Flight risk prediction method, electronic device, and storage medium integrating Pelican algorithm

Abstract

The present invention discloses a flight risk prediction method, electronic device, and storage medium that integrate the Pelican algorithm. The method includes: S1, constructing a flight risk prediction model, inputting flight risk sample data into the constructed flight risk prediction model for model training, and the flight risk prediction model includes a random forest model integrated with the Pelican algorithm; S2, optimizing the parameters trained on the random forest model using the Pelican algorithm, thereby obtaining the fitness value of each Pelican and storing it in x best In this step, the optimal values ​​of the random forest model parameters are simultaneously obtained; S3, flight data is collected and input into the flight risk prediction model, and the prediction results are output. This invention improves the Pelican algorithm by introducing a Q function, or probability factor, to increase the probability that the Pelican population will search at the global optimal position. The slope of the positions around the current search point is used as the search speed of the Pelican algorithm. This method has the advantages of strong optimization capability, good generalization performance, fast convergence speed, strong robustness, and high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of flight risk random forest model prediction, and in particular to a flight risk prediction method, electronic device, and storage medium integrating a Pelican algorithm. Background Art

[0002] The Random Forest model (RF) is an ensemble learning method proposed by Leo Breiman and Adele Cutler in 2001. By constructing multiple decision trees for classification or regression tasks, it combines the simplicity and interpretability of decision trees with their ability to model complex data, and improves the model's robustness and generalization capabilities by introducing randomness and ensemble concepts. The Random Forest model can effectively handle problems such as nonlinearity, small sample sizes, and overfitting, and is therefore widely used in fields such as text and image classification, data mining, and pattern recognition. Its advantage lies in its ability to model complex data and more accurately capture nonlinear relationships between data. Furthermore, the Random Forest model's ensemble concept and random sampling techniques enable it to better handle small sample sizes and reduce the risk of overfitting. These characteristics have made the Random Forest model one of the preferred algorithms in many fields.

[0003] In the application of random forest models, model parameters and loss functions are key factors affecting their learning and generalization performance. Therefore, determining random forest parameters is extremely important and has become a current research hotspot. Currently, methods for optimizing random forest model parameters include cross-validation, particle swarm optimization, gradient descent, and genetic algorithms. However, these methods all have limitations. Cross-validation is time-consuming and unsuitable for large datasets; gradient descent can be unstable and prone to getting stuck in local optima; and genetic algorithms (GAs) are prone to being limited by local optima. While particle swarm optimization (PSO) offers the advantage of fast convergence, it is also prone to getting stuck in local optima. Cross-validation is time-consuming; gradient descent is not very stable; genetic algorithms are prone to getting stuck in local optima; and particle swarm optimization converges quickly but is also prone to getting stuck in local optima. Therefore, current methods for determining random forest model parameters still have certain drawbacks, requiring further research and improvement to improve their performance and robustness. The Pelican Algorithm (PA) is a heuristic optimization algorithm based on the fishing behavior of pelicans in nature. It simulates the predation behavior of pelicans and has the advantages of strong adaptability, fast convergence, and strong robustness. Currently, the Pelican Algorithm has not been combined with the random forest model. Summary of the Invention

[0004] The purpose of the present invention is to solve the technical problems pointed out in the background technology, provide a flight risk prediction method, electronic device and storage medium integrating the Pelican algorithm, adopt the random forest model integrated with the improved Pelican algorithm to construct a flight risk prediction model, and introduce the Q function, i.e., the probability factor, into the improved Pelican algorithm. The probability of the pelican population searching at the global optimal position is increased, and the slope of the positions around the current search point is used as the search speed of the pelican algorithm. It has the advantages of good model performance, strong optimization ability, good generalization performance, and high prediction accuracy.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A flight risk prediction method integrating the Pelican algorithm, the method comprising:

[0007] S1. Construct a flight risk prediction model and input flight risk sample data into the flight risk prediction model for model training. The flight risk prediction model includes a random forest model integrated with the Pelican algorithm. The random forest model parameters include the number of decision trees, the maximum split feature, the maximum depth of the decision tree, the minimum sample partition, and the minimum sample of the leaf node. The Pelican algorithm parameters include the probability factor Maximum number of iterations T;

[0008] S2. Use the Pelican algorithm to optimize the parameters of the random forest model training. The optimization expression is as follows:

[0009]

[0010] in:

[0011] in represents the position of the i-th pelican and the j-th dimension feature after the final update of the Pelican algorithm; rand represents a random number in the range [0, 1]; R is a random integer of 0 or 2; x i，j is the position of the i-th pelican and the j-th dimension feature when the maximum number of iterations T is processed, F(x i，j ) represents x i，j The corresponding objective function value; Q is the increased optimization search factor; x i，j ′ represents x i，j The position of the i-th pelican and the j-th dimension feature in the last iteration, F(x i，j ′) represents x i，j ′ corresponds to the objective function value; x best，j Indicates the current global optimal position of the j-th dimension feature;

[0012] The fitness value of each pelican is stored in x bestAt the same time, the optimal values ​​of the parameters of the random forest model are obtained;

[0013] S3. Collect flight data and input it into the flight risk prediction model, which then outputs the prediction results.

[0014] In order to better implement the present invention, the random forest model of the present invention uses n1 decision trees to make decisions together:

[0015] where Y q represents the decision value of the qth decision tree, and Y represents the final prediction result obtained by the random forest model.

[0016] Preferably, the feature vector set of the flight risk sample data input into the flight risk prediction model is n is the total number of eigenvectors, represents the nth eigenvector of sample k, the nth eigenvector corresponds to the nth dimension feature; k represents the sample, k = 1, 2, 3, ...···N, N is the total number of samples, x k Represents the set of feature vectors of sample k; define two regions R1 and R2 for the division of feature space, and obtain the predicted output values ​​of the two regions as c1 and c2 respectively. Construct the following loss function of the random forest model and find the optimal feature variable x (j) And the value s:

[0017]

[0018] where y k Indicates the output value corresponding to the flight risk sample data, represents the error and minimum value of region R1, Indicates the regional R2 error and minimum value.

[0019] Preferably, in step S2, the optimization process of the Pelican algorithm includes two stages: an exploration stage and an exploitation stage. The exploration stage of the Pelican algorithm constructs the following formula model according to the prey approximation strategy:

[0020]

[0021] in is the j-th dimension position of the i-th pelican after the exploration phase update; rand is a random number in the range [0, 1]; I is a random integer of 1 or 2; p j is the j-th dimension position of the prey; F p is the objective function value of the prey; F i is the objective function value of the i-th pelican;

[0022] The development phase of the Pelican algorithm is the final phase of the Pelican algorithm in step S2.

[0023] Preferably, the flight risk sample data of step S1 is preprocessed before being input into the flight risk prediction model. The data preprocessing includes normalizing the data to the interval (0, 1). The normalization formula is as follows:

[0024] Where h represents the eigenvalue; y h ′、y h are the values ​​before and after normalization respectively; min(y h )、max(y b ) are the minimum and maximum values ​​of the h-th dimension feature before normalization.

[0025] Preferably, the pelican population initialization expression in the pelican algorithm is as follows:

[0026] x i，j =l j +rand·(u j -l j ); where u j 、l j are the upper and lower boundaries of the j-th dimension feature respectively.

[0027] An electronic device comprises at least one processor, at least one memory and a data bus; wherein: the processor and the memory communicate with each other via the data bus; the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute steps of the flight risk prediction method of the present invention.

[0028] A storage medium includes a memory and a processor, wherein the memory stores an executable program, and the processor implements the steps of the flight risk prediction method of the present invention when executing the executable program.

[0029] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0030] (1) The present invention adopts a random forest model integrated with an improved Pelican algorithm to construct a flight risk prediction model, and the improved Pelican algorithm introduces the Q function, i.e., the probability factor The probability of the pelican population searching at the global optimal position is increased, and the slope of the positions around the current search point is used as the search speed of the pelican algorithm. It has the advantages of good model performance, strong optimization ability, good generalization performance, and high prediction accuracy.

[0031] (2) The present invention creatively adopts the improved Pelican algorithm to optimize the parameters of the random forest model, which solves the problem that traditional methods are prone to falling into local optimal technology, enhances the probability and speed of searching for the global optimal position, and has the advantages of strong adaptability, fast convergence speed, and strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the flight risk prediction method of the present invention;

[0033] Figure 2 This is a schematic diagram of the principle of the flight risk prediction method integrating the improved Pelican algorithm random forest in the embodiment;

[0034] Figure 3 This is a situation diagram after normalization of the flight risk sample data in the embodiment;

[0035] Figures 4a-4d In the embodiment, the non-optimized RF, GA-RF, PSO-RF and PA-RF were established as the verification prediction situations of the four models. DETAILED DESCRIPTION

[0036] Below in conjunction with embodiment, the present invention is described in further detail:

[0037] Example

[0038] like Figure 1 As shown, a flight risk prediction method integrating the Pelican algorithm includes:

[0039] S1. Construct a flight risk prediction model and input flight risk sample data into the flight risk prediction model for model training. The flight risk sample data includes flight data and risk labels corresponding to the flight data. The flight risk sample data set constitutes a flight risk sample data set. The flight risk sample data set is divided into a training set, a test set, and a validation set. The flight risk prediction model includes a random forest model integrated with the Pelican algorithm. The random forest model parameters include the number of decision trees, the maximum split feature, the maximum depth of the decision tree, the minimum sample partition, and the minimum sample of the leaf node. The Pelican algorithm parameters include the probability factor Maximum number of iterations T; the principle of flight risk prediction method can be found in Figure 2 .

[0040] In some preferred embodiments, the feature vector set of the flight risk sample data input into the flight risk prediction model is n is the total number of eigenvectors, represents the nth eigenvector of sample k, and the nth eigenvector corresponds to the nth dimension feature. k represents the sample, k = 1, 2, 3, ... N, N is the total number of samples, x k Represents the set of feature vectors of sample k. The division of feature space defines two regions R1 and R2 (the division of feature space adopts a heuristic method, each division examines all the values ​​of all features in the current set one by one, and selects the best one as the split point according to the square error minimization criterion; for example, in the training set of the flight risk sample data set, the j-th dimension feature variable x(j) and its value s as the segmentation variable and segmentation point, and define two regions), and obtain the predicted output values ​​of the two regions as c1 and c2 respectively, construct the following loss function of the random forest model and find the optimal feature variable x (j ) and the value s:

[0041]

[0042] where y k Indicates the output value corresponding to the flight risk sample data, represents the error and minimum value of region R1, y in k Indicates the output belonging to the R1 region; represents the regional R2 error and minimum value, y in k Indicates the output belonging to the R2 region.

[0043] In some preferred embodiments, the flight risk sample data of step S1 is preprocessed before being input into the flight risk prediction model. The data preprocessing includes normalizing the data to the interval (0, 1). The normalization formula is as follows:

[0044] Where h represents the eigenvalue; y h ′、y h are the values ​​before and after normalization respectively. min(y h )、max(y h ) are the minimum and maximum values ​​of the h-th dimension feature before normalization; in this embodiment, the situation diagram of the flight risk sample data after normalization is shown as follows Figure 3 shown.

[0045] In some preferred embodiments, the random forest model uses n1 decision trees to jointly make decisions:

[0046] where Y qRepresents the decision value of the qth decision tree, and Y represents the final prediction result obtained by the random forest model. The decision tree in the embodiment adopts a regression decision tree, mainly referring to the CART (classification and regression tree) algorithm. The values ​​of the internal node features are "yes" and "no", which is a binary tree structure (compared to the classification tree, the regression tree is more suitable for the prediction problem of continuous output; it can divide the data into fine-grained parts according to the features, so it can fit the output value of the data more accurately to a certain extent; as the depth of the regression tree increases, the model can better adapt to the training data, and by integrating multiple regression trees to build a more powerful random forest model, the risk of overfitting can be reduced and the prediction performance of the model can be improved); the basic idea of ​​the decision tree is: as the depth of the tree increases, the loss of the node should be rapidly reduced, and the faster the loss is reduced, the better, so that the decision tree with the shortest height is finally obtained. The regression tree used is a decision tree model for regression problems. It mainly divides the feature space into multiple regions and assigns a specific output (or output value) to each divided region. The partitioning of the regression tree is based on the value of the feature vector. When the test data reaches the leaf node, it can be assigned to the corresponding leaf node according to its characteristics and the corresponding output (or output value) is obtained. For example, if X and Y are input and output respectively, and Y is a continuous variable, the data input to the random forest model can be expressed as (x k ,y k ), where k represents the sample, x k Represents the input sample data (or input), the input sample data contains n features, y k Indicates the risk label (or output) corresponding to the input sample data.

[0047] S2. Optimize the parameters of the random forest model training using the Pelican algorithm (the present invention adopts the improved Pelican algorithm). The initialization expression of the Pelican population in the Pelican algorithm is as follows:

[0048] x i，j =l j +rand·(u j -l j ); where u j 、l j are the upper and lower boundaries of the j-th dimension feature respectively.

[0049] The optimization process of the Pelican algorithm includes two stages: exploration and development. The exploration stage of the Pelican algorithm is constructed according to the prey approximation strategy to build the following formula model:

[0050]

[0051] in is the j-th dimension position of the i-th pelican after the exploration phase update; rand is a random number in the range [0, 1]; I is a random integer of 1 or 2; p j is the j-th dimension position of the prey; F p is the objective function value of the prey; F i is the objective function value of the i-th pelican; else represents other (or F p ≥F i The development phase of the Pelican algorithm is the final phase of the Pelican algorithm in step S2. The optimization process expression of the development phase of the Pelican algorithm is as follows:

[0052]

[0053] in:

[0054] in The position of the i-th pelican and the j-th dimension feature after the final update of the Pelican algorithm. rand represents a random number in the range [0, 1]. The i-th pelican is the i-th pelican in the Pelican algorithm, and the j-th dimension feature is the j-th dimension feature in the random forest model. R is a random integer of 0 or 2. i，j is the position of the i-th pelican and the j-th dimension feature when the maximum number of iterations T is processed, F(x i，j ) represents x i，j The corresponding objective function value. Q is the increased optimization search factor. x i，j ′ represents x i，j The position of the i-th pelican and the j-th dimension feature in the last iteration, F(x i，j ′) represents x i，j ′ corresponds to the objective function value. best，j Indicates the current global optimal position of the j-th dimension feature.

[0055] In order to improve the global search capability of the Pelican Algorithm, the present invention expands the search area of ​​each pelican to the search range of the entire flock (the Q function is introduced in the development stage of the Pelican Algorithm, ) to help the Pelican algorithm jump out of the local optimal solution; Therefore, the present invention uses the slope of the position around the current search point as the speed of the Pelican algorithm search, introducing the probability factor The probability of the pelican population searching at the global optimal position is increased, and the search area of ​​each pelican is expanded to the search range of the entire flock to help the pelican algorithm escape from the local optimal solution.

[0056] The fitness value of each pelican is stored in x best At the same time, the optimal values ​​of the parameters of the random forest model are obtained.

[0057] S3. Collect flight data and input it into the flight risk prediction model, which then outputs the prediction results.

[0058] To verify the model parameter optimization and prediction effect of the improved Pelican algorithm on the random forest model in the flight risk prediction model of the present invention (including the random forest model integrated with the improved Pelican algorithm), the present invention uses a flight risk sample data set from the Civil Aviation Safety Information Network flight risk data set. A total of L data items from January 2017 to December 2022 are selected as an example and randomly divided into a training set, a test set, and a validation set. The flight risk sample data in the flight risk sample data set are preprocessed, and the data preprocessing includes data normalization to the interval (0, 1). The normalization formula is as follows:

[0059] Where h represents the eigenvalue; y h ′、y h are the values ​​before and after normalization respectively; min(y h )、max(y h ) are the minimum and maximum values ​​of the h-th dimension feature before normalization; the situation diagram after L data normalization is as follows Figure 3 As shown. GA-RF (a random forest model including a genetic algorithm), PSO-RF (a random forest model including a particle swarm algorithm), PA-RF (i.e., the technical solution of the present invention: a random forest model integrated with an improved Pelican algorithm, with model parameters optimized by the improved Pelican algorithm), and RF without optimization (a single random forest model) were constructed as four models. The four models were trained and their predictions were verified using flight risk sample data. The results are shown in the figure below. Figure 4a -d shows that there is no optimized RF ( Figure 4a As shown), PA-RF( Figure 4d As shown), GA-RF( Figure 4b As shown), PSO-RF( Figure 4c shown).

[0060] In the example of selecting L data from January 2017 to December 2022, the model parameters of the technical solution of the present invention are as follows:

[0061]

[0062] like Figures 4a to 4dAs shown, the flight risk prediction of the random forest model integrated with the improved Pelican algorithm of the present invention is consistent with the actual value the best, and the degree of consistency is significantly better than the three models of GA-RF, PSO-RF and unoptimized RF. The random forest model integrated with the improved Pelican algorithm of the present invention has higher prediction accuracy (smaller prediction error) and has a better effect on flight risk prediction. To further verify the superiority of the flight risk prediction model of the present invention (including the random forest model integrated with the improved Pelican algorithm), the performance of the four models was evaluated using the root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient (R), as shown in the following table:

[0063] Model Unoptimized RF PA-RF GA-RF PSO-RF RMSE metric 0.126765 0.011009 0.083117 0.106581 MSE indicator 0.106413 0.009451 0.07213 0.08423 R indicator 0.7021 0.9163 0.7804 0.7602

[0064] Among the evaluation indicators of root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (R), the smaller the RMSE and MAE values ​​and the larger the R, the better the prediction performance of the model. By comparison, it can be concluded that the flight risk prediction model of the present invention (i.e., PA-RF) has the best prediction performance and the highest prediction accuracy.

[0065] A storage medium includes a memory and a processor, wherein the memory stores an executable program, and the processor implements the steps of the flight risk prediction method of the present invention when executing the executable program.

[0066] 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 and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A flight risk prediction method integrating the Pelican algorithm, characterized by: The methods include: S1. Construct a flight risk prediction model and input flight risk sample data into the flight risk prediction model for model training. The flight risk prediction model includes a random forest model integrated with the Pelican algorithm. The random forest model parameters include the number of decision trees, the maximum split feature, the maximum depth of the decision tree, the minimum sample partition, and the minimum sample of the leaf node. The Pelican algorithm parameters include the probability factor Maximum number of iterations T; S2. Use the Pelican algorithm to optimize the parameters of the random forest model training. The optimization expression is as follows: in: in represents the position of the i-th pelican and the j-th dimension feature after the final update of the Pelican algorithm; rand represents a random number in the range [0, 1]; R is a random integer of 0 or 2; x i，j is the position of the i-th pelican and the j-th dimension feature when the maximum number of iterations T is processed, F(x i，j ) represents x i，j The corresponding objective function value; Q is the increased optimization search factor; x i，j ′ represents x i，j The position of the i-th pelican and the j-th dimension feature in the last iteration, F(x i，j ′) represents x i，j ′ corresponds to the objective function value; x best，j Indicates the current global optimal position of the j-th dimension feature; The fitness value of each pelican is stored in x best At the same time, the optimal values ​​of the parameters of the random forest model are obtained; S3. Collect flight data and input it into the flight risk prediction model, which then outputs the prediction results.

2. The flight risk prediction method integrating the Pelican algorithm according to claim 1, characterized in that: The random forest model uses n1 decision trees to make decisions together: where Y q represents the decision value of the qth decision tree, and Y represents the final prediction result obtained by the random forest model.

3. The flight risk prediction method integrating the Pelican algorithm according to claim 1 or 2, characterized in that: The feature vector set of flight risk sample data input into the flight risk prediction model n is the total number of eigenvectors, represents the nth eigenvector of sample k, the nth eigenvector corresponds to the nth dimension feature; k represents the sample, k = 1, 2, 3, ... N, N is the total number of samples, x k Represents the set of feature vectors of sample k; The feature space is divided into two regions R1 and R2, and the predicted output values ​​of the two regions are c1 and c2 respectively. The following loss function of the random forest model is constructed and the optimal feature variable x is found. (j) And the value s: where y k Indicates the output value corresponding to the flight risk sample data, represents the error and minimum value of region R1, Indicates the regional R2 error and minimum value.

4. The flight risk prediction method integrating the Pelican algorithm according to claim 1, characterized in that: In step S2, the optimization process of the Pelican algorithm includes two stages: the exploration stage and the development stage. The exploration stage of the Pelican algorithm constructs the following formula model according to the prey approximation strategy: in is the j-th dimension position of the i-th pelican after the exploration phase update; rand is a random number in the range [0, 1]; I is a random integer of 1 or 2; p j is the j-th dimension position of the prey; F p is the objective function value of the prey; F i is the objective function value of the i-th pelican; The development phase of the Pelican algorithm is the final phase of the Pelican algorithm in step S2.

5. The flight risk prediction method integrating the Pelican algorithm according to claim 1, characterized in that: The flight risk sample data of step S1 is preprocessed before being input into the flight risk prediction model. The data preprocessing includes normalizing the data to the interval (0, 1). The normalization formula is as follows: Where h represents the eigenvalue; y h ′、y h are the values ​​before and after normalization respectively; min(y b )、max(y h ) are the minimum and maximum values ​​of the h-th dimension feature before normalization.

6. The flight risk prediction method integrating the Pelican algorithm according to claim 1, characterized in that: The initialization expression of the pelican population in the pelican algorithm is as follows: x i，j =l j +rand·(u j -l j ); where u j 、l j are the upper and lower boundaries of the j-th dimension feature respectively.

7. An electronic device, characterized in that: The invention comprises at least one processor, at least one memory and a data bus; wherein: the processor and the memory communicate with each other via the data bus; the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the flight risk prediction method according to any one of claims 1 to 6.

8. A storage medium comprising a memory and a processor, wherein the memory stores an executable program, characterized in that: When the processor executes the executable program, the steps of the flight risk prediction method according to one of claims 1 to 6 are implemented.

Images