Intelligent traffic flow statistics and prediction platform based on multi-source data fusion

Abstract

The invention discloses an intelligent traffic flow statistics and prediction platform based on multi-source data fusion, and relates to the technical field related to traffic prediction, and the platform comprises a data collection layer which is used for collecting road traffic flow data, meteorological data, public traffic operation data and data including traffic condition description on a social media platform; a data preprocessing layer; a data fusion layer; a traffic flow calculation module; and the traffic flow prediction module constructs a prediction model framework based on the space-time convolutional neural network, optimizes hyper-parameters of the ST-CNN, and applies the optimized model to traffic flow prediction. According to the invention, data collected by the annular induction coil and the camera focuses on actual traffic conditions of specific roads and intersections at a microscopic level, and meteorological data, public traffic operation data and social media data reflect traffic influence factors of the whole city or the region from a more macroscopic perspective, so that macroscopic and microscopic aspects are combined, and the traffic influence factors of the whole city or the region are reflected. And the operation condition of the urban traffic system can be known more comprehensively.

Description

Technical Field

[0001] The present invention relates to the technical field related to traffic prediction, and in particular to an intelligent traffic flow statistics and prediction platform integrating multi-source data. Background Art

[0002] With the acceleration of urbanization and the rapid growth of the number of motor vehicles, traffic congestion has become increasingly serious, which has had a negative impact on people's travel efficiency and quality of life. In order to effectively deal with traffic congestion and improve the operating efficiency of urban transportation systems, it is necessary to accurately count and efficiently predict traffic flow so that traffic management departments can take corresponding regulatory measures in a timely manner.

[0003] Traditional traffic flow statistics methods mainly rely on data from a single source, such as data collected by fixed detection equipment such as ground induction coils and video surveillance. However, these single data sources have many limitations. For example, ground induction coils can only obtain vehicle passing information on a specific section of the road, and do not fully reflect the traffic flow distribution in the entire area; although video surveillance can cover a large area, the data processing is complex and is easily affected by environmental factors, such as weather and light changes, which will cause image recognition accuracy to decrease. In addition, a single data source is difficult to cope with complex and changeable traffic scenarios, and cannot provide high-precision traffic flow statistics, which in turn affects the accuracy of traffic forecasts.

[0004] In terms of traffic flow prediction, most of the early prediction methods were based on historical data and simple mathematical models, such as time series analysis. These methods did not fully consider the real-time changes in traffic conditions and the complex relationship between various factors affecting traffic flow, resulting in large deviations between the prediction results and the actual traffic conditions. With the development of technology, some machine learning algorithms have begun to be applied to traffic flow prediction, but due to the lack of full utilization of multi-source data and effective algorithm optimization, its prediction accuracy still needs to be improved.

[0005] Therefore, it is necessary to propose an intelligent traffic flow statistics and prediction platform with multi-source data fusion to solve the above problems. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides an intelligent traffic flow statistics and prediction platform with multi-source data fusion, which solves the problems that the existing single data source can only obtain vehicle passing information on a specific road section, and does not fully reflect the traffic flow distribution in the entire area, and the prediction method does not fully consider the real-time changes in traffic conditions and the complex relationship between various factors affecting traffic flow, resulting in the prediction results often having a large deviation from the actual traffic conditions.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] An intelligent traffic flow statistics and prediction platform for multi-source data fusion, the platform includes:

[0009] Data acquisition layer: Collect road traffic flow data, meteorological data, public transportation operation data, and data containing traffic condition descriptions on social media platforms through various types of sensors and data interfaces, and transmit the data to the platform server through the network;

[0010] Data preprocessing layer: The platform server performs cleaning, denoising, format unification, and standardization processing on the collected raw data;

[0011] Data fusion layer: Use the biogeography-based optimization algorithm to fuse the preprocessed data. Consider each data source as a species, construct a fitness function based on the characteristics of the data, and find a data fusion method that can effectively integrate multi-source traffic data through the migration (information sharing) and mutation (exploring new data combinations) operations of the species;

[0012] Traffic flow calculation module: Based on the result of multi-source data fusion, use the weighted average method to calculate the traffic flow at the current moment;

[0013] Traffic flow prediction module: Construct a prediction model framework based on the spatio-temporal convolutional neural network (ST-CNN). Perform convolution and pooling operations on the input data after multi-source data fusion to extract high-level features. Use the biogeography-based optimization algorithm to optimize the hyperparameters of the ST-CNN, and apply the optimized model to the prediction of traffic flow, outputting the traffic flow prediction values of different road sections within the next hour;

[0014] Visualization display platform: Display the fused traffic flow data and traffic flow prediction values in the form of charts, presenting the change trend and distribution of traffic flow.

[0015] Optionally, the traffic flow data is collected through loop induction coils, cameras, and electronic tag readers installed on urban roads and intersections, the meteorological data is collected through weather stations and meteorological satellites, the public transportation operation data includes bus GPS positioning data and subway card swiping data, and the social media traffic-related data includes text, images, and video information containing traffic condition descriptions captured from social media platforms.

[0016] Optionally, the calculation method of the biogeography-based optimization algorithm module is as follows:

[0017] Habitat Initialization: Different types of data sources are used as habitats. Multiple data features included in each habitat are regarded as species. A habitat includes traffic flow data from induction coils, vehicle speed data from cameras, and travel data from mobile terminals. Randomly initialize the species richness (i.e., fitness) of each habitat, which is determined by calculating the deviation between the initial data fusion result and the actual traffic flow.

[0018] Fitness Function Calculation: Design a fitness function to evaluate the quality of each habitat (data fusion scheme). The factors of the fitness function include data accuracy, diversity, correlation, and reliability.

[0019] Migration Operation: Calculate the immigration rate and emigration rate based on the species richness of the habitat, and calculate the migration probability according to the immigration rate and emigration rate. For each habitat, if the migration probability is greater than a randomly generated number between [0,1], then perform the migration operation.

[0020] Mutation Operation: Determine the probability according to specific rules and perform mutation operations on the migrated habitat.

[0021] Iterative Update: Repeat the above steps until the maximum number of iterations is reached or the fitness reaches the preset convergence threshold.

[0022] Optionally, in the traffic flow calculation module, the way of calculating the traffic flow of different road segments at the current moment by the weighted average method is as follows:

[0023] For road segment L, assume that the fused data includes road traffic flow data F1, meteorological data F2, public transportation operation data F3, and social media data F4, and the corresponding weights are w1, w2, w3, and w4 respectively. Then the traffic flow Q of road segment L is calculated as: Q = w1F1 + w2F2 + w3F3 + w4F4. At the same time, the calculation result is corrected by combining the traffic flow data of the same historical period.

[0024] Optionally, the construction method of the prediction model of the spatio-temporal convolutional neural network (ST-CNN) in the traffic flow prediction module is as follows:

[0025] Data Preparation and Preprocessing: The multi-source data fusion input takes the dataset after data cleaning, preprocessing, and multi-source data fusion as the input of the model. Assume that the fused dataset is X, and its dimension is [n, m, k], where n represents the number of time steps, m represents the spatial position, and k represents the feature dimension at each position. Data standardization is performed on the input data.

[0026] Construction of Spatiotemporal Convolutional Neural Network (ST-CNN): The input layer receives the fused input X' of multi-source data after standardization, and its shape is [n, m, k]. Convolution kernels are designed in the spatiotemporal dimension to extract temporal and spatial features simultaneously. Let the convolution kernel size be [h, w, d], where h represents the convolution kernel size in the temporal dimension, w represents the convolution kernel size in the spatial dimension, and d represents the number of channels between the input and output feature maps. For each position (i, j, l) in the input feature map X', where i represents the time step, j represents the spatial position, and l represents the channel, a new feature map is generated through convolution operations. Pooling operations are used to reduce the dimensionality of the data, reduce the computational amount, and retain important features at the same time. Max pooling with a size of [a, b] is used for downsampling in the temporal and spatial dimensions. The fully connected layer unfolds the pooled feature map into a one-dimensional vector and performs a linear transformation through the fully connected layer to map it to the dimension of the predicted output. According to the specific prediction task, the activation function and structure of the output layer are designed.

[0027] Optionally, the way the biogeography-based optimization algorithm in the traffic flow prediction module optimizes the hyperparameters of ST-CNN is as follows:

[0028] S1. Initialize the population: Randomly generate a group of hyperparameter individuals of ST-CNN, and each individual contains a combination of hyperparameters such as convolution kernel size, learning rate, and number of network layers.

[0029] S2. Fitness evaluation: For each individual, use its corresponding hyperparameter combination to construct an ST-CNN model and train and evaluate it on the validation set.

[0030] S3. Migration operation: Define the migration rate P i , and the calculation formula of the migration rate is designed as: For each individual θ i , with a fixed probability P i select other individuals θ i for migration operations, and replace the h i value of individual θ i with the h j value of individual θ j ;

[0031] S4. Iterative update: Repeat steps 2 and 3 until the preset number of iterations is reached. During each iteration, record the optimal hyperparameter combination and its corresponding fitness value.

[0032] S5. Selection of optimal hyperparameters: After multiple iterations, select the hyperparameter combination that makes the ST-CNN model achieve the highest accuracy on the validation set as the optimal hyperparameters of ST-CNN.

[0033] Optionally, the model training and prediction methods in the traffic flow prediction module are as follows:

[0034] For model training, the ST-CNN model is reconstructed using the optimized hyperparameter combination and trained on the training set. The gradient descent algorithm is adopted to minimize the loss function. During the training process, the weights and bias terms of the model are continuously updated.

[0035] For traffic flow prediction, the trained ST-CNN model is applied to the test set. By inputting the new multi-source data fusion dataset, the model output is the predicted traffic flow values for different road segments within the next hour.

[0036] Optionally, the platform further includes a real-time monitoring and dynamic adjustment module, which monitors various indicators of the traffic system in real time, including the road congestion index, average vehicle speed, and traffic accident incidence rate. When a sudden change in traffic conditions is detected, a dynamic adjustment mechanism is triggered. This mechanism re-evaluates the importance of each data source according to the current traffic scenario and change trend, and adjusts the data fusion strategy and prediction model parameters.

[0037] (III) Beneficial effects

[0038] The present invention provides an intelligent traffic flow statistics and prediction platform for multi-source data fusion, which has the following

[0039] beneficial effects:

[0040] 1. The multi-source data acquisition method of the present invention comprehensively obtains various information affecting traffic flow from different dimensions, providing a solid foundation for subsequent comprehensive analysis. The data collected by the loop inductive coils and cameras focuses on the actual traffic conditions of specific roads and intersections at the micro level, while the meteorological data, public transportation operation data, and social media data reflect the traffic influencing factors of the whole city or region from a more macroscopic perspective. Combining the macroscopic and microscopic aspects helps to more comprehensively understand the operation status of the urban traffic system.

[0041] 2. The biogeography optimization algorithm of the present invention can dynamically adjust the weights of different data sources by simulating the species migration and mutation mechanisms in biogeography. This mechanism enables the algorithm to automatically find the optimal data fusion scheme according to the characteristics and real-time conditions of the data, thereby improving the flexibility and adaptability of data fusion. By constructing a fitness function, the algorithm can quantify the accuracy and contribution degree of each data source to traffic flow prediction, which helps the system to more accurately identify which data sources have the greatest impact on the prediction results during specific periods or conditions, and thus make more effective use of these data to improve the overall prediction accuracy.

[0042] 3. The ST-CNN of the present invention can simultaneously consider the temporal and spatial dimensional information of traffic flow data, automatically mine the inherent features and laws therein. The ST-CNN can adaptively adjust the model parameters through the learning and training of a large amount of historical data to fit various complex traffic patterns, thereby improving the accuracy and robustness of prediction.

[0043] 4. The platform of the present invention can real-time monitor various indicators of the traffic system such as road congestion index, average vehicle speed, and traffic accident incidence rate. When a sudden situation such as a traffic accident occurs, it can quickly capture the abnormal changes in relevant indicators. Once a sudden change in traffic conditions is detected, it re-evaluates the importance of each data source, thereby providing a basis for subsequent accurate traffic flow prediction and reasonable traffic control. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a schematic diagram of the modules of the intelligent traffic flow statistics and prediction platform of the present invention;

[0045] Figure 2 is a schematic diagram of the operation process of the intelligent traffic flow statistics and prediction platform of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] Embodiment 1

[0048] Please refer to Figure 1 , an intelligent traffic flow statistics and prediction platform for multi-source data fusion, the platform includes:

[0049] Data acquisition layer: Collect road traffic flow data, meteorological data, public transportation operation data, and data on social media platforms containing traffic condition descriptions (such as road congestion conditions, traffic accident scenarios) through various types of sensors and data interfaces. The traffic flow data is collected by loop induction coils, cameras, and electronic tag readers installed on urban roads and intersections. The meteorological data is collected by weather stations and meteorological satellites. The public transportation operation data includes bus GPS positioning data and subway card swiping data. The social media traffic-related data includes text, images, and video information containing traffic condition descriptions (such as road congestion conditions, traffic accident scenarios) captured from social media platforms, and the data is transmitted to the platform server through the network;

[0050] Data Preprocessing Layer: The platform server cleans, denoises, unifies the format, and standardizes the collected raw data to eliminate outliers and noise interference in the data, making data from different sources comparable and consistent. It performs filtering on road traffic flow data to remove abnormally fluctuating data points, interpolation and missing value filling operations on meteorological data, and text tokenization and sentiment analysis preprocessing on social media data;

[0051] Data Fusion Layer: The biogeography-based optimization algorithm is used to fuse the preprocessed data. Each data source is regarded as a species, and a fitness function is constructed based on the characteristics of the data. Through the migration (information sharing) and mutation (exploration of new data combinations) operations of the species, a data fusion method that can effectively integrate multi-source traffic data is sought. For example, during the morning rush hour, if the road traffic flow data has a high correlation with public transportation operation data and the meteorological data is relatively stable, then the algorithm increases the weights of the road traffic flow data and public transportation operation data while reducing the weight of the meteorological data;

[0052] Traffic Flow Calculation Module: Based on the result of multi-source data fusion, the weighted average method is used to calculate the traffic flow at the current moment;

[0053] Traffic Flow Prediction Module: A prediction model framework based on spatio-temporal convolutional neural network (ST-CNN) is constructed. ST-CNN can automatically learn the spatio-temporal characteristics of traffic flow data, capture the correlations between different times and different road segments, perform convolution and pooling operations on the input data after multi-source data fusion to extract high-level features, and use the biogeography-based optimization algorithm to optimize the hyperparameters of ST-CNN (such as the size of the convolutional kernel). During the optimization process, by evaluating the performance of the model on the validation set (such as the mean absolute error metric), through a specific hyperparameter adjustment strategy, a parameter combination that can effectively improve the model performance is found. Finally, the model is trained and predicted, and the optimized model is applied to the prediction of traffic flow, outputting the traffic flow prediction values for different road segments within the next hour;

[0054] Visualization Display Platform: The fused traffic flow data and traffic flow prediction values are displayed in the form of charts (such as line charts, heat maps) to present the change trends and distribution of traffic flow.

[0055] In this embodiment, by installing loop inductive coils and camera sensors on main roads and intersections in the city to collect road traffic flow data, the real-time traffic conditions of vehicles on the road can be directly reflected, including detailed information such as the volume of traffic flow, vehicle speed, vehicle type classification, and traffic flow direction. At the same time, meteorological data such as temperature, humidity, rainfall, snowfall, and visibility are obtained from the meteorological department. These meteorological factors have a significant impact on traffic flow. For example, bad weather will reduce the driving speed and increase the accident rate, thus affecting traffic flow. Public transportation operation data is obtained from bus companies and subway operation units, covering the passenger volume, departure frequency, and on-time rate of bus lines, as well as the number of passengers on subway lines and the train operation interval information, comprehensively presenting the operation status of urban public transportation. The web crawler technology is used to capture traffic-related data from social media platforms, such as traffic congestion sharing, accident reports, and road construction information, providing supplementary information on traffic conditions from the public perspective. This multi-source data collection method comprehensively obtains various information affecting traffic flow from different dimensions, providing a solid foundation for subsequent comprehensive analysis; the data collected by loop inductive coils and cameras focuses on the actual traffic conditions of specific roads and intersections at the micro level, while meteorological data, public transportation operation data, and social media data reflect the traffic impact factors of the whole city or region from a more macroscopic perspective. Combining the macroscopic and microscopic aspects helps to understand the operation status of the urban traffic system more comprehensively;

[0056] The sensors collect road traffic flow data in real time, and can capture the changes in the number of vehicles and traffic speed on the road in the first time, such as the sharp increase in traffic flow during peak hours in the morning and evening, and the sudden decrease in traffic flow caused by emergencies. The public transportation operation data can also reflect the operation status of buses and subways in real time, such as the departure delay of a certain bus line due to vehicle failure, or the flow-limiting measures taken at a certain subway station due to excessive passenger flow. The real-time weather changes in meteorological data, such as sudden heavy rain or fog, will have an immediate impact on traffic flow. The real-time information on social media platforms can enable the platform to obtain emergencies such as sudden road congestion and temporary traffic control in a timely manner. This real-time data collection enables the platform to dynamically track the evolution of traffic conditions, providing the possibility to detect traffic problems in a timely manner and adjust traffic management strategies;

[0057] Traffic flow itself is highly dynamic and is affected by various factors such as time (e.g., weekdays and weekends, day and night), location (commercial areas and residential areas, city centers and suburbs), special events (large-scale events, festivals), etc. Multi-source data collection can adapt to this dynamic change. For example, when a large-scale sports event is held, the traffic volume on the roads around the venue will increase significantly, and the passenger volume of public transportation will also rise significantly. At the same time, there will be a large number of discussions and navigation information about event travel on social media. At this time, through the dynamic collection of each data source, the platform can timely adjust the monitoring and prediction focus of the traffic flow in this area to meet the traffic needs during special periods;

[0058] The data collected from different data sources can be mutually verified. For example, when the traffic volume data collected by the loop induction coil is abnormal, it can be visually verified through the images captured by the camera to see if the data deviation is caused by equipment failure or special circumstances. There is also a correlation between meteorological data and road traffic volume data. For example, in rainy or snowy weather, if the road traffic volume data shows that the vehicle speed generally decreases but the traffic flow does not decrease significantly, it may be because drivers are more cautious in bad weather. In this case, the meteorological data provides a basis for explaining the change in traffic volume data. The public transportation operation data can also be mutually verified with the road traffic volume data. If the passenger volume of a certain bus line increases, it may be reflected in the road traffic volume data of the corresponding section, and vice versa. This data complementarity helps to improve the accuracy and reliability of the data and reduce the errors that may exist in a single data source;

[0059] Single-type data often can only reflect one aspect of the traffic system, while the collection of multi-source data integrates data from different aspects, providing a richer dimension for in-depth analysis of traffic flow. For example, by combining meteorological data and road traffic volume data, the change rules of traffic flow under different weather conditions can be analyzed, providing a basis for formulating traffic plans to cope with bad weather; by combining public transportation operation data and road traffic volume data, the mutual relationship between public transportation and private transportation can be studied, and the attractiveness and sharing rate of public transportation can be evaluated, so as to optimize the allocation of public transportation resources and route planning.

[0060] Embodiment 2

[0061] This embodiment makes the following optimizations on the basis of Embodiment 1. Specifically, the calculation method of the biogeography optimization algorithm module is as follows:

[0062] Habitat Initialization: Different types of data sources are used as habitats. Multiple data features contained in each habitat are regarded as species. A habitat includes traffic flow data from induction coils, vehicle speed data from cameras, and travel data from mobile terminals. The species richness (i.e., fitness) of each habitat is randomly initialized and determined by calculating the deviation between the initial data fusion result and the actual traffic flow.

[0063] Fitness Function Calculation: A fitness function is designed to evaluate the quality of each habitat (data fusion scheme). The factors of the fitness function include data accuracy, diversity, correlation, and reliability.

[0064] Data Accuracy: The mean square error (MSE) is used to measure the difference between the fused data and the actual traffic flow data. Assume the fused data is the actual data is y, and the size of the dataset is n. Then the MSE calculation formula is:

[0065] Data Diversity: It measures the richness of different types of data in the fused data and represents diversity by calculating the entropy of data features. For data feature x, its entropy H(x) calculation formula is:

[0066] Data Correlation: It measures the correlation by calculating the correlation coefficient between different data sources. For two data features x and y, the correlation coefficient r xy calculation formula is:

[0067]

[0068] Data Reliability: Considering the credibility of data sources, the data of fixed detection devices may be relatively more reliable. Different weights are assigned according to experience. Considering the above factors, the fitness function F is expressed as:

[0069] F = w1(1 - MSE) + w2×H + w3×r + w4×R;

[0070] where w1, w2, w3, and w4 are the weights of each factor, and w1 + w2 + w3 + w4 = 1;

[0071] Migration Operation: Calculate the immigration rate λ i and the emigration rate μ i for the habitat i. The immigration rate The emigration rate where I and E are the maximum rates of immigration and emigration, k i is the species richness of habitat i, and K is the maximum species richness; Calculate the migration probability P i = min(λi , 1-μ i ); Migration operation execution: For each habitat, if the migration probability is greater than a randomly generated number between [0, 1], the migration operation is performed. The species to be migrated move from one habitat to another and mutate in the new habitat;

[0072] Mutation operation: Determine the probability according to specific rules and perform mutation operations on the migrated habitats. Mutation randomly modifies the values of data features or changes the connection methods between data features;

[0073] Iterative update: Repeat the above steps until the maximum number of iterations is reached or the fitness reaches the preset convergence threshold.

[0074] In this embodiment, the biogeography optimization algorithm can dynamically adjust the weights of different data sources by simulating the species migration and mutation mechanisms in biogeography. This mechanism enables the algorithm to automatically find the optimal data fusion solution according to the characteristics and real-time conditions of the data, thereby improving the flexibility and adaptability of data fusion;

[0075] By constructing a fitness function, the algorithm can quantify the accuracy and contribution degree of each data source to traffic flow prediction. This helps the system more accurately identify which data sources have the greatest impact on the prediction results during specific periods or conditions, so as to utilize these data more effectively and improve the overall prediction accuracy. The mutation operation in the biogeography optimization algorithm helps to explore new data combination methods, which can not only increase the diversity of the model but also improve the model's resistance to abnormal data. When some data sources show abnormalities or noise, the algorithm can reduce their negative impact on the prediction results by adjusting the weights or trying new data combinations;

[0076] Since the algorithm can dynamically adjust the fusion strategy according to real-time data, it can adapt to changes in traffic conditions faster. For example, during the morning rush hour, if the road traffic flow data and public transportation operation data are highly correlated while the meteorological data is relatively stable, the algorithm can quickly increase the weights of the former two and reduce the dependence on meteorological data, so as to more accurately reflect the current traffic conditions. The biogeography optimization algorithm finds the optimal solution through parallel search and local optimization. Compared with traditional exhaustive methods or other optimization methods, it can find a satisfactory solution in a shorter time, thereby improving the operating efficiency of the entire system; This algorithm framework has good scalability and can easily add new data sources or adjust the processing methods of existing data sources. At the same time, due to its rule-based optimization mechanism, the algorithm has high maintainability and interpretability, which is convenient for subsequent technical improvement and upgrading.

[0077] Example 3

[0078] This embodiment makes the following optimizations based on Example 1 or Example 2. Specifically, in the traffic flow calculation module, the weighted average method calculates the traffic flow of different road segments at the current moment as follows:

[0079] For road segment L, assuming that the fused data includes road traffic flow data F1, meteorological data F2, public transportation operation data F3, and social media data F4, with corresponding weights w1, w2, w3, and w4 respectively, the traffic flow Q of road segment L is calculated as: Q = w1F1 + w2F2 + w3F3 + w4F4. At the same time, the calculation result is corrected by combining the traffic flow data of the same historical period to obtain a more accurate current traffic flow value;

[0080] In the traffic flow prediction module, the prediction model of the spatio-temporal convolutional neural network (ST-CNN) is constructed as follows: Data preparation and preprocessing: The multi-source data fusion input takes the dataset after data cleaning, preprocessing, and multi-source data fusion as the input of the model. Let the fused dataset be X, with a dimension of [n, m, k], where n represents the number of time steps (such as data collected at different time points), m represents the spatial position (i.e., the number of different road segments or regions), and k represents the feature dimension at each position (such as multi-source data features of traffic flow, vehicle speed, and meteorological information). Data standardization: The input data is standardized to facilitate model training and convergence. The Min-Max standardization method is used. For each element x in the dataset X ijk , the standardization formula is: where min(x :ijk ) and max(x :ijk ) represent the minimum and maximum values over all time steps and spatial positions respectively;

[0081] Construction of Spatio-Temporal Convolutional Neural Network (ST-CNN): The input layer receives the normalized multi-source data fusion input X', whose shape is [n, m, k]; design convolutional kernels in the spatio-temporal dimension to extract temporal and spatial features simultaneously. Let the convolutional kernel size be [h, w, d], where h represents the convolutional kernel size in the temporal dimension, w represents the convolutional kernel size in the spatial dimension, and d represents the number of channels between the input and output feature maps; for each position (i, j, l) in the input feature map X', where i represents the time step, j represents the spatial position, and l represents the channel, generate a new feature map through convolutional operations; use pooling operations to reduce the dimensionality of the data, reduce the computational amount, and retain important features at the same time. Use max pooling with a size of [a, b] for downsampling in the temporal and spatial dimensions; the fully connected layer unfolds the pooled feature map into a one-dimensional vector and performs a linear transformation through the fully connected layer to map it to the dimension of the predicted output. According to the specific prediction task (such as predicting the traffic flow in multiple future time steps), design the activation function and structure of the output layer. For example, if it is a regression prediction task, no activation function is used and the predicted value is directly output; if it is a classification task, the Softmax activation function is used to output the probability distribution;

[0082] The biogeography optimization algorithm in the traffic flow prediction module optimizes the hyperparameters of ST-CNN as follows:

[0083] S1. Initialize the population: Randomly generate a group of ST-CNN hyperparameter individuals, each individual containing a combination of hyperparameters such as convolutional kernel size, learning rate, and number of network layers. Let the population size be N, and the hyperparameter vector of each individual be θ i ;

[0084] S2. Fitness evaluation: For each individual, use its corresponding hyperparameter combination to construct an ST-CNN model, and train and evaluate it on the validation set. Select the mean squared error (MSE) metric as the fitness function. The fitness function f(θ i ) is defined as: where M represents the number of samples in the validation set, y t represents the true value, represents the model predicted value;

[0085] S3. Migration operation: According to biogeography theory, species will migrate from places with high habitat suitability to places with low suitability. In the BGO algorithm, define the migration rate P i , which is related to the fitness value of the individual. The higher the fitness, the lower the migration rate; the lower the fitness, the higher the migration rate. The calculation formula for the migration rate is designed as: For each individual θ i , with a fixed probability P i select other individuals θi A migration operation is performed, where the migration operation is to exchange some hyperparameter values or mutate the hyperparameters. For example, a hyperparameter dimension (such as the convolutional kernel size h) is randomly selected, and the h i value of the individual θ i is replaced with the h j value of the individual θ j value, or h i is mutated within a certain range: h' j = h i + Δh, where Δh represents a random mutation amount;

[0086] S4. Iterative update: Repeat steps 2 and 3 until the preset number of iterations is reached. During each iteration, record the optimal hyperparameter combination and its corresponding fitness value;

[0087] S5. Optimal hyperparameter selection: After multiple iterations, select the hyperparameter combination that makes the ST-CNN model achieve the highest accuracy on the validation set as the optimal hyperparameters of the ST-CNN.

[0088] The model training and prediction methods in the traffic flow prediction module are as follows:

[0089] Model training: Reconstruct the ST-CNN model using the optimized hyperparameter combination and train it on the training set. The gradient descent algorithm is used to minimize the loss function. During the training process, the weights and bias terms of the model are continuously updated to improve the prediction performance;

[0090] Traffic flow prediction: Apply the trained ST-CNN model to the test set, input a new multi-source data fusion dataset (using the same data preprocessing method), and the model output is the traffic flow prediction values for different road segments within the next hour.

[0091] In this embodiment, ST-CNN can simultaneously consider the temporal and spatial dimensional information of traffic flow data, automatically mine the inherent features and patterns therein. For example, it can capture the changing trends of traffic flow at different times and the mutual influence relationships between adjacent road segments, which is difficult to achieve by traditional methods. Through operations such as convolution and pooling, it can effectively extract features at different granularities, from local short-term fluctuations to overall long-term trends. Traffic flow data is highly complex and non-linear, affected by various factors such as weekdays / weekends, weather, special events, etc. ST-CNN can adaptively adjust the model parameters through learning and training on a large amount of historical data to fit various complex traffic patterns, thereby improving the prediction accuracy and robustness; it can identify the correlations between different road segments, such as the traffic flow interaction between adjacent main roads and the influence of branch roads on the main road, which is very important for comprehensively understanding the operation status of the urban traffic network and can help optimize traffic signal control, plan road construction, etc.;

[0092] Using the biogeography optimization algorithm to optimize the hyperparameters of ST-CNN can efficiently search for the optimal combination in the huge parameter space. By evaluating the performance metrics (such as mean square error, mean absolute error, etc.) of the model on the validation set, continuously adjust the values of the hyperparameters to ensure that the model achieves the best prediction effect. Compared with traditional manual tuning or grid search methods, the BGO algorithm can converge to the global optimal solution faster, improving the generalization ability and prediction accuracy of the model. Different traffic flow datasets may have different characteristics and distributions. By optimizing the hyperparameters, the ST-CNN model can better adapt to the changes in various datasets. Whether it is the scale of the data, the sparsity of the data, or the noise level of the data, it can find the most suitable model configuration for the dataset, thereby improving the stability and reliability of the model in different scenarios. Appropriate hyperparameters can prevent the model from overfitting during the training process, that is, the model is too complex and performs well on the training set but has a performance decline on the test set. By optimizing the hyperparameters, while ensuring the appropriate complexity of the model, it can fully utilize the relevant information of the data, improving the computational efficiency and prediction speed of the model;

[0093] Multi-source data fusion provides a richer information source for traffic flow prediction. Combining the powerful feature learning ability of ST-CNN and the high-precision model optimized by the BGO algorithm can more accurately predict the traffic flow of different road segments in the future. These prediction results have important reference value for traffic management departments to formulate traffic control strategies and optimize traffic signal timing, which helps to alleviate traffic congestion and improve road traffic efficiency. Whether it is urban road traffic, highway traffic or rail transit and other different traffic scenarios, this prediction module can be adjusted and applied according to the actual situation. By replacing or adjusting some input data and model structures, it can adapt to different traffic patterns and requirements, with strong versatility and flexibility.

[0094] Example 4

[0095] In this embodiment, the following optimizations are made on the basis of Example 3. Specifically, the platform further includes a real-time monitoring and dynamic adjustment module, which monitors various indicators of the traffic system in real time, including the road congestion index, average vehicle speed, and accident incidence rate. When a sudden change in traffic conditions is detected (such as a sudden traffic accident causing road congestion), the dynamic adjustment mechanism is triggered. This mechanism re-evaluates the importance of each data source according to the current traffic scenario and change trend, and adjusts the data fusion strategy and prediction model parameters. Parameter adjustment: According to the new importance of the data source, adjust the data fusion strategy and prediction model parameters. For example, in the data fusion process, the weighted average method is used to assign different weights according to the importance of the data source. For example, when a traffic accident causes a road section to be blocked, the platform will increase the weight of the traffic flow data of the road sections around the accident, and at the same time reduce the weight of the data of other non-critical road sections, and quickly adjust the parameters of the prediction model to achieve a quick response and accurate prediction of the impact of the accident. When a traffic accident is detected, immediately start the dynamic adjustment mechanism to re-evaluate the importance of each data source. For example, increase the weight of the traffic flow data of the road sections around the accident to 0.6, reduce the weight of the traffic flow data of other road sections to 0.2, and at the same time adjust the weights of the meteorological data and public transportation operation data to 0.15 and 0.05 respectively. Then, quickly adjust the parameters of the prediction model, recalculate the traffic flow prediction value, and timely feedback the results to the traffic management department for taking corresponding measures, such as dispatching police force and dredging traffic.

[0096] In this embodiment, the platform can monitor various indicators of the transportation system in real time, such as the road congestion index, average vehicle speed, and traffic accident incidence rate. When a sudden situation such as a traffic accident occurs, it can quickly capture abnormal changes in relevant indicators. For example, the vehicle speed on the accident section will drop sharply, and the congestion index will rise rapidly. Through these changes in real-time data, the platform can know the occurrence of the traffic accident and its impact on the traffic flow at the first time. Once a sudden change in traffic conditions is detected, the dynamic adjustment mechanism will be immediately activated. It will re-evaluate the importance of each data source according to the current specific traffic scenario and change trend. For example, in the case of a traffic accident, the platform can recognize that the traffic flow data of the surrounding sections of the accident is of greatly increased importance for the current traffic condition analysis, thus providing a basis for subsequent accurate traffic flow prediction and reasonable traffic control;

[0097] There are various uncertain factors in the transportation system. Traffic condition changes of different types, at different locations, and at different times may all affect the entire traffic network. The dynamic adjustment mechanism can flexibly adjust the data fusion strategy and prediction model parameters according to the specific situation. Whether it is the congestion of a single section or regional traffic chaos, it can find appropriate coping methods;

[0098] When a certain section is blocked due to a traffic accident, the weight of the traffic flow data of the surrounding sections of the accident is increased to 0.6, making it dominant in the data fusion process, which can more accurately reflect the actual traffic conditions around the accident. The fused data can provide more representative and targeted inputs for the subsequent prediction model, improving the accuracy of the prediction. Correspondingly, the data weight of other non-critical sections is reduced to 0.2, which can avoid the interference of the data of these sections on the overall prediction, making the model more focused on the key areas related to the accident and improving the efficiency and performance of the model;

[0099] At the same time, the weights of meteorological data and public transportation operation data are adjusted to 0.15 and 0.05 respectively, taking into account more traffic-related factors. Meteorological data may affect the rescue and evacuation speed after an accident and the driving speed of drivers; public transportation operation data is closely related to citizens' travel choices and traffic flow distribution. By reasonably adjusting the weights of these data, the prediction model can better adapt to complex traffic environment changes;

[0100] Real-time monitoring and dynamic adjustment enable the prediction model to closely follow the changing rhythm of traffic conditions, avoiding prediction biases caused by using fixed parameters. As traffic conditions continue to change, the model parameters can be continuously updated, always maintaining a high-precision prediction of traffic flow. With accurate predictions and timely feedback, traffic management departments can more efficiently take measures to respond to various sudden changes in traffic conditions. Whether it is alleviating congestion caused by traffic accidents or coping with the impact of bad weather on traffic, they can make decisions more proactively and scientifically, improving the operation efficiency and safety of the entire urban traffic system.

[0101] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent traffic flow statistics and prediction platform for multi-source data fusion, characterized in that: The platform includes: Data acquisition layer: Collect road traffic flow data, meteorological data, public transportation operation data, and data containing traffic condition descriptions on social media platforms through various types of sensors and data interfaces, and transmit the data to the platform server. Data preprocessing layer: The platform server performs cleaning, denoising, format unification, and standardization processing on the collected raw data. Data fusion layer: Use the biogeography-based optimization algorithm to fuse the preprocessed data. Consider each data source as a species, construct a fitness function based on the characteristics of the data, and find a data fusion method that can effectively integrate multi-source traffic data through the migration and mutation operations of the species. Traffic flow calculation module: Based on the result of multi-source data fusion, use the weighted average method to calculate the traffic flow at the current moment. Traffic flow prediction module: Construct a prediction model framework based on the spatio-temporal convolutional neural network (ST-CNN). Perform convolution and pooling operations on the input data after multi-source data fusion to extract high-level features. Use the biogeography-based optimization algorithm to optimize the hyperparameters of ST-CNN, and apply the optimized model to the prediction of traffic flow to output the traffic flow prediction values of different road sections within the next hour.

2. The intelligent transportation flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The traffic flow data is collected through loop induction coils, cameras, and electronic tag readers installed on urban roads and intersections. The meteorological data is collected through weather stations and meteorological satellites. The public transportation operation data includes bus GPS positioning data and subway card swiping data. The social media traffic-related data includes text, images, and video information containing traffic condition descriptions captured from social media platforms.

3. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The calculation method of the biogeography-based optimization algorithm module is as follows: Habitat initialization: Consider different types of data sources as habitats, and consider multiple data features contained in each habitat as species. A habitat includes traffic flow data of induction coils, vehicle speed data of cameras, and travel data of mobile terminals. Randomly initialize the species richness of each habitat, which is determined by calculating the deviation between the initial data fusion result and the actual traffic flow. Fitness function calculation: Design a fitness function to evaluate the quality of each habitat. The factors of the fitness function include data accuracy, diversity, correlation, and reliability. Migration operation: Calculate the immigration rate and emigration rate based on the species richness of the habitat, and calculate the migration probability based on the immigration rate and emigration rate. Mutation operation: Determine the probability according to specific rules and perform mutation operations on the migrated habitat. Iterative update: Repeat the above steps until the maximum number of iterations is reached or the fitness reaches the preset convergence threshold.

4. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 3, characterized in that: The data accuracy: Use the mean square error to measure the difference between the fused data and the actual traffic flow data. Data diversity: Represent the diversity by calculating the entropy of data features. Data correlation: Measure the correlation by calculating the correlation coefficient between different data sources. Data reliability: Consider the credibility of the data source and assign different weights according to experience.

5. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: In the traffic flow calculation module, the weighted average method calculates the traffic flow of different road sections at the current moment as follows: For road segment L, assume that the fused data includes road traffic flow data F1, meteorological data F2, public transportation operation data F3, and social media data F4, with corresponding weights w1, w2, w3, and w4 respectively. Then the traffic flow Q of road segment L is calculated as: Q = w1F1 + w2F2 + w3F3 + w4F4. At the same time, the calculation result is corrected by combining the traffic flow data of the same historical period.

6. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The construction method of the prediction model of the spatio-temporal convolutional neural network in the traffic flow prediction module is as follows: Data preparation and preprocessing: The input of multi-source data fusion will use the data set after data cleaning, preprocessing, and multi-source data fusion as the input of the model. Assume that the fused data set is X, and its dimension is [n, m, k], where n represents the number of time steps, m represents the spatial position, and k represents the feature dimension at each position. The data is standardized to perform standardized processing on the input data; Construction of spatio-temporal convolutional neural network (ST-CNN): The input layer accepts the standardized multi-source data fusion input X', whose shape is [n, m, k]; Design convolutional kernels in the spatio-temporal dimension to simultaneously extract time and space features. Assume that the convolutional kernel size is [h, w, d], where h represents the convolutional kernel size in the time dimension, w represents the convolutional kernel size in the spatial dimension, and d represents the number of channels between the input and output feature maps; For each position (i, j, l) in the input feature map X', where i represents the time step, j represents the spatial position, and l represents the channel, a new feature map is generated through convolutional operations; Max pooling with a size of [a, b] is used for downsampling in the time dimension and spatial dimension; The fully connected layer unfolds the pooled feature map into a one-dimensional vector and performs a linear transformation through the fully connected layer to map it to the dimension of the prediction output. According to the specific prediction task, the activation function and structure of the output layer are designed.

7. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The method for optimizing the hyperparameters of ST-CNN by the biogeography-based optimization algorithm in the traffic flow prediction module is as follows: S1. Initialize the population: Randomly generate a group of individuals of ST-CNN hyperparameters, and each individual contains a combination of hyperparameters such as convolutional kernel size, learning rate, and number of network layers; S2. Fitness evaluation: For each individual, use the corresponding hyperparameter combination to construct an ST-CNN model and perform training and evaluation on the validation set; S3. Migration operation: Define the migration rate P i , and the calculation formula of the migration rate is designed as follows: For each individual θ i , with a fixed probability P i , select another individual θ i to perform the migration operation, and replace the h i value of individual θ i with the h j value of individual θ j according to the predetermined probability; S4. Iterative update: Repeat steps 2 and 3 until the preset number of iterations is reached. During each iteration, record the optimal hyperparameter combination and its corresponding fitness value; S5. Selection of optimal hyperparameters: After multiple iterations, select the hyperparameter combination that makes the ST-CNN model achieve the highest accuracy on the validation set as the optimal hyperparameters of ST-CNN.

8. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The method of model training and prediction in the traffic flow prediction module is as follows: Model training: Use the optimized hyperparameter combination to reconstruct the ST-CNN model and perform training on the training set. The gradient descent algorithm is used to minimize the loss function. During the training process, the weights and bias terms of the model are continuously updated; Traffic flow prediction applies the trained ST-CNN model to the test set, inputs a new multi-source data fusion dataset, and the model output is the predicted traffic flow values for different road segments within the next hour.

9. An intelligent traffic flow statistics and prediction platform for multi-source data fusion according to claim 1, characterized in that: The platform also includes a real-time monitoring and dynamic adjustment module that monitors various indicators of the traffic system in real time, including the road congestion index, average vehicle speed, and accident incidence rate. When a sudden change in traffic conditions is detected, a dynamic adjustment mechanism is triggered. This mechanism re-evaluates the importance of each data source based on the current traffic scenario and change trend, and adjusts the data fusion strategy and prediction model parameters.

Images