Reliability Evaluation Method for Predicting UAV Flight Parameters Based on Entropy Value

Abstract

The present invention discloses a reliability evaluation method for predicting the flight parameters of an unmanned aerial vehicle based on entropy value. Multisource and multimodal flight data of the same aircraft in the same time period are collected; the collected data is cleaned and normalized by an autoencoder network; the cleaned data is input into a long short-term memory network (LSTM) flight parameter prediction network to obtain a prediction result matrix; the obtained prediction matrix is calculated with the original data after dimensionality reduction using entropy value for reliability evaluation; finally, the reliability of the prediction result is obtained through the entropy value calculation result. The present invention can give a certain reliability evaluation to the prediction result of the flight parameters, judge the rationality of the prediction, and then reasonably adjust the flight parameters, avoid unreasonable data, which is of great significance for the healthy operation of the aircraft.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a method for evaluating the reliability of unmanned aerial vehicle flight parameter prediction based on entropy value. Background Technique

[0002] Flight parameters are key data for ensuring flight safety, guiding flight operations, optimizing flight performance, and achieving precise navigation and positioning. Accurately predicting flight parameters can ensure flight safety, optimize flight paths, reduce energy consumption, and improve flight efficiency. By precisely predicting key parameters such as flight speed, altitude, and heading, pilots and ground control centers can make timely and effective decisions to handle various emergencies.

[0003] Conducting reliability analysis on the predicted flight parameter data is a crucial link for ensuring flight safety, improving the accuracy of the prediction model, and the scientific nature of decision-making. It helps to identify and correct biases and uncertainties in the prediction, enhance data confidence, and provide a solid basis for flight control, path planning, and risk assessment. Through reliability analysis, the prediction model can be optimized, flight efficiency can be improved, and potential risks can be reduced, which is very important for ensuring the smooth implementation of flight missions.

[0004] Therefore, it is of great significance to evaluate the reliability of aircraft flight parameter prediction. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides a method for evaluating the reliability of unmanned aerial vehicle flight parameter prediction based on entropy value, which collects multi-source and multi-modal flight data of the same aircraft in the same time period; uses an autoencoder network to clean and normalize the collected data; inputs the cleaned data into a long short-term memory network (LSTM) flight parameter prediction network to obtain a prediction result matrix; uses entropy value to calculate the obtained prediction matrix and the original data after dimensionality reduction for reliability evaluation; finally, obtains the reliability of the prediction result through the entropy value calculation result. The present invention can give a certain reliability evaluation to the prediction result of flight parameters, judge the rationality of the prediction, and then reasonably adjust the flight parameters, avoid unreasonable data, which is of great significance for the healthy operation of the aircraft.

[0006] The technical solutions adopted by the present invention to solve its technical problems are as follows:

[0007] Step 1: Collect multi-source and multi-modal flight data of the same aircraft in the same time period;

[0008] Step 2: For the collected multi-source and multi-modal flight data, use an autoencoder network to perform data cleaning and normalization processing to achieve normalization and dimensionality reduction of the multi-source data, and obtain the aircraft parameter time matrix as: , where , Indicates a column of time series after dimensionality reduction within the time period , and there are a total of columns of such time series;

[0009] Step 3: Input the aircraft parameter time matrix into the LSTM flight parameter prediction network to obtain the predicted flight parameters for the future time period , and the predicted flight parameter matrix is , where represents the flight parameters corresponding to the obtained future time period , and there are a total of M columns of such predicted sequences;

[0010] Step 4: Calculate the entropy values of the aircraft parameter time matrix separately according to the time series to obtain initial entropy values; Concatenate the predicted flight parameter matrix obtained in Step 3 with the aircraft parameter time matrix to form a new matrix , where , and calculate the entropy value of the matrix ;

[0011] Step 5: Obtain the reliability through the entropy values of the matrices and .

[0012] Furthermore, the normalization process in Step 2 is specifically as follows: For the multi-source multi-modal flight data in the same time period , obtain the maximum value of this time period, divide the collected original flight parameter data by its mean to normalize all the original flight parameter data, and divide the data range into -1 to 1 to obtain the normalized data , Indicates a column of time series of the same data source within the time period , and there are a total of columns of such time series.

[0013] Furthermore, the input of the autoencoder network in Step 2 is , and the process of compressing and reconstructing the input learns effective data representations to obtain the aircraft parameter time matrix as:

[0014] .

[0015] Furthermore, the objective function of the autoencoder network in Step 2 is: , where Denote the dist function, which calculates the distance between each vector in X and each vector in Y. The returned result Minimizeloss is a matrix of size N×M, where N is the number of vectors in X and M is the number of vectors in Y.

[0016] Furthermore, the LSTM flight parameter prediction network is used to obtain the predicted flight parameters for a future time period The predicted flight parameter matrix is The input is The specific structure of the LSTM flight parameter prediction network is as follows:

[0017] The long short-term memory network LSTM includes three gates: the input gate, the forget gate, and the output gate, which are used to control the flow and update of information; there are four groups of weights in the LSTM neuron, namely the weights for the input gate, the output gate, the forget gate, and generating candidate memories; in this network, these four groups of weights are updated by the backpropagation algorithm to minimize the prediction error:

[0018] Step 3-1: The input gate determines the degree to which the current input information is written into the memory cell;

[0019] Step 3-2: The forget gate determines whether to forget the historical information stored in the memory cell, thus allowing the long short-term memory network to selectively clear irrelevant information;

[0020] Step 3-3: The output gate determines the output of the information in the memory cell at the current time step.

[0021] Furthermore, the loss function of the LSTM flight parameter prediction network is the mean square error.

[0022] Furthermore, the LSTM flight parameter prediction network uses the stochastic gradient descent method to update the weight matrix, optimize the model parameters, and reduce the value of the loss function.

[0023] Furthermore, the LSTM flight parameter prediction network calculates the entropy value using permutation entropy, obtaining a column of length of the entropy value of matrix and a new matrix a column of length of the entropy value .

[0024] Furthermore, the magnitude of the permutation entropy characterizes the randomness of the time series. The smaller the value of the permutation entropy, the more regular the time series. Conversely, the more random the time series; the rationality of the time series prediction is calculated by statistically analyzing the change in the degree of disorder of the time series before and after prediction, and then the reliability of the flight parameter prediction is deduced. ​

[0025] Further, step 5 is specifically as follows:

[0026] Input the flight parameter time matrix X, and the obtained entropy value sequence of this matrix is: ; For the new time matrix Z formed by splicing before and after prediction obtained in step 4, the entropy value sequence obtained after calculating its entropy value is ; The overall reliability obtained based on the above entropy value sequence is:

[0027] ;

[0028] Step 5-1: By statistically calculating the entropy values of the time series before and after prediction, judge the change in the degree of disorder, calculate whether the time series is reasonably predicted, and then infer the prediction reliability of the flight parameters; use permutation entropy to calculate the entropy value, and the calculation method of permutation entropy is as follows:

[0029] Step 5-1-1: Reconstruct the time series into sequences , where , is the time delay parameter, is the embedding dimension;

[0030] Step 5-1-2: Arrange all data points inside each in ascending order, and set a symbol vector to represent the order of magnitude, with the smallest element corresponding to , and the largest element corresponding to . At this time, if two identical values are generated, then arrange them according to their subscript order; contains dimensions, so there are a total of permutation patterns;

[0031] Step 5-1-3: Use the symbol vector to represent , and at this time, each element in is represented by their order of magnitude;

[0032] Step 5-1-4: Calculate the probability of each order appearing ;

[0033] Step 5-1-5: According to the form of Shannon entropy, define the permutation entropy of different permutation methods as follows:

[0034]

[0035] Step 5-1-6: Calculate the permutation entropy for the data before prediction to obtain a column of permutation entropy values:

[0036] ;

[0037] Step 5-2: For the predicted flight parameter matrix calculate the permutation entropy to obtain a column of permutation entropy values .

[0038] The beneficial effects of the present invention are as follows:

[0039] 1. The present invention can give a certain reliability evaluation to the prediction result of flight parameters, judge the rationality of the prediction, and then reasonably adjust the flight parameters, avoid unreasonable data, which is of great significance to the healthy operation of the aircraft.

[0040] 2. The present invention adopts the data dimensionality reduction technology based on the autoencoder, faces complex multi-source and multi-modal flight parameters, fully extracts its useful information, processes complex flight parameter information with relatively simple data, and explores deeper data connections. The LSTM prediction network is adopted to face the time series information, avoid the gradient disappearance and gradient explosion, and better predict the time series signal.

[0041] 3. The present invention calculates the reliability of the prediction by measuring the change in signal complexity before and after the prediction and measuring whether there is a drastic change in signal complexity. The entropy value method is simple and rapid, and can calculate the change in reliability in real time, which helps to realize the real-time update of aircraft parameters. Brief Description of the Drawings

[0042] Figure 1 is the flowchart of the method of the present invention;

[0043] Figure 2 is the schematic diagram of the autoencoder of the present invention;

[0044] Figure 3 is the schematic diagram of the long short-term memory neural network of the present invention;

[0045] Figure 4 is the schematic diagram of the LSTM reliability prediction network of the present invention. Detailed Embodiments

[0046] The present invention will be further described below with reference to the drawings and embodiments.

[0047] Flight parameters, as important indicators to measure the flight state of an aircraft, cover multiple aspects such as axial, lateral and normal overloads, pitch, roll, yaw rate, angle of attack, sideslip angle, flight path angle, track inclination angle, true course angle, pitch angle, roll angle, as well as true airspeed, ground speed, etc. These parameters provide important references for the decisions of pilots and ground control personnel. When the navigation system of the aircraft encounters anomalies or failures, the flight system will transmit incorrect flight data to the flight control system, which poses a serious threat to flight safety.

[0048] Flight parameters are crucial data for ensuring flight safety, guiding flight operations, optimizing flight performance, and achieving precise navigation and positioning. Accurately predicting flight parameters can guarantee flight safety, optimize flight paths, reduce energy consumption, and improve flight efficiency. By precisely estimating key parameters such as flight speed, altitude, and heading, pilots and ground control centers can make timely and effective decisions to handle various emergencies.

[0049] Conducting reliability analysis on the predicted flight parameter data is a key step in ensuring flight safety, enhancing the accuracy of the prediction model, and the scientific nature of decision-making. It helps identify and correct biases and uncertainties in the prediction, enhance data confidence, and provide a solid basis for flight control, path planning, and risk assessment. Through reliability analysis, the prediction model can be optimized, flight efficiency can be improved, and potential risks can be reduced, which is of great significance for ensuring the smooth implementation of flight missions.

[0050] Based on the above analysis, the present invention provides a method for measuring the reliability of flight parameter prediction in a short time, as Figure 1 shown, which specifically includes the following steps:

[0051] Step S1: Collect multi-source and multi-modal flight data of the same aircraft in the same time period;

[0052] Step S2: For the collected multi-source and multi-modal flight data, use an autoencoder network for data cleaning and normalization processing to achieve the normalization and dimensionality reduction of multi-source data, and obtain the aircraft parameter time matrix as: , where , represents a column of time series after dimensionality reduction within the time period , and there are a total of such columns;

[0053] Step S3: Input the aircraft parameter time matrix into the LSTM flight parameter prediction network to obtain the predicted flight parameters for the future time period , and its predicted flight parameter matrix is , where represents the flight parameters corresponding to for the future time period , and a total of M columns of such predicted sequences are obtained;

[0054] Step S4: Calculate the entropy values of the aircraft parameter time matrix respectively according to the time series to obtain initial entropy values; splice the predicted flight parameter matrix obtained in the previous step with the aircraft parameter time matrix according to the source to form a new matrix , where , calculate the entropy value of matrix .

[0055] Step S5: Obtain the reliability through the entropy values of matrices and .

[0056] Furthermore, the normalization process in step 2 is specifically as follows: For multi-source and multi-modal flight data in the same time period , obtain the maximum value of this time period, divide the collected original flight parameter data by the mean value to normalize all the original flight parameter data, and divide the data range into -1 to 1 to obtain the normalized data , represents a column of time series of the same data source within the time period , and there are such columns of time series in total.

[0057] In step S2, an autoencoder is constructed as shown in Figure 2 , with the input being . Through the process of compressing and reconstructing the input , an effective data representation is learned to obtain the aircraft parameter time matrix as:

[0058] .

[0059] S21. Build an autoencoder, including the following steps:

[0060] S211. Build an encoder;

[0061] S212. Build a decoder;

[0062] S213. Set a loss function;

[0063] S214. Train.

[0064] In step S2, an autoencoder network is constructed, which has only one hidden layer. For the sample , the activation value of the middle hidden layer of the autoencoder is the encoding of , that is, . The output of the autoencoder for reconstructing the data is

[0065] The objective function of the autoencoder network constructed in step S2 is .

[0066] In step S3, an LSTM flight parameter prediction network is constructed, as shown in Figure 3 , which is used to obtain the predicted future time period The flight parameters, and the predicted flight parameter matrix is , and the input is . The specific structure of the LSTM flight parameter prediction network is as follows:

[0067] S31. The long short-term memory neural network LSTM mainly includes three key gates: the input gate, the forget gate, and the output gate. These gates are used to control the flow and update of information. The model updates its weights through the backpropagation algorithm to minimize the prediction error.

[0068] S311. Set the hidden layer size of the LSTM unit. The larger the value, the stronger the expression ability of the model, but the computational complexity and the risk of overfitting also increase. Start testing from 50 or 100.

[0069] S312. Select an appropriate sequence length, which usually needs to be selected according to the temporal characteristics of the data. Too short a sequence may lead to insufficient information, and too long a sequence will increase the computational overhead. According to the actual situation, it is recommended to select more than 1024 data points.

[0070] S313. A smaller batch_size can help the model converge more stably, but the training speed will decrease; a larger batch_size will accelerate the training speed, but may cause the model to oscillate. Common values are 16, 32, or 64.

[0071] S314. The learning rate is usually set to 0.001 or less, and can be adjusted according to the convergence speed of the model. You can also try using a learning rate scheduler to dynamically adjust the learning rate.

[0072] S315. You can try stacking multiple layers of LSTM (such as 2 - 3 layers) to increase the model depth to improve performance, but it will also increase the training difficulty and the risk of overfitting.

[0073] Step S3 constructs an LSTM flight parameter prediction network, and the loss function is the mean squared error.

[0074] Step S3 constructs an LSTM flight parameter prediction network, uses the stochastic gradient descent method to update the weight matrix, optimize the model parameters, and reduce the value of the loss function.

[0075] In step S4, permutation entropy is used for entropy value calculation. The permutation entropy has the advantages of simple algorithm, fast calculation speed, strong anti-noise ability, etc., and is widely used in fields such as mechanical fault diagnosis and biological signal processing. The matrix with a column length of entropy value and the new matrix with a column length of entropy value are obtained.

[0076] Among them, the permutation entropy is used to calculate the matrices before and after prediction. The magnitude of the permutation entropy characterizes the randomness of the time series. The smaller the value, the more regular the time series is. On the contrary, the more random the time series is. Whether the sequence is reasonably predicted is calculated by statistically analyzing the change in the degree of disorder of the sequence before and after prediction, and then the reliability of the flight parameter prediction is deduced.

[0077] Input the original flight parameter time matrix X, and the entropy value sequence of this matrix obtained is ; For the new time matrix Z spliced before and after prediction obtained in step 4, the entropy value sequence obtained after calculating its entropy value is . The overall reliability obtained based on the above entropy value sequences is:

[0078] ;

[0079] S41. The calculation method using permutation entropy is as follows:

[0080] S411. Reconstruct the time series into sequences , where , is the time delay parameter, is the embedding dimension;

[0081] S412. Arrange all the data points inside each in ascending order, and set a symbol vector to represent the order of magnitude. The smallest element corresponds to , and the largest element corresponds to . At this time, if two identical values are generated, they are arranged according to their subscript order; contains dimensions, so there are a total of permutation patterns;

[0082] S413. Use the symbol vector to represent , and at this time, each element in is represented by their order of magnitude;

[0083] S414. Calculate the probability of each order appearing;

[0084] S415. According to the form of Shannon entropy, define the permutation entropy of different permutation methods as follows:

[0085]

[0086] S416. Calculate the permutation entropy for the data before prediction to obtain a column of permutation entropy values:

[0087] ;

[0088] S42, for the predicted flight parameter matrix calculate the permutation entropy to obtain a column of permutation entropy values .

Claims

1. A method for evaluating the reliability of UAV flight parameter prediction based on entropy value, characterized in that: The steps include: Step 1: Collect multi-source and multi-modal flight data of the same aircraft in the same time period; Step 2: The collected multi-source and multi-modal flight data is cleaned and normalized using an autoencoder network to achieve normalization and dimensionality reduction of the multi-source data. The obtained aircraft parameter time matrix is: X = {X1, X2, …X i ,…X N },in X i Represents a time series after dimension reduction in time period T1. There are N columns of such time series. The normalization process is specifically as follows: for the multi-source multi-modal flight data in the same time period T1, the maximum value of the time period is obtained. The collected raw flight parameter data are divided by their mean value to normalize all the raw flight parameter data, and the data range is divided into -1 to 1 to obtain the normalized data R = {R1, R2, ... R i ,…R N }, R i Represents a time series of the same data source within time period T1. There are N columns of such time series in total. Step 3: Input the aircraft parameter time matrix into the LSTM flight parameter prediction network to obtain the predicted flight parameters for the future time period T2. The predicted flight parameter matrix is ​​Y = {Y1, Y2, ... Y i ,…Y M }, where Y i Indicates that it corresponds to X i The flight parameters of the future time period T2 are obtained, and the prediction sequence obtained in this way has a total of M columns; Step 4: For the aircraft parameter time matrix X = {X1, X2, ... X i ,…X N } Calculate the entropy values ​​according to the time series to obtain N initial entropy values; The predicted flight parameter matrix Y obtained in step 3 is concatenated with the aircraft parameter time matrix X to form a new matrix Z = {Z1, Z2, ... Z i ,…Z N+M }, where Z = X + Y, calculate the entropy value of the matrix Z; Step 5: Obtain reliability through the entropy values ​​of matrices X and Z; Input the flight parameter time matrix X, and get the entropy value sequence of the matrix: E(X) = {E(X1), E(X2), … E(X i ),…E(X N )}; The new time matrix Z obtained in step 4 is concatenated before and after the prediction. The entropy value sequence obtained after calculating its entropy value is E(Z)={E(Z1),E(Z2),…E(Z i ),…E(Z N )}; The overall reliability obtained based on the above entropy value sequence is: Step 5-1: By statistically analyzing the entropy values ​​of the time series before and after the prediction, we can determine the change in the disorder degree, calculate whether the time series is reasonably predicted, and then deduce the reliability of the flight parameter prediction; the entropy value is calculated using the permutation entropy, where the calculation method of the permutation entropy is as follows: Step 5-1-1: Convert the time series X i ={x i，1 ,x i，2 ,…x i，n } is reconstructed into k = n-(m-1)*t sequences {A 1，1 ,A 1，2 ,…,A 1,i ,...,A 1，k }, where A 1,i ={x 1,i ,x 1,i+1 ,…x 1,i+(m-1)t }, t is the time delay parameter, m is the embedding dimension; Step 5-1-2: Place each A 1,i All internal data points are arranged in ascending order, and a symbol vector π={r0,r1,…,r m-1 }={0,1,...,m-1} represents the order of size, the smallest element corresponds to r0, the largest element corresponds to r m-1 , if two identical values ​​are generated, they are arranged in order of their subscripts; A 1,i Contains m dimensions, so there are m! permutation patterns; Step 5-1-3: Use symbolic vectors to represent A 1,i , at this time each A 1,i The elements in are expressed in order of their magnitude; Step 5-1-4: Calculate the probability of each sequence appearing P1, P2, ... P m！ ; Step 5-1-5: According to the form of Shannon entropy, define the permutation entropy of m! different permutations as follows: Step 5-1-6: Calculate the permutation entropy of the data before prediction and obtain a column of permutation entropy values: E(X)={E(X1),E(X2),…E(X i ),…FORMER N )}; Step 5-2: The predicted flight parameter matrix is ​​Y = {Y1, Y2, ... Y i ,…Y N }Calculate the permutation entropy and get a column of permutation entropy values ​​E(Z) = {E(Z1), E(Z2), …E(Z i ),…E(Z N )}.

2. The method for predicting reliability of unmanned aerial vehicle flight parameters based on entropy value according to claim 1 is characterized in that: The autoencoder network in step 2 has an input of R = {R1, R2, ... R i ,…R N }, the process of compressing and reconstructing the input R learns an effective data representation to obtain the aircraft parameter time matrix as: X={X1,X2,...X i ,…X N }。 3. The method for predicting reliability of UAV flight parameters based on entropy value according to claim 2 is characterized in that: The objective function of the autoencoder network in step 2 is: Minimizeloss=dist(X,Y), where dist(.) represents the dist function, which calculates the distance between each vector in X and each vector in Y. The returned result Minimizeloss is a matrix of size N×M, where N is the number of vectors in X and M is the number of vectors in Y.

4. The method for predicting reliability of unmanned aerial vehicle flight parameters based on entropy value according to claim 3 is characterized in that: The LSTM flight parameter prediction network is used to obtain the predicted flight parameters for the future time period T2. The predicted flight parameter matrix is ​​Y = {Y1, Y2, ... Y i ,…Y M }, the input is X={X1,X2,…X i ,…X N }, the specific structure of the LSTM flight parameter prediction network is as follows: The long short-term memory network LSTM consists of three gates: input gate, forget gate and output gate, which are used to control the flow and update of information; there are four sets of weights in the LSTM neuron, namely input gate, output gate, forget gate, and generation of candidate memory; in this network, these four sets of weights are updated through the back propagation algorithm to minimize the prediction error: Step 3-1: Input gate determines the extent to which the current input information is written into the memory cell; Step 3-2: Forget gate, which determines whether to forget the historical information stored in the memory unit, thereby allowing the long short-term memory network to selectively clear irrelevant information; Step 3-3: Output gate determines the output of the information in the memory unit at the current time step.

5. The method for predicting reliability of UAV flight parameters based on entropy value according to claim 4 is characterized in that: The loss function of the LSTM flight parameter prediction network is the mean square error.

6. The method for predicting reliability of unmanned aerial vehicle flight parameters based on entropy value according to claim 5 is characterized in that: The LSTM flight parameter prediction network uses the stochastic gradient descent method to update the weight matrix, optimize the model parameters, and reduce the value of the loss function.

7. The method for predicting reliability of UAV flight parameters based on entropy value according to claim 6 is characterized in that: The LSTM flight parameter prediction network uses permutation entropy to calculate entropy values, and obtains an entropy value E1 of a column length of N in the matrix X and an entropy value E2 of a column length of M+N in the new matrix Z.

8. The method for predicting reliability of UAV flight parameters based on entropy value according to claim 7 is characterized in that: The size of the permutation entropy represents the randomness of the time series. The smaller the value of the permutation entropy, the more regular the time series is. Conversely, the more random the time series is. By statistically predicting the change in the disorder of the time series before and after the prediction, it is calculated whether the time series is reasonably predicted, and then the reliability of the flight parameter prediction is inferred.

Images