Deep Learning-Based Route Planning Evaluation Method, Device, Equipment, Medium and Product

Abstract

The present invention discloses a method, device, equipment, medium and product for evaluating route planning based on deep learning. The method includes: obtaining flight plan data and actual flight data of a to-be-tested route planning result to form an input feature set and an error label set; wherein, the input feature set and the error label set have corresponding time series relationships; training the input feature set and the error label set according to a long short-term memory network to establish an error prediction model and a residual distribution model; generating a dynamic tolerance interval according to the error prediction model, the residual distribution model and reference standard data; and evaluating the accuracy of the to-be-tested route planning result according to the dynamic tolerance interval. The present invention deeply mines the time series relationship between flight plan data and actual flight data by using a long short-term memory network, and constructs a dynamic tolerance interval for a planning target, so as to more accurately evaluate the accuracy of route planning, thereby improving the safety and efficiency of aviation operations.

Description

Technical Field

[0001] The present invention relates to the field of aviation technology, and in particular, to a route planning evaluation method, device, equipment, medium and product based on deep learning. Background Art

[0002] In the field of aviation, route planning is an important part of flight operations, and its accuracy directly affects flight time, fuel consumption and operating costs. At present, although numerous algorithms have emerged in the field of route planning, air traffic has characteristics that are significantly different from ground traffic, that is, it is impossible to compare and verify the algorithm results through actual flight. Instead, the accuracy is often verified by simply comparing the planned results with reference standard data. However, this evaluation method has significant limitations. The evaluation process ignores the rationality and acceptability of errors under different routes or flight conditions. For example, the traditional method sets a single error standard for all routes uniformly, without fully considering the dynamic impact of factors such as route distance, flight time, and weather on the error tolerance. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a route planning evaluation method, device, equipment, medium and product based on deep learning, which uses a long short-term memory network to deeply mine the temporal relationship between flight plan data and actual flight data, and constructs a dynamic tolerance interval for the planning target, so as to more accurately evaluate the accuracy of route planning, thereby improving the safety and efficiency of aviation operations.

[0004] To achieve the above object, an embodiment of the present invention provides a route planning evaluation method based on deep learning, including:

[0005] Obtain the flight plan data and actual flight data of the route planning result to be tested, and form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding temporal relationship;

[0006] Train the input feature set and the error label set according to the long short-term memory network to establish an error prediction model and a residual distribution model;

[0007] Generate a dynamic tolerance interval according to the error prediction model, the residual distribution model and the reference standard data;

[0008] Evaluate the accuracy of the route planning result to be tested according to the dynamic tolerance interval.

[0009] As an improvement of the above solution, the obtaining the flight plan data and actual flight data of the route planning result to be tested, and forming an input feature set and an error label set includes:

[0010] Acquire flight plan data and actual flight data of the route planning result to be tested; wherein the flight plan data and the actual flight data are both time series data;

[0011] Performing feature extraction and feature standardization processing on the flight plan data and the actual flight data to obtain an input feature set; wherein the features include flight plan features, actual flight features and environmental features; the flight plan features include planned distance, planned time and planned fuel consumption; the actual flight features include actual distance, actual time and actual fuel consumption; the environmental features include wind speed, wind direction, temperature and air pressure;

[0012] Error annotation is performed on the planned target according to the flight plan data and the actual flight data to obtain an error label set; wherein the planned target includes flight distance, flight time and flight fuel consumption.

[0013] As an improvement of the above solution, the input feature set and the error label set are trained according to the long short-term memory network to establish an error prediction model and a residual distribution model, including:

[0014] Dividing the input feature set and the error label set into a training set, a validation set and a test set for training, validation and performance evaluation, respectively;

[0015] Constructing a long short-term memory network architecture, establishing an error prediction model, and training using the input features and error labels in the training set;

[0016] Taking minimizing the mean square error as the objective function, calculating the deviation between the predicted error and the actual error, and calculating the gradient by back propagation to update the parameters of the error prediction model;

[0017] Testing the performance of the error prediction model through the validation set, and evaluating the fit of the error prediction model on the validation data;

[0018] Evaluating the trained error prediction model through the test set to verify the generalization performance of the error prediction model;

[0019] Residuals are calculated based on the test set, distribution characteristics of the residuals are analyzed, and a residual distribution model is established based on distribution fitting.

[0020] As an improvement of the above solution, the long short-term memory network architecture includes a convolutional layer, a long short-term memory network layer, a fully connected layer and a residual analysis layer.

[0021] As an improvement of the above solution, generating a dynamic tolerance interval according to the error prediction model, the residual distribution model and the reference standard data includes:

[0022] Input the reference standard data into the error prediction model to obtain the predicted error output by the error prediction model;

[0023] According to the residual distribution model, determine the tolerance coefficient based on the quantile regression method;

[0024] Construct a dynamic tolerance interval according to the predicted error and the tolerance coefficient;

[0025] Wherein, the dynamic tolerance interval is ; In the formula, represents the reference standard data of the planning objective; and both represent the tolerance coefficient; represents the predicted error.

[0026] As an improvement of the above solution, the accuracy evaluation of the to-be-tested route planning result according to the dynamic tolerance interval includes:

[0027] Calculate the error between the to-be-tested route planning result and the reference standard data;

[0028] Obtain the accuracy evaluation result of the to-be-tested route planning result according to whether the error falls into the dynamic tolerance interval.

[0029] An embodiment of the present invention further provides a route planning evaluation device based on deep learning, including:

[0030] A data acquisition module, configured to acquire the flight plan data and the actual flight data of the to-be-tested route planning result to form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding time sequence relationship;

[0031] A model construction module, configured to train the input feature set and the error label set according to a long short-term memory network to establish an error prediction model and a residual distribution model;

[0032] An interval generation module, configured to generate a dynamic tolerance interval according to the error prediction model, the residual distribution model, and the reference standard data;

[0033] An accuracy evaluation module, configured to perform accuracy evaluation on the to-be-tested route planning result according to the dynamic tolerance interval.

[0034] An embodiment of the present invention further provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the above-mentioned route planning evaluation method based on deep learning is implemented.

[0035] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned deep learning-based route planning evaluation method according to any one of the above.

[0036] An embodiment of the present invention further provides a computer program product, which includes a computer program or computer instructions. When the computer program or the computer instructions are executed by a processor, they implement the above-mentioned deep learning-based route planning evaluation method according to any one of the above.

[0037] Compared with the prior art, the beneficial effects of a deep learning-based route planning evaluation method, device, equipment, medium and product provided by an embodiment of the present invention are as follows: by obtaining the flight plan data and actual flight data of the to-be-tested route planning result, an input feature set and an error label set are formed; wherein, the input feature set and the error label set have a corresponding time sequence relationship; according to the long short-term memory network, the input feature set and the error label set are trained to establish an error prediction model and a residual distribution model; according to the error prediction model, the residual distribution model and the reference standard data, a dynamic tolerance interval is generated; according to the dynamic tolerance interval, the accuracy of the to-be-tested route planning result is evaluated. By deeply mining the time sequence relationship between the flight plan data and the actual flight data through the long short-term memory network, an error prediction model and a residual distribution model are established, and a corresponding dynamic tolerance interval is constructed for each planning target to evaluate the accuracy of the route planning result, which can comprehensively consider the dynamic influence of different routes, environments and other characteristics, more accurately evaluate the accuracy of the route planning result, and thus improve the safety and efficiency of aviation operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic flowchart of a preferred embodiment of a deep learning-based route planning evaluation method provided by the present invention;

[0039] Figure 2 is a schematic structural diagram of a preferred embodiment of a deep learning-based route planning evaluation device provided by the present invention;

[0040] Figure 3 is a schematic structural diagram of a preferred embodiment of a terminal device provided by the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a preferred embodiment of a route planning evaluation method based on deep learning provided by the present invention. The route planning evaluation method based on deep learning includes:

[0043] S1. Obtain the flight plan data and actual flight data of the route planning result to be tested, and form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding time sequence relationship;

[0044] S2. Train the input feature set and the error label set according to the long short-term memory network to establish an error prediction model and a residual distribution model;

[0045] S3. Generate a dynamic tolerance interval according to the error prediction model, the residual distribution model and the reference standard data;

[0046] S4. Evaluate the accuracy of the route planning result to be tested according to the dynamic tolerance interval.

[0047] Specifically, the embodiment of the present invention provides a route planning evaluation method based on deep learning. First, flight plan data and actual flight data of the to-be-tested route planning result are obtained to form an input feature set and an error label set. Among them, both the flight plan data and the actual flight data are time series data, and the input feature set and the error label set have a corresponding time series relationship. For example, the time, flight speed, flight altitude, wind speed and direction of the planned route points, etc., these features and the actual errors (such as time deviation, speed deviation) are point-by-point matched based on the time axis. The error label is calculated point by point based on the planning target and the actual data (such as the time deviation between the planned route point and the actual route point), so each input feature (such as wind speed, route point time) has its corresponding error label. Then, the input feature set and the error label set are trained according to the long short-term memory network to establish an error prediction model and a residual distribution model. According to the error prediction model, the residual distribution model and the reference standard data of the planning target, a dynamic tolerance interval for each planning target is generated. Among them, the reference standard data can be derived from the actual flight data and environmental data corresponding to the to-be-tested route planning result; or the flight plan and environmental data corresponding to the to-be-tested route planning result; or other standard data recognized by the source or the to-be-tested party (such as an airline). Finally, according to the dynamic tolerance interval, the accuracy of the to-be-tested route planning result is evaluated.

[0048] The embodiment of the present invention deeply mines the time series relationship between the flight plan data and the actual flight data through the long short-term memory network, establishes an error prediction model and a residual distribution model, and constructs a corresponding dynamic tolerance interval for each planning target to evaluate the accuracy of the route planning result, which can comprehensively consider the dynamic influence of different routes, environments and other characteristics, more accurately evaluate the accuracy of the route planning result, and thus improve the safety and efficiency of air operations.

[0049] In another preferred embodiment, the step S1 of obtaining the flight plan data and the actual flight data of the to-be-tested route planning result to form an input feature set and an error label set includes:

[0050] S101, obtaining the flight plan data and the actual flight data of the to-be-tested route planning result; among them, both the flight plan data and the actual flight data are time series data;

[0051] S102, performing feature extraction and feature standardization processing on the flight plan data and the actual flight data to obtain an input feature set; among them, the features include flight plan features, actual flight features and environmental features; the flight plan features include planned distance, planned time and planned fuel consumption; the actual flight features include actual distance, actual time and actual fuel consumption; the environmental features include wind speed, wind direction, temperature and air pressure;

[0052] S103, performing error labeling on the planned targets according to the flight plan data and the actual flight data to obtain an error label set; wherein the planned targets include flight distance, flight time, and flight fuel consumption.

[0053] Specifically, the embodiment of the present invention obtains the flight plan data and actual flight data of the route planning result to be tested. Among them, the flight plan data and the actual flight data are both time series data. Feature extraction and feature standardization processing are performed on the flight plan data and the actual flight data to obtain an input feature set. Among them, the features include flight plan features, actual flight features and environmental features. Flight plan features include planned distance, planned time and planned fuel consumption; actual flight features include actual distance, actual time and actual fuel consumption; environmental features include wind speed, wind direction, temperature and air pressure. Feature standardization processing is ; In the formula, represents the original eigenvalue; represents the standardized feature value; and Represent the mean and standard deviation of the features respectively. According to the flight plan data and the actual flight data, the planning goals are annotated with errors to obtain the error label set. Among them, the planning goals include flight distance, flight time and flight fuel consumption. The error label is calculated point by point based on the planning goals and actual data (for example, the time deviation between the planned waypoint and the actual waypoint), so each input feature (such as wind speed, waypoint time) has its corresponding error label. The error label is ; In the formula, represents the error of the planning target; Indicates the actual value of the planning target; Represents the planned value of the planning target.

[0054] In another preferred embodiment, the step S2, training the input feature set and the error label set according to a long short-term memory network to establish an error prediction model and a residual distribution model, includes:

[0055] S201, dividing the input feature set and the error label set into a training set, a validation set and a test set, which are used for training, validation and performance evaluation respectively;

[0056] S202, constructing a long short-term memory network architecture, establishing an error prediction model, and using the input features and error labels in the training set for training;

[0057] S203, taking minimizing the mean square error as the objective function, calculating the deviation between the predicted error and the actual error, and calculating the gradient by back propagation to update the parameters of the error prediction model;

[0058] S204. Test the performance of the error prediction model through the validation set and evaluate the fitness of the error prediction model on the validation data;

[0059] S205. Evaluate the trained error prediction model through the test set and examine the generalization performance of the error prediction model;

[0060] S206. Calculate the residuals based on the test set, analyze the distribution characteristics of the residuals, and establish a residual distribution model based on distribution fitting.

[0061] Specifically, the embodiments of the present invention establish an error prediction model and a residual distribution model based on the long short-term memory network architecture. The long short-term memory network architecture includes a convolutional layer, a long short-term memory network layer, a fully connected layer, and a residual analysis layer. Among them, the convolutional layer, the long short-term memory network layer, and the fully connected layer are used to construct the error prediction model and output the predicted error; the residual analysis layer is used to construct the residual distribution model and output the residual distribution. The embodiments of the present invention divide the input feature set and the error label set into a training set, a validation set, and a test set according to a ratio, which are respectively used for training, validation, and performance evaluation. Construct a long short-term memory network architecture, establish an error prediction model, and use the input features and error labels in the training set for training. Taking the minimization of the mean square error as the objective function, calculate the deviation between the predicted error and the actual error, and calculate the gradient through backpropagation to update the parameters of the error prediction model. Test the performance of the error prediction model through the validation set and evaluate the fitness of the error prediction model on the validation data. Evaluate the trained error prediction model through the test set and examine the generalization performance of the error prediction model. Calculate the residuals based on the test set, analyze the distribution characteristics of the residuals, and establish a residual distribution model based on distribution fitting.

[0062] Exemplarily, the long short-term memory network architecture includes:

[0063] A convolutional layer that extracts local time features of the input feature set through one-dimensional convolution ; where t represents the time step, j represents the filter index, represents the weight of the convolution kernel, represents the bias, represents the activation function, k represents the element index in the convolution kernel, d represents the number of elements included in the convolution kernel.

[0064] A long short-term memory network layer that captures long-term dependencies in the time series, and its forget gate, input gate, memory cell update, and output gate are respectively:

[0065] Forget gate: ;

[0066] Input gate: , ;

[0067] Memory cell update: ;

[0068] Output gate: ; ;

[0069] Wherein, represents the Sigmoid activation function, represents the hidden state at the current time step, represents the memory cell state at the current time step, represents the input at the current time step, and represent the weight and bias matrices.

[0070] Fully connected layer, taking the final hidden state of the long short-term memory network layer as the input, predicting and outputting the prediction error through the fully connected layer; wherein, represents the weight matrix, represents the bias term.

[0071] Residual analysis layer, calculating the residual between the prediction error and the actual error , analyzing the distribution characteristics of the residual, and fitting the residual distribution.

[0072] In yet another preferred embodiment, in S3, according to the error prediction model, the residual distribution model, and the reference standard data, a dynamic tolerance interval is generated, including:

[0073] S301, inputting the reference standard data into the error prediction model to obtain the prediction error output by the error prediction model;

[0074] S302, determining the tolerance coefficient based on the quantile regression method according to the residual distribution model;

[0075] S303, constructing a dynamic tolerance interval according to the prediction error and the tolerance coefficient;

[0076] Wherein, the dynamic tolerance interval is ; wherein, represents the reference standard data of the planning target; and both represent the tolerance coefficients; represents the prediction error.

[0077] Specifically, in the embodiment of the present invention, the reference standard data is input into the error prediction model, and the prediction error output by the error prediction model is obtained. According to the residual distribution model, the tolerance coefficient is determined based on the quantile regression method. and According to the prediction error and the tolerance coefficient , , a dynamic tolerance interval is constructed as ; in the formula, represents the reference standard data of the planning target; and both represent the tolerance coefficient; represents the prediction error.

[0078] It should be noted that and are the tolerance coefficients determined according to the quantile regression results of the residual distribution, and are set by the lower and upper quantiles of the quantile regression. Exemplarily, assume that in a planning task with distance as the target, the flight distance of the reference standard data is D = 1000 nautical miles, and the prediction error output by the error prediction model is = 20 nautical miles. Based on the residual distribution model, through the quantile regression method, the 10% quantile (lower quantile) error and 90% quantile (upper quantile) error determined are -4 nautical miles and 6 nautical miles respectively, then the tolerance coefficients are: = |-4| / 20 = 0.2, = 6 / 20 = 0.3, then the dynamic tolerance interval = [1000 - 0.2 * 20, 1000 + 0.3 * 20] nautical miles = [996, 1006] nautical miles.

[0079] In another preferred embodiment, in S4, according to the dynamic tolerance interval, the accuracy of the planned route to be measured is evaluated, including:

[0080] S401, calculating the error between the planned route to be measured and the reference standard data;

[0081] S402, obtaining the accuracy evaluation result of the planned route to be measured according to whether the error falls within the dynamic tolerance interval.

[0082] Specifically, in the embodiment of the present invention, the error between the planned route to be measured and the reference standard data is calculated as ; in the formula, represents the planned route to be measured. The accuracy of the planned route to be measured is determined, and it is checked whether the error falls within the dynamic tolerance interval. Exemplarily, the accuracy determination is:

[0083] 。

[0084] Calculate the overall accuracy rate of the planned route to be measured as ; In the formula, N represents the total number of samples, represents the i th sample's accuracy.

[0085] Correspondingly, the present invention also provides a route planning evaluation device based on deep learning, which can implement all processes of the route planning evaluation method based on deep learning in the above embodiments.

[0086] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of a preferred embodiment of a route planning evaluation device based on deep learning provided by the present invention. The route planning evaluation device based on deep learning includes:

[0087] A data acquisition module 201, configured to acquire flight plan data and actual flight data of the planned route to be measured, and form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding time sequence relationship;

[0088] A model construction module 202, configured to train the input feature set and the error label set according to a long short-term memory network, and establish an error prediction model and a residual distribution model;

[0089] An interval generation module 203, configured to generate a dynamic tolerance interval according to the error prediction model, the residual distribution model, and reference standard data;

[0090] An accuracy evaluation module 204, configured to evaluate the accuracy of the planned route to be measured according to the dynamic tolerance interval.

[0091] Preferably, the data acquisition module 201 is specifically configured to:

[0092] Acquire flight plan data and actual flight data of the planned route to be measured; wherein, both the flight plan data and the actual flight data are time series data;

[0093] Perform feature extraction and feature standardization processing on the flight plan data and the actual flight data to obtain an input feature set; wherein, the features include flight plan features, actual flight features, and environmental features; the flight plan features include planned distance, planned time, and planned fuel consumption; the actual flight features include actual distance, actual time, and actual fuel consumption; the environmental features include wind speed, wind direction, temperature, and air pressure;

[0094] Perform error annotation on the planning objectives according to the flight plan data and the actual flight data to obtain an error label set; wherein, the planning objectives include flight distance, flight time, and flight fuel consumption.

[0095] Preferably, the model construction module 202 is specifically configured to:

[0096] Divide the input feature set and the error label set into a training set, a validation set, and a test set, which are respectively used for training, validation, and performance evaluation;

[0097] Construct a long short-term memory network architecture, establish an error prediction model, and use the input features and error labels in the training set for training;

[0098] Take minimizing the mean square error as the objective function, calculate the deviation between the predicted error and the actual error, and calculate the gradient through backpropagation to update the parameters of the error prediction model;

[0099] Test the performance of the error prediction model through the validation set, and evaluate the fitting degree of the error prediction model on the validation data;

[0100] Evaluate the trained error prediction model through the test set to test the generalization performance of the error prediction model;

[0101] Calculate the residuals based on the test set, analyze the distribution characteristics of the residuals, and establish a residual distribution model based on distribution fitting.

[0102] Preferably, the long short-term memory network architecture includes a convolutional layer, a long short-term memory network layer, a fully connected layer, and a residual analysis layer.

[0103] Preferably, the interval generation module 203 is specifically configured to:

[0104] Input the reference standard data into the error prediction model to obtain the predicted error output by the error prediction model;

[0105] Determine the tolerance coefficient based on the quantile regression method according to the residual distribution model;

[0106] Construct a dynamic tolerance interval according to the predicted error and the tolerance coefficient;

[0107] Wherein, the dynamic tolerance interval is ; in the formula, represents the reference standard data of the planning objective; and both represent the tolerance coefficient; represents the predicted error.

[0108] Preferably, the accuracy evaluation module 204 is specifically configured to:

[0109] Calculate the error between the to-be-tested route planning result and the reference standard data;

[0110] Obtain the accuracy evaluation result of the to-be-tested route planning result according to whether the error falls within the dynamic tolerance interval.

[0111] In specific implementation, the working principle, control flow, and achieved technical effects of the route planning evaluation device based on deep learning provided in the embodiments of the present invention are the same as those of the route planning evaluation method based on deep learning in the above embodiments, and will not be elaborated herein.

[0112] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a preferred embodiment of a terminal device provided by the present invention. The terminal device includes a processor 301, a memory 302, and a computer program stored in the memory 302 and configured to be executed by the processor 301. When the processor 301 executes the computer program, the route planning evaluation method based on deep learning described in any of the above embodiments is implemented.

[0113] Preferably, the computer program can be divided into one or more modules / units (such as computer program 1, computer program 2,...). The one or more modules / units are stored in the memory 302 and executed by the processor 301 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0114] The processor 301 can be a central processing unit (CPU), or can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 301 can also be any conventional processor. The processor 301 is the control center of the terminal device and connects various parts of the terminal device through various interfaces and lines.

[0115] The memory 302 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory 302 can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory 302 can also be other volatile solid-state storage devices.

[0116] It should be noted that the above terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 3 the structural schematic diagram is only an example of the above terminal device, and does not constitute a limitation on the above terminal device. It may include more or fewer components than shown in the figure, or combine some components, or different components.

[0117] An embodiment of the present invention also provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program. Among them, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the deep learning-based route planning evaluation method described in any one of the above embodiments.

[0118] An embodiment of the present invention also provides a computer program product. The computer program product includes a computer program or computer instructions. When the computer program or the computer instructions are executed by a processor, they implement the deep learning-based route planning evaluation method described in any one of the above embodiments.

[0119] An embodiment of the present invention provides a deep learning-based route planning evaluation method, device, equipment, medium and product. By obtaining the flight plan data and actual flight data of the to-be-tested route planning result, an input feature set and an error label set are formed; among them, the input feature set and the error label set have a corresponding time sequence relationship; according to the long short-term memory network, the input feature set and the error label set are trained to establish an error prediction model and a residual distribution model; according to the error prediction model, the residual distribution model and the reference standard data, a dynamic tolerance interval is generated; according to the dynamic tolerance interval, the accuracy of the to-be-tested route planning result is evaluated. By deeply mining the time sequence relationship between the flight plan data and the actual flight data through the long short-term memory network, an error prediction model and a residual distribution model are established, and a corresponding dynamic tolerance interval is constructed for each planning target to evaluate the accuracy of the route planning result. The dynamic influence of different routes, environments and other characteristics can be comprehensively considered, and the accuracy of the route planning result can be evaluated more precisely, so as to improve the safety and efficiency of aviation operations.

[0120] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the system embodiments provided by the present invention, the connection relationships between the modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0121] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A route planning evaluation method based on deep learning, characterized in that Including: Obtain the flight plan data and actual flight data of the flight route planning result to be measured, and form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding time series relationship; the error labels in the error label set are calculated point by point based on the planning target and the actual flight data; Train the input feature set and the error label set according to the long short-term memory network to establish an error prediction model and a residual distribution model; Generate a dynamic tolerance interval according to the error prediction model, the residual distribution model and the reference standard data; Evaluate the accuracy of the flight route planning result to be measured according to the dynamic tolerance interval; Among them, the training of the input feature set and the error label set according to the long short-term memory network to establish an error prediction model and a residual distribution model includes: Divide the input feature set and the error label set into a training set, a validation set and a test set, which are used for training, validation and performance evaluation respectively; Construct a long short-term memory network architecture, establish an error prediction model, and train it using the input features and error labels in the training set; Take the minimization of the mean square error as the objective function, calculate the deviation between the predicted error and the actual error, and calculate the gradient through backpropagation to update the parameters of the error prediction model; Test the performance of the error prediction model through the validation set, and evaluate the fitting degree of the error prediction model on the validation data; Evaluate the trained error prediction model through the test set to test the generalization performance of the error prediction model; Calculate the residuals based on the test set, analyze the distribution characteristics of the residuals, and establish a residual distribution model based on distribution fitting; Among them, the generation of the dynamic tolerance interval according to the error prediction model, the residual distribution model and the reference standard data includes: Input the reference standard data into the error prediction model to obtain the predicted error output by the error prediction model; Determine the tolerance coefficient based on the quantile regression method according to the residual distribution model; Construct a dynamic tolerance interval according to the predicted error and the tolerance coefficient.

2. The method for evaluating route planning based on deep learning according to claim 1, wherein The obtaining of the flight plan data and actual flight data of the flight route planning result to be measured and forming an input feature set and an error label set includes: Obtain the flight plan data and actual flight data of the flight route planning result to be measured; wherein, both the flight plan data and the actual flight data are time series data; Perform feature extraction and feature standardization processing on the flight plan data and the actual flight data to obtain an input feature set; wherein, the features include flight plan features, actual flight features and environmental features; the flight plan features include planned distance, planned time and planned fuel consumption; the actual flight features include actual distance, actual time and actual fuel consumption; the environmental features include wind speed, wind direction, temperature and air pressure; Perform error annotation on the planning target according to the flight plan data and the actual flight data to obtain an error label set; wherein, the planning target includes flight distance, flight time and flight fuel consumption.

3. The method for evaluating route planning based on deep learning according to claim 2, characterized in that, The long short-term memory network architecture includes a convolutional layer, a long short-term memory network layer, a fully connected layer, and a residual analysis layer.

4. The method for evaluating route planning based on deep learning according to claim 2, characterized in that The dynamic tolerance interval is ; in the formula, represents the reference standard data of the planning target; and both represent tolerance coefficients; represents the prediction error.

5. The method for evaluating route planning based on deep learning according to claim 4, characterized in that, The accuracy evaluation of the to-be-tested route planning result according to the dynamic tolerance interval includes: Calculating the error between the to-be-tested route planning result and the reference standard data; Obtaining the accuracy evaluation result of the to-be-tested route planning result according to whether the error falls into the dynamic tolerance interval.

6. An evaluation device for route planning based on deep learning, characterized in that, It includes: A data acquisition module, configured to acquire the flight plan data and the actual flight data of the to-be-tested route planning result to form an input feature set and an error label set; wherein, the input feature set and the error label set have a corresponding time series relationship; the error labels in the error label set are calculated point by point based on the planning target and the actual flight data; A model construction module, configured to train the input feature set and the error label set according to the long short-term memory network to establish an error prediction model and a residual distribution model; An interval generation module, configured to generate a dynamic tolerance interval according to the error prediction model, the residual distribution model, and the reference standard data; An accuracy evaluation module, configured to perform accuracy evaluation on the to-be-tested route planning result according to the dynamic tolerance interval; Among them, the model construction module is specifically configured to: Divide the input feature set and the error label set into a training set, a validation set, and a test set, which are respectively used for training, validation, and performance evaluation; Construct a long short-term memory network architecture, establish an error prediction model, and train it using the input features and error labels in the training set; Taking the minimization of the mean square error as the objective function, calculating the deviation between the predicted error and the actual error, and calculating the gradient through backpropagation to update the parameters of the error prediction model; Testing the performance of the error prediction model through the validation set and evaluating the fitting degree of the error prediction model on the validation data; Evaluating the trained error prediction model through the test set to test the generalization performance of the error prediction model; Calculating the residual based on the test set, analyzing the distribution characteristics of the residual, and establishing a residual distribution model based on distribution fitting; Among them, the interval generation module is specifically configured to: Input the reference standard data into the error prediction model to obtain the predicted error output by the error prediction model; Determining the tolerance coefficient based on the quantile regression method according to the residual distribution model; Constructing a dynamic tolerance interval according to the predicted error and the tolerance coefficient.

7. A terminal device, characterized in that, It includes a processor and a memory, and a computer program is stored in the memory, and the computer program is configured to be executed by the processor. When the processor executes the computer program, it implements the deep learning-based route planning evaluation method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program. When the device where the computer-readable storage medium is located executes the computer program, it implements the deep learning-based route planning evaluation method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, and when the computer program or the computer instructions are executed by a processor, the method for evaluating route planning based on deep learning according to any one of claims 1 to 5 is implemented.

Images