Track prediction method based on error reverse neural network

Abstract

The invention discloses a track prediction method based on an error reverse neural network, and relates to the field of track prediction, and the method comprises the following steps: obtaining historical track influence data and historical track data, and constructing an influence coefficient list according to the historical track influence data and the historical track data; constructing a track prediction model according to the influence coefficient list, the historical track influence data and the historical track data; acquiring flight influence data, flight path data and corresponding timestamps in the flight process, inputting the flight influence data, the flight path data and the corresponding timestamps into the flight path prediction model, outputting predicted flight path data, performing comparative analysis on the predicted flight path data and the flight path data, and performing correction processing on the flight path prediction model according to a comparison result to obtain the flight path prediction model. Obtaining a corresponding sequence error contribution value according to the timestamp; performing error analysis on the flight influence data to obtain an influence error contribution value; performing error updating on the track prediction model according to the error contribution value, and outputting predicted track data according to the error updating; according to the invention, the accuracy of track prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of flight path prediction, and specifically to a flight path prediction method based on an error backpropagation neural network. Background Art

[0002] With the development of society and the improvement of people's living standards, people's work and lifestyle are constantly changing. With the rapid development of the air transportation industry, air traffic flow has also increased, and the exposed problems of airspace operation efficiency and safety have become increasingly serious. Therefore, to ensure airspace operation safety and improve operation efficiency, predicting and adjusting the operation process of aircraft is an urgent problem for us to solve;

[0003] An invention with publication number CN111292563A, a flight path prediction method for flights, discloses a flight path prediction method for flights. Considering the meteorological environment in which the flight is located, it uses a convolutional neural network to extract meteorological features, and uses a recurrent neural network to model the flight path features, and finally outputs a series of future flight trajectories. Specifically, it includes flight path data preprocessing, meteorological feature extraction, defining a time series model, model compilation and training, and flight path prediction. Through the technical solution of the present invention, it is possible to accurately predict the long-term flight path of a flight in an actual environment.

[0004] An invention with publication number CN111047914A, an FMS flight path prediction method based on four-dimensional flight path operation, belongs to the technical field of flight path operation, and particularly relates to an FMS flight path prediction method completed by using a method for predicting four-dimensional horizontal and vertical flight paths. The specific steps of the method are as follows: S1) Obtain the aircraft planned segment file and planned point file according to the flight plan of the aircraft in the flight segment it is in; S2) According to the planned segment file, obtain the prediction result of the horizontal flight path through the horizontal prediction method; obtain the prediction result of the vertical flight path through the vertical prediction method; S3) According to the prediction results of S2), obtain the RNP value of the flight segment where it is located to complete the precise flight of the flight segment. Due to the above technical solution, the prediction method of the present invention has a high-precision flight path prediction ability, can improve the operation efficiency of the aircraft, and is an effective guarantee for the green and safe flight of the aircraft.

[0005] However, in real life, there are many possible influencing factors in the process of flight path prediction, including meteorological factors, aircraft performance factors, human factors, etc., which affect the accuracy in the process of flight path prediction. And static analysis of various types of influencing factors has an impact on the process of flight path prediction. Therefore, how to dynamically analyze various types of influencing factors during the flight of an aircraft, reduce the influence of other factors during flight, and improve the accuracy of flight path prediction is a problem we need to solve. For this reason, a flight path prediction method based on an error backpropagation neural network is provided. Summary of the Invention

[0006] To solve the above technical problems, the purpose of the present invention is to provide a track prediction method based on an error backpropagation neural network.

[0007] The purpose of the present invention can be achieved by the following technical solutions: A track prediction method based on an error backpropagation neural network, comprising the following steps:

[0008] Step S1: Set up a data management center, pre-store the historical track influence data and historical track data corresponding to each type of aircraft, obtain the influence coefficients corresponding to each type of data in the historical track influence data and the importance data of different types of data under different circumstances, and respectively construct an influence coefficient list and a dynamic attention mechanism for the historical track influence data;

[0009] Step S2: Set up a multi-channel input layer according to the influence coefficient list, historical track influence data and historical track data, extract the feature data from the data information in the multi-channel input layer, and construct a track prediction model based on it;

[0010] Step S3: Obtain the flight influence data, flight track data and corresponding timestamps during the flight of the aircraft, input the obtained flight influence data into the track prediction model, output the predicted track data, and compare and analyze the predicted track data with the flight track data in the corresponding data set to determine whether to analyze and process the track prediction model;

[0011] Step S4: Set up a time series period according to the timestamps corresponding to the flight influence data and flight track data, and obtain the sequence error contribution values corresponding to the difference values between the predicted track data and the flight track data within each time series period;

[0012] Step S5: Preset an error threshold, compare and analyze the sequence error contribution value with the error threshold. When the sequence error contribution value is greater than the error threshold, perform error analysis on the corresponding type of flight influence data within this time series period to obtain the influence error contribution value;

[0013] Step S6: Update the error of the track prediction model according to the influence error contribution value, obtain the track prediction model that has completed error processing, and output the predicted track data through the track prediction model that has completed error processing.

[0014] Further, the process of constructing the influence coefficient list and the dynamic attention mechanism includes:

[0015] Set up a data management center, which is provided with an external interface. Through the external interface, the staff inputs the historical track influence data and historical track data of each type of aircraft;

[0016] Obtain the model information of the aircraft, and obtain the corresponding historical flight track impact data and historical track data according to its model information. The historical flight track impact data includes meteorological data, aircraft performance data, human control data, and the corresponding impact timestamps; the historical track data includes standard track data, position data, heading data, and the corresponding track timestamps; according to the impact of the data of the corresponding type in the historical flight track impact data of the corresponding model aircraft on the historical track data, set the corresponding time difference coefficient, and perform correction processing on the impact timestamp and the track timestamp according to the time difference coefficient; based on the multi-variable algorithm with the same timestamp, obtain the impact coefficient of the impact of the change of different types of data in the historical flight track impact data on other data, store the obtained impact coefficients in each historical flight track impact data, and construct an impact coefficient list;

[0017] Obtain the historical flight track impact data and historical track data stored in the data management center, respectively extract the single-variable data sets of meteorological data, aircraft performance data, and human control data regarding the historical track data, integrate the obtained single-variable data sets, obtain the correlation data between the corresponding single variable and the historical track data, record the obtained correlation data as importance data, and construct a dynamic attention mechanism according to the importance data corresponding to the historical flight track impact data.

[0018] Further, the process of setting the multi-channel input layer according to the impact coefficient list, historical flight track impact data, and historical track data includes:

[0019] Obtain the meteorological data, aircraft performance data, and human control data in the historical flight track impact data, obtain the impact coefficient of the meteorological data on the other two types of data according to the impact coefficient list, and perform correction processing on the aircraft performance data and human control data according to the obtained impact coefficient;

[0020] Set multiple channel input layers according to the type of historical flight track impact data, and record them as the first channel input layer, the second channel input layer, and the third channel input layer respectively; obtain the importance data of the corresponding type of data in each historical flight track impact data according to the dynamic attention mechanism, and set the sorting from high to low according to the importance data of the type of data, and input them into the corresponding channel input layer in turn.

[0021] Further, the process of constructing a flight track prediction model according to the integrated processing of the multi-channel input layer includes:

[0022] Obtain historical track impact data of corresponding types through each channel input layer, extract feature data of corresponding type data information that affects flight trajectory data, set corresponding feature data sets according to the obtained feature data and flight trajectory data, construct a track prediction model based on a convolutional neural network, use the feature data sets in each channel input layer as training sets and test sets, use the dynamic attention mechanism as an additional input feature, train the track prediction model according to the training set, and adjust its training process according to the additional input feature until the loss function is stably trained, save the network parameters corresponding to the model, test the track prediction model through the test set until it meets the preset requirements, and output the track prediction model.

[0023] Further, the process of determining whether to perform analysis and processing on the track prediction model includes:

[0024] Set a data acquisition terminal in the aircraft. The data acquisition terminal is used to collect flight impact data and flight trajectory data during the flight of the aircraft in real time, and set corresponding timestamps during the acquisition process; obtain the corresponding time difference coefficient and list of impact coefficients, and perform correction processing on the obtained flight impact data according to them;

[0025] Input the obtained flight impact data into the track prediction model, output predicted track data, compare and analyze the obtained predicted track data with the collected flight trajectory data, obtain similarity data between the two, preset a similarity threshold, and compare and analyze the similarity data with the similarity threshold;

[0026] When the similarity data is greater than the similarity threshold, no analysis and processing is performed on the track prediction model; when the similarity data is less than or equal to the similarity threshold, analysis and processing is performed on the track prediction model.

[0027] Further, the process of obtaining the sequence error contribution value includes:

[0028] Obtain the flight impact data, flight trajectory data after correction processing, and their corresponding timestamps, set a flight time series according to the order of the corresponding timestamps, preset a time series period, segment the flight time series according to the time series period, obtain the flight curves corresponding to the predicted track data and flight trajectory data within each time series period, based on the loss function of mean square error, calculate the difference between the points where the predicted track data and the flight trajectory data are located at the corresponding timestamps within the two flight curves, and obtain the total sum of the difference values corresponding to all timestamps within each time series period; obtain the timing error value of the two flight curves within the time series period based on the statistical analysis algorithm;

[0029] Preset corresponding error weight coefficients, and perform weighted calculations on the sum of the difference values and the timing error values obtained within each time series period according to the error weight coefficients to obtain the sequence error contribution values within each time series period.

[0030] Further, the process of obtaining the influence error contribution value includes:

[0031] Obtain the sequence error contribution values within each time series period, preset an error threshold, and compare and analyze the sequence error contribution value with the error threshold. When the sequence error contribution value is greater than the error threshold, perform error analysis on the flight influence data of the corresponding type within this time series period;

[0032] Based on the sequence error contribution value, correct the timestamps of the data information in the obtained flight influence data and historical track influence data. Combine the various types of data in the flight influence data obtained within the time series period to form a validation data set. Calculate the validation data set sequentially through a multi-channel input layer to obtain the loss values of the corresponding type of data. Set weights for the channel data layers where different types of data in the flight influence data are located to obtain their weight factors. Obtain the influence error contribution values of the corresponding type of data according to the weight factors and their corresponding loss values.

[0033] Further, the process of updating the error of the track prediction model according to the influence error contribution value includes:

[0034] Calculate the network parameters corresponding to the track prediction model through the backpropagation algorithm according to the influence error contribution value to obtain its gradient data. Update the track prediction model based on the stochastic gradient descent algorithm. Preset an update data set, perform iterative training on the track prediction model according to the update data set, output the track prediction model after the update process is completed, input the obtained flight influence data into the track prediction model after the update process is completed, and output the predicted track data.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] 1. By obtaining different types of flight influence data, and obtaining the importance data of different types of flight influence data in the process of influencing flight trajectory data according to the flight influence data, setting a dynamic attention mechanism according to the importance data, and using the dynamic attention mechanism as an additional input feature to adjust during the analysis and training of the prediction trajectory model, thereby improving the accuracy of the prediction trajectory model to a certain extent;

[0037] 2. By obtaining flight impact data and flight trajectory data to set corresponding timestamps, and correcting the flight image data corresponding to the timestamps according to the errors caused by possible response errors during the flight of the aircraft, the influence of time errors on the track prediction process can be avoided to a certain extent. In addition, by analyzing and processing to obtain the sequence error contribution value, the influence of time factors on the track prediction process is reduced during the process of adjusting the track prediction model. Brief Description of the Drawings

[0038] Figure 1 This is the schematic diagram of a track prediction method based on an error reverse neural network according to an embodiment of the present application. Detailed Embodiment

[0039] As Figure 1 shown, a track prediction method based on an error reverse neural network includes the following steps:

[0040] Step S1: Set up a data management center, pre-store the historical track impact data and historical trajectory data corresponding to each type of aircraft, obtain the impact coefficients corresponding to various types of data in the historical track impact data and the importance data of different types of data under different circumstances, and respectively construct an impact coefficient list and a dynamic attention mechanism for the historical track impact data;

[0041] Step S2: Set up a multi-channel input layer according to the impact coefficient list, historical track impact data and historical trajectory data, extract the feature data from the data information in the multi-channel input layer, and construct a track prediction model according to it;

[0042] Step S3: Obtain the flight impact data, flight trajectory data and corresponding timestamps during the flight of the aircraft, input the obtained flight impact data into the track prediction model, output the predicted track data, and compare and analyze the predicted track data with the flight trajectory data in the data corresponding set to determine whether to analyze and process the track prediction model;

[0043] Step S4: Set a time series period according to the timestamps corresponding to the flight impact data and flight trajectory data, and obtain the sequence error contribution value corresponding to the difference value between the predicted track data and the flight trajectory data within each time series period;

[0044] Step S5: Preset an error threshold, compare and analyze the sequence error contribution value with the error threshold. When the sequence error contribution value is greater than the error threshold, perform error analysis on the corresponding type of flight impact data within this time series period to obtain the impact error contribution value;

[0045] Step S6: Update the error of the track prediction model according to the influence error contribution value, obtain the track prediction model after error processing, and output the predicted track data through the track prediction model after error processing.

[0046] Set up a data management center, pre-store the historical track influence data and historical track data corresponding to each type of aircraft, obtain the influence coefficients corresponding to different types of data in the historical track influence data and the importance data of different types of data under different conditions, and respectively construct an influence coefficient list and a dynamic attention mechanism for the historical track influence data. The specific implementation process includes:

[0047] Set up a data management center, which is provided with an external interface. Through the external interface, the staff inputs the historical track influence data, historical track data consistent with the aircraft model, and the preset flight track during the corresponding flight process, and stores the input historical track influence data and historical track data in the data management center.

[0048] The data management center is interconnected with the data acquisition terminal and data processing terminal of the corresponding aircraft. The data acquisition terminal is used to collect data information during the flight of the aircraft; the data processing terminal is used to analyze and process the historical track influence data, historical track data stored in the data management center, and the data information collected by the data acquisition terminal.

[0049] The data processing terminal obtains the historical track influence data and historical track data; the historical track influence data includes meteorological data, aircraft performance data, human control data, and the corresponding influence timestamps; the historical track data includes position data, heading data, and the corresponding track timestamps.

[0050] Based on the same timestamp, obtain the influence coefficients of the influence of the changes in different types of data in the historical track influence data on other data through a multivariate algorithm, store the influence coefficients obtained in each historical track influence data, and construct an influence coefficient list.

[0051] It should be further noted that in the specific implementation process, the influence coefficient is the influence between different data types in different flight influence data. For example, the influence of meteorological data on aircraft performance data, the influence of meteorological data on human control data, the influence of aircraft performance data on human control data, etc.

[0052] Obtain the historical track impact data and historical trajectory data stored in the data management center, extract the single variable data sets of meteorological data, aircraft performance data and human control data about the historical trajectory data respectively, integrate the obtained single variable data sets, analyze and process the track impact data and historical trajectory data in the single variable data sets based on the Spearman correlation coefficient, obtain the correlation data between the two, record the obtained correlation data as importance data, and build a dynamic attention mechanism according to the importance data corresponding to the historical track impact data;

[0053] It should be further explained that, in the specific implementation process, the single variable data sets include a meteorological single variable data set, a performance single variable data set, and a control single variable data set. The elements in the single variable data set are denoted as x, the historical trajectory data therein are denoted as y, and the correlation coefficients corresponding to the meteorological single variable data set, the performance single variable data set, and the control single variable data set are denoted as ρ. 1 , 2 and ρ 3 ;in

[0054] A multi-channel input layer is set according to the influence coefficient list, historical track influence data and historical track data, and feature extraction is performed on the data information in the multi-channel input layer to obtain feature data, and a track prediction model is constructed based on the feature data. The specific implementation process includes:

[0055] Obtaining meteorological data, aircraft performance data and human control data from historical flight track influence data, obtaining the influence coefficient of meteorological data on the other two types of data according to the influence coefficient list, and correcting the aircraft performance data and human control data according to the obtained influence coefficient;

[0056] According to the types of historical track impact data, multiple channel input layers are set, which are respectively recorded as the first channel input layer, the second channel input layer and the third channel input layer; according to the dynamic attention mechanism, the importance data of the corresponding type of data in each historical track impact data is obtained, and the order of the type data is set from high to low according to the importance data, and input into the corresponding channel input layer in sequence;

[0057] Obtain historical track impact data of corresponding types through each channel input layer, extract feature data of corresponding type data information that affects flight trajectory data, set corresponding feature data sets according to the obtained feature data and flight trajectory data, construct a track prediction model based on a convolutional neural network, use the feature data sets in each channel input layer as training sets and test sets, use the dynamic attention mechanism as an additional input feature, train the track prediction model according to the training set, and adjust its training process according to the additional input feature until the loss function training is stable, save the network parameters corresponding to the model, test the track prediction model through the test set until it meets the preset requirements, and output the track prediction model.

[0058] Obtain flight impact data, flight trajectory data, and corresponding timestamps during the flight of the aircraft, input the obtained flight impact data into the track prediction model, output predicted track data, compare and analyze the obtained predicted track data with the flight trajectory data in the data corresponding set, and determine whether to perform analysis and processing on the track prediction model. Its specific implementation process includes:

[0059] Set a data acquisition terminal in the aircraft. The data acquisition terminal is used to collect flight impact data and flight trajectory data during the flight of the aircraft in real time, and set corresponding timestamps during the collection process; obtain its corresponding time difference coefficient and impact coefficient list, and correct the obtained flight impact data according to them respectively;

[0060] It should be further noted that in the specific implementation process, the flight impact data and flight trajectory data collected by the data acquisition terminal include the following types;

[0061] The flight impact data includes meteorological data, aircraft performance data, and human control data;

[0062] The meteorological data is data information that affects the flight of the aircraft during the track prediction process, including temperature, humidity, wind speed, and wind direction data;

[0063] The aircraft performance data is data information that affects the aircraft during its flight due to its own factors, including aircraft load, mechanical failure, and fuel consumption data;

[0064] The human control data is data information that affects the aircraft during its flight due to factors such as pilot operation and air traffic controllers, including human operation data and air traffic control data;

[0065] The flight trajectory data includes standard trajectory data, position data, speed data, and heading data;

[0066] The position data is the longitude and latitude information at different time points during the flight of the aircraft;

[0067] The heading data is the heading information at different time points during the flight of the aircraft;

[0068] In the flight impact data and the corresponding data information in the flight trajectory data, an impact timestamp and a trajectory timestamp are set during the data acquisition process.

[0069] Input the obtained flight impact data into the flight path prediction model to output the predicted flight path data. Compare and analyze the obtained predicted flight path data with the collected flight trajectory data to obtain the similarity data between the two. Preset a similarity threshold, and compare and analyze the similarity data with the similarity threshold;

[0070] When the similarity data is greater than the similarity threshold, no analysis and processing are performed on the flight path prediction model; when the similarity data is less than or equal to the similarity threshold, analysis and processing are performed on the flight path prediction model.

[0071] Set a time series period according to the timestamps corresponding to the flight impact data and the flight trajectory data, and obtain the corresponding sequence error contribution values of the difference values between the predicted flight path data and the flight trajectory data within each time series period. The specific implementation process includes:

[0072] Obtain the corrected flight impact data, flight trajectory data, and their corresponding timestamps. Set a flight time series according to the order of the corresponding timestamps, preset a time series period, segment the flight time series according to the time series period, and obtain the flight curves corresponding to the predicted flight path data and the flight trajectory data within each time series period. Based on the loss function of the mean square error, calculate the difference between the points where the predicted flight path data and the flight trajectory data are located corresponding to the timestamps within the two flight curves, and obtain the total sum of the difference values corresponding to all timestamps within each time series period; obtain the timing error value of the two flight curves within the time series period based on the statistical analysis algorithm;

[0073] Preset the corresponding error weight coefficient, and perform weighted calculation on the total sum of the difference values and the timing error value obtained within each time series period according to the error weight coefficient to obtain the sequence error contribution value within each time series period.

[0074] Preset an error threshold, and compare and analyze the sequence error contribution value with the error threshold. When the sequence error contribution value is greater than the error threshold, perform error analysis on the flight impact data of the corresponding type within this time series period to obtain the impact error contribution value. The specific implementation process includes:

[0075] Obtain the sequence error contribution values within each time series period, preset an error threshold, and conduct a comparative analysis of the sequence error contribution values and the error threshold. When the sequence error contribution value is greater than the error threshold, perform an error analysis on the flight impact data of the corresponding type within this time series period;

[0076] Based on the sequence error contribution values, correct the timestamps of the data information in the obtained flight impact data and historical track impact data. Compose the data of each type in the flight impact data obtained within the time series period into a validation dataset. Calculate the validation dataset obtained successively for the multi-channel input layer to obtain the loss values of the corresponding type of data. Set the weights for the channel data layers where the different types of data in the flight impact data are located to obtain their weight factors, and obtain the impact error contribution values of the corresponding type of data according to the weight factors and their corresponding loss values.

[0077] Calculate the network parameters corresponding to the track prediction model according to the impact error contribution values through the backpropagation algorithm to obtain their gradient data. Update the track prediction model based on the stochastic gradient descent algorithm. Preset an update dataset, and perform iterative training on the track prediction model according to the update dataset. Output the track prediction model that has completed the update process. Input the obtained flight impact data into the track prediction model that has completed the update process, and output the predicted track data.

[0078] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A track prediction method based on error inverse neural network, characterized in that: The following steps are involved: Step S1: Setting up a data management center, pre-storing historical track impact data and historical trajectory data corresponding to each model of aircraft, obtaining the corresponding impact coefficients between various types of data in the historical track impact data and the importance data of different types of data in different situations, and respectively constructing an impact coefficient list and a dynamic attention mechanism for the historical track impact data; Step S2: setting a multi-channel input layer according to the influence coefficient list, historical track influence data and historical track data, extracting features from the data information in the multi-channel input layer, obtaining feature data, and building a track prediction model based on the feature data; Step S3: obtaining flight impact data, flight trajectory data and corresponding timestamps during the flight of the aircraft, inputting the obtained flight impact data into the trajectory prediction model, outputting the predicted trajectory data, comparing and analyzing the obtained predicted trajectory data with the flight trajectory data in the data corresponding set, and determining whether to perform analysis and processing on the trajectory prediction model; Step S4: setting a time series period according to the timestamps corresponding to the flight impact data and the flight trajectory data, and obtaining a sequence error contribution value corresponding to the difference value between the predicted track data and the flight trajectory data in each time series period; Step S5: Preset an error threshold, compare and analyze the sequence error contribution value with the error threshold, and when the sequence error contribution value is greater than the error threshold, perform error analysis on the corresponding type of flight impact data within the time series period to obtain the impact error contribution value; Step S6: updating the track prediction model according to the contribution value of the influencing error, obtaining the track prediction model after the error processing, and outputting the predicted track data through the track prediction model after the error processing.

2. A track prediction method based on error inverse neural network according to claim 1, characterized in that: The process of building the influence coefficient list and dynamic attention mechanism includes: A data management center is set up, in which an external port is set up, through which the staff can input the historical track impact data and historical trajectory data of various types of aircraft; Obtain the model information of the aircraft, and obtain the corresponding historical track influence data and historical trajectory data according to the model information, wherein the historical track influence data includes meteorological data, aircraft performance data, human control data and corresponding influence timestamps; the historical trajectory data includes standard trajectory data, position data, heading data and corresponding trajectory timestamps; according to the influence of the corresponding type of data in the historical track influence data of the corresponding model aircraft on the historical trajectory data, set the corresponding time difference coefficient, and perform correction processing on the influence timestamp and trajectory timestamp according to the time difference coefficient; according to the same timestamp, obtain the influence coefficient of the influence of different types of data in the historical track influence data on other data during the change process based on the multivariate algorithm, store the influence coefficients obtained in each historical track influence data, and construct an influence coefficient list; The historical track impact data and historical trajectory data stored in the data management center are obtained, and the single variable data sets of meteorological data, aircraft performance data and human control data about the historical trajectory data are respectively extracted. The single variable data sets include meteorological single variable data set, performance single variable data set and control single variable data set. The obtained single variable data sets are integrated to obtain the correlation data between the corresponding single variable and the historical trajectory data. The obtained correlation data is recorded as importance data, and a dynamic attention mechanism is constructed according to the importance data corresponding to the historical track impact data.

3. A track prediction method based on error inverse neural network according to claim 2, characterized in that: The process of setting the multi-channel input layer according to the influence coefficient list, the historical track influence data and the historical track data includes: Obtaining meteorological data, aircraft performance data and human control data from historical flight track influence data, obtaining the influence coefficient of meteorological data on the other two types of data according to the influence coefficient list, and correcting the aircraft performance data and human control data according to the obtained influence coefficient; According to the types of historical track impact data, multiple channel input layers are set, and they are respectively recorded as the first channel input layer, the second channel input layer and the third channel input layer; the importance data of the corresponding type data in each historical track impact data is obtained according to the dynamic attention mechanism, and the type data is sorted from high to low according to the importance data, and is input into the corresponding channel input layer in sequence.

4. A track prediction method based on error inverse neural network according to claim 3, characterized in that: According to the multi-channel input layer integration processing, the process of building a track prediction model includes: Obtain the corresponding type of historical track influence data through each channel input layer, extract the feature data of the corresponding type of data information on the impact of the flight track data, set the corresponding feature data set according to the obtained feature data and the flight track data, build a track prediction model based on the convolutional neural network, use the feature data sets in each channel input layer as training sets and test sets, use the dynamic attention mechanism as an additional input feature, train the track prediction model according to the training set, and adjust its training process according to the additional input features until the loss function training is stable, save the network parameters corresponding to the model, test the track prediction model through the test set until it meets the preset requirements, and output the track prediction model.

5. A track prediction method based on error inverse neural network according to claim 4, characterized in that: The process of determining whether to analyze and process the track prediction model includes: A data acquisition terminal is set in the aircraft, and the data acquisition terminal is used to collect flight impact data and flight trajectory data of the aircraft in real time during flight, and set corresponding timestamps during the collection process; obtain the corresponding time difference coefficients and influence coefficient lists, and perform correction processing on the obtained flight impact data according to the corresponding time difference coefficients and influence coefficient lists; Input the obtained flight impact data into the track prediction model, output the predicted track data, compare and analyze the obtained predicted track data with the collected flight trajectory data, obtain similarity data between the two, preset a similarity threshold, and compare and analyze the similarity data with the similarity threshold; When the similarity data is greater than the similarity threshold, the track prediction model is not analyzed and processed; when the similarity data is less than or equal to the similarity threshold, the track prediction model is analyzed and processed.

6. A track prediction method based on error inverse neural network according to claim 5, characterized in that: The process of obtaining the sequence error contribution value includes: Obtain corrected flight impact data, flight trajectory data and their corresponding timestamps, set the flight time series according to the order of the corresponding timestamps, preset the time series period, segment the flight time series according to the time series period, obtain the flight curves corresponding to the predicted track data and the flight trajectory data in each time series period, calculate the difference between the points where the predicted track data and the points where the flight trajectory data of the corresponding timestamps in the two flight curves are located based on the loss function of the mean square error, and obtain the sum of the difference values ​​of the corresponding differences of all timestamps in each time series period; obtain the time series error value of the two flight curves in the time series period based on the statistical analysis algorithm; The corresponding error weight coefficient is preset, and the sum of the difference values ​​and the timing error value obtained in each time series period are weightedly calculated according to the error weight coefficient to obtain the sequence error contribution value in each time series period.

7. A track prediction method based on error inverse neural network according to claim 6, characterized in that: The process of obtaining the influence error contribution value includes: Obtain the sequence error contribution value in each time series period, preset the error threshold, compare and analyze the sequence error contribution value with the error threshold, and when the sequence error contribution value is greater than the error threshold, perform error analysis on the corresponding type of flight impact data in the time series period; Based on the sequence error contribution value, the timestamps of the data information in the obtained flight impact data and historical track impact data are corrected, and the various types of data in the flight impact data obtained within the time series period are composed of a verification data set. The obtained verification data set is used to calculate the multi-channel input layer in turn to obtain the loss value of the corresponding type of data, and the channel data layers where the different types of data in the flight impact data are located are weighted to obtain their weight factors. According to the weight factors and their corresponding loss values, the impact error contribution values ​​of the corresponding types of data are obtained.

8. A track prediction method based on error inverse neural network according to claim 7, characterized in that: The process of updating the error of the track prediction model according to the influencing error contribution value includes: According to the contribution value of the influencing error, the network parameters corresponding to the track prediction model are calculated through the back propagation algorithm to obtain its gradient data, and the track prediction model is updated based on the stochastic gradient descent algorithm. An updated data set is preset, and the track prediction model is iteratively trained according to the updated data set. The updated track prediction model is output, the obtained flight impact data is input into the updated track prediction model, and the predicted track data is output.

Images