Low-altitude traffic flow prediction method and system based on artificial intelligence

Abstract

The invention discloses a low-altitude traffic flow prediction method and system based on artificial intelligence. The method comprises the following steps: acquiring a complete traffic flow data set; obtaining traffic flow weight coefficients corresponding to various low-altitude aircrafts based on the complete traffic flow data set; obtaining a weather influence weight coefficient based on the complete traffic flow data set and the corresponding weather factor data set; obtaining a holiday activity traffic flow data set based on the complete traffic flow data set, and further calculating to obtain a holiday influence weight coefficient; acquiring multi-factor historical data of the target area, and correcting the multiple weight coefficients by calculating the variance of the historical data corresponding to each factor; constructing a traffic flow prediction sub-model corresponding to each factor according to the corrected weight coefficient, and training the traffic flow prediction sub-model; and obtaining a low-altitude traffic flow predicted value based on the trained model. According to the low-altitude traffic flow prediction method, more accurate low-altitude traffic flow prediction is realized by comprehensively considering various factors such as aircraft types, weather conditions, holiday and festival activities and the like.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and particularly relates to a low-altitude traffic flow prediction method and system based on artificial intelligence. Background Art

[0002] Low-altitude traffic flow prediction is a complex problem involving multiple influencing factors and uncertainties. First, there are various types of low-altitude traffic participants, including helicopters, drones, light aircraft, etc., and their flight characteristics are different. Second, low-altitude traffic is easily affected by weather conditions, such as visibility, wind speed, air pressure, etc., and these factors will affect the operation status and route selection of aircraft. Moreover, factors such as holidays and major events will also cause fluctuations in low-altitude traffic flow. Therefore, to accurately predict future low-altitude traffic flow, the influence of these complex factors must be fully considered.

[0003] In addition, the historical data mastered by low-altitude traffic management departments and service providers is often incomplete, with a limited time span, and problems such as data missing and noise are relatively prominent, which brings difficulties to the training of prediction models. Moreover, the low-altitude traffic characteristics in different regions vary greatly. For example, the situations over urban areas and suburban areas are completely different, which requires the prediction model to be able to adapt to the characteristics of different regions. At the same time, the timeliness requirement for low-altitude traffic flow prediction is very high, and the traffic flow prediction results for a future period need to be given within a short time, which poses a challenge to the computing efficiency. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a low-altitude traffic flow prediction method and system based on artificial intelligence to solve the problems existing in the above prior art.

[0005] To achieve the above object, the present invention provides a low-altitude traffic flow prediction method based on artificial intelligence, including:

[0006] Obtaining historical traffic flow data and performing preprocessing to obtain a complete traffic flow data set;

[0007] Classifying low-altitude aircraft, and obtaining traffic flow weight coefficients corresponding to various low-altitude aircraft based on the complete traffic flow data set;

[0008] Obtaining a weather factor data set corresponding to the complete traffic flow data set; obtaining weather influence weight coefficients based on the weather factor data set and the complete traffic flow data set;

[0009] Extracting a holiday activity traffic flow data set during holidays and major events from the complete traffic flow data set, and obtaining holiday influence weight coefficients based on the holiday activity traffic flow data set;

[0010] Obtain the multi-factor historical data of the target area, and calculate the variance of the historical data corresponding to each factor; correct the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient based on the variance of each factor; construct and train a traffic flow prediction sub-model corresponding to each factor according to the corrected weight coefficient; obtain the low-altitude traffic flow prediction value based on the trained traffic flow prediction sub-model.

[0011] Optionally, the process of obtaining the complete traffic flow data set includes:

[0012] Analyze whether there is data missing in the historical traffic flow data. If so, extract the corresponding missing time period and area; estimate the traffic flow data within the time period according to the historical flow data by area to obtain the complemented historical traffic flow data; perform outlier processing on the complemented historical traffic flow data to obtain the complete traffic flow data set.

[0013] Optionally, the process of obtaining the traffic flow weight coefficient corresponding to each type of low-altitude aircraft includes:

[0014] Obtain the flight characteristic parameters of the low-altitude aircraft, use the K-means clustering algorithm to perform clustering analysis on the flight characteristic parameters to obtain different types of low-altitude aircraft sets; divide the complete traffic flow data set according to the types of the low-altitude aircraft sets, and determine the traffic flow weight coefficient by calculating the proportion of the data corresponding to each type of low-altitude aircraft in the complete traffic flow data set.

[0015] Optionally, the process of obtaining the weather impact weight coefficient includes:

[0016] Extract the time period and area corresponding to the complete traffic flow data set, and obtain the weather factor data set corresponding to the time period and area; after preprocessing the complete traffic flow data set and the weather factor data set, calculate the correlation coefficient between each weather factor and the traffic flow data, and perform normalization processing on the correlation coefficient, and obtain the weather impact weight coefficient of each weather factor on the traffic flow data based on the normalized correlation coefficient.

[0017] Optionally, the process of obtaining the holiday impact weight coefficient includes:

[0018] Extract the holiday activity traffic flow data set and the normal data set during holidays and major events in the complete traffic flow data set, calculate the traffic flow change rate based on the holiday activity traffic flow data set and the normal data set, and perform normalization processing on the traffic flow change rate to obtain the holiday impact weight coefficient.

[0019] Optionally, obtain the multi-factor historical data of the target area and perform preprocessing. Use the analysis of variance method to analyze the historical data corresponding to each factor, obtain the variance corresponding to each factor, and correct the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient according to the variance. Among them, the types of the factors include aircraft category, weather condition, and holiday activity; the weather conditions include but are not limited to temperature, humidity, wind speed, visibility, and precipitation.

[0020] Optionally, the process of obtaining the low-altitude traffic flow prediction value includes:

[0021] Construct multiple traffic flow prediction sub-models, use the corrected traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient as the initial parameters of the traffic flow prediction sub-models respectively, and train the multiple traffic flow prediction sub-models using an adaptive learning rate adjustment strategy; obtain the multi-factor data in real time, split it and input it into the corresponding traffic flow prediction sub-models, obtain the prediction value corresponding to each factor, and use the weighted average method to fuse the prediction values corresponding to each factor to obtain the final low-altitude traffic flow prediction value.

[0022] Optionally, when using the weighted average method to fuse the prediction values corresponding to each factor, allocate the fusion weights for each factor according to the proportion of the coefficients corresponding to each factor in the coefficient sum among the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient.

[0023] Optionally, it further includes updating the complete traffic flow data set through the multi-factor data and the corresponding traffic flow data obtained in real time to obtain a model training set, and updating the traffic flow prediction sub-model through the online gradient descent method.

[0024] The present invention also provides an artificial intelligence-based low-altitude traffic flow prediction system, including:

[0025] A data preprocessing module, configured to obtain historical traffic flow data and perform preprocessing to obtain a complete traffic flow data set;

[0026] An aircraft classification module, configured to classify the low-altitude aircraft and obtain the traffic flow weight coefficient corresponding to each type of low-altitude aircraft based on the complete traffic flow data set;

[0027] A weather impact analysis module, configured to obtain a weather factor data set corresponding to the complete traffic flow data set; obtain a weather impact weight coefficient based on the weather factor data set and the complete traffic flow data set;

[0028] A holiday event impact analysis module is used to extract the holiday event traffic flow dataset during holidays and major events from the complete traffic flow dataset, and obtain the holiday impact weight coefficient based on the holiday event traffic flow dataset;

[0029] An adaptive weight adjustment module is used to obtain the multi-factor historical data of the target area and calculate the variance of the historical data corresponding to each factor; correct the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient based on the variance of each factor; construct and train the traffic flow prediction sub-model corresponding to each factor according to the corrected weight coefficient; obtain the low-altitude traffic flow prediction value based on the trained traffic flow prediction sub-model;

[0030] A model update and fusion prediction module is used to update the complete traffic flow dataset through the multi-factor data and the corresponding traffic flow data obtained in real time to obtain a model training set, and update the traffic flow prediction sub-model by the online gradient descent method.

[0031] Compared with the prior art, the present invention has the following advantages and technical effects:

[0032] Aiming at the problem of missing historical traffic data, the present invention uses interpolation and extrapolation methods to complete the data. Cluster and classify according to the characteristic parameters of the aircraft, and calculate the proportion of each category as the traffic flow weight coefficient. By calculating the correlation coefficient between the weather factor and the historical data, the weight coefficient of the weather impact is obtained. Analyze the traffic flow change rate of holidays and major events to obtain their impact weights. Combining with the regional characteristics, an adaptive weight adjustment strategy is adopted to dynamically adjust the weights according to the variances of each factor in the historical data. The incremental learning algorithm is used to update the model parameters in real time, and the weighted average method is used to fuse the prediction results of various impact factors. By comprehensively considering various factors such as aircraft categories, weather conditions, and holiday events, and combining with regional characteristics for adaptive weight adjustment, the present invention realizes more accurate low-altitude traffic flow prediction, effectively improves the prediction accuracy, and provides important decision-making support for low-altitude traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0034] Figure 1 is the method flow chart of the embodiment of the present invention;

[0035] Figure 2 is the system schematic diagram of the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The following will describe the present application in detail with reference to the drawings and in combination with the embodiments.

[0037] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0038] Embodiment 1

[0039] As Figure 1 shown, in this embodiment, a low-altitude traffic flow prediction method based on artificial intelligence is provided, including:

[0040] Obtain historical traffic flow data and perform preprocessing to obtain a complete traffic flow data set;

[0041] As a preferred implementation manner, the process of obtaining the complete traffic flow data set includes:

[0042] Analyze whether there is data missing in the historical traffic flow data. If so, extract the corresponding missing time period and area; estimate the traffic flow data within the time period according to the historical flow data by region to obtain the complemented historical traffic flow data; perform outlier processing on the complemented historical traffic flow data to obtain the complete traffic flow data set.

[0043] Specifically, according to the historical traffic flow data, determine the time period and road section with data missing; for the determined data missing time period, obtain the complete traffic flow data before and after the missing time period; according to the obtained complete traffic flow data before and after, use the linear interpolation algorithm to estimate the traffic flow data within the missing time period; if the missing time period is near the start or end time of the historical data, use the extrapolation algorithm to estimate the traffic flow data within the missing time period according to the complete traffic flow data near the missing time period; insert the estimated traffic flow data into the corresponding missing time period in the original historical traffic flow data set to obtain the complemented complete traffic flow data set; for the complemented complete traffic flow data set, use the anomaly detection algorithm to determine whether there are outliers. If there are outliers, replace the outliers with the normal traffic flow mean of adjacent time points.

[0044] Exemplarily, the integrity and accuracy of traffic flow data are crucial for traffic prediction and management. First, it is necessary to determine the time periods and road segments with missing data in the historical data. For example, due to equipment failures, data was missing for a certain urban arterial road from January 1st to January 3rd, 2023. In response to this situation, complete data before and after the missing time period can be obtained. For instance, the traffic flow data on December 31st and January 4th can be obtained as a reference. The linear interpolation algorithm can be used to estimate the data during the missing period. Suppose the traffic flow at a certain time period on December 31st was 1000 vehicles per hour, and at the same time period on January 4th was 1200 vehicles per hour. Then, the traffic flow data from January 1st to 3rd can be estimated proportionally. For missing data near the start or end of the dataset, extrapolation methods need to be adopted. If the missing data occurred before January 1st, the data at the beginning of January can be used for retrospective estimation. When extrapolating, periodic changes such as the traffic flow differences between weekdays and weekends need to be considered. Insert the estimated data into the corresponding positions in the original dataset to obtain a complete dataset. This step is crucial for subsequent analysis because the continuity of the data can improve the accuracy of the prediction model. Next, outlier detection needs to be carried out. Outliers may be caused by factors such as equipment failures and special events. For example, if the traffic flow at a certain time period suddenly surges to 10 times the normal level, this may be an outlier. Statistical methods such as the 3σ principle or machine learning algorithms such as Isolation Forest can be used to detect outliers. After detecting outliers, they can be replaced with the mean of adjacent normal time periods. For example, if it is found that the traffic flow at 14:00 is abnormally high, it can be replaced with the average of 13:00 and 15:00. This method can smooth the data and reduce the impact of outliers on the model. After completing data cleaning and completion, a traffic flow prediction model can be constructed. Commonly used models include time series models (such as ARIMA), machine learning models (such as Random Forest), and deep learning models (such as LSTM). Selecting an appropriate model requires considering factors such as data characteristics, prediction accuracy requirements, and computing resources. Taking the ARIMA model as an example, it can capture the time correlation and periodicity of traffic flow. By analyzing the trends, seasonality, and random fluctuations in historical data, ARIMA can predict the traffic flow in the short term in the future. For example, it can predict the traffic flow in the next hour, providing decision-making support for traffic management departments. Deep learning models such as LSTM are more suitable for dealing with long-term dependencies. It can learn complex traffic patterns, such as morning and evening rush hours, holiday effects, etc. The LSTM model can consider multiple relevant factors simultaneously, such as weather and events, to improve the accuracy of prediction. When constructing the model, the dataset is usually divided into a training set, a validation set, and a test set. For example, the data in 2022 can be used as the training set, the data from January to June in 2023 as the validation set, and the data from July to December as the test set. This can evaluate the generalization ability of the model. After the model training is completed, performance evaluation is required.Common evaluation metrics include the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), etc. For example, if the MAE of the model on the test set is 50 vehicles per hour and the RMSE is 70 vehicles per hour, this indicates that the model has good predictive ability. The application scope of the prediction model is extensive. The traffic management department can use the prediction results to optimize signal timing and alleviate congestion. Navigation software can recommend the best route based on the prediction results. Urban planners can optimize the road network based on long-term prediction results. In short, through data completion, outlier handling, and model construction, historical traffic flow data can be fully utilized to provide strong support for traffic management and planning. This process not only improves the data quality but also lays a foundation for subsequent traffic flow analysis and prediction, contributing to enhancing the overall efficiency of the urban traffic system.

[0045] Classify low-altitude aircraft, and obtain the traffic flow weight coefficients corresponding to various types of low-altitude aircraft based on the complete traffic flow data set;

[0046] As a preferred implementation manner, the process of obtaining the traffic flow weight coefficients corresponding to various types of low-altitude aircraft includes:

[0047] Obtain the flight characteristic parameters of low-altitude aircraft, use the K-means clustering algorithm to perform clustering analysis on the flight characteristic parameters to obtain different types of low-altitude aircraft sets; divide the complete traffic flow data set according to the types of the low-altitude aircraft sets, and determine the traffic flow weight coefficients by calculating the proportion of the data corresponding to various types of low-altitude aircraft in the complete traffic flow data set.

[0048] Exemplarily, the flight characteristic parameters of an aircraft are key indicators for evaluating its performance and applicability. Flight speed reflects the maneuverability of the aircraft, load capacity determines its transportation efficiency, range distance embodies its endurance capacity, and aircraft type covers different types of aircraft for various purposes. For example, a certain model of a passenger aircraft has a cruising speed of 900 km / h, a maximum load of 20 tons, and a range of 8000 km, belonging to the category of large passenger aircraft. The K-means clustering algorithm is a commonly used unsupervised learning method that can group aircraft with similar characteristics into one category. Suppose aircraft are divided into three categories: short-range small, medium-range medium, and long-range large. After clustering, short-range small aircraft may have a lower flight speed and load capacity but good maneuverability; medium-range medium aircraft are more balanced in various parameters; long-range large aircraft have a higher flight speed, load capacity, and range distance. Counting the number of occurrences of each type of aircraft in historical data can reflect the usage frequency of different types of aircraft. For example, in the historical data of an airport, short-range small aircraft appear 1000 times, medium-range medium aircraft appear 2000 times, and long-range large aircraft appear 500 times. These data can be used to calculate the traffic flow weight coefficients of each type of aircraft. The calculation of traffic flow weight coefficients can adopt a simple proportion method. Based on the above example, the weight coefficient of short-range small aircraft is 0.286 (1000 / 3500), that of medium-range medium aircraft is 0.571 (2000 / 3500), and that of long-range large aircraft is 0.143 (500 / 3500). These weight coefficients reflect the contribution degree of different types of aircraft to the airport traffic flow. When establishing an aircraft traffic flow prediction model, historical data and weight coefficients can be combined. For example, if it is predicted that the total traffic flow of the airport on a certain day is 100 flights, then it can be estimated that there are approximately 29 short-range small aircraft, about 57 medium-range medium aircraft, and about 14 long-range large aircraft. This prediction method takes into account the historical usage of different types of aircraft and can provide a relatively accurate prediction of traffic flow distribution. When predicting the traffic flow of a new aircraft, it is first necessary to determine its category. Suppose a new type of aircraft has a flight speed of 700 km / h, a load of 10 tons, and a range of 5000 km. By comparing with the characteristics of existing categories, it may be classified as a medium-range medium aircraft. According to the traffic flow weight coefficient of 0.571 for this category, it can be predicted that this new type of aircraft will account for a relatively high proportion in the total traffic flow, about 57%. The prediction results can provide important references for aircraft operation scheduling and route planning. For example, if the prediction shows that the traffic flow of medium-range medium aircraft accounts for a relatively high proportion, the airport may need to increase the corresponding number of parking positions and ground crew. Airlines may consider increasing the number of flights on medium-range routes to meet market demand. This data-driven decision-making method can improve resource utilization efficiency, optimize the route network structure, and thus enhance the overall operation efficiency.

[0049] Obtain the weather factor dataset corresponding to the complete traffic flow dataset; obtain the weather impact weight coefficient based on the weather factor dataset and the complete traffic flow dataset;

[0050] As a preferred implementation, the process of obtaining the weather impact weight coefficient includes:

[0051] The time period and area corresponding to the complete traffic flow data set are extracted, and the weather factor data set corresponding to the time period and area is obtained; after preprocessing the complete traffic flow data set and the weather factor data set, the correlation coefficient between each weather factor and the traffic flow data is calculated, and the correlation coefficient is normalized. Based on the normalized correlation coefficient, the weather influence weight coefficient of each weather factor on the traffic flow data is obtained.

[0052] Specifically, obtain complete historical traffic flow data and weather data for the corresponding time period, including weather factors such as temperature, humidity, wind speed, visibility and precipitation. Preprocess the historical traffic flow data and weather factor data to remove outliers and missing values ​​to ensure the integrity and accuracy of the data. Calculate the Pearson correlation coefficient between each weather factor and the traffic flow data to obtain a coefficient value that reflects the strength of their correlation. Normalize the calculated Pearson correlation coefficient and map the coefficient value to the [0, 1] interval to eliminate the influence of differences in dimension and numerical range. According to the normalized correlation coefficient value, determine the weight coefficient of each weather factor on traffic flow. The larger the weight coefficient, the more significant the impact of the factor on traffic flow.

[0053] Exemplarily, in traffic flow prediction, it is crucial to consider the impact of weather factors. First, it is necessary to obtain complete historical traffic flow data and weather data for the corresponding time period. For example, the daily traffic volume data of a city in a year, as well as weather factor data such as temperature, humidity, wind speed, visibility, and precipitation during the same period. These raw data may contain outliers and missing values, and preprocessing is required to ensure data quality. During the preprocessing process, the moving average method can be used to handle outliers, and the missing values can be filled with the average value of adjacent data. The processed data is more complete and accurate, laying a foundation for subsequent analysis. Next, calculate the Pearson correlation coefficient between each weather factor and traffic flow to reflect the strength of their correlation. The value range of the Pearson correlation coefficient is [-1, 1], and the larger the absolute value, the stronger the correlation. In order to eliminate the differences in dimension and numerical range between different weather factors, it is necessary to normalize the correlation coefficient. The maximum-minimum normalization method can be used to map the coefficient value to the interval [0, 1]. The normalized coefficient value can be directly used as the weight coefficient for the impact of each weather factor on traffic flow. For example, the normalized coefficient of temperature is 0.8, humidity is 0.6, wind speed is 0.4, visibility is 0.7, and precipitation is 0.9, indicating that precipitation has the greatest impact on traffic flow, followed by temperature and visibility.

[0054] Extract the holiday activity traffic flow dataset during holidays and major events from the complete traffic flow dataset, and obtain the holiday impact weight coefficient based on the holiday activity traffic flow dataset;

[0055] As a preferred implementation manner, the process of obtaining the holiday impact weight coefficient includes:

[0056] Extract the holiday activity traffic flow dataset and the normal dataset during holidays and major events from the complete traffic flow dataset, calculate the traffic flow change rate based on the holiday activity traffic flow dataset and the normal dataset, normalize the traffic flow change rate, and obtain the holiday impact weight coefficient.

[0057] Exemplarily, obtaining traffic flow data during historical holidays and major events is an important basis for predicting future traffic conditions. For example, the large-scale population movement across the country during the Spring Festival will lead to a sharp increase in traffic flow, and the traffic pressure in tourist hot cities will also increase significantly during the National Day Golden Week. For major events, such as during the Beijing Olympics, the traffic flow pattern in the host city will change significantly. During the data preprocessing stage, outliers and missing values need to be removed to ensure data quality. For example, abnormally low flow records caused by equipment failures or data missing due to system maintenance need to be identified and processed. The moving average method or interpolation method can be used to fill in the missing data to ensure the continuity and reliability of the data. When calculating the traffic flow change rate, the average daily traffic flow during holidays or major events can be compared with the average flow on normal working days. For example, assuming that the average traffic flow in a certain city on weekdays is 100,000 vehicles per day, and this figure may rise to 150,000 vehicles per day during the Spring Festival, then the change rate is 50%. The normalization process can adopt the minimum-maximum normalization method to map the change rates of different holidays and events to between 0 and 1. This can eliminate the influence of different scales, enabling the impacts of different festivals such as the Spring Festival, National Day, and New Year's Day to be compared under the same standard. When establishing the influence weight calculation model, various factors can be considered. For example, for the Spring Festival, factors such as the length of the holiday, the proportion of returning home population, and the city level can be used as feature inputs to the model.

[0058] Obtain multi-factor historical data of the target area, and calculate the variance of the historical data corresponding to each factor; based on the variance of each factor, correct the traffic flow weight coefficient, weather influence weight coefficient, and holiday influence weight coefficient;

[0059] As a preferred implementation manner, obtain multi-factor historical data of the target area and perform preprocessing, analyze the historical data corresponding to each factor by using the analysis of variance method to obtain the variance corresponding to each factor, and correct the traffic flow weight coefficient, weather influence weight coefficient, and holiday influence weight coefficient according to the variance. Among them, the types of factors include aircraft category, weather conditions, and holiday activities; the weather conditions include but are not limited to temperature, humidity, wind speed, visibility, and precipitation.

[0060] Specifically, obtain multi-dimensional historical data such as aircraft category, weather conditions, and holiday activities of the target area, and perform data cleaning and preprocessing on the data for each dimension. Use the analysis of variance method to calculate the variance of the data for each dimension, determine the weight coefficient of each dimension according to the size of the variance, and assign a higher weight to the dimension with a larger variance. Use the weight coefficients of each dimension as the initial parameters of the neural network model.

[0061] Exemplarily, multi-dimensional historical data such as the aircraft category, weather conditions, and holiday activities in the target area are obtained, and data cleaning and preprocessing are performed on the data for each dimension. For example, it is necessary to collect the traffic flow data in the core business district of a certain city in the past three years, and at the same time collect the aircraft category data in the same period, such as the number of civil airliners, cargo planes, private planes, etc. passing through the area every day; and weather data, including temperature, humidity, rainfall, wind speed, whether it is rainy or snowy weather, etc.; and holiday and major event information, such as National Day, Spring Festival, large-scale concerts, sports events, etc. Data cleaning and preprocessing refer to checking, correcting, and supplementing the collected raw data to eliminate errors, inconsistencies, and incompleteness in the data, and ensure the quality and usability of the data. For example, if it is found that the traffic flow data for a certain day is missing, it can be estimated by interpolation using the data of the previous and next days; if it is found that the weather data for a certain day is recorded as "clear turning to cloudy", but there is no specific temperature range, it can be supplemented by querying a professional meteorological database. Doing so can improve the accuracy of subsequent analysis and modeling. The analysis of variance method is used to calculate the variance of the data for each dimension, and the weight coefficient for each dimension is determined according to the size of the variance. The dimension with a larger variance is given a higher weight. The analysis of variance is a statistical method used to compare whether the differences in data between different groups are significant. Here, each dimension such as the aircraft category, weather conditions, and holiday activities can be regarded as a group, and the variance of the data for each group is calculated. For example, it is found that the traffic flow data fluctuates greatly and the variance is large during holidays, while the temperature data in the weather conditions is relatively stable and the variance is small. Then, when determining the weight coefficient, a higher weight will be given to the holiday dimension and a lower weight will be given to the temperature dimension. This is because the larger the variance, the more significant the change in the data of this dimension and the greater the impact on the traffic flow.

[0062] Construct traffic flow prediction sub-models corresponding to each factor according to the corrected weight coefficients and train them; obtain the low-altitude traffic flow prediction value based on the trained traffic flow prediction sub-models.

[0063] As a preferred implementation manner, the process of obtaining the low-altitude traffic flow prediction value includes:

[0064] Construct multiple traffic flow prediction sub-models, use the corrected traffic flow weight coefficients, weather impact weight coefficients, and holiday impact weight coefficients as the initial parameters of the traffic flow prediction sub-models respectively, and train the multiple traffic flow prediction sub-models using an adaptive learning rate adjustment strategy; obtain the prediction values corresponding to each factor by inputting the multi-factor data obtained in real time into the corresponding traffic flow prediction sub-models after splitting, and use the weighted average method to fuse the prediction values corresponding to each factor to obtain the final low-altitude traffic flow prediction value.

[0065] As a preferred implementation, when using the weighted average method to fuse the predicted values corresponding to each factor, the fusion weights are assigned to each factor according to the proportion of the coefficients corresponding to each factor in the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient in the total coefficient.

[0066] Specifically, the predicted results of factors such as aircraft type, weather conditions, holiday activities, historical data, and regional characteristics are fused using the weighted average method to obtain the final predicted value of low-altitude traffic flow. The weights of each factor in the weighted average method are determined according to the weight coefficients calculated in the previous steps.

[0067] Exemplarily, for the feature data of each dimension, an independent sub-prediction model is constructed. Using machine learning algorithms such as decision trees and support vector machines, the influence weight coefficients of each dimension feature on low-altitude traffic flow are trained. In the prediction stage, the sub-prediction models of each dimension are used respectively to obtain the predicted results of factors such as aircraft type, weather conditions, holiday activities, historical data, and regional characteristics. According to the weight coefficients of each dimension feature calculated in the previous steps, the predicted results of each factor are fused using the weighted average method. By assigning different weights to different features, the predicted value of low-altitude traffic flow considering multiple factors is obtained. If the prediction error exceeds the preset threshold, the incremental learning process is triggered.

[0068] As a preferred implementation, it also includes updating the complete traffic flow data set by using the multi-factor data obtained in real time and the corresponding traffic flow data to obtain a model training set, and updating the traffic flow prediction sub-model by the online gradient descent method.

[0069] Specifically, multi-dimensional feature data including aircraft type, weather conditions, holiday activities, historical data, and regional characteristics are obtained, and a training data set for the low-altitude traffic flow prediction model is constructed. The prediction model is trained using the incremental learning algorithm, and the model parameters are updated in real time by the online gradient descent method. The model is dynamically adjusted according to the new data to improve the prediction accuracy.

[0070] Exemplarily, when building a low-altitude traffic flow prediction model, it is first necessary to obtain multi-dimensional feature data. Taking the types of aircraft as an example, historical flight data of different types such as drones, small helicopters, and light fixed-wing aircraft can be collected. In terms of weather conditions, factors such as temperature, wind speed, and visibility can be included. Holiday activity data can cover traffic flow changes during major festivals such as the Spring Festival and National Day. Historical data includes low-altitude traffic flow statistics over the past few years. Regional characteristics can consider urban density, topography, etc. Using an incremental learning algorithm to train the prediction model can enable the model to continuously update and optimize as new data arrives. For example, when new types of aircraft or special weather conditions appear, the model can quickly adapt to these changes. The online gradient descent method can update the model parameters in real time, making the prediction results more accurate. For instance, if a sudden foggy weather on a certain day leads to a sharp decrease in low-altitude traffic flow, the model can quickly adjust the parameters to reflect this change. Constructing independent sub-prediction models for each dimension can capture the impact of each factor on low-altitude traffic flow more precisely. For example, for the types of aircraft, a decision tree algorithm can be used to construct a sub-model, which can learn the activity patterns of different types of aircraft. For weather conditions, a support vector machine algorithm can be adopted. This algorithm is good at dealing with non-linear relationships and can effectively capture the complex impact of weather changes on traffic flow. In the prediction stage, each dimension sub-model gives a prediction result. Suppose the sub-model for the types of aircraft predicts frequent drone activities on a certain day, the sub-model for weather conditions predicts clear and suitable flying conditions on that day, the sub-model for holiday activities predicts an ordinary working day, the historical data sub-model shows medium traffic flow in this period in previous years, and the regional characteristics sub-model indicates that the terrain of this area is flat. These prediction results need to be weighted and averaged to obtain the final low-altitude traffic flow prediction value. The determination of the weight coefficients is crucial. For example, if historical data shows that weather conditions have the greatest impact on low-altitude traffic flow, then the weight of the weather conditions sub-model should be higher. Suppose after training, the weight of weather conditions is 0.3, the weight of the types of aircraft is 0.25, the weight of holiday activities is 0.2, the weight of historical data is 0.15, and the weight of regional characteristics is 0.1. Then the final prediction value will be the weighted average of the prediction results of these factors. When the prediction error exceeds the preset threshold, an incremental learning process is triggered. For example, if the difference between the predicted value and the actual value exceeds 20%, the actual data will be added to the training set, and the online gradient descent method will be used to update the model parameters. This method can enable the model to continuously adapt to new trends and patterns, such as traffic flow changes brought about by newly opened air routes or emerging low-altitude tourism projects. Continuous monitoring and regular triggering of the incremental learning process can enable the model to always maintain sensitivity to the latest data. For example, the model is updated once a weekend to ensure that the model can capture periodic change patterns.Through this dynamic update mechanism, the prediction model can adapt to the long-term trend changes in low-altitude traffic flow, such as seasonal fluctuations or changes in traffic patterns brought about by urban development, thereby maintaining the timeliness and accuracy of the prediction.

[0071] Embodiment 2

[0072] As Figure 2 shown, this embodiment provides an artificial intelligence-based low-altitude traffic flow prediction system, including:

[0073] A data preprocessing module, which is used to obtain historical traffic flow data and perform preprocessing to obtain a complete traffic flow data set; for the missing problem of historical traffic flow data, data interpolation and extrapolation methods are used for estimation and completion to obtain a complete and continuous data set;

[0074] An aircraft classification module, which is used to classify low-altitude aircraft and obtain the traffic flow weight coefficients corresponding to various types of low-altitude aircraft based on the complete traffic flow data set; according to the obtained aircraft flight characteristic parameters, including flight speed, load, range, aircraft type, etc., a clustering algorithm is used to classify the aircraft, and the proportion of each category in the complete historical traffic flow data is calculated, and the proportion is used as the traffic flow weight coefficient of the aircraft in this category;

[0075] A weather impact analysis module, which is used to obtain the weather factor data set corresponding to the complete traffic flow data set; obtain the weather impact weight coefficient based on the weather factor data set and the complete traffic flow data set; by calculating the Pearson correlation coefficient between the weather factors and the complete historical traffic flow data and normalizing the correlation coefficient, the impact weight coefficient of each weather factor on the traffic flow is obtained, and the weather factors include temperature, humidity, wind speed, visibility, precipitation;

[0076] A holiday activity impact analysis module, which is used to extract the holiday activity traffic flow data set during holidays and major events from the complete traffic flow data set, and obtain the holiday impact weight coefficient based on the holiday activity traffic flow data set;

[0077] An adaptive weight adjustment module, which is used to obtain the multi-factor historical data of the target area, calculate the variance of the historical data corresponding to each factor; correct the traffic flow weight coefficient, the weather impact weight coefficient, and the holiday impact weight coefficient based on the variance of each factor; construct a traffic flow prediction sub-model corresponding to each factor according to the corrected weight coefficient and train it; obtain the low-altitude traffic flow prediction value based on the trained traffic flow prediction sub-model;

[0078] The model update and fusion prediction module is used to update the complete traffic flow data set through the multi-factor data obtained in real time and the corresponding traffic flow data to obtain a model training set, and update the traffic flow prediction sub-model through the online gradient descent method. The parameters of the prediction model are updated in real time using the incremental learning algorithm. The incremental learning algorithm adopts the online gradient descent method and continuously adjusts the model parameters according to the newly added data. The weighted average method is used to fuse the prediction results of factors such as aircraft types, weather conditions, holiday activities, historical data, and regional characteristics to obtain the final low-altitude traffic flow prediction value.

[0079] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A low-altitude traffic flow prediction method based on artificial intelligence, characterized in that: The following steps are involved: Obtain historical traffic flow data and preprocess it to obtain a complete traffic flow data set; Classify low-altitude aircraft and obtain the traffic flow weight coefficients corresponding to each type of low-altitude aircraft based on the complete traffic flow data set; Get the weather factor dataset corresponding to the complete traffic flow dataset; Obtaining a weather impact weight coefficient based on the weather factor dataset and the complete traffic flow dataset; Extracting holiday activity traffic flow data sets during holidays and major events from the complete traffic flow data set, and obtaining holiday impact weight coefficients based on the holiday activity traffic flow data sets; Obtain multi-factor historical data of the target area and calculate the variance of each factor corresponding to the historical data; Based on the variance of each factor, the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient are corrected; based on the corrected weight coefficient, a traffic flow prediction sub-model corresponding to each factor is constructed and trained; The low-altitude traffic flow prediction value is obtained based on the trained traffic flow prediction sub-model.

2. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: The process of obtaining a complete traffic flow dataset includes: Analyze whether there is any missing data in the historical traffic flow data. If so, extract the corresponding missing time period and area; estimate the traffic flow data within the time period according to the historical traffic flow data and the area to obtain the completed historical traffic flow data; perform outlier processing on the completed historical traffic flow data to obtain a complete traffic flow data set.

3. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: The process of obtaining the traffic flow weight coefficients corresponding to various types of low-altitude aircraft includes: The flight characteristic parameters of low-altitude aircraft are obtained, and the flight characteristic parameters are clustered and analyzed by using the K-means clustering algorithm to obtain different types of low-altitude aircraft sets; the complete traffic flow data set is divided according to the type of the low-altitude aircraft set, and the traffic flow weight coefficient is determined by calculating the proportion of data corresponding to each type of low-altitude aircraft in the complete traffic flow data set.

4. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: The process of obtaining the weather impact weight coefficient includes: The time period and area corresponding to the complete traffic flow data set are extracted, and the weather factor data set corresponding to the time period and area is obtained; after preprocessing the complete traffic flow data set and the weather factor data set, the correlation coefficient between each weather factor and the traffic flow data is calculated, and the correlation coefficient is normalized, and the weather influence weight coefficient of each weather factor on the traffic flow data is obtained based on the normalized correlation coefficient.

5. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: The process of obtaining the holiday impact weight coefficient includes: The holiday activity traffic flow data set and the normal data set during holidays and major events are extracted from the complete traffic flow data set, the traffic flow change rate is calculated based on the holiday activity traffic flow data set and the normal data set, and the traffic flow change rate is normalized to obtain the holiday impact weight coefficient.

6. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: Obtain and preprocess the historical data of multiple factors in the target area, use variance analysis to analyze the historical data corresponding to each factor, obtain the variance corresponding to each factor, and calibrate the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient based on the variance. The types of factors include aircraft types, weather conditions, and holiday activities; the weather conditions include but are not limited to temperature, humidity, wind speed, visibility, and precipitation.

7. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: The process of obtaining low-altitude traffic flow forecasts includes: Construct multiple traffic flow prediction sub-models, use the calibrated traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient as the initial parameters of the traffic flow prediction sub-models, and use an adaptive learning rate adjustment strategy to train multiple traffic flow prediction sub-models; obtain multi-factor data in real time and split them into corresponding traffic flow prediction sub-models to obtain the prediction value corresponding to each factor, use the weighted average method to fuse the prediction value corresponding to each factor, and obtain the final low-altitude traffic flow prediction value.

8. The artificial intelligence-based low-altitude traffic flow prediction method according to claim 7 is characterized in that: When the weighted average method is used to fuse the predicted values ​​corresponding to each factor, a fusion weight is assigned to each factor according to the proportion of the coefficient corresponding to each factor in the total coefficient of the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient.

9. The low-altitude traffic flow prediction method based on artificial intelligence according to claim 1 is characterized in that: It also includes updating the complete traffic flow data set through the multi-factor data and corresponding traffic flow data acquired in real time, obtaining a model training set, and updating the traffic flow prediction sub-model through an online gradient descent method.

10. A low-altitude traffic flow prediction system based on artificial intelligence, characterized in that: include: Data preprocessing module, used to obtain historical traffic flow data and preprocess it to obtain a complete traffic flow data set; An aircraft classification module is used to classify the low-altitude aircraft and obtain the traffic flow weight coefficient corresponding to each type of low-altitude aircraft based on the complete traffic flow data set; A weather impact analysis module, used to obtain a weather factor data set corresponding to a complete traffic flow data set; and obtain a weather impact weight coefficient based on the weather factor data set and the complete traffic flow data set; A holiday activity impact analysis module, used to extract holiday activity traffic flow data sets during holidays and major events from the complete traffic flow data set, and obtain a holiday impact weight coefficient based on the holiday activity traffic flow data set; The adaptive weight adjustment module is used to obtain the historical data of multiple factors in the target area and calculate the variance of the historical data corresponding to each factor; based on the variance of each factor, the traffic flow weight coefficient, weather impact weight coefficient, and holiday impact weight coefficient are corrected; and the traffic flow prediction sub-model corresponding to each factor is constructed and trained according to the corrected weight coefficient; Obtain low-altitude traffic flow prediction values ​​based on the trained traffic flow prediction sub-model; The model updating and fusion prediction module is used to update the complete traffic flow data set through the multi-factor data and corresponding traffic flow data acquired in real time, obtain the model training set, and update the traffic flow prediction sub-model through the online gradient descent method.

Images