Civil aviation service score prediction method and system based on graph neural network
Through the adaptive graph neural network selection model, the problems of high complexity of model selection and unstable performance in civil aviation service data analysis are solved, efficient and accurate civil aviation service score prediction are achieved, and computing efficiency and applicability are improved.
Patent Information
- Application Number
- CN202510341053.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2025-07-08
AI Technical Summary
In the analysis of civil aviation service data, the existing graph neural network has high complexity, unstable performance, high computing resources, lacks adaptability, and is difficult to meet different data characteristics and task requirements.
The civil aviation service score prediction method based on graph neural network is adopted, and the most suitable graph neural network model is dynamically selected through the neural network selection model. Combined with streaming data processing and feature extraction, an adaptive end-to-end system is built, including data acquisition, preprocessing, model training and selection.
It improves prediction accuracy and generalization capabilities, reduces manual intervention, improves model flexibility and applicability, reduces computing resource consumption, and achieves efficient civil aviation service score prediction.
Smart Images
Figure CN120277323A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of data processing, and particularly relates to a civil aviation service score prediction method and system based on graph neural networks. Background Art
[0002] With the rapid development of the civil aviation industry and the improvement of the informatization level, a large amount of structured and unstructured data has been continuously accumulated in the industry, such as flight information, airport data, passenger feedback, route network, etc. These data have complex correlations and spatio-temporal characteristics, and traditional data processing and analysis methods are difficult to effectively mine their potential value. Graph Neural Networks (GNNs) have become an effective tool for processing these large-scale data because they can capture complex relationships and dependencies in the data. The application of GNNs in civil aviation service data analysis is becoming increasingly widespread, especially in the fields of flight delay prediction, airport service optimization, route network optimization, and passenger behavior prediction, showing significant advantages.
[0003] Although graph neural networks have broad application prospects in civil aviation service data analysis, the existing technologies still face many challenges in the model selection process. First, the complexity of model selection is high, and it is necessary to make detailed adjustments according to different data characteristics and task requirements; second, the performance of existing GNN models in different scenarios is difficult to guarantee, and overfitting or underfitting may occur; third, most models lack adaptive capabilities and cannot automatically adjust model parameters according to data changes; finally, existing GNN models usually require high computing resources, resulting in excessive computing consumption and affecting their popularization in practical applications. To overcome these problems, a system that can automatically select the most suitable model is needed to improve the performance and computing efficiency of the model. Summary of the Invention
[0004] The present invention aims to overcome the defect of the above-mentioned prior art that the model needs to be changed or adjusted according to different data or task requirements, and proposes a civil aviation service score prediction method and system based on graph neural networks.
[0005] To achieve the above technical effects, the technical solution of the present invention is as follows:
[0006] A civil aviation service score prediction method based on graph neural networks, comprising the following steps:
[0007] Collect civil aviation service data and perform preprocessing;
[0008] Construct a neural network selection model and multiple graph neural network models;
[0009] Train the graph neural network model with the civil aviation service data, and use the performance metrics generated during the training process as supervision labels to perform parallel training on the neural network selection model;
[0010] After inputting the civil aviation service data to be predicted into the trained neural network selection model to select a suitable graph neural network model, input the civil aviation service data to be predicted into the target graph neural network model to output the prediction result. The present invention also proposes a civil aviation service score prediction system based on a graph neural network, and the system includes:
[0011] Data acquisition module: used to collect civil aviation service data and perform preprocessing;
[0012] Neural network selection module: on which a neural network selection model is installed, used to select a suitable graph neural network for prediction for the input data;
[0013] Prediction module: on which multiple graph neural networks are installed, used to output the prediction result based on the input civil aviation service data.
[0014] The present invention also proposes an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the civil aviation service score prediction method based on a graph neural network as described in the present invention.
[0015] The present invention also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the civil aviation service score prediction method based on a graph neural network as described in the present invention.
[0016] Compared with the prior art, the beneficial effects of the present invention:
[0017] By introducing a model selection agent, the present invention realizes the dynamic analysis of input data and the adaptive selection of the optimal GNN model, can automatically select the most suitable model according to different data sets and task requirements, thereby improving the prediction accuracy and generalization ability. Dynamic model selection avoids the uncertainty and low efficiency brought by traditional manual model selection, significantly enhances the flexibility and applicability, and has broad practical application potential. In addition, the system adopts an end-to-end workflow, including data preprocessing, model training, and model selection, effectively improving the overall efficiency and performance. Description of the drawings
[0018] Figure 1 It is a flowchart of a civil aviation service score prediction method based on a graph neural network for Embodiment 1.
[0019] Figure 2 It is an architecture diagram of a civil aviation service score prediction system based on a graph neural network for Embodiment 2.
[0020] Figure 3 It is the implementation flowchart of Example 3.
[0021] Figure 4 It is the architecture diagram of the neural network selection model. Detailed implementation manners
[0022] The accompanying drawings are only for illustrative purposes and should not be construed as limitations on the present invention.
[0023] For those skilled in the art, it is understandable that some well-known descriptions in the accompanying drawings may be omitted.
[0024] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0025] Example 1
[0026] This example proposes a civil aviation service score prediction method based on graph neural network. As Figure 1 shown, it is the flowchart of a civil aviation service score prediction method based on graph neural network in this example.
[0027] A civil aviation service score prediction method based on graph neural network proposed in this example includes the following steps:
[0028] Collect civil aviation service data and perform preprocessing;
[0029] Construct a neural network selection model and multiple graph neural network models;
[0030] Train the graph neural network models with the civil aviation service data, and use the performance indicators generated during the training process as supervision labels to perform parallel training on the neural network selection model;
[0031] After inputting the civil aviation service data to be predicted into the trained neural network selection model to select a suitable graph neural network model, input the civil aviation service data to be predicted into the target graph neural network model to output the prediction result.
[0032] In this example, a neural network selection model is introduced to dynamically select the most suitable graph neural network model during the model training process. Based on different features in the civil aviation service data, this selection model judges and selects the best model, thus effectively solving the problems of cumbersome model selection and unstable performance in traditional methods, significantly improving the prediction accuracy and the generalization ability of the model, and also being able to effectively avoid the problems of overfitting and underfitting, thereby ensuring the prediction performance of the civil aviation service score.
[0033] As an illustrative example, the graph neural network models include GAT, GCN or GraphSAGE.
[0034] In an alternative embodiment, the step of collecting and preprocessing civil aviation service data includes:
[0035] Collect multi-dimensional civil aviation service data through a streaming data processing engine, remove null values and duplicate values from the data, perform standardization processing on numerical data in the data, perform encoding processing on label data in the data, and perform normalization processing on text data in the data.
[0036] In this embodiment, real-time collection and processing of civil aviation service data through a streaming data processing engine can effectively ensure data quality. Secondly, by removing null values and duplicate values, data redundancy and interference from invalid information are avoided, improving the accuracy and integrity of the data. Different processing of different types of data helps to unify features with different dimensions and ranges to the same standard, thereby enhancing the convergence and stability during model training, and thus improving the prediction accuracy and generalization ability of the model.
[0037] In an alternative embodiment, the step of training the graph neural network model with the civil aviation service data includes:
[0038] Extract features from the civil aviation service data and construct feature vectors; during feature extraction, continuous features are standardized or normalized, and categorical features are one-hot encoded or processed using an embedding layer; when constructing feature vectors, different fields in each piece of civil aviation service data are combined into a feature vector, and each piece of civil aviation service data is used as a single node, and its average feature vector is used as the node feature;
[0039] Construct an adjacency matrix between nodes based on the feature vectors;
[0040] Input the node features and the adjacency matrix of the civil aviation service data into the graph neural network model for training.
[0041] In this embodiment, joint training using the node information and adjacency matrix of the graph neural network fully explores the complex relationships contained in the data, can effectively capture the spatio-temporal features and non-linear dependence relationships in the civil aviation service data, and improves the prediction accuracy.
[0042] In an alternative embodiment, the step of constructing an adjacency matrix based on the feature vectors includes:
[0043] Calculate the similarity between each pair of feature vectors, regard the feature vector pairs whose similarity meets the preset conditions as adjacent feature vectors, and construct an adjacency matrix.
[0044] In this embodiment, through the construction of the adjacency matrix, the problem of topological mismatch caused by the fixed graph structure in the traditional method is effectively solved, and the model training efficiency is improved.
[0045] In an alternative embodiment, the neural network selection model includes an input layer, a hidden layer, and an output layer; wherein, both the hidden layer and the output layer are fully connected layers; the input layer includes a multi-modal data adapter, which extracts features from different types of data in the input data and integrates the multi-modal feature vectors through a feature fusion layer; the hidden layer includes a weight adjustment unit, which dynamically adjusts the feature weights of different modalities through an attention mechanism.
[0046] Furthermore, in the input layer, for the time series data in the civil aviation service data, an LSTM layer is used to extract time features, an embedding layer is used to compress the dimensions of the classification data, and a pre-trained language model is used to extract semantic features for the text data. In the hidden layer, the preset key dimension weights of civil aviation services (such as the on-time rate accounting for 30% and the service attitude accounting for 25%, etc.) dynamically adjust the feature weights of each dimension through an attention mechanism and combine the flight historical score fluctuation coefficients (such as seasonal fluctuations and the impact of special events) for weight correction.
[0047] Even further, the deviation between the actual score and the prediction result after each prediction of the neural network selection model is used as a reward signal, and a reinforcement learning algorithm is used to dynamically update the selection model strategy. Independent policy networks are established for different time periods (such as the Spring Festival / peak summer travel seasons) and different routes (international / domestic); and by designing a two-layer selection model, the first layer is a rough selection module: based on the basic attributes of the flight (aircraft type, route type), candidate models are quickly filtered; the second layer is a refined selection module: the final decision is made through the confidence score output by the graph neural network, and dynamic switching between levels is achieved through a gating mechanism (such as a Gating Network).
[0048] In this embodiment, the fully connected structure of the hidden layer enables the model to effectively capture the high-order features and potential associations of the input data, improving the representational ability of the model. The fully connected structure of the output layer enables the model to more accurately output the final model selection probability distribution, thereby achieving precise model selection. The neural network structure using fully connected layers simplifies the network architecture and improves the convergence and efficiency during the training process.
[0049] In an alternative embodiment, the step of using the performance metrics generated during the training process as supervision labels to perform parallel training on the neural network selection model includes:
[0050] The accuracy rate or F1 score collected in real time during the training process of the graph neural network is encoded into a unified time series feature vector in a serialized manner and input into the input layer of the neural network selection model;
[0051] In the hidden layer, perform non - linear transformation and feature extraction on the temporal feature vector;
[0052] In the output layer, output the final probability distribution of the performance trend of the graph neural network through the Softmax activation function, and select the graph neural network with the highest probability score as the output result.
[0053] In this embodiment, a probability distribution is output through the Softmax activation function, representing the selection probabilities of different models. During the prediction process, the neural network selection model automatically selects the most suitable graph neural network model for civil aviation service scoring prediction. This kind of adaptive model selection strategy can reduce manual intervention, improve the automation and efficiency of the prediction process, and also avoid the problem of poor applicability of a single model.
[0054] In an alternative embodiment, the neural network selection model uses the cross - entropy loss function as the objective function during training to measure the gap between the model output and the true label. The cross - entropy loss function uses the flight service complexity coefficient as the loss weight; the model parameters are optimized through the Adam optimizer, and the Adam optimizer adjusts the learning rate based on the flight timestamp and geographical location perception.
[0055] As an exemplary illustration, the complexity coefficient includes aircraft type complexity or the number of time zones crossed by the route, and dynamically adjusts the positive and negative sample weights based on the historical score fluctuation index.
[0056] Furthermore, the loss function also has constraints, including: service fluctuation smoothing constraint: penalty for the difference in predicted values of adjacent flights; aircraft type adaptation constraint: regularization of the difference in scoring distributions for different aircraft types; policy compliance constraint: penalty for the deviation in scoring prediction during major holidays. The loss function deals with outliers (such as extreme weather events) in civil aviation data; the M - estimate loss function is used to replace the standard cross - entropy.
[0057] Even further, the Adam optimizer performs learning rate scheduling through flight timestamp and geographical location perception.
[0058] In this embodiment, the neural network selection model uses the cross - entropy loss function as the objective function during training and optimizes the model parameters through the Adam optimizer. The cross - entropy loss function can effectively measure the difference between the model output and the true label. Using cross - entropy as the objective function helps improve the accuracy of the model. The Adam optimizer combines the momentum method and the adaptive learning rate, which can effectively accelerate the gradient descent process, avoid the problems of too large or too small learning rates in traditional optimization algorithms, and enable the model to converge faster and more stably. In addition, when facing large - scale data and complex models, the Adam optimizer can reduce the risk of gradient vanishing or gradient explosion, ensuring the efficiency and stability of the training process.
[0059] Example 2
[0060] This example proposes a civil aviation service score prediction system based on a graph neural network, applying the method for predicting the civil aviation service score based on a graph neural network proposed in Example 1. As Figure 2 shown, it is the architecture diagram of a civil aviation service score prediction system based on a graph neural network in this example.
[0061] This example proposes a civil aviation service score prediction system based on a graph neural network, including:
[0062] Data acquisition module: used to collect civil aviation service data and perform preprocessing;
[0063] Neural network selection module: equipped with a neural network selection model, used to select a suitable graph neural network for prediction for the input data;
[0064] Prediction module: equipped with multiple graph neural networks, used to output prediction results based on the input civil aviation service data.
[0065] It can be understood that the system in this example corresponds to the method in the above Example 1, and the optional items in the above Example 1 also apply to this example, so they will not be repeated here.
[0066] Example 3
[0067] This example specifically applies the method for predicting the civil aviation service score based on a graph neural network proposed in Example 1.
[0068] As Figure 3 shown, it is the implementation process of this example.
[0069] In this example, a streaming data processing engine such as Apache Flink or Apache Spark is used to collect civil aviation service data including passenger evaluations, flight information, etc., and the data is processed and analyzed in real time through streaming computing.
[0070] Subsequently, data preprocessing is carried out; first, data cleaning is performed to remove duplicate data and missing values; at the same time, outliers and inconsistent data records are detected and processed to ensure the integrity and accuracy of the data.
[0071] The numerical data is standardized to ensure that different data has a unified scale to eliminate the influence of different features on model training; the label data (such as satisfaction scores, service types, etc.) is encoded; the text data is preprocessed, including operations such as word segmentation, removal of stop words (such as "de", "le", etc.), stemming or lemmatization.
[0072] Subsequently, feature extraction is performed. Continuous features (such as time periods, scores, etc.) are standardized or normalized so that they are within the same numerical range; categorical features (such as departure location, destination, aircraft type, etc.) are one-hot encoded or processed using an embedding layer.
[0073] Construct a feature vector. Each service evaluation data is regarded as a "superpixel", and its features are extracted. Each review contains multiple fields (such as overall satisfaction, comments, service scores for various items, etc.). These fields are combined into a feature vector. Calculate the average feature vector of each service evaluation data as the node feature.
[0074] Subsequently, construct an adjacency matrix. Using methods such as cosine similarity or Euclidean distance, calculate the similarity between each pair of feature vectors. Based on the calculated similarity values, construct an adjacency matrix. By setting a threshold or using the KNN method, determine which reviews have an adjacent relationship (i.e., construct edges).
[0075] As an exemplary illustration, there are the following 3 reviews
[0076] Review ID 1: [Overall satisfaction: 4, check-in method: 1, security check satisfaction: 3,...]
[0077] Review ID 2: [Overall satisfaction: 5, check-in method: 2, security check satisfaction: 4,...]
[0078] Review ID 3: [Overall satisfaction: 3, check-in method: 1, security check satisfaction: 2,...]
[0079] Standardize the score field:
[0080] Standardize the score of overall satisfaction from 1 to 5 to the range [0, 1].
[0081] The overall satisfaction of Review ID 1 is standardized to: (4 - 1) / (5 - 1) = 0.75.
[0082] The one-hot encoding of the check-in method of Review ID 1 is: [1, 0].
[0083] Combine Review ID1 into a feature vector: [0.75, 1, 0, 0.75,...].
[0084] After obtaining the feature vectors of all reviews, construct the adjacency matrix: Based on the calculated similarity values, construct the adjacency matrix. Assume that the similarity threshold is set to 0.7. Then, edges are constructed between reviews with a similarity greater than 0.7. Table 1 below shows the similarity between review IDs.
[0085] Table 1 Similarity between review IDs
[0086] Comment ID 1 Comment ID 2 Comment ID 3 Comment ID 1 1.0 0.85 0.7 Comment ID 2 0.85 1.0 0.75 Comment ID 3 0.7 0.75 1.0
[0087] Treat each review as a node in the graph, and use the constructed adjacency matrix to represent the connection relationship between nodes. Train the GAT, GCN, and GraphSAGE models on multiple civil aviation datasets, and record their performance metrics (such as accuracy, F1 score, etc.). Use these performance metrics as the training labels for the neural network to select the model, and train to select the most suitable model on different datasets.
[0088] As Figure 4 shown, it is the architecture diagram of the neural network to select the model.
[0089] The neural network to select the model includes:
[0090] 1. Input layer
[0091] Input data: The feature vector output by the feature engineering module.
[0092] Input dimension: The dimension of node features.
[0093] 2. Hidden layer
[0094] Hidden layer 1: A fully connected layer that maps the input data to a high-dimensional space.
[0095] Input dimension: The dimension of node features.
[0096] Output dimension: The hidden layer dimension (such as 64).
[0097] Activation function: ReLU.
[0098] Hidden layer 2: A fully connected layer that further processes the data.
[0099] Input dimension: The hidden layer dimension.
[0100] Output dimension: The hidden layer dimension.
[0101] Activation function: ReLU.
[0102] 3. Output layer
[0103] Output layer: A fully connected layer that maps the data to the output dimension.
[0104] Input dimension: The hidden layer dimension.
[0105] Output dimension: The number of models (such as the number is 3, corresponding to GAT, GCN, and GraphSAGE).
[0106] Activation function: Softmax, used to output the probability distribution.
[0107] 4. Loss function
[0108] Loss function: Cross Entropy Loss.
[0109] Optimizer: Adam.
[0110] Extract node features and adjacency matrix from the civil aviation dataset to be predicted and input them into the neural network selection model. The model will learn the complex relationships between nodes, perform global semantic understanding or task execution, and output the most suitable GNN model (GAT, GCN, or GraphSAGE). Use the selected model for task prediction (such as flight delay prediction, airport service satisfaction prediction, etc.).
[0111] Example 4
[0112] This example proposes a computer device, including a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, the processor executes the steps of the civil aviation service score prediction method based on graph neural network proposed in Example 1.
[0113] Example 5
[0114] This example proposes a storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of the civil aviation service score prediction method based on graph neural network proposed in Example 1 are implemented. The storage medium includes, but is not limited to, various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.
[0115] Exemplarily, the instructions, programs, code sets, or instruction sets can be implemented using conventional programming languages.
[0116] Exemplarily, the processor includes, but is not limited to, smartphones, personal computers, servers, network devices, etc., and is used to execute all or part of the steps of the civil aviation service score prediction method based on graph neural network described in Example 1.
Claims
1. A civil aviation service score prediction method based on graph neural network, characterized in that, It includes the following steps: Collect civil aviation service data and perform preprocessing; Construct a neural network selection model and multiple graph neural network models; Train the graph neural network models with the civil aviation service data, and use the performance metrics generated during the training process as supervision labels to perform parallel training on the neural network selection model; After inputting the civil aviation service data to be predicted into the trained neural network selection model to select a suitable graph neural network model, input the civil aviation service data to be predicted into the target graph neural network model to output the prediction result.
2. The civil aviation service score prediction method based on a graph neural network according to claim 1, wherein, The step of collecting civil aviation service data and performing preprocessing includes: Collect multi-dimensional civil aviation service data through a streaming data processing engine, remove null values and duplicate values in the data, perform standardization processing on the numerical data in the data, perform encoding processing on the label data in the data, and perform normalization processing on the text data in the data.
3. A civil aviation service score prediction method based on a graph neural network according to claim 1, characterized in that, The step of training the graph neural network models with the civil aviation service data includes: Extract features from the civil aviation service data and construct feature vectors; when extracting features, standardize or normalize continuous features, and perform one-hot encoding or use an embedding layer to process categorical features; when constructing feature vectors, combine different fields in each civil aviation service data into a feature vector, use each civil aviation service data as a single node, and use its average feature vector as the node feature; Construct an adjacency matrix between nodes based on the feature vectors; Input the node features of the civil aviation service data and the adjacency matrix into the graph neural network models for training.
4. The civil aviation service score prediction method based on graph neural network according to claim 3, wherein The step of constructing an adjacency matrix based on the feature vectors includes: Calculate the similarity between each pair of feature vectors, regard the feature vector pairs whose similarity meets the preset conditions as adjacent feature vectors, and construct an adjacency matrix.
5. A civil aviation service score prediction method based on a graph neural network according to claim 1, characterized in that, The neural network selection model includes an input layer, a hidden layer, and an output layer; among them, both the hidden layer and the output layer are fully connected layers; the input layer includes a multi-modal data adapter, which extracts features from different types of data in the input data and integrates the multi-modal feature vectors through a feature fusion layer; the hidden layer includes a weight adjustment unit, which dynamically adjusts the feature weights of different modalities through an attention mechanism.
6. The civil aviation service score prediction method based on graph neural network according to claim 5, characterized in that, The step of using the performance metrics generated during the training process as supervision labels to perform parallel training on the neural network selection model includes: Encode the accuracy rate or F1 score collected in real time during the graph neural network training process into a unified time-series feature vector in a serialized manner and input it into the input layer of the neural network selection model; In the hidden layer, perform non-linear transformation and feature extraction on the time-series feature vector; In the output layer, output the final probability distribution of the graph neural network performance trend through the Softmax activation function, and select the graph neural network with the highest probability score as the output result.
7. A civil aviation service score prediction method based on a graph neural network according to claim 5, characterized in that The neural network selection model uses the cross-entropy loss function as the objective function to measure the gap between the model output and the true label during the training process, and the cross-entropy loss function uses the flight service complexity coefficient as the loss weight; The model parameters are optimized by the Adam optimizer, and the Adam optimizer adjusts the learning rate based on flight timestamps and geographical location awareness.
8. A civil aviation service score prediction system based on a graph neural network, which applies the civil aviation service score prediction method based on a graph neural network according to any one of claims 1 to 7, characterized in that, It includes: Data acquisition module: used to collect civil aviation service data and perform preprocessing; Neural network selection module: equipped with a neural network selection model thereon, used to select a suitable graph neural network for prediction for the input data; Prediction module: equipped with multiple graph neural networks thereon, used to output prediction results based on the input civil aviation service data.
9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a civil aviation service score prediction method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a civil aviation service score prediction method according to any one of claims 1 - 7.