Aircraft taxiing time prediction method based on deep space-time diagram convolutional neural network
Through the method of deep space-time graph convolutional neural network combined with gated loop units, the problem of insufficient accuracy of aircraft taxi time prediction in the prior art is solved, and more efficient taxi time prediction is achieved, which is suitable for complex airport network environments.
Patent Information
- Application Number
- CN202411599483.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-02-28
AI Technical Summary
The prior art is insufficiently accurate in aircraft taxi-time prediction, making it difficult to deal with complex airport taxi-way networks and dynamic factors.
The method based on deep space-time graph convolution neural network is adopted, combining the airport taxiway network structure, aircraft sliding dynamic data and a variety of real-time factors, and the spatial dependencies are captured through the graph convolutional neural network, and the gated cycle unit is used to model the dynamic changes in time.
It significantly improves the accuracy and robustness of aircraft taxi time prediction, can more effectively capture aircraft taxi characteristics under complex airport network structures, and improves prediction accuracy and stability.
Smart Images
Figure CN119541277A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of aviation transportation and intelligent transportation, and specifically to an aircraft taxiing time prediction method based on a deep spatiotemporal graph convolutional neural network. Background Art
[0002] As the global demand for air transport continues to grow, the complexity of airport ground operations and the challenges of aircraft taxiing management are increasing. During the taxiing process of aircraft at the airport, accurate prediction of taxiing time is a key factor in ensuring airport operation efficiency, reducing delays, optimizing resource allocation and improving passenger experience.
[0003] Traditional aircraft taxi time prediction methods are usually based on rule models or statistical methods, which have many limitations when faced with complex airport network structures and dynamically changing operating environments. Traditional methods are difficult to effectively capture the spatial dependencies in the airport taxiway network and the mutual influence between different taxi paths; the temporal dynamic change characteristics in taxi time, such as aircraft queues and fluctuations in taxi speed affected by weather or traffic flow, are often difficult to accurately describe through simple statistical models. Therefore, the accuracy and robustness of taxi time prediction in existing technologies, especially for large-scale complex airport operation scenarios, still have a lot of room for improvement.
[0004] With the development of artificial intelligence and data science, especially the widespread application of graph convolutional neural networks (GCN) in processing complex network structure data, spatio-temporal graph neural networks (ST-GCN) combined with time series modeling have shown good application prospects in fields with strong spatiotemporal dependence. This type of model can effectively combine spatial graph structure and time series data for joint modeling, providing a new idea and method for time prediction in complex dynamic systems (such as airport taxi networks).
[0005] However, how to use deep learning technology, especially the spatiotemporal graph convolutional neural network model, to more accurately capture the spatial and temporal dependence characteristics of aircraft taxiing is still a research hotspot and technical problem in the current field of aircraft taxiing time prediction. Therefore, the present invention designs an aircraft taxiing time prediction method that can combine the airport taxiway network structure, aircraft taxiing dynamic data and multiple real-time factors. Summary of the invention
[0006] In order to solve the problems of insufficient accuracy in aircraft taxi time prediction in the existing technology and difficulty in dealing with the complexity and dynamic factors of airport taxiway networks, an aircraft taxi time prediction method based on deep spatiotemporal graph convolutional neural network was proposed.
[0007] To achieve the above object, the present invention adopts the following technical solutions:
[0008] A method for predicting aircraft taxiing time based on a deep spatiotemporal graph convolutional neural network comprises the following steps:
[0009] Step 1: Input the original data set consisting of the airport adjacency matrix, aircraft operation characteristics, and spatiotemporal data of historical taxiing time into the prediction model;
[0010] Step 2: Standardize the input raw data set to avoid the influence of outliers and extreme values on the results;
[0011] Step 3: Construct the airport operation network topology map and use the graph convolutional neural network to model the spatial dependency of the airport taxiway network. According to the network topology of the airport taxiway, perform convolution operations on the information of the node and its neighboring nodes to capture the spatial correlation between the nodes on the taxiway.
[0012] Step 4: Combine the gated recurrent unit to iterate and optimize the aircraft taxi time prediction model, calculate the prediction error of the model by setting the loss function, and use the optimization algorithm to adjust the model parameters, including weights and biases, to minimize the loss function;
[0013] Step five: Conduct a systematic evaluation of the prediction results to verify the accuracy and reliability of the model, analyze the prediction error through different evaluation indicators, and ensure the applicability of the model in actual scenarios.
[0014] Furthermore, the specific content of step one is:
[0015] The airport adjacency matrix is used to describe the connection relationship between nodes in the airport taxiway network. The values in the adjacency matrix represent the mutual relationship between the taxiway path nodes;
[0016] Aircraft operation characteristics, including aircraft type, model, and flight operation time period;
[0017] Historical taxi time: obtain the taxi time records of aircraft at the airport in different time periods through the airport detection system, and provide the historical characteristics of taxi time for the model.
[0018] Furthermore, the specific steps of step 2 are:
[0019] (2.1) Standardize the initial data set using the mean and standard deviation to ensure that all input features are in the same numerical range to reduce the bias of the model caused by different magnitudes of certain features;
[0020] (2.2) Remove or limit outliers and extreme values to reduce their impact on model training. This can improve the adaptability of the model to the input data and make the model more stable for normal data and abnormal data.
[0021] Furthermore, the specific steps of step three are:
[0022] (3.1) Taking the airport taxiways as nodes and the direct connections between taxiways as edges, a topological graph of the airport taxiway path is constructed. The nodes on the topological graph represent the key locations in the taxiway network, and the edges represent the direct connections between taxiways.
[0023] (3.2) Use GCN to perform convolution operations on the nodes in the topological graph. Through graph convolution operations, the spatial correlation information in the airport taxi network is extracted. The convolution operation transmits information through the adjacency matrix, so that each node can aggregate the information of its neighboring nodes and learn the spatial dependency between the nodes on the taxi path.
[0024] (3.3) In each layer of graph convolution, the nodes update their feature values and gradually learn deeper spatial information to improve the prediction ability of the gliding time.
[0025] Furthermore, the specific steps of step 4 are:
[0026] (4.1) Through the gated recurrent unit, the adjustment model adjusts the parameters according to the feedback of the loss function in each round of iteration to reduce the error of the prediction results.
[0027] (4.2) Perform multiple rounds of iterative training on the training set, and gradually optimize the model parameters to ensure that the model fits the training data well and has good generalization effect on the validation set.
[0028] Furthermore, the specific content of step five is:
[0029] Different evaluation indicators are used to systematically analyze the prediction results and verify the accuracy and reliability of the model. The specific evaluation indicators include: mean square error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE).
[0030] The innovation of the present invention is that by combining the graph convolutional neural network with the time series model, the spatiotemporal characteristics of the aircraft taxiing time are modeled. The graph convolutional neural network is used to capture the spatial dependencies of nodes in the airport taxiway network, while the gated recurrent unit is responsible for modeling the temporal dynamic changes. The combination of the two can effectively capture the aircraft taxiing characteristics under the complex network structure of the airport. In addition, a spatial attention mechanism is added to identify and weight the nodes in the taxiing path that have a greater impact on the taxiing time, and a temporal attention mechanism can focus on the key time steps in the taxiing process, thereby improving the model's ability to focus on key taxiing features and further optimizing the model's prediction accuracy and stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a flow chart of the spatiotemporal graph convolutional neural network prediction method in the present invention;
[0032] Figure 2 It is a module block diagram of the glide time prediction model in the present invention;
[0033] Figure 3 is a state transition diagram of a gated cycle unit in the present invention;
[0034] Figure 4 A comparison diagram of the predicted taxiing time and the actual taxiing time of the present invention;
[0035] Figure 5 It is a schematic diagram of loss value in the prediction process of the present invention. DETAILED DESCRIPTION
[0036] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.
[0037] like Figure 1 As shown, an aircraft taxiing time prediction method based on a deep spatiotemporal graph convolutional neural network provided by an embodiment of the present invention includes the following steps:
[0038] Step 1: Input the original data set consisting of the airport adjacency matrix, aircraft operation characteristics, and spatiotemporal data of historical taxiing time into the prediction model. The specific contents of each data set are as follows:
[0039] (1.1) Airport adjacency matrix, which is used to describe the connection relationship between nodes in the airport taxiway network. The values in the adjacency matrix represent the mutual relationship between the taxiway path nodes;
[0040] (1.2) Aircraft operation characteristics, including aircraft type, model, and flight operation time period;
[0041] (1.3) Historical taxi time: The taxi time records of aircraft at the airport in different time periods are obtained through the airport detection system to provide the historical characteristics of taxi time for the model.
[0042] Step 2: Standardize the input raw data set to avoid the influence of outliers and extreme values on the results. The specific contents are as follows:
[0043] (2.1) The feature data is normalized by using the mean and standard deviation to convert the data into a standard normal distribution with a mean of 0 and a variance of 1. The purpose of standardization is to scale all input features to the same numerical range to prevent some features from having too much impact on model training due to their large magnitude. The specific formula for data normalization is as follows:
[0044]
[0045] Among them, x′ i is the standardized eigenvalue, is the mean of the feature, is the standard deviation of the feature.
[0046] (2.2) The purpose of outlier processing is to remove or limit extreme data points to reduce the adverse effects of outliers on model training and ensure that the prediction model is more stable for normal data and abnormal data. The data processing method based on standard deviation outlier removal sets an outlier detection threshold and regards data whose distance from the mean exceeds the threshold as an outlier. If the eigenvalue x i Satisfy|x i -u|>kσ, then x i It is an outlier and the data is removed.
[0047] Step 3: Construct the airport operation network topology map and use the graph convolutional neural network to model the spatial dependency of the airport taxiway network. According to the network topology of the airport taxiway, perform convolution operations on the information of the node and its neighboring nodes to capture the spatial correlation between the nodes on the taxiway. The calculation formulas used include:
[0048] H I+1 =f(H (I) ,A)
[0049] f(H (I) ,A)=σ(AH (I) W (I) )
[0050] Among them, A represents the adjacency matrix, H (0) =X and H L=Z, L is the number of layers, σ is a nonlinear activation function. After transformation, it is finally expressed as follows:
[0051]
[0052] Among them, H (I) represents the feature representation of the lth layer, A is the adjacency matrix, D is the degree matrix, and W (I) It represents the weight matrix of layer l, and σ is a nonlinear activation function.
[0053] Step 4: Iterate and optimize the aircraft taxi time prediction model in combination with the gated recurrent unit, calculate the prediction error of the model by setting a loss function, and adjust the model parameters, including weights and biases, using an optimization algorithm to minimize the loss function. The calculation formula used by the gated recurrent unit includes:
[0054] z t =σ(W z ·[h t-1 ,x t ])
[0055] z t =σ(W z ·[h t-1 ,x t ])
[0056]
[0057] Among them, W z is a weight matrix for x t and h t-1 The concatenated matrix is linearly transformed, and then the larger value of the multiplication of the two matrices is input into the sigmoid function, and r is obtained. t Value, which is used in the formula for candidate hidden states and is also considered as the memory state at each time step. represents the candidate hidden state, h t Represents the hidden state passed to the next moment, r t represents the reset gate, z t Represents the update gate, σ is the sigmoid function, through which the data can be converted into a numerical value within a range.
[0058] Step 5: Conduct a systematic evaluation of the prediction results to verify the accuracy and reliability of the model. Use different evaluation indicators to analyze the prediction error to ensure the applicability of the model in actual scenarios. The specific formulas of the analysis indicators are as follows:
[0059]
[0060] Reference Figure 2 , showing the overall process of the aircraft taxiing time prediction model in the present invention, the specific steps are as follows
[0061] Step 1: Extract spatial feature data. The spatial structure of the airport taxi network is modeled and features are extracted using a graph convolutional neural network. The GCN module extracts the spatial features of the airport taxi network through the adjacency relationship of each node on the taxi path to capture the spatial dependency of the taxi path.
[0062] Step 2: Extract time feature data. The gated recurrent unit module is used to model the time series characteristics of the aircraft taxiing process. GRU extracts the time dynamic changes of the node states during the taxiing process and obtains the time series dependency characteristics of the taxiing time.
[0063] Step 3: Feature data fusion encoding: The spatial features extracted by the graph convolutional neural network are fused with the temporal features extracted by the gated recurrent unit to generate a combined high-dimensional feature representation as the input of the subsequent modules.
[0064] Step 4: Attention mechanism processing. The attention mechanism module is used to perform weighted calculation on the fused features, and the key nodes and important time steps that have a greater impact on the taxiing time prediction are highlighted through appropriate weight allocation.
[0065] Step 5: Decoding the prediction results. The features processed by the attention mechanism are input into the decoding layer to generate the final aircraft taxi time prediction results.
[0066] Step 6: Result output: The final prediction result of the model, i.e., the aircraft taxiing time, is output through the decoding layer to provide data support for airport taxiing scheduling management and resource allocation decisions.
[0067] Reference Figure 3 , shows the state transition diagram of the gated cycle unit in the present invention, the specific contents are as follows:
[0068] The hidden state of the current layer It will be passed to the next layer of GRU units. In this way, the next layer of GRU can use the information passed by the previous layer to calculate its own hidden state and output, and pass it layer by layer until it reaches the last layer. The inter-layer state transfer process of GRU dynamically weights historical information and new information through update gates and reset gates, updates hidden states layer by layer and passes them downward, so that the entire GRU network can capture the long-term and short-term dependencies of sequence data in a multi-layer structure.
[0069] Reference Figure 4 , showing a comparison chart between the predicted taxiing time and the actual taxiing time in the present invention, the specific contents are as follows:
[0070] The upper part shows the curve of the actual taxiing time of the aircraft over time, reflecting the time changes affected by various factors during the taxiing process. The lower part shows the curve of the taxiing time predicted by the model over time. Compared with the actual taxiing time curve, the predicted taxiing time curve generally has a similar trend, indicating that the model can capture the changing pattern of taxiing time to a certain extent. At some time points, there is a certain deviation between the predicted taxiing time and the actual taxiing time, which may be related to factors such as weather and congestion on the taxiing path. The distribution of this error on the entire timeline can help identify the direction of model improvement.
[0071] Reference Figure 5 , shows a schematic diagram of the loss value in the prediction process of the present invention, the specific contents are as follows:
[0072] In the initial stage of training (about 0 to 200 iterations), the loss value drops rapidly, indicating that the model effectively reduces the prediction error by continuously adjusting weights and parameters in the early stage.
[0073] In the mid-term stage (about 200 to 600 iterations), the loss value continues to decrease, but the rate of decrease gradually slows down, indicating that the model is gradually approaching the optimal state and the prediction accuracy is gradually improving.
[0074] In the later stage (after about 600 iterations), the loss value tends to stabilize, indicating that the model is close to convergence, the loss value fluctuates in a small range, and the prediction accuracy reaches a stable level.
[0075] The loss value diagram intuitively shows the learning process and performance of the model in the taxiing time prediction task, indicating that the model can stably provide relatively accurate taxiing time prediction results after training. This provides data support for the reliability and accuracy of the model in subsequent practical applications.
[0076] In the evaluation of aircraft taxi time prediction, the missing values in the prediction directly affect the evaluation results of the prediction model, and are therefore the key to determining whether the model is suitable for practical applications. Smaller loss values not only reflect the accuracy of the model, but also reflect the value of the model in multiple actual operational indicators, such as punctuality, fuel efficiency, operational efficiency, and cost control. Therefore, optimizing the loss value of the taxi time prediction model is the key to improving the overall operations of airports and airlines.
[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network, characterized in that: The following steps are involved: Step 1: Input the original data set consisting of the airport adjacency matrix, aircraft operation characteristics, and spatiotemporal data of historical taxiing time into the prediction model; Step 2: Standardize the input raw data set to avoid the influence of outliers and extreme values on the results; Step 3: Construct the airport operation network topology map and use the graph convolutional neural network to model the spatial dependency of the airport taxiway network. According to the network topology of the airport taxiway, perform convolution operations on the information of the node and its neighboring nodes to capture the spatial correlation between the nodes on the taxiway. Step 4: Combine the gated recurrent unit to iterate and optimize the aircraft taxi time prediction model, calculate the prediction error of the model by setting the loss function, and use the optimization algorithm to adjust the model parameters, including weights and biases, to minimize the loss function; Step five: Conduct a systematic evaluation of the prediction results to verify the accuracy and reliability of the model, analyze the prediction error through different evaluation indicators, and ensure the applicability of the model in actual scenarios.
2. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 1, characterized in that: In the step 1, the input original data specifically includes: (1.1) Airport adjacency matrix, which is used to describe the connection relationship between nodes in the airport taxiway network. The values in the adjacency matrix represent the mutual relationship between the taxiway path nodes; (1.2) Aircraft operation characteristics, including aircraft type, model, and flight operation time period; (1.3) Historical taxi time: The taxi time records of aircraft at the airport in different time periods are obtained through the airport detection system to provide the historical characteristics of taxi time for the model.
3. The aircraft taxiing time prediction method based on deep spatiotemporal graph convolutional neural network according to claim 1, characterized in that: The specific steps of step 2 are: (2.1) Standardize the initial data set using the mean and standard deviation to ensure that all input features are in the same numerical range to reduce the bias of the model caused by different magnitudes of certain features; (2.2) Remove or limit outliers and extreme values to reduce their impact on model training. This can improve the adaptability of the model to the input data and make the model more stable for normal data and abnormal data.
4. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 1, characterized in that: The calculation formula used in step 3 includes: H I+1 =f(H (I) ,A) f(H (I) ,A)=σ(AH (I) W (I) ) Among them, A represents the adjacency matrix, H (0) =X and H L =Z, L is the number of layers, σ is a nonlinear activation function. After transformation, it is finally expressed as follows: Among them, H (I) represents the feature representation of the lth layer, A is the adjacency matrix, D is the degree matrix, and W (I) It represents the weight matrix of layer l, and σ is a nonlinear activation function.
5. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 1, characterized in that: The specific steps of step 4 are: (4.1) The gated recurrent unit is used to process the characteristics of each time point during the taxiing process of the aircraft. The update gate and reset gate are set to capture the dependency in the time dimension, so that the model can effectively retain and transmit the key information in the taxiing time series data, thereby accurately predicting the taxiing time of the aircraft; (4.2) Add an attention mechanism to the aircraft taxi time prediction process and dynamically weight the features of each time point on the taxi path to improve the ability to focus on key time points and features. By setting an appropriate weight distribution strategy, the time nodes and information features on the taxi path that have a greater impact on the prediction results are identified and focused on, thereby improving the prediction accuracy of the model.
6. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 5, characterized in that: The calculation formula used by the gated recurrent unit includes: z t =σ(W z ·[h t-1 ,x t ]) z t =σ(W z ·[h t-1 ,x t ]) Among them, W z is a weight matrix for x t and h t-1 The concatenated matrix is linearly transformed, and then the larger value of the multiplication of the two matrices is input into the sigmoid function, and r is obtained. t Value, which is used in the formula for candidate hidden states and is also considered as the memory state at each time step. represents the candidate hidden state, h t Represents the hidden state passed to the next moment, r t represents the reset gate, z t Represents the update gate, σ is the sigmoid function, through which the data can be converted into a numerical value within a range.
7. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 5, characterized in that: The calculation formula used by the attention mechanism includes: in, They represent t i The time step and t j All spatiotemporal features of the time step are represented. After obtaining the attention score, the future time step adaptively selects the relevant features of the past time step and converts the historical time series into the future time series representation. is the future time step t j Attention score.
8. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 1, characterized in that: In step five, different evaluation indicators are used to systematically analyze the prediction results and verify the accuracy and reliability of the model. The specific evaluation indicators include: mean square error, mean absolute error, mean absolute percentage error and root mean square error.
9. The method for predicting aircraft taxiing time based on deep spatiotemporal graph convolutional neural network according to claim 8, characterized in that: The specific formula of the evaluation index is as follows: Among them, y i is the ith actual taxiing time, is the i-th predicted taxiing time, n is the number of prediction samples, and MSE reflects the overall size of the prediction error. Predictions with large errors will significantly increase this value, so it is more sensitive to outliers; MAE provides the average absolute size of the deviation between the predicted value and the actual value. Compared with MSE, MAE is less sensitive to outliers and can more directly reflect the average level of prediction error. MAPE indicates the relative size of the forecast error and can provide an intuitive measure of the error size in percentage form, but it is more sensitive to data points with smaller actual values, so it is necessary to balance the forecast results; RMSE is similar to MSE in that it can amplify the impact of larger errors. At the same time, it maintains the same dimension as the data, making it easy to interpret. This indicator is used to comprehensively evaluate the overall error of this model.