sample, and N is the sample size.

[0111] The optimization suggestion unit generates specific optimization suggestions, including readjusting the parameters of the AI model, increasing the amount of training data or characteristic factors, when the error exceeds the preset threshold based on the results of the prediction error calculation unit.

[0112] The traffic control module specifically includes:

[0113] The result receiving unit receives the prediction results, including the future traffic volume, passenger volume, bus volume and traffic signal state of each road segment;

[0114] The control strategy generation unit calculates the traffic pressure value of each road segment based on the received prediction results, specifically represented by the formula:

[0115] P i i = a · V car,i i + b · V ped,i i + g · V bus,i ,

[0116] where P i i is the traffic pressure value of the i-th road segment, V car,i i, V ped,i i and V bus,i i are the predicted values of the traffic volume, passenger volume and bus volume of the i-th road segment, and a, b and g are the weights of each traffic type, which are preset as fixed values;

[0117] The traffic signal adjustment unit automatically adjusts the traffic signal of a road segment when its P i i exceeds the preset upper limit value of 1000 units of pressure value, for example, extends the green light time or shortens the red light time, to ensure that the traffic can pass smoothly.

[0118] The present application is intended to cover all such alternatives, modifications and variations as fall within the broad scope of the appended claims. Accordingly, any and all such modifications, variations, omissions, etc. are intended to be encompassed by the present application.

Claims

1. An AI-based multi-modal traffic flow prediction system, characterized in that, The system comprises a data collection module, a data preprocessing module, a data analysis and feature extraction module, an AI prediction model module, a real-time event response module, a prediction result analysis module, and a traffic control module. The data collection module is used to collect multi-modal traffic data, including vehicle flow, pedestrian flow, bus flow, traffic signal status, and weather conditions. The data preprocessing module receives raw data from the data collection module, cleans, standardizes, and integrates the data, generates a data quality report, and passes it to the real-time event response module. The data analysis and feature extraction module receives processed data from the data preprocessing module and performs data analysis to extract key features. The AI prediction model module receives key features from the data analysis and feature extraction module and performs traffic flow prediction based on a pre-trained AI model. The real-time event response module receives the data quality report and key features to identify factors that can affect prediction accuracy in real time and adjust the parameters of the AI prediction model module. The prediction result analysis module receives the prediction results from the AI prediction model module and performs result analysis and optimization. The traffic control module receives the prediction results from the prediction result analysis module and adjusts traffic control measures accordingly. The data analysis and feature extraction module specifically includes: The time series analysis unit performs time series analysis on the received vehicle flow, pedestrian flow, and bus flow data to identify the periodicity, trend, and potential seasonal impact of traffic flow. The correlation analysis unit analyzes the relationships between traffic signal status, vehicle flow, pedestrian flow, and bus flow. The environmental factor analysis unit analyzes the relationship between weather conditions and traffic flow, specifically the impact of rain or snow on vehicle flow and pedestrian flow. The feature extraction unit automatically extracts key feature vectors, including peak traffic times, major traffic bottleneck sections, and weather impact indices on traffic. The real-time event response module includes a data quality monitoring unit, a feature deviation identification unit, an influence factor analysis unit, and a parameter adjustment recommendation unit. The data quality monitoring unit receives and reads the data quality report from the data preprocessing module and monitors the data integrity and consistency by setting thresholds for missing data (not more than 2%) and abnormal data (not more than 1%). The feature deviation identification unit compares the received key features with historical average feature values and triggers a deviation alert when any key feature deviates from the historical average by more than 10%. The influence factor analysis unit uses pre-set logical rules to analyze the results from the data quality monitoring unit and the feature deviation identification unit. The parameter adjustment recommendation unit automatically adjusts the model weights when a high-risk situation is identified. The traffic control module specifically includes: A result receiving unit receives a prediction result, the prediction result including future traffic volume, people flow, bus flow and traffic signal state of each road segment; A control strategy generating unit calculates a traffic pressure value of each road segment based on the received prediction result, and a specific formula is as follows: ， wherein, is the traffic pressure value of the th road segment, , and are the vehicle flow, the pedestrian flow and the bus flow prediction values of the th road segment, respectively, and , and are the weights of each flow type, which are preset as fixed values; Traffic signal adjusting unit: when the pressure value of a certain road segment exceeds the preset upper limit value 1000 units, the traffic signal adjusting unit will automatically adjust the traffic signal of the road segment. Traffic signal adjusting unit: when the pressure value of a certain road segment exceeds the preset upper limit value 1000 units, the traffic signal adjusting unit will automatically adjust the traffic signal of the road segment.

2. The AI-based multi-modal traffic flow prediction system of claim 1, wherein, The data collection module includes a traffic volume data collection unit, a people flow data collection unit, a bus flow data collection unit, a traffic signal state data collection unit and a weather condition data collection unit, wherein The traffic volume data collection unit is equipped with a vehicle counter for monitoring vehicle flow of each intersection and main road in real time, and the collected traffic volume data is marked in the form of a time stamp; The people flow data collection unit uses ground and elevated human body sensors or CCTV cameras to monitor people flow of sidewalks and intersections in real time, and associates these data with time and location information; The bus flow data collection unit connects with an API interface of a bus system or a GPS tracking system to collect position, speed, direction and passenger capacity of each bus; The traffic signal state data collection unit uses sensors or interfaces with a traffic signal control system to obtain the state of traffic lights in real time; The weather condition data collection unit connects with a weather station or an online weather service interface to collect weather information such as local temperature, humidity, wind speed and precipitation in real time.

3. The AI-based multi-modal traffic flow prediction system of claim 2, wherein, The data preprocessing module includes a data cleaning unit, a data standardization unit, a data integration unit and a data quality report generating unit, wherein The data cleaning unit can identify and delete abnormal values, repeated values and irrelevant information in the data, and process missing or incomplete data entries to ensure the accuracy of subsequent analysis; The data standardization unit converts various data into a unified standard format through a preset algorithm, and converts the data of vehicle flow and people flow into hourly flow to ensure consistency of the data in subsequent analysis; The data integration unit is used to automatically identify and integrate data from different data collection units into a structured data set, and specifically integrates the data of vehicle flow, people flow, bus flow, traffic signal state and weather condition according to time and location; The data quality report generating unit is used to generate a data quality report based on original data and processed data, which lists details of data cleaning, standards of data standardization and structure description of data integration.

4. The AI-based multi-modal traffic flow prediction system of claim 1, wherein, The AI prediction model module includes a model training unit, a model optimization unit and a real-time prediction unit, wherein The model training unit uses historical traffic data as a training set, combines extracted key features such as traffic peak time, main traffic bottleneck road segment and weather influence index on traffic, and trains the model, and the training process is based on a deep learning regression model, and a specific algorithm is as follows: ， wherein, is a loss function representing the mean squared error between predicted and true values; is the number of training samples; is the true value of the th sample; is the input feature of the th sample; is a prediction function associated with model parameters . The model optimization unit optimizes model parameters using a stochastic gradient descent algorithm, and a specific update rule for the stochastic gradient descent is as follows: ， wherein, is a model parameter at time , is a learning rate, is a loss function is a gradient at time ; Real-time prediction unit: receives key features from the data analysis and feature extraction module, inputs the key features into the pre-trained AI model, and obtains real-time prediction results of traffic flow.

5. The AI-based multi-modal traffic flow prediction system of claim 1, wherein, The formula for automatically adjusting the model weight by the parameter adjustment recommendation unit is: ， where, is the new weight, is the original weight, is a predefined learning rate set to 0.01, is the gradient of the loss function J with respect to the weight.

6. The AI-based multi-modal traffic flow prediction system of claim 5, wherein, The prediction result analysis module specifically includes: Result analysis unit: used to receive prediction results from the A1 prediction model module and to structure the prediction results, specifically, the quality and accuracy of the prediction results are comprehensively analyzed by using explicit index standards including average error rate, maximum error, and error distribution; Prediction error calculation unit: calculates the error between the prediction and the actual data, and the specific formula is: error 7. wherein, is the predicted value of the th sample, is the true value of the th sample, is the number of samples; Optimization suggestion unit: based on the results of the prediction error calculation unit, when the error exceeds the preset threshold, the optimization suggestion unit will generate specific optimization suggestions, including re-adjusting the parameters of the AI model, increasing the amount of training data or feature factors.

Images