Non-tower air route trajectory prediction method, system and device under environmental disturbance and medium

Abstract

The invention discloses a non-control tower air route trajectory prediction method, system and device under environmental disturbance and a medium, belongs to air route trajectory prediction in the technical field of aviation, and solves the technical problem of low accuracy of air route trajectory prediction of an aircraft in a non-control tower terminal airspace in the prior art. The method comprises the following steps: designing a conditional variation auto-encoder to perform modeling on uncertainty of an external environment, and predicting an expected value of an environment factor of a future time node through Monte Carlo simulation; the method comprises the following steps: designing an air route prediction model based on a Transform architecture, taking wind direction and wind speed expectation, wind direction and wind speed time sequence data and air route time sequence position data after future P seconds as input, processing the input data through a self-attention mechanism, capturing spatio-temporal characteristics, and finally outputting position, speed and course information of an aircraft at a future time node. By combining the airway data, the environment data and the future environment prediction information, richer environment information is provided for the airway prediction model, so that the accuracy and robustness of airway prediction are improved.

Description

Technical Field

[0001] The present invention belongs to the field of aviation technology, relates to route trajectory prediction, and in particular to a non-tower route trajectory prediction method, system, equipment and medium under environmental disturbance. Background Art

[0002] Short-term trajectory prediction refers to the prediction of the flight trajectory of an aircraft (mainly an airplane) in the short term (usually seconds or minutes) in the future, including its position, speed, heading and other information. It is of great significance to improve aviation safety, optimize flight scheduling, reduce navigation costs, etc., and is one of the key technologies in air transportation. Especially in non-towered terminal airspace, short-term trajectory prediction plays a vital role.

[0003] Non-tower terminal airspace refers to airspace that is not directly controlled by a ground air traffic control tower. It usually occurs when an aircraft is about to enter the vicinity of an airport or in the early stages after taking off from an airport. In this airspace, the flight route of an aircraft is usually more flexible, affected by meteorological factors, air traffic flow, pilot operation and other factors, and ground control is more difficult. The operation of the aircraft often relies more on prediction and autonomous judgment. Therefore, accurate short-term route trajectory prediction can effectively help air traffic controllers, pilots and flight management systems make decisions in advance, which is extremely important for avoiding mid-air collisions, ensuring flight safety and improving the efficiency of airspace resource utilization.

[0004] However, one of the difficulties in short-term trajectory prediction is the high uncertainty of the flight environment. The flight path of an aircraft is not only constrained by the aircraft's own dynamics, but also affected by external factors such as the environment. In particular, changes in environmental factors such as wind speed and airflow have a great impact on the flight trajectory, and this change is highly random and time-varying.

[0005] At present, short-term route trajectory prediction methods mainly include physical modeling, statistical methods and artificial intelligence-based prediction methods.

[0006] The physical modeling method calculates the route trajectory through flight mechanics formulas, taking into account factors such as the aircraft's dynamic model and meteorological conditions, and is suitable for route prediction under certain conditions. However, this type of method is mostly based on simplified assumptions and is often difficult to cope with complex and dynamic flight environments, especially in non-tower terminal airspace. The rapid changes in weather and the complexity of airspace make the physical modeling method less adaptable.

[0007] Statistical methods (such as regression analysis, time series analysis, etc.) predict future flight trajectories by mining the patterns in historical data. Although these methods can capture some typical trajectory patterns through a large amount of data, they mainly rely on the regularity of historical data and lack sufficient modeling capabilities for random changes (such as wind speed, air current changes, etc.) and uncertainties in the external environment. Especially in the case of sudden meteorological conditions or large changes in airspace, the accuracy of the prediction results is relatively low.

[0008] In recent years, with the development of artificial intelligence technology, deep learning methods have gradually become a research hotspot for short-term air route prediction. Models based on neural networks (such as Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), etc.) can learn temporal features from historical data, capture the potential patterns of aircraft trajectories, and effectively make predictions.

[0009] The invention patent application with the application number 202411316802.3 discloses a multi-UAV trajectory dynamic planning and cooperative collision avoidance method based on reinforcement learning, which includes: 1. Initial environment construction: constructing a three-dimensional grid map of the urban low-altitude environment and planning an initial four-dimensional trajectory for each UAV flight; 2. Conflict perception: during the operation of the UAV, calculating the conflict information between the UAV and surrounding obstacles and other UAVs in real time; 3. Conflict resolution: using a time series neural network model and a reinforcement learning algorithm to train an intelligent agent to learn the optimal avoidance strategy and dynamically adjust the trajectory; 4. Trajectory adjustment: after successful conflict resolution, backtracking the UAV action sequence and recording the position of the UAV at each time step to obtain a new four-dimensional trajectory of the UAV. Although this invention patent can achieve the trajectory adjustment of the UAV, its main purpose is collision avoidance, that is, mainly considering the conflict information between the UAV and surrounding obstacles and other UAVs; it cannot solve the problem of short-term air route prediction (in non-tower terminal airspace).

[0010] The invention patent application with application number 202410575372.0 also discloses a method for trajectory prediction in autonomous operation mode based on time-frequency analysis. According to the principle of autonomous operation mode, it generates multiple sets of flight trajectory data in autonomous operation mode in combination with flight plan, and performs data preprocessing on the trajectory data; then, based on the wavelet transform theory method, a flight trajectory prediction model in autonomous operation mode is established, wavelet coefficients are generated by establishing a deep learning neural network architecture, and the dynamic characteristics of real-time changes of aircraft flight intentions in autonomous operation mode are captured in combination with time-frequency analysis, and multi-scale autonomous operation flight trajectory prediction is performed; pattern recognition of different time and frequency scales is performed from the perspective of time-frequency analysis, thereby improving the flight trajectory prediction performance. Although this method can realize trajectory prediction, it mainly performs trajectory prediction based on flight plan (block removal time, payload, fuel, weather data, waypoint data, altitude / speed profile, flight number, aircraft model, call sign, route code, navigation notice data, waypoint name, altitude, speed, time, distance, heading, etc.).

[0011] As mentioned in the invention patent application, although these methods in the prior art can realize the prediction of the aircraft's route trajectory to a certain extent, they still have some shortcomings, especially in the modeling of environmental uncertainty. There is a lack of effective modeling capabilities for uncertain environmental factors such as meteorological changes, wind direction and wind speed changes. Especially in non-tower terminal airspace, aircraft are affected by multiple uncertain factors. Traditional deep learning methods are difficult to take into account the changes in uncertain environmental factors, which ultimately leads to low accuracy in trajectory prediction of aircraft routes (especially short-term routes) in non-tower terminal airspace, bringing uncertainty to flight safety. Summary of the invention

[0012] The purpose of the present invention is to solve the technical problem of low accuracy in trajectory prediction of an aircraft in a non-tower terminal airspace (especially a short-term route) in the prior art, and to provide a method, system, device and medium for predicting the trajectory of a non-tower route under environmental disturbance.

[0013] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions: A non-tower route trajectory prediction method under environmental disturbance includes the following steps: Step 1, obtain sample data; Obtain historical wind direction and wind speed sample data, historical meteorological condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T; Step 2, constructing a route trajectory prediction model; Constructing a route trajectory prediction model, which includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule; Step 3, training the uncertainty environmental factor prediction module; The historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in step 1 are used as the input of the CVAE encoder; the latent space distribution output by the CVAE encoder is reparameterized and sampled to obtain the latent space vector; the latent space vector and the historical meteorological state condition sample data are used as the input of the CVAE decoder of the CVAE decoding submodule; the historical wind direction and wind speed sample data are input as label data into the CVAE decoder, and the CVAE encoder and CVAE decoder are trained; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained; Step 4: Real-time prediction of environmental factors; The current meteorological state data is obtained, and the trained latent space distribution is sampled to obtain the latent space vector; the current meteorological state data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed, and the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; Step 5, training the route prediction module; Obtain the wind direction and wind speed time series sample data in a time window of size T, the route time series position sample data in a time window of size T, and obtain the meteorological state condition data corresponding to the time window; input the meteorological state condition data corresponding to the time window and the trained latent space vector into the CVAE decoder to obtain N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window; input the N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the wind direction and wind speed expectation under the meteorological state condition of the time window; use the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series sample data in a time window of size T, and the route time series position sample data in a time window of size T as inputs of the route prediction module, and input the trajectory information after P seconds as label data into the route prediction module; train the route prediction module, and after the training is completed, obtain the trained route prediction module; Step 6: Real-time prediction of route trajectory; The wind direction and wind speed time series data in the current time window of size T and the route time series position data in the current time window of size T are obtained to obtain the current meteorological status data; the current meteorological status data and the latent space vector are input into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

[0014] Furthermore, the historical wind direction and wind speed sample data include historical longitude direction wind speed, historical latitude direction wind speed and historical altitude direction wind speed; The sample data of historical meteorological conditions include air temperature, dew point temperature, relative humidity, visibility, gust wind speed, gust wind speed direction, peak wind speed, and peak wind speed direction; The historical route sample data is the flight trajectory of the aircraft in the past period of time, including the time series position data of the aircraft's longitude, latitude and altitude information.

[0015] Furthermore, the CVAE encoder includes an input layer, a first hidden layer, a second hidden layer and an output layer. The input layer includes 11 neurons, the first hidden layer includes 128 neurons, the second hidden layer includes 64 neurons, and the output layer includes 20 neurons. 10 neurons in the output layer are used to output 10 latent space distribution means, and the remaining 10 neurons are used to output 10 latent space distribution logarithmic variances; each neuron in the previous layer is connected to each neuron in the next layer; The CVAE decoder includes an input layer, a first hidden layer, a second hidden layer and an output layer. The input layer includes 18 neurons, the first hidden layer includes 64 neurons, the second hidden layer includes 128 neurons, and the output layer includes 3 neurons; each neuron in the previous layer is connected to each neuron in the next layer.

[0016] Furthermore, the route prediction module includes an embedding layer, multiple stacked Transformer encoders and a fully connected output layer; the embedding layer maps the input of the route prediction module to be consistent with the model dimension inside the Transformer encoder, and the output of the embedding layer is input into the first Transformer encoder after adding the position code; the feature representation of the Transformer encoder after self-attention and feedforward network processing is used as the output of the current Transformer encoder; the last time step feature in the output of the last Transformer encoder is used as the input of the fully connected output layer, and the fully connected output layer outputs the trajectory information after P seconds in the future, and the trajectory information after P seconds in the future includes the position prediction and heading speed prediction of the aircraft after P seconds, the position prediction includes longitude, latitude and altitude, and the heading speed prediction includes longitude speed, latitude speed and altitude speed.

[0017] Furthermore, the Transformer encoder includes a self-attention mechanism submodule, a first layer normalization, a feedforward neural network and a second layer normalization; the output of the embedding layer is added with the position encoding and then input into the self-attention mechanism submodule, the feature obtained after the output of the embedding layer is added with the position encoding is residually connected with the output of the self-attention mechanism submodule and then input into the first layer normalization, the output of the first layer normalization is input into the feedforward neural network, the output of the first layer normalization is residually connected with the output of the feedforward neural network and then input into the second layer normalization, and the output of the second layer normalization is input into the next Transformer encoder or the fully connected output layer.

[0018] A non-tower route trajectory prediction system under environmental disturbance, comprising: A sample data acquisition module is used to acquire historical wind direction and wind speed sample data, historical meteorological state condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T; A route trajectory prediction model construction module is used to construct a route trajectory prediction model. The route trajectory prediction model includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule. The uncertain environmental factor prediction module training module is used to use the historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in the sample data acquisition module as the input of the CVAE encoder; re-parameterize the sampling of the latent space distribution output by the CVAE encoder to obtain the latent space vector; use the latent space vector and the historical meteorological state condition sample data as the input of the CVAE decoder of the CVAE decoding submodule; input the historical wind direction and wind speed sample data as label data into the CVAE decoder to train the CVAE encoder and CVAE decoder; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained; The real-time prediction module of environmental factors is used to obtain the current meteorological status data, sample the trained latent space distribution, and obtain the latent space vector; the current meteorological status data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed, and the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; The route prediction module training module is used to obtain the wind direction and wind speed time series sample data in a time window of size T, the route time series position sample data in a time window of size T, and obtain the meteorological state condition data corresponding to the time window; the meteorological state condition data corresponding to the time window and the trained latent space vector are input into the CVAE decoder to obtain N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window; the N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the wind direction and wind speed expectation under the meteorological state condition of the time window; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series sample data in a time window of size T, and the route time series position sample data in a time window of size T are used as inputs of the route prediction module, and the trajectory information after P seconds is used as label data to input the route prediction module; the route prediction module is trained, and after the training is completed, the trained route prediction module is obtained; The real-time route trajectory prediction module is used to obtain the wind direction and wind speed time series data in the current time window of size T, the route time series position data in the current time window of size T, and the current meteorological status data; the current meteorological status data and the latent space vector are input into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

[0019] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0020] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to execute the steps of the above method.

[0021] The beneficial effects of the present invention are as follows: 1. In the present invention, a route trajectory prediction model is constructed and trained based on a conditional variational autoencoder and a Transformer model. The route trajectory prediction model can take into account the changes in uncertain environmental factors and can effectively predict the trajectory of routes (especially short-term routes) in non-tower terminal airspace to cope with the uncertainty of the flight link and improve the accuracy and robustness of route prediction.

[0022] 2. In this invention, by introducing the conditional variational autoencoder (CVAE) to model the probability distribution of the external uncertainty environment, the limitation of the existing route trajectory prediction method based on regression analysis that is difficult to handle the uncertainty of the external environment is innovatively overcome. Based on the modeling of the external uncertainty environment, while capturing the potential pattern of environmental uncertainty, the probability distribution of the uncertain environment is dynamically generated based on the real-time meteorological status, and the environmental uncertainty is effectively quantified.

[0023] 3. In the present invention, a single model is used to model the probability distribution of the external uncertainty environment under different meteorological conditions. Compared with the existing method that requires the use of multiple independent models to deal with meteorological conditions in different situations, the present invention only uses a single model to model the uncertainty of all meteorological conditions, which reduces the complexity of the model and improves the efficiency and flexibility of the overall system.

[0024] 4. In the present invention, by integrating future environmental uncertainty with historical route information, the accuracy of the route prediction model is improved, breaking through the limitation of the existing technology that mainly relies on historical route information, providing richer environmental information for the route prediction model, and improving the reliability and accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of the flow of the uncertainty environmental factor prediction module in the present invention; Figure 3 It is a schematic diagram of the flow of the route prediction module in the present invention; Figure 4 It is a schematic diagram of the structure of the CVAE encoder in the present invention; Figure 5 It is a schematic diagram of the structure of the CVAE decoder in the present invention; Figure 6 is a schematic diagram of the connection of neurons in the present invention; Figure 7 It is a schematic diagram of the structure of the route prediction module in the present invention (wherein the structure of the first Transformer encoder is fully illustrated, and the second Transformer encoder omits some structures for simple illustration). DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0027] Therefore, based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0028] Example 1 This embodiment provides a non-tower route trajectory prediction method under environmental disturbance, which is used to predict the trajectory of routes (especially short-term routes) in non-tower terminal airspace. Environmental disturbance in this application refers to the impact of external environmental factors on the aircraft route trajectory during flight, and these factors usually show randomness and time-varying properties. Specifically, environmental disturbances include but are not limited to meteorological factors (such as wind speed, wind direction, airflow, etc.), air traffic flow, and changes in other dynamic factors on the ground or in the air. Especially in non-tower terminal airspace, the flight path of the aircraft is greatly affected by environmental disturbances, and due to the lack of ground control, the operation and decision-making of the aircraft are more dependent on the prediction of environmental disturbances. Therefore, the existence of environmental disturbances increases the complexity and uncertainty of short-term route trajectory prediction, which needs to be processed by advanced prediction methods to improve flight safety.

[0029] When predicting the non-tower route trajectory, this embodiment includes the following steps: Step 1, obtain sample data; Obtain historical wind direction and wind speed sample data, historical meteorological status condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T.

[0030] The historical wind direction and wind speed sample data is based on the observation data at the historical time point, which includes three wind direction components, namely, longitude wind speed (u), latitude wind speed (v) and altitude wind speed (w). The data dimension is M*3, where M represents the number of samples and 3 represents the three wind speed components (longitude, latitude and altitude wind speed). That is, the historical wind direction and wind speed sample data includes historical longitude wind speed, historical latitude wind speed and historical altitude wind speed.

[0031] The sample data of historical meteorological conditions is also based on the observation data at historical time points. It can be regarded as a combination of various meteorological factors, including air temperature (tmpf), dew point temperature (dwpf), relative humidity (relh), visibility (vsby), gust wind speed (gust), gust wind speed direction (gust_drct), peak wind speed (peak_wind_gust), and peak wind speed direction (peak_wind_drct). It contains the meteorological state information corresponding to each historical external environmental data (taking wind direction and wind speed as an example). It can provide a comprehensive description of the weather, and assumes that these conditions remain stable in a short period of time, that is, different meteorological states, their wind direction and wind speed should have corresponding probability distribution patterns, guiding the CVAE encoder and CVAE decoder to find the probability distribution of wind direction and wind speed in different meteorological states. The data dimension is M*8, where M represents the number of samples and 8 represents eight meteorological factors. These eight meteorological factors are: Air temperature (tmpf): refers to the temperature in the atmosphere, usually used to describe how hot or cold the current environment is, in degrees Fahrenheit; Dew point temperature (DWPF): refers to the temperature at which water vapor begins to condense into water droplets under a certain pressure in the air, reflecting the humidity level in the air, and the unit is Fahrenheit; Relative humidity (relh): refers to the ratio of the water vapor content in the air to the maximum amount of water vapor that can be contained at that temperature; Visibility (vsby): refers to the maximum distance that an observer can see clearly, reflecting the degree of scattering or absorption of light by particles in the air (such as water droplets, dust, smoke, etc.), and the unit is miles; Gust: A brief, sudden, rapid change in wind speed, usually a strong, stormy wind, lasting for a relatively short time, usually a few seconds to a few minutes, measured in knots; Gust wind speed direction (gust_drct): refers to the wind direction of the gust, usually expressed in degrees (°); Peak wind speed (peak_wind_gust): refers to the maximum instantaneous wind speed reached within a certain period of time. It is the strongest wind speed recorded by meteorological observations and is usually measured in knots. Peak wind speed direction (peak_wind_drct): refers to the wind direction when the wind speed reaches its maximum value when measuring the peak wind speed, usually expressed in degrees (°).

[0032] Historical route sample data refers to the movement trajectory data of the aircraft in the past period of time, including the longitude, latitude and altitude of the aircraft. That is, historical route sample data includes the time series position data of the longitude, latitude and altitude information of the aircraft.

[0033] Longitude (H_Lon): Indicates the longitude position of the aircraft at the current moment.

[0034] Latitude (H_Lat): indicates the latitude of the aircraft at the current moment.

[0035] Altitude (H_Alti): indicates the current flight altitude of the aircraft.

[0036] This data is usually provided in the form of a time series, such as one data point per second, forming a three-dimensional array.

[0037] Data format: The dimensions of historical route data are N×T×3, where: N represents the batch size during training; T represents the time window size, which indicates the number of time series data points input to the model; 3 represents the three location features of each data point: longitude, latitude, and altitude.

[0038] The historical wind direction and wind speed sample data and the historical meteorological condition sample data are paired and concatenated to obtain an 11-dimensional input vector, which is used as the input of the uncertainty environmental factor prediction module.

[0039] Step 2, constructing a route trajectory prediction model; A route trajectory prediction model is constructed. The route trajectory prediction model includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule.

[0040] The function of the CVAE encoder is to compress the input data into a latent space and learn the potential probability distribution of wind direction and speed under the condition of meteorological conditions. Specifically, after nonlinear transformations of multiple network layers, the encoder outputs the mean and logarithmic variance of a latent space. These outputs are used to describe the randomness of wind direction and speed under different meteorological conditions and their corresponding probability distribution patterns.

[0041] The input of the CVAE encoder is the input vector of length 11 concatenated with historical wind direction and wind speed sample data and historical meteorological state condition sample data. The output of the CVAE encoder is the latent space distribution. The latent space distribution includes the mean vector ( , dimension is 10), logarithmic standard deviation vector ( , with dimension 10).

[0042] The specific structure of the CVAE encoder is as follows: it includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 11 neurons, which are used to receive historical wind direction and speed data (3D) and historical meteorological state condition data (8D); the first hidden layer includes 128 neurons, and the second hidden layer includes 64 neurons, and both hidden layers are fully connected layers; the output layer includes 20 neurons, of which 10 neurons are used to output the means of 10 latent space distributions, and the remaining 10 neurons are used to output the logarithmic variances of 10 latent space distributions. Each neuron in the previous layer is connected to each neuron in the next layer.

[0043] The output of the CVAE encoder is the latent space distribution, including 10 latent space distribution means and 10 latent space distribution logarithmic variances.

[0044] The latent space sampling submodule based on reparameterization uses the reparameterization technique to sample the latent space distribution obtained after the CVAE encoder is trained. This includes: first sampling the noise vector obtained from the standard normal distribution ( , dimension is 10), and then sample from the distribution to obtain the latent space vector Z. The sampling process is performed by the following formula: ; in, represents the mean value of the CVAE encoder output, represents the logarithmic variance of the CVAE encoder output (i.e., the standard deviation in the logarithmic variance), represents a noise vector sampled from a standard normal distribution.

[0045] Through the reparameterization technology (existing technology), the sampling process of the latent space sampling submodule is made continuous and differentiable ("differentiable" here means that the derivative of the function at a certain point exists and is continuous everywhere, and back propagation can be performed only when the derivative exists), ensuring that the model can be optimized through back propagation during training.

[0046] The CVAE decoding submodule includes a CVAE decoder and performs sampling processing on the output of the CVAE decoder.

[0047] The input of the CVAE decoder is an input vector of length 18, which is a concatenation of the latent space vector Z (dimension 10) output by the latent space sampling submodule based on the re-parameterization and the current meteorological status data (8 dimensions).

[0048] The specific structure of the CVAE decoder is as follows: it includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 18 neurons, which are used to receive the latent space vector Z (10 dimensions) and the current meteorological status data (8 dimensions); the first hidden layer includes 64 neurons, and the second hidden layer includes 128 neurons, and both hidden layers are fully connected layers; the output layer includes 3 neurons, which are used to output the predicted value that conforms to the probability distribution pattern of future wind direction and speed, with a dimension of 3 (corresponding to the wind speed in longitude, latitude, and altitude). Each neuron in the previous layer is connected to each neuron in the next layer.

[0049] The output of the CVAE decoder is to fit the probability distribution pattern of future wind direction and wind speed. After sampling, the fitted probability distribution pattern of future wind direction and wind speed is obtained to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed. The N samples that conform to the probability distribution pattern of future wind direction and wind speed are used as the output of the entire CVAE decoding submodule.

[0050] The Monte Carlo simulation submodule performs Monte Carlo prediction on N samples that conform to the probability distribution pattern of future wind direction and wind speed to obtain the expected wind direction and wind speed in the future P seconds.

[0051] Each sampling obtains a wind direction and wind speed forecast, and multiple forecast results are averaged to obtain the expected value of the wind direction and wind speed close to the future: ; in, Indicates Sub-sampled predictions, Indicates the number of sampling times.

[0052] The final output is a 3D vector, which represents the expected wind direction and speed after P seconds:

[0053] in, represents the expected wind speed in the longitude direction, represents the expected wind speed in the latitudinal direction, Indicates the expected wind speed in the height direction.

[0054] The expected wind direction and wind speed after P seconds in the future will be used as one of the inputs of the route prediction module to further predict the route trajectory after P seconds in the future.

[0055] The route prediction module uses the expected wind direction and speed in the future P seconds, the time series data of wind direction and speed in a time window of size T, and the time series position data of the route in a time window of size T as input to perform route prediction. Specifically, the route prediction module uses the encoder structure of Transformer to process the time series data, specifically: The input matrix of the route prediction module consists of the expected wind direction and speed in the future P seconds, the wind direction and speed time series data in the time window of size T, and the route time series position data in the time window of size T. Among them, the wind direction and speed time series data in the time window of size T refers to: the external environment wind direction and speed time series data of the aircraft in the time window of size T close to the current time, including T longitude direction wind speed, latitude direction wind speed, and altitude direction wind speed; the route time series position data in the time window of size T refers to: the flight trajectory of the aircraft in the time window of size T close to the current time, including the aircraft's T longitude, latitude and altitude information time series position data. The input data dimension each time is (N, T, 9). N represents the batch size during training; T represents the time window size, which represents the number of time series data points input to the model; 9 represents the number of input features, including 3 time series route features, 6 wind direction and speed features (including 3 direction wind speeds of the time series wind direction and speed and 3 direction wind speed predictions of the future wind speed expectations output by the uncertainty environmental factor prediction module). When inputting, vectors are constructed by time. The input contains a total of T vectors, one vector for each time. The dimension of each vector is 9, including all the 9 features mentioned above.

[0056] The route prediction module includes an embedding layer, multiple stacked Transformer encoders, and a fully connected output layer. The number of stacked Transformer encoder layers can be set as needed ( Figure 7The embedding layer (i.e., input layer) maps the input of the route prediction module to the same model dimension as the Transformer encoder, i.e., (N, T, 9), through an embedding layer (linear layer). After linear transformation, it is mapped to (N, T, d_model), where d_model=64 is the dimension size of the Transformer internal representation. The output of the embedding layer is added with the position encoding and input to the first Transformer encoder. The dimension of the position encoding is consistent with the input d_model, i.e., (N, d_model); the generated position encoding is added to the input data element by element to obtain the final input representation (for input to the first Transformer encoder). The feature representation of the Transformer encoder after self-attention and feedforward network processing is used as the output of the current Transformer encoder; the last time step feature in the output of the last Transformer encoder is used as the input of the fully connected output layer, and the fully connected output layer outputs the trajectory information after P seconds in the future. The trajectory information after P seconds in the future includes the position prediction and heading speed prediction of the aircraft after P seconds. The position prediction includes longitude, latitude, and altitude. The heading speed prediction includes longitude speed, latitude speed, and altitude speed.

[0057] The Transformer encoder includes a self-attention mechanism submodule, a first-layer normalization, a feedforward neural network, and a second-layer normalization. The output of the embedding layer is input into the self-attention mechanism submodule after adding the position encoding. The self-attention mechanism submodule allows the model to focus on the parts of all time steps in the input sequence, helping to capture long-term dependencies; the relationship between each input vector and all other vectors is calculated, thereby generating a weighted context vector for each time step. The output of the embedding layer is input into the first-layer normalization after adding the position encoding feature and the output of the self-attention mechanism submodule. The layer normalization is used for standardization to make the model training more stable; the gradient is prevented from disappearing by establishing a residual connection between the self-attention and feedforward networks; multi-head attention is used to calculate the weighted sum of multiple attentions, so that the model can learn information in different subspaces. The output of the first-layer normalization is input into the feedforward neural network, which includes two fully connected layers with a ReLU activation function, which can perform independent nonlinear transformations on the output of each time step. The normalized output of the first layer is residually connected to the output of the feedforward neural network and then input into the second layer for normalization. The normalized output of the second layer is input into the next Transformer encoder or fully connected output layer.

[0058] The output of the previous Transformer encoder is used as the input of the next Transformer encoder, and the last time step feature in the output of the last Transformer encoder is used as the input of the fully connected output layer.

[0059] The fully connected output layer maps the dimension d_model to the prediction space (the longitude, latitude and altitude of the aircraft after P seconds, the speed in the longitude direction, the speed in the latitude direction, and the speed in the altitude direction), and the output dimension is 6.

[0060] The output of the fully connected output layer is the output of the route prediction module, which is the predicted trajectory information of the aircraft in the next P seconds, including: Flight position prediction (longitude Lon, latitude Lat, altitude Alti): The aircraft position prediction for the next P seconds, including longitude, latitude and altitude, with a dimension of (N,3).

[0061] Flight heading prediction (longitude speed u, latitude speed v, altitude speed w): The heading prediction for the next P seconds, including the speeds in the three directions of longitude, latitude, and altitude, with a dimension of (N,3).

[0062] Step 3, training the uncertainty environmental factor prediction module; The historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in step 1 are used as the input of the CVAE encoder; the latent space distribution output by the CVAE encoder is reparameterized and sampled to obtain a latent space vector; the latent space vector and the historical meteorological state condition sample data are used as the input of the CVAE decoder of the CVAE decoding submodule; the historical wind direction and wind speed sample data are input as label data into the CVAE decoder, and the CVAE encoder and CVAE decoder are trained; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained.

[0063] Step 4: Real-time prediction of environmental factors; The current meteorological status data is obtained, and the trained latent space distribution is sampled to obtain the latent space vector. The current meteorological status data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed. The N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future.

[0064] Step 5, training the route prediction module; Obtain the time series sample data of wind direction and wind speed in a time window of size T, the time series position sample data of the route in a time window of size T, and obtain the meteorological state condition data corresponding to the time window; input the meteorological state condition data corresponding to the time window and the trained latent space vector into the CVAE decoder to obtain N samples of wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window; input the N samples of wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed under the meteorological state condition of the time window; use the output of the Monte Carlo simulation submodule, the time series sample data of wind direction and wind speed in a time window of size T, and the time series position sample data of the route in a time window of size T as inputs of the route prediction module, and input the trajectory information after P seconds as label data into the route prediction module; train the route prediction module, and after the training is completed, obtain the trained route prediction module.

[0065] Step 6: Real-time prediction of route trajectory; The wind direction and wind speed time series data in the current time window of size T and the route time series position data in the current time window of size T are obtained to obtain the current meteorological status data; the current meteorological status data and the latent space vector are input into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

[0066] Example 2 A non-tower route trajectory prediction system under environmental disturbance, comprising: A sample data acquisition module is used to acquire historical wind direction and wind speed sample data, historical meteorological state condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T; A route trajectory prediction model construction module is used to construct a route trajectory prediction model. The route trajectory prediction model includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule. The uncertain environmental factor prediction module training module is used to use the historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in the sample data acquisition module as the input of the CVAE encoder; re-parameterize the sampling of the latent space distribution output by the CVAE encoder to obtain the latent space vector; use the latent space vector and the historical meteorological state condition sample data as the input of the CVAE decoder of the CVAE decoding submodule; input the historical wind direction and wind speed sample data as label data into the CVAE decoder to train the CVAE encoder and CVAE decoder; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained; The real-time prediction module of environmental factors is used to obtain the current meteorological status data, sample the trained latent space distribution, and obtain the latent space vector; the current meteorological status data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed, and the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; The route prediction module training module is used to obtain the wind direction and wind speed time series sample data in a time window of size T, the route time series position sample data in a time window of size T, and obtain the meteorological state condition data corresponding to the time window; the meteorological state condition data corresponding to the time window and the trained latent space vector are input into the CVAE decoder to obtain N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window; the N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state condition of the time window are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the wind direction and wind speed expectation under the meteorological state condition of the time window; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series sample data in a time window of size T, and the route time series position sample data in a time window of size T are used as inputs of the route prediction module, and the trajectory information after P seconds is used as label data to input the route prediction module; the route prediction module is trained, and after the training is completed, the trained route prediction module is obtained; The real-time route trajectory prediction module is used to obtain the wind direction and wind speed time series data in the current time window of size T, the route time series position data in the current time window of size T, and the current meteorological status data; the current meteorological status data and the latent space vector are input into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; the output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

[0067] Example 3 A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of a method for predicting a non-tower route trajectory under environmental disturbance.

[0068] The computer device may be a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with the user through a keyboard, a mouse, a remote control, a touch pad, or a voice control device.

[0069] The memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or D interface display memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory can be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory can also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory is often used to store the operating system and various application software installed on the computer device, such as the program code of the non-tower route trajectory prediction method under environmental disturbance, etc. In addition, the memory can also be used to temporarily store various types of data that have been output or are to be output.

[0070] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor is generally used to control the overall operation of the computer device. In this embodiment, the processor is used to run the program code stored in the memory or process data, such as running the program code of the non-tower route trajectory prediction method under the environmental disturbance.

[0071] Example 4 A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of a method for predicting a non-tower route trajectory under environmental disturbance.

[0072] Wherein, the computer-readable storage medium stores an interface display program, and the interface display program can be executed by at least one processor so that the at least one processor executes the steps of the non-tower route trajectory prediction method under environmental disturbance as described above.

[0073] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the non-tower route trajectory prediction method under environmental disturbances described in the embodiment of the present application.

Claims

1. A non-tower route trajectory prediction method under environmental disturbance, characterized in that: The following steps are involved: Step 1, obtain sample data; Obtain historical wind direction and wind speed sample data, historical meteorological condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T; Step 2, constructing a route trajectory prediction model; Constructing a route trajectory prediction model, which includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule; Step 3, training the uncertainty environmental factor prediction module; The historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in step 1 are used as the input of the CVAE encoder; the latent space distribution output by the CVAE encoder is reparameterized and sampled to obtain the latent space vector; the latent space vector and the historical meteorological state condition sample data are used as the input of the CVAE decoder of the CVAE decoding submodule; the historical wind direction and wind speed sample data are input as label data into the CVAE decoder, and the CVAE encoder and CVAE decoder are trained; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained; Step 4: Real-time prediction of environmental factors; The current meteorological state data is obtained, and the trained latent space distribution is sampled to obtain the latent space vector; the current meteorological state data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed, and the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; Step 5, training the route prediction module; Obtain the time series sample data of wind direction and wind speed in a time window of size T, the time series position sample data of the route in a time window of size T, and obtain the corresponding meteorological state condition data in the time window; input the corresponding meteorological state condition data in the time window and the trained latent space vector into the CVAE decoder to obtain N samples of the probability distribution mode of wind direction and wind speed under the meteorological state conditions of the time window; input the N samples of the probability distribution mode of wind direction and wind speed under the meteorological state conditions of the time window into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed under the meteorological state conditions of the time window; The output of the Monte Carlo simulation submodule, the wind direction and wind speed time series sample data in a time window of size T, and the route time series position sample data in a time window of size T are used as inputs of the route prediction module, and the trajectory information after P seconds is used as label data to input the route prediction module; The route prediction module is trained, and after the training is completed, a trained route prediction module is obtained; Step 6: Real-time prediction of route trajectory; Obtain the wind direction and wind speed time series data in the current time window of size T, the route time series position data in the current time window of size T, and obtain the current meteorological status data; input the current meteorological status data and the latent space vector into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; N samples that meet the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; The output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

2. The method for predicting non-tower airway trajectory under environmental disturbance according to claim 1, characterized in that: The historical wind direction and wind speed sample data include historical longitude wind speed, historical latitude wind speed and historical altitude wind speed; The sample data of historical meteorological conditions include air temperature, dew point temperature, relative humidity, visibility, gust wind speed, gust wind speed direction, peak wind speed, and peak wind speed direction; The historical route sample data is the flight trajectory of the aircraft in the past period of time, including the time series position data of the aircraft's longitude, latitude and altitude information.

3. The method for predicting non-tower airway trajectory under environmental disturbance according to claim 1, characterized in that: The CVAE encoder includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 11 neurons, the first hidden layer includes 128 neurons, the second hidden layer includes 64 neurons, and the output layer includes 20 neurons. 10 neurons in the output layer are used to output the means of 10 latent space distributions, and the remaining 10 neurons are used to output the logarithmic variances of 10 latent space distributions. Each neuron in the previous layer is connected to each neuron in the next layer. The CVAE decoder includes an input layer, a first hidden layer, a second hidden layer and an output layer. The input layer includes 18 neurons, the first hidden layer includes 64 neurons, the second hidden layer includes 128 neurons, and the output layer includes 3 neurons; each neuron in the previous layer is connected to each neuron in the next layer.

4. The method for predicting non-tower airway trajectory under environmental disturbance according to claim 1, characterized in that: The route prediction module includes an embedding layer, multiple stacked Transformer encoders, and a fully connected output layer. The embedding layer maps the input of the route prediction module to the same model dimension as the Transformer encoder, and the output of the embedding layer is input into the first Transformer encoder after adding the position encoding. The feature representation of the Transformer encoder after self-attention and feedforward network processing is used as the output of the current Transformer encoder; the last time step feature in the output of the last Transformer encoder is used as the input of the fully connected output layer, and the fully connected output layer outputs the trajectory information after P seconds in the future. The trajectory information after P seconds in the future includes the position prediction and heading speed prediction of the aircraft after P seconds. The position prediction includes longitude, latitude, and altitude. The heading speed prediction includes longitude speed, latitude speed, and altitude speed.

5. The method for predicting non-tower airway trajectory under environmental disturbance according to claim 4, characterized in that: The Transformer encoder includes a self-attention mechanism submodule, a first layer normalization, a feedforward neural network and a second layer normalization; the output of the embedding layer is added with the position encoding and then input into the self-attention mechanism submodule, the feature obtained after the output of the embedding layer is added with the position encoding is residually connected with the output of the self-attention mechanism submodule and then input into the first layer normalization, the output of the first layer normalization is input into the feedforward neural network, the output of the first layer normalization is residually connected with the output of the feedforward neural network and then input into the second layer normalization, and the output of the second layer normalization is input into the next Transformer encoder or the fully connected output layer.

6. A non-tower route trajectory prediction system under environmental disturbance, characterized in that: include: A sample data acquisition module is used to acquire historical wind direction and wind speed sample data, historical meteorological state condition sample data, wind direction and wind speed time series sample data within a time window of size T, and route time series position sample data within a time window of size T; A route trajectory prediction model construction module is used to construct a route trajectory prediction model. The route trajectory prediction model includes an uncertainty environmental factor prediction module and a route prediction module. The uncertainty environmental factor prediction module includes a CVAE encoder, a latent space sampling submodule based on heavy parameters, a CVAE decoding submodule, and a Monte Carlo simulation submodule. The uncertain environmental factor prediction module training module is used to use the historical wind direction and wind speed sample data and the historical meteorological state condition sample data obtained in the sample data acquisition module as the input of the CVAE encoder; re-parameterize the sampling of the latent space distribution output by the CVAE encoder to obtain the latent space vector; use the latent space vector and the historical meteorological state condition sample data as the input of the CVAE decoder of the CVAE decoding submodule; input the historical wind direction and wind speed sample data as label data into the CVAE decoder to train the CVAE encoder and CVAE decoder; after the training is completed, the trained CVAE encoder, latent space distribution, and CVAE decoder are obtained; The real-time prediction module of environmental factors is used to obtain the current meteorological status data, sample the trained latent space distribution, and obtain the latent space vector; the current meteorological status data and the latent space vector are input into the CVAE decoder, and the CVAE decoder outputs N samples that conform to the probability distribution pattern of future wind direction and wind speed, and the N samples that conform to the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; The route prediction module training module is used to obtain the wind direction and wind speed time series sample data in a time window of size T, the route time series position sample data in a time window of size T, and the corresponding meteorological state condition data in the time window; the corresponding meteorological state condition data in the time window and the trained latent space vector are input into the CVAE decoder to obtain N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state conditions of the time window; the N samples of the wind direction and wind speed probability distribution mode that conforms to the meteorological state conditions of the time window are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed under the meteorological state conditions of the time window; The output of the Monte Carlo simulation submodule, the wind direction and wind speed time series sample data in a time window of size T, and the route time series position sample data in a time window of size T are used as inputs of the route prediction module, and the trajectory information after P seconds is used as label data to input the route prediction module; The route prediction module is trained, and after the training is completed, a trained route prediction module is obtained; The real-time prediction module of the route trajectory is used to obtain the time series data of wind direction and wind speed in the current time window of size T, the time series position data of the route in the current time window of size T, and the current meteorological status data; the current meteorological status data and the latent space vector are input into the CVAE decoder to obtain N samples that conform to the probability distribution pattern of future wind direction and wind speed; N samples that meet the probability distribution pattern of future wind direction and wind speed are input into the Monte Carlo simulation submodule, and the Monte Carlo simulation submodule outputs the expected wind direction and wind speed after P seconds in the future; The output of the Monte Carlo simulation submodule, the wind direction and wind speed time series data in the current time window of size T, and the route time series position data in the current time window of size T are used as the input of the route prediction module, and the route prediction module outputs the trajectory information after P seconds in the future.

7. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

Images