Bird activity prediction system and method based on weather radar data
By combining weather radar and meteorological data, and utilizing machine learning and deep learning models, the problems of accuracy and timeliness in bird migration prediction have been solved, enabling precise prediction of bird migration activities, reducing bird strike incidents, and ensuring aviation safety and ecological research.
Patent Information
- Application Number
- CN202411317369.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2026-03-20
AI Technical Summary
Existing technologies are insufficient to effectively combine meteorological factors for accurate prediction of bird migration activities, leading to frequent bird strikes and affecting the accuracy and timeliness of aviation safety and ecological research.
By acquiring bird migration data and multi-source meteorological data through weather radar, and combining machine learning and deep learning models, especially convolutional neural networks (CNN), long short-term memory networks (LSTM) and attention mechanisms, data processing and prediction can be performed to achieve accurate prediction of bird migration.
It enables real-time, dynamic prediction of bird migration activities, effectively reducing the probability of bird strikes, ensuring aviation safety, and supporting ecological protection.
Smart Images
Figure CN121703772A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of aviation safety and ecological protection, particularly a bird activity prediction technology based on weather radar data. By combining machine learning and meteorological data analysis, this technology can monitor and predict the timing, route, and intensity of bird migration in real-time, mainly applied to airport bird strike prevention and flight safety management. Meanwhile, this technology can also provide data support for ecological research on bird migration. BACKGROUND
[0002] During the bird migration season, the risk of collision between birds and aircrafts significantly increases, leading to frequent bird strike incidents, which seriously threatens aviation safety. Traditional methods of monitoring bird migration mainly rely on manual observation and statistical data, which are difficult to cope with complex and changing environments, resulting in insufficient timeliness and accuracy of prediction. With the development of radar technology, weather radar has been widely used to monitor bird migration activities. By obtaining large-scale radar echo data, bird activity can be accurately recorded. However, current technology cannot effectively combine meteorological factors to accurately predict bird activity, and lacks deep analysis and modeling capabilities for data. Therefore, there is an urgent need for an efficient prediction system that combines weather radar and meteorological data to address the safety risks posed by bird migration and provide scientific decision support. SUMMARY
[0003] The present application aims to provide a bird activity prediction system and method based on weather radar data, overcoming the shortcomings of existing technology in predicting bird migration activities, particularly in terms of accuracy, timeliness, and adaptability to complex weather conditions. By combining bird migration data obtained by weather radar and multi-source meteorological data, machine learning models, and deep learning technology, the present application can accurately predict bird migration, thereby effectively reducing the probability of airport bird strike incidents, providing scientific basis for aviation safety management, and supporting ecological protection and bird migration research.
[0004] The technical purpose of the present application is achieved by the following technical solutions.
[0005] A bird activity prediction method based on weather radar data, according to the following steps:
[0006] Step 1, data collection. Obtain bird activity data and related meteorological factor data through weather radar and meteorological monitoring equipment. The core data collection of this system includes the following parts:
[0007] (1) Weather Radar Data Collection: Real-time monitoring of bird migration activities in the target area through Doppler weather radar to obtain radar echo data. The main role of weather radar is to monitor the flight height, direction and speed of birds and other information. The system uses multiple modes to accurately capture bird migration, including scanning different heights of air flow and biological reflection signals, thereby achieving dynamic capture of bird migration.
[0008] (2) Weather Data Collection: Using weather monitoring equipment or platforms to collect weather data related to bird activities, including wind speed, temperature, humidity, air pressure and solar radiation. These data can be obtained through weather stations, satellites or third-party weather services, and used to analyze the impact of weather factors on bird migration behavior.
[0009] (3) Data synchronization and storage: After data collection, the system synchronously stores radar data and weather data to ensure that data from different sources can be analyzed on a unified time axis, ensuring the accuracy of the prediction.
[0010] Step 2, data processing. In order to ensure the accuracy and stability of the subsequent prediction model, the data processing module performs a series of preprocessing on the collected data, including the following processing methods (or processing steps):
[0011] (1) Data cleaning: During data collection, there may be data noise, outliers or missing values. The system first removes invalid data and abnormal data through data cleaning algorithms. Abnormal data includes invalid radar echoes caused by radar failure, weather equipment false positives or environmental interference.
[0012] (2) Data alignment: Since the time resolution of radar data and weather data is not consistent, the system aligns them on the same time axis through data alignment technology. This step ensures that weather conditions and bird migration activity data can be accurately matched to improve the timeliness and accuracy of the prediction.
[0013] (3) Missing value processing: In some cases, there may be missing data during data collection. The system uses machine learning algorithms to interpolate missing values. The specific method uses an interpolation algorithm based on machine learning to ensure that the processed data is complete and reliable. XGBoost, LightGBM and CatBoost integrated learning models are used for missing value interpolation.
[0014] (4) Data standardization: Since different weather factors have different dimensions (such as temperature and wind speed units), the system needs to standardize the data to convert different units of data to a unified numerical range. Standardization helps the model better predict and analyze, ensuring the stability and consistency of the model input.
[0015] Step 3, Model Training and Prediction. After data processing is complete, the system enters the core model training and prediction phase, using a variety of machine learning and deep learning algorithms to analyze and predict the processed data.
[0016] (1) Deep Learning Model Prediction: To further improve the accuracy of time series prediction, the system introduces convolutional neural networks (CNN), long short-term memory networks (LSTM), and attention mechanism (Attention) models to specifically handle time series data. Through these deep learning models, the system can capture the long-term and short-term effects of weather factors on bird migration. By identifying complex temporal dependencies, the model can achieve dynamic prediction of bird activity.
[0017] (2) Fourier Transform and Data Smoothing: As there may be noise interference in the collected data, the system uses Fourier transform to smooth the data to eliminate high-frequency noise signals. The main purpose of this step is to ensure the stability of the model prediction, making the prediction results smoother and more reliable, and avoiding prediction errors caused by noise.
[0018] (3) Model Optimization and Validation: The system optimizes and validates the model through cross-validation methods to ensure its robustness and accuracy. Specifically, the system selects the best model structure through parameter tuning and uses training data and validation data sets to validate the model, ensuring its wide applicability under different weather conditions.
[0019] Step 4, Result Output: After model prediction, the system visualizes the results and provides real-time warning information to users.
[0020] The bird activity prediction system based on weather radar data using the above prediction method includes a data acquisition module, a data processing module, a prediction module, and an output module, wherein:
[0021] The data acquisition module acquires large-scale, multi-dimensional bird migration-related data through weather radar; uses meteorological monitoring equipment or platforms to collect meteorological data related to bird activity, and synchronously stores radar data and meteorological data;
[0022] The data processing module standardizes, cleans, and handles missing values of the collected radar data to ensure the accuracy and integrity of the data;
[0023] The prediction module uses the processed data to establish a bird activity intensity prediction model based on deep learning models, combining meteorological data such as wind speed, temperature, humidity, radiation, and other influencing factors to predict bird migration activity intensity;
[0024] The output module outputs the prediction results through a visual interface, providing information such as the time, route and intensity of bird activity.
[0025] The present application realizes accurate prediction of bird migration activity through the bird activity prediction system and method based on weather radar data, especially provides real-time and dynamic prediction information by combining multi-source meteorological data and advanced machine learning algorithms, effectively reduces the occurrence of bird strike events and ensures aviation safety. BRIEF DESCRIPTION OF DRAWINGS
[0026] Figure 1 is a flow chart of a bird activity prediction system and method based on weather radar data of the present application.
[0027] Figure 2 is a XGBoost model hyperparameter search diagram in Zhangjiakou City in the present application.
[0028] Figure 3 is a XGBoost model hyperparameter search diagram in Changsha City in the present application.
[0029] Figure 4 is a CNN-LSTM-Attention model structure diagram in the present application.
[0030] Figure 5 is a comparison diagram of the predicted value and the true value of the spring migration MTR value in Zhangjiakou City in the present application.
[0031] Figure 6 is a comparison diagram of the predicted value and the true value of the spring migration MTR value in Changsha City in the present application. DETAILED DESCRIPTION
[0032] The present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.
[0033] From the attached Figure 1 It can be seen that the present application provides a bird activity prediction method based on weather radar data, which is carried out according to the following steps:
[0034] Step 1, Data Collection. The data collection module is used to collect bird activity data and meteorological data, which are then input into the system for analysis. The bird activity data is extracted from weather radar data using the bioRad software package, a weather radar data analysis and visualization tool designed for biological research. The meteorological data includes temperature, humidity, barometric pressure, precipitation, meridional wind, zonal wind, surface wind speed, wind direction, surface horizontal radiation, normal direct radiation, and scattered radiation. The data collected from Zhangjiakou, Shijiazhuang, Zhengzhou, Yichang, and Changsha from March to May 2023 is used as the data collection object. The following tables show the data collected from Zhangjiakou and Changsha.
[0035] Table 1. Partial climate and bird activity data in Zhangjiakou
[0036]
[0037]
[0038] Table 2. Partial climate and bird activity data in Changsha
[0039]
[0040] Step 2, Data Processing. To ensure the accuracy and stability of the subsequent prediction model, the data processing module performs a series of preprocessing on the collected data, including the following processing methods (or processing steps):
[0041] Step 21, Data Cleaning: During data collection, data noise, outliers, or missing values may occur. The system first removes invalid data and outliers through data cleaning algorithms. Outliers include invalid radar echoes caused by radar malfunction, false reports from meteorological equipment, or environmental interference.
[0042] Step 22, Data Alignment: Due to the inconsistent time resolution of radar data and meteorological data, the system ensures that the two are aligned on the same time axis through data alignment technology. This step ensures that meteorological conditions and bird migration activity data can be accurately matched to improve the timeliness and accuracy of the prediction.
[0043] Step 23, Missing Value Processing: In some cases, there may be missing data during data collection. The system uses machine learning algorithms to interpolate missing values. The specific method uses an interpolation algorithm based on machine learning to ensure that the processed data is complete and reliable.
[0044] The system uses XGBoost, LightGBM, CatBoost and other ensemble learning models to train the data processed by steps 21 and 22, that is, to use the interpolation algorithm of machine learning to fill in the missing values. These models can effectively handle complex nonlinear relationships, especially in large-scale, multi-dimensional data, and can accurately identify the complex relationship between meteorological factors and bird migration intensity. Preliminary reading and deleting irrelevant columns and converting categorical features are performed by the Pandas library built into Python. The two categorical variables of month and time are independently encoded to meet the needs of model regression training. Adjust the column order of the data and apply a circular shift operation to ensure that the data structure meets the model input requirements. During model training, the data set is divided into a training set and a test set, where the training set is further divided into a training subset and a validation subset, with a specific ratio of 70:15:15. The test set is used as the final evaluation standard to test the model's performance in a real environment and ensure the reliability and practical value of the model. The initial search range and step size of the hyperparameters are set based on experience and refined gradually during optimization to balance the search space and computational efficiency. In the experimental process, the Optuna library built into Python is used for hyperparameter tuning. Optuna is a hyperparameter tuning framework based on Bayesian optimization, and its optimization objective is as shown in the formula.
[0045] y = f(x) + ε
[0046] where y is the target function value, x is the hyperparameter, f(x) is the target function, and ε is the noise term.
[0047] To ensure the approximation of the optimal hyperparameters as much as possible, the search number of Optuna is set to 150, and the target function is set to the R 2 score of the model on the validation subset. The search range of the parameters of each model is shown in Table 4, Table 5, and Table 6. Table 7 shows the best parameters and the best R 2 score of the XGBoost model experiment in each city. The visualization of the non-optimal model's hyperparameter search process in the XGBoost interpolation model experiment in Zhangjiakou City and Changsha City is shown in Figure 2 、 Figure 3 .
[0048] Table 4 XGBoost model hyperparameter range
[0049]
[0050]
[0051] Table 5 LGBM model hyperparameter range
[0052] Parameter Search range Step size num_leaves 20-120 1 max_depth 5-20 1 n_estimators 100-1500 50 learning_rate 0.01-0.15 0.01 subsample 0.5-0.1 0.1 colsample_bytree 0.5-0.1 0.1 min_child_samples 1-20 1 min_child_weight 1-10 1 reg_lambda 10 -5 -10]]> 10-fold change values reg_alpha 10 -5 -10]]> 10-fold change values
[0053] Table 6 CatBoost model hyperparameter ranges
[0054] Parameter Search range Step size depth 1-10 1 iterations 1000-2000, step size 100 100 learning_rate 0.15-0.3 0.01 l2_leaf_reg 5-15 1 colsample_bylevel 0.5-0.1 0.1 random_strength 0-50 1 bagging_temperature 0-10 1
[0055] Table 7 XGBoost model experiment best parameters and best R 2 Score
[0056] Zhangjiakou Shijiazhuang Zhengzhou Yichang Changsha max_depth 12 7 3 8 18 n_estimators 800 1700 1500 1050 1250 subsample 0.01 0.11 0.0699 0.04 0.05 colsample_bytree 6 2 7 5 4 learning_rate 0.0028 0.5402 0.0569 8.17*10-5 0.1861 min_child_weight 0.0090 0.0003 7.3166 0.2255 0.0026 reg_alpha 0.8808 0.5778 0.7268 0.7579 0.6273 reg_lambda 0.9599 0.9872 0.8018 0.6647 0.7408 Best R 2 score]]> 0.9239 0.9276 0.9262 0.8951 0.8305
[0057] Step 24, data standardization: due to the different dimensions of different meteorological factors (such as the units of temperature and wind speed are different), the system needs to standardize the data, and convert the data of different units into a unified numerical range. Standardization helps the model to better predict and analyze, and ensures the stability and consistency of the model input.
[0058] Step 3, model training and prediction. After data processing is completed, the system enters the core model training and prediction stage, and uses various machine learning and deep learning algorithms to analyze and predict the processed data.
[0059] Step 31, deep learning model prediction: in order to further improve the accuracy of time series prediction, the system introduces convolutional neural network (CNN), long short-term memory network (LSTM) and attention mechanism (Attention) model, which are specially designed to process time series data. The CNN-LSTM-Attention model structure diagram is as follows Figure 4The system can capture both long-term and short-term effects of meteorological factors on bird migration through these deep learning models. The models can achieve dynamic prediction of bird activity by identifying complex time-dependent relationships. In the initial model selection stage, step sizes of 24, 48, and 72 are chosen to balance the complexity and computational cost of the model while ensuring that the model can capture the main features in the data. Additionally, the CLA model architecture first uses convolutional layers for feature extraction, downsamples through max-pooling layers, then uses LSTM layers to capture time series information, and finally introduces an attention mechanism layer to dynamically allocate weights. Convolutional layers are used to extract local features from input data, and two layers of convolutional layers are used in the model. Some parameter settings are as follows: 32 convolutional kernels are set in the first layer to extract 32 different features. The convolutional kernel size is 3, i.e., three time steps of data are covered in each convolution operation. The output dimension is kept the same as the input to avoid loss of edge information. The ReLU activation function is used to enhance non-linear characteristics. The second layer sets 64 convolutional kernels to extract more features, and the convolutional kernel size is 2, with other parameters the same as the first layer of convolutional layers. Pooling layers are used to reduce feature dimensions and computational load to prevent overfitting. The two pooling layers have the same parameters, with a pooling window size of 2, i.e., two time steps of data are taken for pooling each time. Two LSTM layers are used to capture long-term dependencies in time series, and the ReLU activation function is used to enhance non-linear characteristics. The attention mechanism layer is used to calculate the attention weight of each time step. The number of memory cells in the LSTM layer and the attention mechanism layer is part of the debugging. Finally, two fully connected layers are set to further process features and output prediction results.
[0060] Step 32, Fourier transform and data smoothing: To eliminate high-frequency noise signals, the system uses Fourier transform to smooth the data. The main purpose of this step is to ensure the stability of the model prediction, making the prediction result smoother and more reliable, and avoiding prediction errors caused by noise.
[0061] Step 33, model optimization and verification: During model training, the Adam optimizer is used to optimize model parameters, and the mean square error (MSE) is chosen as the loss function. Through 50 training iterations, the training loss and validation loss of each iteration are recorded to evaluate the training effect and generalization ability of the model. The evaluation results of each city are shown in Table 8.
[0062] Table 8
[0063]
[0064] Step 4, result output. After model prediction, the system visualizes the results and provides real-time warning information to users.
[0065] Step 41, visualization output: the system provides an intuitive display of the prediction results through the user interface. The output includes important information such as the intensity, route, and time of bird migration, and correlation analysis with meteorological data. The system supports multiple graphical display methods, such as migration heat maps, time series trend charts, route maps, etc., to help users intuitively understand the spatiotemporal changes of bird activity.
[0066] Step 42, real-time warning and feedback: in aviation safety management, the system can generate real-time warning information based on the prediction results, especially in areas and time periods with high bird activity intensity. The system will prompt users to take necessary preventive measures, such as adjusting flight times, planning flight routes, and starting bird repelling devices. In addition, the system also supports feedback on the prediction results, and users can modify the prediction model based on actual conditions to further improve the accuracy of the prediction.
[0067] As shown in Figure 5 and 6 , the predicted and actual values of the spring migration MTR values in Zhangjiakou and Changsha are very close.
[0068] The above is an exemplary description of the present application. It should be noted that any simple modification, modification or other equivalent replacement that does not deviate from the core of the present application and can be easily replaced by those skilled in the art without creative effort falls within the scope of protection of the present application.
Claims
1. A method for predicting bird activity based on weather radar data, characterized in that, Follow these steps: Step 1, Data Acquisition: Obtain bird activity data and related meteorological factor data through weather radar and meteorological monitoring equipment, and store them synchronously; Step 2: Perform the following preprocessing on the collected data: (1) Data cleaning: Remove invalid and abnormal data using data cleaning algorithms; (2) Data alignment: Data alignment technology is used to ensure that radar data and meteorological data are aligned on the same time axis, and to ensure that meteorological conditions and bird migration data can be accurately matched to improve the timeliness and accuracy of forecasts; (3) Missing value handling: A machine learning-based interpolation algorithm was used to impute missing values; (4) Data standardization: Convert data from different units into a unified numerical range to ensure the stability and consistency of model input; Step 3, Model Training and Prediction After data processing is complete, the model training and prediction phase begins, where various machine learning and deep learning algorithms are used to analyze and predict the processed data. (1) Deep learning model prediction: A model combining convolutional neural networks, long short-term memory networks and attention mechanisms is introduced to specifically process time series data; by learning the model, the long-term and short-term effects of meteorological factors on bird migration can be captured, and by identifying complex time dependencies, dynamic prediction of bird activities can be achieved. (2) Fourier transform and data smoothing: The system uses Fourier transform to smooth the data in order to eliminate high-frequency noise signals. (3) Model optimization and validation: Select the best model structure through parameter tuning, and validate the model using training data and validation datasets to ensure its wide applicability under different meteorological conditions. Step 4: Results Output. After model prediction, the system visualizes the results.
2. The bird activity prediction method based on weather radar data according to claim 1, characterized in that, By using Doppler weather radar to monitor bird migration activities in the target area in real time, radar echo data is obtained to acquire bird activity data, such as the birds' flight altitude, direction, and speed; relevant meteorological factor data include wind speed, temperature, humidity, air pressure, and solar radiation.
3. The method for predicting bird activity based on weather radar data according to claim 1, characterized in that, Missing value imputation was performed using an ensemble learning model of XGBoost, LightGBM, and CatBoost.
4. The method for predicting bird activity based on weather radar data according to claim 1, characterized in that, Bird activity intensity is predicted by combining temporal and meteorological characteristics. The temporal characteristics include the diurnal cycle, seasonal variation and solar terms, while the meteorological characteristics include wind speed (V), air pressure (P), temperature (T), humidity (H) and radiation (R).
5. The method for predicting bird activity based on weather radar data according to claim 1, characterized in that, The model uses two convolutional layers. The first layer has 32 convolutional kernels to extract 32 different features, with a kernel size of 3, meaning each convolution operation covers data from three time steps. The ReLU activation function is used to enhance non-linear characteristics. The second layer has 64 convolutional kernels with a kernel size of 2, and other parameters are the same as the first convolutional layer. The two pooling layers have the same parameters, with a pooling window size of 2, meaning that data from two time steps is pooled each time. Two LSTM layers are used to capture long-term dependencies in the time series, and the ReLU activation function is used to enhance non-linear characteristics.
6. A bird activity prediction system based on weather radar data using the prediction method described in any one of claims 1-5, characterized in that, It includes a data acquisition module, a data processing module, a prediction module, and an output module, among which: The data acquisition module obtains large-scale, multi-dimensional bird migration-related data through weather radar; it also collects meteorological data related to bird activities using meteorological monitoring equipment or platforms, and stores radar data and meteorological data synchronously. The data processing module standardizes, cleans, and handles missing values in the collected radar data to ensure the accuracy and integrity of the data. The prediction module is based on a deep learning model. It uses the processed data to build a prediction model for the intensity of bird activity and combines meteorological data and other influencing factors to predict the intensity of bird migration activities. The output module outputs the prediction results through a visualization interface, providing information such as the time, route, and intensity of bird activity.
7. The bird activity prediction system based on weather radar data according to claim 6, characterized in that, By using Doppler weather radar to monitor bird migration activities in the target area in real time, radar echo data is obtained to acquire bird activity data, such as the birds' flight altitude, direction, and speed; relevant meteorological factor data include wind speed, temperature, humidity, air pressure, and solar radiation.
8. The bird activity prediction system based on weather radar data according to claim 6, characterized in that, Missing value imputation was performed using an ensemble learning model of XGBoost, LightGBM, and CatBoost.
9. The bird activity prediction system based on weather radar data according to claim 6, characterized in that, This deep learning model incorporates a combination of convolutional neural networks, long short-term memory networks, and attention mechanisms. It employs two convolutional layers: the first layer has 32 kernels to extract 32 different features, with a kernel size of 3, meaning each convolution operation covers data from three time steps, and uses the ReLU activation function to enhance non-linearity; the second layer has 64 kernels with a kernel size of 2, and other parameters are the same as the first layer. The two pooling layers have identical parameters, with a pooling window size of 2, meaning pooling is performed on data from two time steps at a time. Two LSTM layers are used to capture long-term dependencies in the time series, and the ReLU activation function is used to enhance non-linearity.