Bird strike risk prediction method and system based on improved weasel optimization model

Abstract

The invention discloses a bird strike risk prediction method and system based on an improved dwarf optimization model, and the method comprises the steps: S1, calculating the unit bird strike risk of each time span group for historical bird strike event data based on a bird strike risk evaluation index system, and collecting the unit bird strike risks according to a time sequence to form a bird strike risk data set; s2, constructing a bird strike risk prediction combination model and performing model training; s3, performing model parameter optimization processing on the bird strike risk prediction combination model by using the improved dwarf weasel optimization algorithm model; and S4, the bird strike risk prediction combination model outputs unit bird strike risks of a plurality of continuous time span groups after the research prediction day according to the time sequence. According to the method, the trend characteristics, seasonal characteristics, time sequence characteristics and nonlinear relations of the historical time sequence bird strike risk data set can be better captured, the continuous unit bird strike risks after the prediction day can be researched, and prediction technical support is provided for the airport bird strike risks.

Description

Technical Field

[0001] The present invention relates to the field of bird strike event risk prediction, and particularly to a bird strike risk prediction method and system based on an improved dwarf mongoose optimization model. Background Art

[0002] Birds near airports have an important impact on aircraft takeoff and landing. During the takeoff and landing process of aircraft, birds may collide with the aircraft, posing a safety hazard to aircraft takeoff and landing. Therefore, bird strike risk prediction near airports is a key technical direction for airport safety guarantee. Bird strike events have become a safety issue in the world's aviation industry, and bird strike risk prediction is an important topic in the field of aviation safety. With the increase in air transportation volume, bird strike risk is generally predicted based on manual prediction or a single model. Existing models cannot effectively capture the temporal characteristics and temporal dependence relationships of bird strikes, and the model training iteration is not optimized, etc., which result in low accuracy of bird strike prediction. Summary of the Invention

[0003] The purpose of the present invention is to provide a bird strike risk prediction method and system based on an improved dwarf mongoose optimization model, construct a bird strike risk prediction combined model including an autoregressive moving average model, an LSTM model, and an XGBoost model, use the bird strike risk data set as a sample for model training to capture the trends, seasonality, temporal characteristics, and non-linear relationships in the bird strike risk data, and use the improved dwarf mongoose optimization algorithm model for model parameter optimization processing, so as to be able to predict the average risk of wind shear in a continuous number of time span groups after the research prediction date.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A bird strike risk prediction method based on an improved dwarf mongoose optimization model, the method comprising:

[0006] S1. Construct a bird strike risk assessment index system, obtain historical bird strike event data before the research prediction date, divide it into several time span groups according to a natural year, calculate the unit bird strike risk of each time span group based on the bird strike risk assessment index system, and collect the unit bird strike risks according to the time series to form a bird strike risk data set;

[0007] S2. Construct a bird strike risk prediction combined model including an autoregressive moving average model, an LSTM model, and an XGBoost model, and input the bird strike risk data set into the bird strike risk prediction combined model for model training;

[0008] S3. Construct an improved dwarf mongoose optimization algorithm model, and use the improved dwarf mongoose optimization algorithm model to optimize the model parameters of the bird strike risk prediction combined model;

[0009] S4. After the model parameters are optimized, the combined bird strike risk prediction model outputs the unit bird strike risk for several consecutive time span groups after the research prediction date according to the time series.

[0010] To better implement the present invention, the improved dwarf meerkat optimization algorithm model includes an initialization stage, a foraging stage, a social stage, and an update stage. The expression of the initialization stage is as follows:

[0011] X m = X min + rand·(X max - X min ), where X m is the initial position of meerkat individual m, X max and X min are the upper and lower bounds of the search range respectively, and rand is a random number of the model between [0, 1];

[0012] In the foraging stage, a dynamic weight ω(t) and an adaptive perturbation η(t) are introduced to search and update the position. The improved iterative update expression is as follows:

[0013] X m (t1 + 1)= X m (t1)+ ω(t1)·(X best - X m (t1))+ η(t1)·(X rand - X m (t1)), where X m (t1 + 1), X m (t1) are the positions of meerkat individual m at iterations t1 + 1 and t1 respectively; X best is the global optimal position of the meerkat population, X rand is the position of a randomly selected meerkat individual, ω(t1) is the dynamic weight at iteration t1, and η(t1) is the adaptive perturbation at iteration t1;

[0014] In the social stage, the position is updated based on distance-based social behavior, and the search direction is adjusted according to the distance. The expression is as follows:

[0015] where β is a parameter controlling the intensity of social behavior, ε is a preset minimum value, and D mn is the distance matrix between meerkat individuals m and n;

[0016] In the update stage, the global optimal position X best of the meerkat population is the goal.

[0017] Preferably, the autoregressive moving average model of the bird strike risk prediction combined model is a time series prediction model combined by an autoregressive model AR, a moving average model MA, and a differencing method model, and is used to capture the short-term dependencies of the time series of the bird strike risk dataset.

[0018] Preferably, the LSTM model of the bird strike risk prediction combined model consists of an input gate, a forget gate, and an output gate.

[0019] Preferably, the XGBoost model classifies each time span group based on the bird strike risk dataset. The XGBoost model constructs several weak learners. Each weak learner takes the fitting error of the previous weak learner as the learning target and continues to fit. Then, all the base learners are accumulated to obtain a better classification performance than a single model. The classification expression is as follows:

[0020] is the prediction result of sample i after the t2 - th tree iteration of the XGBoost model; is the prediction result of sample i after the (t2 - 1)-th tree iteration of the XGBoost model; f i (x i ) is the prediction result of sample i of the t2 - th tree.

[0021] Preferably, the objective function obj of the XGBoost model (t2) The expression is as follows:

[0022] Where is the loss function corresponding to the real result and the prediction result, is the regularization term.

[0023] Preferably, the method for obtaining the unit bird strike risk is as follows:

[0024] Where R is the unit bird strike risk corresponding to the time span group. The number of landings of the time span group represents the total number of landings of the aircraft within the time span group, and ∑ the severity of bird strike events in the time span group represents the sum of all the severity classification data of bird strike events within the time span group. The method for classifying the severity of bird strike events is as follows:

[0025] The classification data for slightly affected bird strike events corresponds to 1, the classification data for moderately affected bird strike events corresponds to 2, and the classification data for severely affected bird strike events corresponds to 3;

[0026] When normalizing during the method S1 for aggregating the unit bird strike risk, it is stored in the bird strike risk dataset.

[0027] Preferably, the time span groups are divided by day, and the time span of the time span groups is N1 days; or the time span groups are divided by month, and the time span of the time span groups is one natural month.

[0028] A bird strike risk prediction system based on an improved dwarf mongoose optimization model comprises a bird strike risk assessment index system, a data acquisition calculation module, a bird strike risk data set, a bird strike risk prediction combination model and an improved dwarf mongoose optimization algorithm model. The data acquisition calculation module is used to obtain historical bird strike event data before a research prediction date, divide the data into a plurality of time span groups according to a natural year, calculate the unit bird strike risk of each time span group based on the historical bird strike event data based on the bird strike risk assessment index system, and aggregate the unit bird strike risks according to time series and store them in the bird strike risk data set; the bird strike risk prediction combination model comprises an autoregressive moving average model, an LSTM model and an XGBoost model; the bird strike risk prediction combination model uses the bird strike risk data set for model training, and uses the improved dwarf mongoose optimization algorithm model to perform model parameter optimization processing on the bird strike risk prediction combination model; the bird strike risk prediction combination model after the model parameter optimization processing outputs the unit bird strike risk of a plurality of consecutive time span groups after the research prediction date according to time series.

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

[0030] (1) The present invention innovatively constructs a bird strike risk assessment index system, obtains historical bird strike event data before the research prediction date, divides the data into several time span groups according to a natural year, calculates the unit bird strike risk of each time span group based on the historical bird strike event data based on the bird strike risk assessment index system, and aggregates the unit bird strike risk according to the time series to form a bird strike risk data set; constructs a bird strike risk prediction combined model including an autoregressive moving average model, an LSTM model and an XGBoost model, uses the bird strike risk data set as a sample for model training to capture the trend, seasonality, time series characteristics and nonlinear relationship in the bird strike risk data, and uses the improved dwarf mongoose optimization algorithm model to optimize the model parameters, so as to predict the average wind shear risk of the span group for several consecutive time span groups after the research prediction date.

[0031] (2) The present invention uses an improved dwarf mongoose optimization algorithm model to optimize the hyperparameters of the bird strike risk prediction combined model, ensuring the optimal performance of the model. In terms of global exploration capability, in early iterations, it can effectively explore the solution space and avoid falling into local optimality; in terms of local development capability, in later iterations, the search range is significantly reduced, and the solution can be finely adjusted to improve convergence accuracy; in terms of convergence speed, this embodiment can find a better solution under the same number of iterations; and the prediction accuracy and robustness of the model bird strike risk are improved.

[0032] (3) Through multi - model fusion and parameter optimization, the model of the present invention can effectively capture the linear and non - linear characteristics of time - series data, significantly improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic flow chart of the bird - strike risk prediction method of the present invention;

[0034] Figure 2 It is a statistical chart of the bird - strike risk of the bird - strike risk data set in the embodiment, which contains 97 time - span groups and is grouped by monthly units in time series;

[0035] Figure 3 It is a schematic comparison diagram of the prediction results and the actual true results for the training set and the test set in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The present invention will be further described in detail below in conjunction with embodiments:

[0037] Embodiment

[0038] As Figure 1 shown, a bird - strike risk prediction method based on an improved dwarf mongoose optimization model, the method includes:

[0039] S1. Construct a bird - strike risk assessment index system, obtain historical bird - strike event data before the research prediction date, divide it into several time - span groups according to a natural year, calculate the unit bird - strike risk of each time - span group based on the bird - strike risk assessment index system for the historical bird - strike event data, and collect the unit bird - strike risks according to the time series to form a bird - strike risk data set.

[0040] In some embodiments, the time - span group is divided by day, and the time - span of the time - span group is N1 days; or the time - span group is divided by month, and the time - span of the time - span group is a natural month; in this embodiment, the time - span group is divided according to natural months. A natural year contains 12 natural months, so 12 time - span groups are divided in a natural year. For example: the research prediction date is February 2025 (the subsequent natural months after February 2025 are the time - span groups to be predicted), obtain the historical bird - strike event data before the research prediction date (that is, before February 2025. Taking the division rule of time - span groups by natural months as an example, before the research prediction date is before February 2025, specifically January 2025 and before January 2025), such as collecting the historical bird - strike event data from January 2017 to January 2025, a total of 97 natural months (that is, 97 time - span groups), and calculate the unit bird - strike risk of each time - span group (97 time - span groups in time series). In some embodiments, the method for obtaining the unit bird - strike risk is as follows:

[0041] Where R is the unit bird strike risk corresponding to the time span group, "number of takeoffs and landings in the time span group" represents the total number of takeoffs and landings of aircraft within the time span group, and "∑severity of bird strike events in the time span group" represents the sum of the severity classification data of all bird strike events within the time span group. The method for classifying the severity of bird strike events is as follows:

[0042] The classification data for minor impacts of bird strike events corresponds to 1 (classification description: minor impact, such as minor damage to the fuselage, does not affect flight), the classification data for medium impacts of bird strike events corresponds to 2 (classification description: medium impact, such as engine damage, requires an emergency return), and the classification data for severe impacts of bird strike events corresponds to 3 (classification description: severe impact, such as engine failure, may lead to an accident).

[0043] Preferably, after normalization (for example, 97 time span groups from January 2017 to January 2025 are first normalized and then stored in the bird strike risk dataset according to the time series) when collecting the unit bird strike risk in method S1, it is stored in the bird strike risk dataset. The normalization in this embodiment uses the following formula:

[0044]

[0045] Where i = 1, 2,..., n, n is the number of eigenvalue; y i ′, y i are the values before and after normalization respectively; min(y i ), max(y i ) are the minimum and maximum values of the feature before normalization respectively. In the example of collecting historical bird strike event data from January 2017 to January 2025 with the prediction date being February 2025, when collecting the unit bird strike risk, it is first normalized and then stored in the bird strike risk dataset. At this time, the data statistics of the bird strike risk dataset are as Figure 2 shown; the abscissa is the time span group sequence from January 2017 to January 2025, and the ordinate is the normalized unit bird strike risk (i.e., the monthly unit bird strike risk). It can be seen from the figure that the bird strike risk is relatively high in April, May in spring and September, October in autumn every year, and the curve as a whole has strong seasonal and time series characteristics.

[0046] S2. Construct a bird strike risk prediction combined model including an autoregressive moving average model, an LSTM model, and an XGBoost model, and input the bird strike risk dataset into the bird strike risk prediction combined model for model training.

[0047] In some embodiments, the autoregressive integrated moving average (ARIMA) model of the bird strike risk prediction combined model is a time series prediction model combined by an autoregressive (AR) model, a moving average (MA) model, and a differencing method model, and is used to capture the short-term dependencies of the time series of the bird strike risk dataset. The autoregressive integrated moving average model is a hybrid model with p-order autoregression and q-order moving average, denoted as ARIMA(p, q). In practice, the data series is a non-stationary time series. Therefore, in this embodiment, differencing is performed d times, denoted as ARIMA(p, d, q). After making it stationary, it will not change over time, and thus the future can be predicted based on the past behavior of the series. In the autoregressive integrated moving average model, the future value of the series is expressed as a linear function of the lag terms and the current and lagged values of the random disturbance term. The parameter configuration of the autoregressive integrated moving average model in this embodiment is: automatically select the optimal orders (p, q), where: p: the order of autoregression (AR), ranging from 0 to 4; q: the order of moving average (MA), ranging from 0 to 4. The autoregressive integrated moving average model (i.e., the ARIMA model) is used to capture the autocorrelation and moving average components in the residuals of the Holt-Winters model. The optimal model is selected by traversing the (p, q) combinations.

[0048] In some embodiments, the LSTM model (full English name: Long Short-Term Memory, Chinese full name: long short-term memory network, English abbreviation: LSTM) of the bird strike risk prediction combined model consists of an input gate, a forget gate, and an output gate. The LSTM model has the ability of long-term memory, and its principle formula is:

[0049] f t = σ(W f [h t-1 x t +b f ) (1)

[0050] i t = σ(W i [h t-1 x t +b i ) (2)

[0051]

[0052] O t = σ(W ｏ [h t-1 x ｔ +b o ) (5)

[0053] h t = O t tanh(C t) (6)

[0054] where x t , h t are the input and hidden states corresponding to time t respectively; f t , i t and O t are the states of the forget gate, input gate, and output gate respectively; C t and h t are the neuron and cell states to be updated respectively; W f , W i , W c , W o and b f , b i , b c , b o are the weight matrices and bias terms of each gate respectively; σ is the Sigmoid activation function in the hidden layer.

[0055] In some preferred embodiments, the XGBoost model classifies each time span group based on the bird strike risk dataset. The XGBoost model (optimizes the model performance through the gradient boosting framework, can handle non-linear relationships, supports feature engineering, and is suitable for complex data) constructs a number of weak learners. Each weak learner uses the fitting error of the previous weak learner as the learning objective to continue fitting, and then accumulates all the base learners to obtain a better classification performance than a single model. The classification expression is as follows:

[0056] is the prediction result of sample i after the t2-th tree iteration of the XGBoost model; is the prediction result of sample i after the (t2 - 1)-th tree iteration of the XGBoost model, and f i (x i ) is the prediction result of sample i of the t2-th tree. The objective function obj (t2) of the XGBoost model is expressed as follows:

[0057] where is the loss function corresponding to the true result and the prediction result, is the regularization term.

[0058] S3. Construct an improved Dwarf Mongoose Optimization Algorithm model (abbreviated as the improved DMOA model, the full English name of the Dwarf Mongoose Optimization Algorithm is Dwarf Mongoose Optimization Algorithm, and the English abbreviation is DMOA), and use the improved Dwarf Mongoose Optimization Algorithm model to optimize the model parameters of the bird strike risk prediction combined model. The improved Dwarf Mongoose Optimization Algorithm model includes an initialization stage, a foraging stage, a social stage, and an update stage, simulating the foraging, social, and group cooperation behaviors of dwarf mongooses (for example, the population size is N, and the position of the dwarf mongoose is represented as X m ), and is used to solve the training optimization problem of the bird strike risk prediction combined model. The expression of the initialization stage (initializing the population) is as follows:

[0059] X m = X min + rand·(X max - X min ), where X m is the initial position of mongoose individual m, X max and X min are the upper and lower bounds of the search range respectively, and rand is a random number of the model between [0, 1].

[0060] In the foraging stage, a dynamic weight ω(t) (by dynamically adjusting the weight ω(t), the algorithm pays more attention to global search in the initial stage and more attention to local exploitation in the later stage, so as to balance exploration and exploitation) and an adaptive perturbation η(t) (by dynamically adjusting the perturbation coefficient η(t), the algorithm introduces a larger random perturbation in the initial stage to enhance the exploration ability and reduces the perturbation in the later stage to enhance the exploitation ability) are introduced to search and update the position. The improved iterative update expression is as follows:

[0061] X m (t1 + 1)= X m (t1)+ ω(t1)·(X best - X m (t1)+ η(t1)·(X rand - X m (t1)), where X m (t1 + 1), X m (t1) are the positions of mongoose individual m at time t1 + 1 and time t1 iteration (time t1 iteration is also the t1-th iteration) respectively; X best is the global optimal position of the mongoose group, X rand is the position of a randomly selected mongoose individual, ω(t1) is the dynamic weight at time t1 iteration, and η(t1) is the adaptive perturbation at time t1 iteration. The dynamic weight ω(t1) in the foraging stage can adopt the following expression:

[0062] where ωmax is the maximum value of the dynamic weight, which is used to control the intensity of global exploration in the initial stage of the algorithm; ω min is the minimum value of the dynamic weight, which is used to control the intensity of local development in the later stage of the algorithm; t max is the maximum number of iterations of the algorithm, that is, one of the termination conditions of the optimization process; t1 is the current number of iterations (the dynamically adjusted weight corresponding to the current number of iterations is ω(t1)). The adaptive perturbation η(t1) in the foraging stage can be expressed as follows:

[0063] η max is the initial maximum value of the perturbation intensity, t max is the maximum number of iterations of the algorithm, that is, one of the termination conditions of the optimization process; t1 is the current number of iterations (the adaptive perturbation η(t1) corresponding to the current number of iterations). In the initial stage of the algorithm iteration, the perturbation intensity is usually set to η max to enhance the global exploration ability; as the iteration progresses, the perturbation intensity gradually decays (adjust the global exploration through the above formula); a larger η max helps the algorithm to widely explore the search space and avoid premature convergence to the local optimal solution. This embodiment adopts dynamic attenuation: usually combined with the number of iterations t1, to achieve the perturbation intensity gradually decreasing from η max to the minimum value, balancing exploration and development. The present invention adopts the adaptive perturbation η(t1)·(X rand -X m (t1)) in the foraging stage to enhance the model exploration ability.

[0064] In the social stage, the position is updated based on the distance-based social behavior, and the search direction is adjusted according to the distance. The expression is as follows:

[0065] where β is a parameter to control the intensity of social behavior, ε is a preset minimum value (preset to 0.001 in this embodiment), D mn is the distance matrix of meerkat individuals m and n; the distance matrix D mn The expression is as follows: D mn =||X m (r)-X n (t)||, X m (t), X n (t) represent the position vectors of meerkat individuals m and n respectively, and ||·|| represents the norm of the vector (Euclidean distance in this embodiment). In the formula represents the weight, indicating that the closer individuals have a greater impact on the current position; for example: if individual n is closer to individual m, then D mn is smaller, is larger, and the position difference X m (t)-X n(t) makes a greater contribution to the update. The improved formula above (the improved formula updates the position of each individual using a weighted average method. Each individual calculates a weighted average update direction based on the distance from other individuals. Specifically, individuals closer in distance have a greater impact on the current position, while those farther away have a smaller impact. The weighted method in the present invention makes individuals more inclined to move closer to individuals with a closer distance, enhancing the local search ability) functions as follows: Individual m tends to move closer to neighbors with a closer distance while weakening the influence of individuals at a greater distance. This mechanism simulates the behavior of "sharing local information" in a dwarf mongoose group, enhancing the local search ability of the algorithm. The improved formula for the social stage in the present invention enhances the local search ability and information sharing ability of the algorithm by introducing a distance matrix and a weighted average update mechanism. The improved formula can more precisely adjust the position of each individual, avoid premature convergence, and improve the global search efficiency of the algorithm.

[0066] The update stage aims at the global optimal position X of the mongoose group best as the target. For example, if f(x m ) is the objective function, then the global optimal position X of the mongoose group best (i.e., the optimal solution in the population) is expressed as follows: X best = arg min f(x m ).

[0067] S4. The bird strike risk prediction combined model after optimizing the model parameters outputs the unit bird strike risks of several consecutive time span groups after the research prediction date according to the time series.

[0068] A bird strike risk prediction system based on an improved dwarf mongoose optimization model, including a bird strike risk assessment index system, a data acquisition and calculation module, a bird strike risk data set, a bird strike risk prediction combined model, and an improved dwarf mongoose optimization algorithm model. The data acquisition and calculation module is used to obtain historical bird strike event data before the research prediction date, divide it into several time span groups according to a natural year, calculate the unit bird strike risks of each time span group based on the bird strike risk assessment index system, and collect and store the unit bird strike risks in the bird strike risk data set according to the time series. The bird strike risk prediction combined model includes an autoregressive moving average model, an LSTM model, and an XGBoost model. The bird strike risk prediction combined model uses the bird strike risk data set for model training and uses the improved dwarf mongoose optimization algorithm model to optimize the model parameters of the bird strike risk prediction combined model. The bird strike risk prediction combined model after optimizing the model parameters outputs the unit bird strike risks of several consecutive time span groups after the research prediction date according to the time series.

[0069] To further verify the prediction performance of the combined model of the present invention, four statistical indicators are selected to evaluate the prediction results, namely, root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R 2 ). RMSE, MSE, and MAE can evaluate the degree of data variation. The smaller their values, the higher the accuracy of the prediction model in describing the experimental data. R 2 is used to evaluate the fitting effect of the model. The closer its value is to 1, the better the fitting effect. In the example of the airport bird strike risk dataset from January 2017 to January 2025, in this embodiment, the training set and the test set are divided for model prediction respectively. After the training set and the test set are predicted respectively, the comparison results between the prediction results and the actual true results are as Figure 3 shown. At the same time, the present invention also conducts a control test on whether to optimize the bird strike risk prediction combined model by using the improved dwarf mongoose optimization model. See the following table:

[0070]

[0071]

[0072] As can be seen from the table, compared with not using the improved dwarf mongoose optimization model for optimization, the bird strike risk prediction combined model optimized by the improved dwarf mongoose optimization model has a 55.67% reduction in MSE and a 34.08% reduction in MAE in the prediction results; R 2 is increased by 9.61%. It can be seen that the improved dwarf mongoose optimization model has a very excellent effect on improving the bird strike risk prediction accuracy of the bird strike risk prediction combined model of the present invention, and all aspects of the index have been greatly improved.

[0073] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A bird strike risk prediction method based on an improved dwarf mongoose optimization model, characterized in that: The method includes: S1. Construct a bird strike risk assessment index system, obtain historical bird strike event data before the research prediction date, divide it into several time span groups according to a natural year, calculate the unit bird strike risk of each time span group based on the bird strike risk assessment index system for the historical bird strike event data, and collect the unit bird strike risks in time series to form a bird strike risk data set; S2. Construct a bird strike risk prediction combined model including an autoregressive moving average model, an LSTM model, and an XGBoost model, and input the bird strike risk data set into the bird strike risk prediction combined model for model training; S3. Construct an improved dwarf mongoose optimization algorithm model, and use the improved dwarf mongoose optimization algorithm model to optimize the model parameters of the bird strike risk prediction combined model; S4. The bird strike risk prediction combined model after model parameter optimization processes outputs the unit bird strike risks of several consecutive time span groups after the research prediction date in time series.

2. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1, wherein: The improved dwarf mongoose optimization algorithm model includes an initialization stage, a foraging stage, a social stage, and an update stage. The expression in the initialization stage is as follows: X m = X min + rand · (X max - X min ), where X m is the initial position of the meerkat individual m, X max and X min are the upper and lower bounds of the search range respectively, and rand is a random number of the model between [0, 1]; In the foraging stage, a dynamic weight ω(t) and an adaptive perturbation η(t) are introduced to search and update the position. The improved iterative update expression is as follows: X m (t1 + 1) = X m (t1) + ω(t1)·(X best -X m (t1)) + η(t1)·(X rand -X m (t1))), where X m (t1 + 1), X m (t1) are the positions of meerkat individual m at iterations of time t1 + 1 and time t1 respectively; X best is the global optimal position of the meerkat population, X rand is the position of a randomly selected meerkat individual, ω(t1) is the dynamic weight at iteration of time t1, and η(t1) is the adaptive perturbation at iteration of time t1; In the social stage, the position is updated based on distance-based social behavior, and the search direction is adjusted according to the distance. The expression is as follows: where β is a parameter controlling the intensity of social behavior, ε is a preset minimum value, and D mn is the distance matrix between meerkat individuals m and n; The update phase aims at the global optimal position X of the mongoose swarm best as the target.

3. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1, wherein: The autoregressive moving average model of the bird strike risk prediction combined model is a time series prediction model combined by an autoregressive model AR, a moving average model MA, and a differencing method model and is used to capture the short-term dependencies of the time series of the bird strike risk data set.

4. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1, wherein: The LSTM model of the bird strike risk prediction combined model consists of an input gate, a forget gate, and an output gate.

5. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1, characterized in that: The XGBoost model classifies each time span group based on the bird strike risk data set. The XGBoost model constructs several weak learners, and each weak learner takes the fitting error of the previous weak learner as the learning target to continue fitting. Then all the base learners are accumulated to obtain a better classification performance than a single model. The classification expression is as follows: is the prediction result of sample i after the t2 - th tree iteration of the XGBoost model; is the prediction result of sample i after the (t2 - 1)-th tree iteration of the XGBoost model; f i (x i ) is the prediction result of sample i of the t2 - th tree.

6. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 5, characterized in that: The objective function obj of the XGBoost model (t2) The expression is as follows: wherein is the loss function corresponding between the true result and the predicted result, is the regularization term.

7. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1, characterized in that: The method for obtaining the unit bird strike risk is as follows: Where R is the unit bird strike risk corresponding to the time span group, the number of landings and takeoffs in the time span group represents the total number of landings and takeoffs of aircraft within the time span group, and the severity of bird strike events in the time span group represents the sum of the severity classification data of all bird strike events within the time span group; the method for classifying the severity of bird strike events is as follows: The grading data of slightly affected bird strike events corresponds to 1, the grading data of moderately affected bird strike events corresponds to 2, and the grading data of severely affected bird strike events corresponds to 3; When collecting the unit bird strike risk in method S1, after normalization processing, it is stored in the bird strike risk data set.

8. The bird strike risk prediction method based on the improved dwarf mongoose optimization model according to claim 1 or 7, characterized in that: The time span group is divided by day, and the time span of the time span group is N1 days; or the time span group is divided by month, and the time span of the time span group is a natural month.

9. A bird strike risk prediction system based on an improved dwarf mongoose optimization model, characterized in that: The invention comprises a bird strike risk assessment index system, a data acquisition calculation module, a bird strike risk data set, a bird strike risk prediction combination model and an improved dwarf mongoose optimization algorithm model. The data acquisition calculation module is used to obtain historical bird strike event data before the research prediction date and divide the data into a plurality of time span groups according to a natural year. The unit bird strike risk of each time span group is calculated based on the bird strike risk assessment index system for the historical bird strike event data, and the unit bird strike risk is aggregated according to the time series and stored in the bird strike risk data set. The bird strike risk prediction combination model comprises an autoregressive moving average model, an LSTM model and an XGBoost model. The bird strike risk prediction combination model uses the bird strike risk data set for model training, and uses the improved dwarf mongoose optimization algorithm model to perform model parameter optimization processing on the bird strike risk prediction combination model. After the model parameters are optimized, the bird strike risk prediction combined model outputs the unit bird strike risk of several consecutive time span groups after the prediction date in time series.

Images