Gated cycle and whale algorithm combined wind shear risk prediction method and system

Abstract

The invention discloses a wind shear risk prediction method and system in combination with a gating cycle and a whale algorithm. The method comprises the following steps: S1, collecting historical wind shear data of all take-off and landing transport aircrafts in a research area for several years according to a time sequence; segmenting into m time slices according to a natural year, sorting out a time slice risk value and time slice data for historical wind shear data according to each time slice, and carrying out associated coding storage; s2, the gating loop unit model takes the time slice risk value as a predicted value and takes the time slice data as data content to carry out prediction learning iteration training; s3, performing model hyper-parameter optimization processing on the gating circulation unit model by adopting a spiral strategy of an improved whale algorithm; and S4, collecting wind shear data of the current time slice, inputting the wind shear data into the gating circulation unit model, and outputting a risk prediction value. The whale optimization algorithm is utilized to optimize the hyper-parameter combination of the gating circulation unit model, and the wind shear risk prediction method shows remarkable performance advantages in a wind shear risk prediction task.

Description

Technical Field

[0001] The present invention relates to the field of wind shear risk prediction for take-off and landing transport aircraft, and in particular to a wind shear risk prediction method and system combining a gated cycle with a whale algorithm. Background Art

[0002] Accurate prediction of wind shear risk during civil aviation flight is crucial to ensure flight safety; wind shear refers to the phenomenon of rapid changes in wind speed or direction over a short distance, especially during aircraft takeoff and landing, which can bring huge risks. Therefore, the prediction of wind shear risk needs to deal with complex nonlinear dynamic characteristics and unstable atmospheric conditions (for example, literature 1: Mirjalili, S., & Lewis, A. (2016). The Whale Optimization Algorithm. Advances in Engineering Software, 95, 51-67; Literature 2: Vaswani, A., Shazeer, N., Parmar, N., et al. (2017). Attention is All You Need. Advances in Neural Information Processing Systems, 30, 5998-6008). In recent years, deep learning technology has been widely used in the field of time series prediction, providing a new solution for wind shear risk prediction. Among them, recurrent neural networks (RNNs) and their variants (such as long short-term memory networks (LSTMs)) have been widely used in fields such as wind shear and meteorological data prediction due to their excellent performance in capturing nonlinear and time series features. However, the above network models rely heavily on the setting of hyperparameters, such as learning rate, number of hidden layer neurons, and batch size. Traditional manual parameter adjustment methods are inefficient and difficult to find the optimal solution (see reference 3: Pahwa, M., & Tiwari, S. (2021). A Study on Hyperparameter Optimization Techniques for Neural Networks. International Journal of Information Technology, 13 (1), 223-228). In model learning training, optimization algorithms can tune hyperparameters, such as reference 4: [8] Liu, S., et a1. (2019). A Survey on Deep Learning with Attention Mechanism. IEEE Access, 7, 42023-42045, studied various hyperparameter optimization methods, emphasizing the importance of intelligent optimization algorithms in deep learning; traditional optimization methods such as particle swarm optimization (PSO) are prone to fall into local optimal solutions in improving model prediction accuracy. In the prediction of wind shear risk for take-off and landing transport aircraft, it is urgent to study a shear risk prediction method and model, and at the same time, match the model hyperparameter optimization method and strategy to achieve effective prediction of wind shear risk. Summary of the invention

[0003] The purpose of the present invention is to provide a wind shear risk prediction method and system that combines gated circulation and whale algorithm. The gated cyclic unit model is used to capture the temporal characteristics of wind shear risk data and the characteristics of various weather and geographical data as factors. At the same time, the whale optimization algorithm is used to optimize the hyperparameter combination of the gated cyclic unit model, thereby effectively improving the generalization ability of the model.

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

[0005] A wind shear risk prediction method combining gated loop and whale algorithm, the method comprising:

[0006] S1. Collect historical wind shear data of several years for all take-off and landing transport aircraft in the study area in chronological order. The historical wind shear data includes wind shear events and corresponding data of the events. The corresponding data of the events include meteorological data, geographical data, and transport aviation time in the study area. Divide the time slices into m time slices according to the natural year, sort out the time slice risk value and time slice data according to each time slice, and store them in associated codes. The time slice data includes the meteorological data and geographic data corresponding to the time slice;

[0007] S2. Construct a gated recurrent unit model and input the time slice risk value and time slice data stored in the associated code. The gated recurrent unit model uses the time slice risk value as the prediction value and the time slice data as the data content to perform prediction learning iterative training;

[0008] S3. The gated recurrent unit model uses the spiral strategy of the improved whale algorithm to optimize the model hyperparameters. The spiral strategy expression of the improved whale algorithm is as follows:

[0009]

[0010] Where X(j+1) represents the parameter output at the j+1th iteration, X * (j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations;

[0011] S4. Collect wind shear data of the current time slice and input it into the trained gated recurrent unit model. The wind shear data of the current time slice includes wind shear events of the current time slice and corresponding data of events of the current time slice. The corresponding data of events of the current time slice includes meteorological data of the current time slice, geographic data of the current time slice, and transportation aviation time within the study area of ​​the current time slice. The gated recurrent unit model outputs the risk prediction value corresponding to the current time slice.

[0012] In order to better implement the present invention, the time slices are time-series encoded, the time slice risk values ​​and the time slice data are further encoded after the time-series encoding, and the encoded time slice risk values ​​and time slice data are associated and stored.

[0013] Preferably, in step S2, the gated recurrent unit model includes a reset gate and an update gate, and the gated recurrent unit model outputs a risk prediction value expression as follows:

[0014] h t =(1-z t )×h t-1 +z t ×g t

[0015] g t =tanh(W t ×[r t ×h t-1 , x t ]+b t )

[0016] where h t Represents the output risk prediction value, z t Indicates that the time slice t updates the gate output data, h t-1 represents the hidden state data of time slice t-1, r t Indicates that the time slice t resets the gate output data; g t represents the memory data of time slice t, and the memory data is the data items and data extracted from the time slice data; x t Indicates the data content of time slice t, b t represents the bias term of time slice t, tanh represents the hyperbolic tangent activation function, W t Represents the weight matrix for time slice t in the model.

[0017] Preferably, the time slice t reset gate output data expression is as follows:

[0018] r t =σ(W r ×[h t-1 , x t ]+b r ), where W r represents the weight matrix of time slice t in the reset gate processing, b r represents the weight matrix of time slice t in the reset gate processing, [h t-1 , x t ] represents the hidden state data of time slice t-1 corresponding to the data content of time slice t, [h t-1 , x t] Take the hidden state data of time slice t-1; σ represents the sigmoid activation function;

[0019] The expression of the gate output data for time slice t is as follows:

[0020] z t =σ(W z ×[h t-1 , x t ]+b z )

[0021] Where W z represents the weight matrix of time slice t in the update gate processing, b z Represents the weight matrix for time slice t in the update gate processing.

[0022] Preferably, in step S3, during the model hyperparameter optimization process, the gated recurrent unit model further selects part of the historical wind shear data as samples and uses the MSE loss function for loss constraint. The MSE loss function expression is as follows:

[0023]

[0024] where y i is the actual value, y 预测，i is the predicted value, and N is the number of samples.

[0025] Preferably, in step S3, the spiral strategy of the improved whale algorithm uses the adjustment of the control parameter α to gradually shrink the search space and optimize the iteration, and the coefficient A is used to determine the position update mode of the individual whale, A=2a×ra, a represents the control parameter, the control parameter is a variable that decreases linearly in the iteration, and the value gradually decreases from 2 to 0, r represents a random number between [0, 1], and the specific method is as follows:

[0026] If |A|<1, execute to make the current individual approach the optimal solution output;

[0027] If |A| ≥ 1, the execution randomly selects another whale individual from the population and updates the current solution.

[0028] Preferably, the spiral strategy of the improved whale algorithm uses the current optimal solution as prey, and other solutions gradually approach the target. The position update expression is as follows:

[0029] D=|2*r*X * -X|

[0030] X 更新 =X * -A*D

[0031] X *represents the location of the current best solution, X represents the current location, D represents the distance between the current solution and the optimal solution, X 更新 Indicates the updated position.

[0032] Preferably, in step S3, when the improved whale algorithm performs model hyperparameter optimization processing on the gated recurrent unit model, the improved whale algorithm also adopts the following dynamic adjustment strategy:

[0033] X(j+1)=X(j)+μ*(X * (j)-X(j))-A*D j

[0034] D j =|2*r*X * (j)-X(j)|

[0035] D j Represents X * (j) is the distance from X(j), and μ represents the hyperparameter sensitivity factor.

[0036] A wind shear risk prediction system combining gated loop and whale algorithm comprises a data collection module, a data sorting module, a gated loop unit model and a real-time data collection input module, wherein the data collection module is used to collect historical wind shear data of several years for all take-off and landing transport aircraft in a study area in chronological order, wherein the historical wind shear data comprises wind shear events and event corresponding data, wherein the event corresponding data comprises meteorological data, geographical data and transport aviation time in the study area; the data sorting module divides the historical wind shear data into m time slices according to natural years, sorts out time slice risk values ​​and time slice data according to each time slice, and stores the associated codes; The time slice data includes the meteorological data and geographic data corresponding to the time slice; the gated recurrent unit model inputs the time slice risk value and the time slice data stored in the association code, and the gated recurrent unit model uses the time slice risk value as the prediction value and the time slice data as the data content for predictive learning iterative training; the gated recurrent unit model uses the spiral strategy of the improved whale algorithm to optimize the model hyperparameters, and the spiral strategy expression of the improved whale algorithm is as follows:

[0037]

[0038] Where X(j+1) represents the parameter output at the j+1th iteration, X * (j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations;

[0039] The real-time data collection input module is used to collect the wind shear data of the current time slice and input it into the trained gated cyclic unit model. The wind shear data of the current time slice includes the wind shear event of the current time slice and the corresponding data of the event of the current time slice. The corresponding data of the event of the current time slice includes the meteorological data of the current time slice, the geographical data of the current time slice, and the transportation aviation time in the study area of ​​the current time slice; the trained gated cyclic unit model processes the wind shear data of the current time slice and outputs the risk prediction value corresponding to the current time slice.

[0040] 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 the steps of the wind shear risk prediction method combining a gated loop with a whale algorithm according to the present invention.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] (1) The present invention captures the time series of wind shear risk data and the characteristics of various weather and geographical data as factors through a gated recurrent unit model, and uses the whale optimization algorithm to optimize the hyperparameter combination of the gated recurrent unit model, thereby effectively improving the generalization ability of the model; the gated recurrent unit model exhibits good performance when processing long-term dependent data, and the whale optimization algorithm provides a powerful global search capability, enabling it to more accurately adjust the model parameter combination; the present invention exhibits significant performance advantages in the task of wind shear risk prediction, and its superiority is mainly reflected in the convergence speed, prediction accuracy and the ability to adapt to complex data.

[0043] (2) The present invention also introduces a dynamic adjustment strategy for constraint conditions, which further improves the prediction accuracy and robustness of the model by adaptively adjusting the parameter search space according to real-time feedback during the training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a method flow chart of the wind shear risk prediction method of the present invention;

[0045] Figure 2 It is a schematic diagram of the spiral strategy principle of improving the whale algorithm in the embodiment;

[0046] Figure 3 Schematic diagram of the principle of iterative processing of the gated recurrent unit model after hyperparameter optimization in the embodiment;

[0047] Figure 4 This is a comparison chart of the convergence curves of the improved whale algorithm and other algorithms in the embodiment;

[0048] Figure 5 The comparison result between the test set and the real value when testing in the embodiment;

[0049] Figure 6 It is a comparison chart of the convergence curves of the embodiment with and without the constraint dynamic adjustment strategy at 1000 iterations;

[0050] Figure 7 A performance comparison chart of the embodiment in which the constraint dynamic adjustment strategy is introduced and the embodiment in which the constraint dynamic adjustment strategy is not introduced in 1000 iterations;

[0051] Figure 8 It is a principle structural block diagram of the wind shear risk prediction system of the present invention. DETAILED DESCRIPTION

[0052] The present invention is further described in detail below in conjunction with embodiments:

[0053] Example

[0054] like Figure 1 As shown, a wind shear risk prediction method combining gated loop and whale algorithm, the method includes:

[0055] S1. Collect historical wind shear data of several years in chronological order for all transport aircraft taking off and landing in the study area (the study area can select transport aircraft taking off and landing in a certain country or province, and the transport aircraft are mainly medium and large passenger and cargo aircraft for transportation). The historical wind shear data include wind shear events and corresponding data of the events. The corresponding data of the events include meteorological data (including wind direction, wind speed, convective weather data, temperature, etc.), geographic data (including topography, surface roughness, interface data, etc., the interface includes water and land, land and mountain, etc.), and transport aviation time in the study area (transport aviation time only intercepts the time when the aircraft track is only located in the study area). Divide the natural year into m time slices (set according to the required prediction time slice accuracy, the time slices are generally segmented by day, or by time periods within a day and sliced ​​into segments), sort the historical wind shear data according to each time slice to obtain time slice risk values ​​and time slice data, and store them in associated codes; preferably, perform time series coding on the time slices, further encode the time slice risk values ​​and time slice data after time series coding, and store the encoded time slice risk values ​​and time slice data in associated codes.

[0056] The time slice data includes the meteorological data and geographic data corresponding to the time slice.

[0057] This embodiment selects a study area and takes a monthly time slice of 1 month as an example to perform historical wind shear data statistics. The historical wind shear data spans from January 2017 to November 2024. When the time slice is divided into natural months, the time slice risk value intercepted part of the data is as follows:

[0058]

[0059] The present invention uses the time slice risk value as the risk value of wind shear occurring in the study area within the time slice, and statistically calculates the time slice risk value of historical wind shear data as the model prediction value. The time slice risk value is used as the prediction value and the time slice data is used as the data content for predictive learning iterative training. Subsequent meteorological data and geographic data (such as October, November or December in 2024 in the above table) are collected in real time and the corresponding time slice risk value is predicted. The above table takes the month as an example. In the specific implementation, time slicing can be performed according to the day or the time period of the day or the hour interval, etc., to achieve the purpose of more accurate risk prediction.

[0060] S2. Construct a gated recurrent unit model and input the time slice risk value and time slice data stored in the associated code. The gated recurrent unit model uses the time slice risk value as the prediction value and the time slice data as the data content to perform predictive learning iterative training.

[0061] In some embodiments, the gated recurrent unit model in step S2 includes a reset gate and an update gate, and the gated recurrent unit model outputs a risk prediction value expression as follows:

[0062] h t =(1-z t )×h t-1 +z t ×g t

[0063] g t =tanh(W t ×[r t ×h t-1 , x t ]+b t )

[0064] where h t Represents the output risk prediction value, z t Indicates that the time slice t updates the gate output data, h t-1 represents the hidden state data of time slice t-1, r t Indicates that the time slice t resets the gate output data. t Represents the memory data of time slice t, and the memory data is the data items and data extracted from the time slice data. t Indicates the data content of time slice t, bt represents the bias term of time slice t, tanh represents the hyperbolic tangent activation function, W t Represents the weight matrix for time slice t in the model.

[0065] The expression of the reset gate output data for time slice t is as follows:

[0066] r t =σ(W r ×[h t-1 , x t ]+b t ), where W r represents the weight matrix of time slice t in the reset gate processing, b r represents the weight matrix of time slice t in the reset gate processing, [h t-1 , x t ] represents the hidden state data of time slice t-1 corresponding to the data content of time slice t, [h t-1 , x t ] Take the hidden state data of time slice t-1. σ represents the sigmoid activation function.

[0067] The expression of the gate output data for time slice t is as follows:

[0068] z t =σ(W z ×[h t-1 , x t ]+b z )

[0069] Where W z represents the weight matrix of time slice t in the update gate processing, b z Represents the weight matrix for time slice t in the update gate processing.

[0070] S3, the gated recurrent unit model uses the spiral strategy of the improved whale algorithm to perform modular hyperparameter optimization. The spiral strategy expression of the improved whale algorithm is as follows:

[0071]

[0072] Where X(j+1) represents the parameter output at the j+1th iteration, X * (j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations. In this embodiment, The value of k is 5; when the search starts, the value of j is very small and T is large, so Close to 0, The value is close to e 5 ; With gradual iteration, j gradually equals T, Close to e -5 , is also close to 0. That is to say, the search range is large at the beginning and gradually becomes smaller, which is equivalent to performing a global large-scale search at the beginning, slowly narrowing the range, and having good convergence ability. In the early stage of the algorithm, this embodiment uses a larger rate to converge to the optimal solution range to improve the global search ability, and reduces its convergence rate in the later stage to enhance the local search ability. In the search process, the whale can dynamically adjust the spiral shape during the search according to the number of iterations, thereby improving the diversity of the search path to obtain the global optimal solution.

[0073] The present invention compares the improved whale algorithm (IWOA), the grey wolf algorithm (GWO), the crested porcupine optimizer (CPO), the standard whale algorithm (WOA), and the particle swarm optimization algorithm (PSO). The results are as follows Figure 4 As shown, the horizontal axis represents the number of iterations, and the vertical axis represents the best fitness value (i.e., the function value, the smaller the function value, the better, that is, the lower the curve, the better). After comparative analysis, the improved whale algorithm (IWOA) of the present invention has the best search and convergence performance. The gated recurrent unit model of the present invention adopts the improved whale algorithm (IWOA) to perform model hyperparameter optimization processing.

[0074] In some embodiments, during the model hyperparameter optimization process, the gated recurrent unit model also selects some historical wind shear data as samples and uses the MSE loss function for loss constraint. The MSE loss function expression is as follows:

[0075]

[0076] where y i is the actual value, y 预测，i is the predicted value, and N is the number of samples. The goal of the improved whale algorithm (IWOA) of the present invention is to minimize the MSE loss function through the optimization of hyperparameters, thereby improving the prediction accuracy of the gated recurrent unit model.

[0077] In some embodiments, the spiral strategy of the improved whale algorithm in step S3 uses the adjustment of the control parameter α to gradually shrink the search space and optimize the iteration, and uses the coefficient A to determine the position update method of the individual whale, A = 2a × ra, a represents the control parameter, the control parameter is a linearly decreasing variable in the iteration, and the value gradually decreases from 2 to 0, r represents a random number between [0, 1], such as Figure 2 As shown, the specific method is as follows:

[0078] If |A|<1, execute to make the current individual approach the optimal solution output;

[0079] If |A| ≥ 1, the execution randomly selects another whale individual from the population and updates the current solution.

[0080] The spiral strategy of the improved whale algorithm takes the current optimal solution as prey, and other solutions gradually approach the target. The position update expression is as follows:

[0081] D=|2*r*X * -X|

[0082] X 更新 =X * -A*D

[0083] X * represents the location of the current best solution, X represents the current location, D represents the distance between the current solution and the optimal solution, X 更新 The spiral strategy of the improved whale algorithm of the present invention is to update the hyperparameter combination (such as learning rate α, number of hidden layer neurons n h , batch size B) to generate model hyperparameter data of a new gated recurrent unit model, and calculate its loss function; after continuous iteration, the improved whale algorithm of the present invention gradually approaches the global optimal solution, thereby obtaining the optimal hyperparameter combination, and improving the performance of the gated recurrent unit model in wind shear risk prediction. Specifically, the optimization process of the improved whale algorithm of the present invention includes two behavioral strategies: "surrounding prey" and "spiral position update". In the stage of surrounding prey, the improved whale algorithm of the present invention assumes that the current optimal solution is the prey position, and other solutions gradually approach the optimal solution. In the spiral position update stage, when the distance between the individual and the optimal solution is small, the improved whale algorithm of the present invention simulates the spiral hunting behavior of the whale, and updates the position through a spiral path to increase the search diversity; the improved whale algorithm of the present invention gradually optimizes the hyperparameter combination of the gated recurrent unit model.

[0084] In some embodiments, when the improved whale algorithm in step S3 performs model hyperparameter optimization processing on the gated recurrent unit model, the improved whale algorithm also adopts the following dynamic adjustment strategy (the present invention can introduce a dynamic adjustment strategy to take into account the sensitivity of different hyperparameters, so that the update amount of each hyperparameter is adaptively adjusted according to its impact on the gated recurrent unit model, such as Figure 3 shown):

[0085] X(j+1)=X(j)+μ*(X * (j)-X(j))-A*D j

[0086] D j =|2*r*X * (j)-X(j)|

[0087] D j Represents X* (j) is the distance between X(j) and X(j), μ represents the hyperparameter sensitivity factor (adjusted according to the contribution of each hyperparameter to MSE); by dynamically adjusting the strategy, it can ensure that the update step size of the hyperparameters that have a greater impact on the model is smaller, while the hyperparameters with less impact can have a larger adjustment space, thereby further accelerating convergence and improving the final model performance.

[0088] In a preferred embodiment, a constrained dynamic adjustment strategy is introduced into the optimization process of the wind shear risk prediction model of the present invention (including the gated cyclic unit model of the improved whale algorithm) to further improve the prediction accuracy, convergence speed and stability of the model. This strategy dynamically narrows the search space in each iteration, thereby effectively guiding the whale optimization algorithm (WOA) to search within a more optimal parameter range. This method is particularly suitable for high-dimensional and complex data sets, and helps to reduce randomness and fluctuations in the optimization process. Thus, the convergence curves of the wind shear risk prediction model with a constrained dynamic adjustment strategy and the wind shear risk prediction model without a constrained dynamic adjustment strategy under 1000 iterations are shown as follows: Figure 6 As shown in the figure, the model with dynamic adjustment strategy showed faster convergence speed in the initial stage, and the MSE value dropped rapidly and stabilized after about 200 iterations, which was significantly earlier than the model without dynamic adjustment. In contrast, although the wind shear risk prediction model without dynamic adjustment also gradually decreased during the iteration process, its convergence speed was slower and fluctuated significantly in the first few hundred iterations; the dynamic adjustment strategy provided stronger convergence guidance during the optimization process, suppressed the disorder of parameter search, and made the model approach the global optimal solution faster. The models with and without constrained dynamic adjustment strategy were compared in MSE, MAE and R. 2 Compare the indicators, such as Figure 7 As shown in Figure 2, the model with dynamic adjustment strategy outperforms the standard model (the model without constrained dynamic adjustment strategy) in all the above evaluation indicators. Specifically, the dynamic adjustment strategy reduces the MSE to 0.01, the MAE index to 0.08, and the R 2The index is significantly improved to 0.95. These results show that the dynamic adjustment strategy can more effectively mine the nonlinear and time series features in the data, thereby improving the generalization ability and prediction accuracy of the model. The reason for these performance improvements is that the dynamic adjustment strategy gradually narrows the parameter search space, so that the optimization process can adaptively adjust the search range, avoiding the "free" phenomenon caused by the whale optimization algorithm in the later stage due to the excessive search range. At the same time, the strategy also significantly improves the smoothness of the convergence curve, making the model more stable in the process of approaching the optimal solution; this convergence smoothness not only improves the training efficiency of the model, but also reduces the randomness and uncertainty in the optimization process, ensuring the efficiency and robustness of the model. In summary, the GRU-WOA model (the wind shear risk prediction model of the present invention, that is, the gated recurrent unit model containing the improved whale algorithm) after introducing the constraint dynamic adjustment strategy shows excellent performance and stability in the wind shear risk prediction task, showing its application potential in the prediction of high-complexity time series data.

[0089] S4. Collect wind shear data of the current time slice and input it into the trained gated recurrent unit model. The wind shear data of the current time slice includes wind shear events of the current time slice and corresponding data of events of the current time slice. The corresponding data of events of the current time slice includes meteorological data of the current time slice, geographic data of the current time slice, and transportation aviation time within the study area of ​​the current time slice. The gated recurrent unit model outputs the risk prediction value corresponding to the current time slice.

[0090] A wind shear risk prediction system combining gated loop and whale algorithm, such as Figure 8 As shown, it includes a data collection module, a data sorting module, a gated cyclic unit model and a real-time data collection input module. The data collection module is used to collect historical wind shear data of several years in chronological order for all take-off and landing transport aircraft in the study area. The historical wind shear data includes wind shear events and event corresponding data. The event corresponding data includes meteorological data, geographical data, and transport aviation time in the study area. The data sorting module divides the time slices into m time slices according to the natural year, sorts out the time slice risk value and time slice data according to each time slice, and stores them in associated codes. The time slice data includes the meteorological data and geographic data corresponding to the time slice. The gated recurrent unit model inputs the time slice risk value and the time slice data stored in the association code, and the gated recurrent unit model uses the time slice risk value as the prediction value and the time slice data as the data content for predictive learning and iterative training. The gated recurrent unit model uses the spiral strategy of the improved whale algorithm to optimize the model hyperparameters. The spiral strategy expression of the improved whale algorithm is as follows:

[0091]

[0092] Where X(j+1) represents the parameter output at the j+1th iteration, X * (j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations.

[0093] The real-time data collection input module is used to collect the wind shear data of the current time slice and input it into the trained gated cyclic unit model. The wind shear data of the current time slice includes the wind shear event of the current time slice and the corresponding data of the event of the current time slice. The corresponding data of the event of the current time slice includes the meteorological data of the current time slice, the geographical data of the current time slice, and the transportation aviation time in the study area of ​​the current time slice. The trained gated cyclic unit model processes the wind shear data of the current time slice and outputs the risk prediction value corresponding to the current time slice; Figure 5 As shown, the historical wind shear data can be divided into a training set and a test set (in order to ensure the applicability of the model in different data environments, the data set of the historical wind shear data is preprocessed, including normalization and missing value filling steps to ensure the consistency of the data and the stability of the model training. In addition, the data set is divided into 70% of the training set and 30% of the test set, which are used for model training and verification, respectively, to ensure the objectivity and comprehensiveness of the model performance evaluation), and then the time slice risk value is predicted. The time slice risk value prediction of the trained gated cyclic unit model (at the same time, the model hyperparameters are optimized by the spiral strategy of the improved whale algorithm) has a good fit with the true value. At the same time, the present invention can also synchronously output the number of wind shears that have occurred in the current time slice and the number of wind shears that may occur in the current time slice. The number of wind shears that may occur in the current time slice = risk prediction value × transportation aviation time in the study area of ​​the current time slice - transportation aviation time in the study area of ​​the current time slice.

[0094] An electronic device includes at least one processor, at least one memory and a data bus. 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 wind shear risk prediction method combining the gated loop and the whale algorithm of the present invention.

[0095] 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 principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A wind shear risk prediction method combining gated loop and whale algorithm, characterized by: The methods include: S1. Collect historical wind shear data of several years for all take-off and landing transport aircraft in the study area in chronological order. The historical wind shear data includes wind shear events and corresponding data of the events. The corresponding data of the events include meteorological data, geographical data, and transport aviation time in the study area. Divide the time slices into m time slices according to the natural year, sort out the time slice risk value and time slice data according to each time slice, and store them in associated codes. The time slice data includes the meteorological data and geographic data corresponding to the time slice; S2. Construct a gated recurrent unit model and input the time slice risk value and time slice data stored in the associated code. The gated recurrent unit model uses the time slice risk value as the prediction value and the time slice data as the data content to perform prediction learning iterative training; S3. The gated recurrent unit model uses the spiral strategy of the improved whale algorithm to optimize the model hyperparameters. The spiral strategy expression of the improved whale algorithm is as follows: Where X(j+1) represents the parameter output at the j+1th iteration, X*(j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations; S4. Collect wind shear data of the current time slice and input it into the trained gated recurrent unit model. The wind shear data of the current time slice includes wind shear events of the current time slice and corresponding data of events of the current time slice. The corresponding data of events of the current time slice includes meteorological data of the current time slice, geographic data of the current time slice, and transportation aviation time within the study area of ​​the current time slice. The gated recurrent unit model outputs the risk prediction value corresponding to the current time slice.

2. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 1 is characterized in that: The time slice is time-series encoded, and the time slice risk value and the time slice data are further encoded after the time-series encoding, and the encoded time slice risk value and time slice data are associated and stored.

3. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 1 is characterized in that: In step S2, the gated recurrent unit model includes a reset gate and an update gate, and the gated recurrent unit model outputs a risk prediction value expression as follows: h t =(1-z t )×h t-1 +z t ×g t g t =tanh(W t ×[r t ×h t-1 ,x t ]+b t ) where h t Represents the output risk prediction value, z t Indicates that the time slice t updates the gate output data, h t-1 represents the hidden state data of time slice t-1, r t Indicates that the time slice t resets the gate output data; g t represents the memory data of time slice t, and the memory data is the data items and data extracted from the time slice data; x t Indicates the data content of time slice t, b t represents the bias term of time slice t, tanh represents the hyperbolic tangent activation function, W t Represents the weight matrix for time slice t in the model.

4. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 3 is characterized in that: The expression of the reset gate output data for time slice t is as follows: r t =σ(W r ×[h t-1 , x t ]+b r ), where W r represents the weight matrix of time slice t in the reset gate processing, b r represents the weight matrix of time slice t in the reset gate processing, [h t-1 , x t ] represents the hidden state data of time slice t-1 corresponding to the data content of time slice t, [h t-1 , x t ] Take the hidden state data of time slice t-1; σ represents the sigmoid activation function; The expression of the gate output data for time slice t is as follows: z t =σ(W z ×[h t-1 ,x t ]+b z ) Where W z represents the weight matrix of time slice t in the update gate processing, b z Represents the weight matrix for time slice t in the update gate processing.

5. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 1, characterized in that: In step S3, during the model hyperparameter optimization process, the gated recurrent unit model also selects some historical wind shear data as samples and uses the MSE loss function for loss constraint. The MSE loss function expression is as follows: where y i is the actual value, y 预测 , i is the predicted value, and N is the number of samples.

6. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 1, characterized in that: In step S3, the spiral strategy of the improved whale algorithm uses the adjustment of the control parameter α to gradually shrink the search space and optimize the iteration. The coefficient A is used to determine the position update method of the individual whale. A = 2a × ra, a represents the control parameter, which is a linearly decreasing variable in the iteration, and its value gradually decreases from 2 to 0. r represents a random number between [0, 1]. The specific method is as follows: If |A|<1, execute to make the current individual approach the optimal solution output; If |A|>1, execute to randomly select another whale individual from the population and update the current solution.

7. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 6, characterized in that: The spiral strategy of the improved whale algorithm uses the current optimal solution as prey, and other solutions gradually approach the target. The position update expression is as follows: D=|2*r*X * -X| X 更新 =X * -A*D X * represents the location of the current best solution, X represents the current location, D represents the distance between the current solution and the optimal solution, X 更新 Indicates the updated position.

8. The wind shear risk prediction method combining gated loop and whale algorithm according to claim 6, characterized in that: In step S3, when the improved whale algorithm performs model hyperparameter optimization processing on the gated recurrent unit model, the improved whale algorithm also adopts the following dynamic adjustment strategy: X(j+1)=X(j)+µ*(x * (j)-X(j))-A*D j D j =|2*r*X * (j)-X(j)| D j Represents X * (j) is the distance from X(j), and μ represents the hyperparameter sensitivity factor.

9. A wind shear risk prediction system combining gated loop and whale algorithm, characterized by: The system comprises a data collection module, a data sorting module, a gated cyclic unit model and a real-time data collection input module. The data collection module is used to collect historical wind shear data of several years in chronological order for all take-off and landing transport aircraft in the study area. The historical wind shear data includes wind shear events and corresponding data of the events. The corresponding data of the events include meteorological data, geographical data and transport aviation time in the study area. The data sorting module divides the time slices into m time slices according to the natural year, sorts out the time slice risk value and time slice data according to each time slice, and stores them in associated codes. The time slice data includes the meteorological data and geographic data corresponding to the time slice; The gated recurrent unit model inputs the time slice risk value and the time slice data stored in the associated code, and the gated recurrent unit model performs prediction learning iterative training with the time slice risk value as the prediction value and the time slice data as the data content; The gated recurrent unit model uses the spiral strategy of the improved whale algorithm to optimize the model hyperparameters. The spiral strategy expression of the improved whale algorithm is as follows: Where X(j+1) represents the parameter output at the j+1th iteration, X * (j) represents the optimal solution parameter output at the jth iteration, X(j) represents the parameter output at the jth iteration, l represents a random number between -1 and 1, k represents a constant, and T represents the maximum number of iterations; The real-time data collection input module is used to collect wind shear data of the current time slice and input it into the trained gated cyclic unit model. The wind shear data of the current time slice includes wind shear events of the current time slice and data corresponding to events of the current time slice. The data corresponding to events of the current time slice include meteorological data of the current time slice, geographical data of the current time slice, and transportation aviation time in the study area of ​​the current time slice. The trained gated recurrent unit model processes the wind shear data of the current time slice and outputs the risk prediction value corresponding to the current time slice.

10. An electronic device, characterized in that: It includes at least one processor, at least one memory and a data bus; wherein: the processor and the memory communicate with each other through 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 wind shear risk prediction method combining the gated loop and the whale algorithm as described in any one of claims 1 to 8.

Images