Light electric helicopter verification machine overall parameter prediction method

Abstract

The invention provides an overall parameter prediction method for a light electric helicopter verification machine. The method comprises the following steps: collecting parameter data information of helicopters of different models; constructing an overall parameter database management system of the light electric helicopter verification machine, and inputting the collected parameter data information of the helicopters of different models into the overall parameter database management system of the light electric helicopter verification machine; constructing a neural network model based on an NRBO-XGboost algorithm, training the neural network model by using data in a database management system, and establishing a corresponding relationship among parameters; and predicting the parameters of the electric helicopter by using the neural network model. According to the light electric helicopter verification machine overall parameter prediction method, construction of a light electric helicopter verification machine overall parameter database is completed, helicopter parameters can be predicted based on the neural network model, and a theoretical basis and technical support can be provided for the actual manufacturing process of electric helicopters in the future.

Description

Technical Field

[0001] The invention belongs to the technical field of the overall design of electric helicopters, and particularly relates to a method for predicting the overall parameters of a light electric helicopter verification aircraft. Background Art

[0002] In recent years, the aviation field has been experiencing unprecedented technological innovation, especially the development of the electric helicopter field is particularly rapid. As an important part of future aviation, the conceptual design in the initial stage of the design process of electric helicopters has become increasingly important. In view of this, aiming to achieve refined management and in-depth analysis of the design of electric helicopters, and provide solid data support and decision-making basis for subsequent design optimization, performance improvement and iterative improvement, it is particularly important to construct a detailed, accurate and comprehensive overall parameter database for light electric helicopter verification aircraft. To actively respond to this demand and fill the gap in the existing technology, it is urgent to develop a management system for the overall parameter database of light electric helicopter verification aircraft by using the graphical user interface (GUI) technology based on Python, covering parameters such as weight, rotor, tail rotor and fuselage, with functions of input, query and correction of electric helicopter parameter data, which can ensure the accuracy, integrity and timeliness of the data, and provide data support for the construction of prediction models in subsequent helicopter conceptual design.

[0003] The conceptual design of a helicopter is the preliminary stage of the design of an electric helicopter. The main task is to form the concept and basic concept of the designed electric helicopter according to the design technical requirements, and obtain a preliminary technical solution that meets the design requirements. The conceptual design has a decisive impact on the overall situation and success or failure of helicopter development. In the initial stage of design when many details are unknown, two methods are usually used to determine the overall parameters in helicopter conceptual design. One is the original design method, that is, selecting a mature aircraft type close to the design requirements as the original reference aircraft, referring to the overall parameters and relevant information of the original reference aircraft, and then determining appropriate overall parameters according to the designer's experience and existing conditions. The other method is the statistical analysis method, that is, making statistical analysis of the parameters of all helicopters of the same type, obtaining a semi-empirical formula through regression algorithms, and using this to determine the overall parameters of the helicopter. In the statistical analysis method, the empirical formulas established by regression analysis and the least squares method have relatively single application scenarios. For problems of processing large uncertain factors and fuzzy information, the neural network method has better non-linear mapping solving ability. However, there is currently no research on establishing an overall parameter database for light electric helicopters and then predicting the overall parameters of electric helicopters by using the neural network method. Summary of the Invention

[0004] In view of this, the invention provides a method for predicting the overall parameters of a light electric helicopter verification aircraft to solve the problems existing in the prior art.

[0005] The present invention provides a method for predicting the overall parameters of a light electric helicopter verification machine, including:

[0006] Step 1: Collect parameter data information of different types of helicopters;

[0007] Step 2: Construct an overall parameter database management system for the light electric helicopter verification machine and input the parameter data information of different types of helicopters collected in Step 1 into the overall parameter database management system of the light electric helicopter verification machine;

[0008] Step 3: Construct a neural network model based on the NRBO-XGboost algorithm, and use the parameter data information of different types of helicopters in the overall parameter database management system of the light electric helicopter verification machine to train the neural network model, and establish the corresponding relationship between various parameters;

[0009] Step 4: Use the trained neural network model in Step 3 to predict the parameters of the electric helicopter.

[0010] Preferably, in Step 2, a graphical user interface technology based on Python is used to construct the overall parameter database management system of the light electric helicopter verification machine.

[0011] More preferably, in Step 2, using the graphical user interface technology based on Python to construct the overall parameter database management system of the light electric helicopter verification machine specifically includes:

[0012] Step 2.1: First, install the PyCharm software on the Windows system, and then configure the corresponding Python environment;

[0013] Step 2.2: Construct the overall parameter database management system of the light electric helicopter verification machine;

[0014] Step 2.3: Program, debug and implement the overall parameter database management system of the light electric helicopter verification machine in the PyCharm compiler.

[0015] More preferably, in Step 2.2, the overall parameter database management system of the light electric helicopter verification machine includes a selection type, data operation and data display modules. Among them, the selection type consists of ultra-light helicopters, small helicopters, medium-small helicopters, medium helicopters, large helicopters and giant helicopters; the data operation consists of data addition, data modification and data deletion; the data display consists of parameter range search, model search and search results.

[0016] More preferably, in Step 3, the training set for training the neural network model includes the following data sets of different types of light electric helicopters:

[0017] "Maximum takeoff weight - empty aircraft weight", "Maximum takeoff weight - disc loading", "Maximum takeoff weight - rotor diameter", "Maximum takeoff weight - rotor solidity", "Maximum takeoff weight - tip speed", "Rotor diameter - overall aircraft length", "Rotor diameter - fuselage length", and "Maximum takeoff weight - tail rotor diameter".

[0018] Further preferably, predicting the parameters of the electric helicopter by using the neural network model trained in step 3 specifically includes: predicting the data of the empty aircraft weight, disc loading, rotor diameter, rotor solidity, tip speed, overall aircraft length, fuselage length, and tail rotor diameter of the electric helicopter by using the maximum takeoff weight and rotor diameter of the electric helicopter.

[0019] The overall parameter prediction method for the light electric helicopter verification aircraft provided by the present invention can construct a management system for the overall parameter database of the light electric helicopter verification aircraft to realize the functions of input, query, and correction of helicopter parameter data by collecting parameter data information of different types of helicopters. Then, by using the collected data, the relationship between each parameter is established through a neural network model, thereby realizing the conceptual design of the light electric helicopter verification aircraft. This prediction method is applied for the first time in the field of electric helicopter design, filling the technical gap of estimating the overall parameters of an electric helicopter based on the overall parameter database of a light electric helicopter and using a neural network method, and providing a theoretical basis and technical support for the actual manufacturing process of future electric helicopters. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It is a flowchart of the overall parameter prediction method for the light electric helicopter verification aircraft provided by the embodiment of the present invention;

[0022] Figure 2 It is an interface diagram of the management system for the overall parameter database of the light electric helicopter verification aircraft provided by the embodiment of the present invention;

[0023] Figure 3 It is an interface diagram of the helicopter model information in the management system for the overall parameter database of the light electric helicopter verification aircraft provided by the embodiment of the present invention;

[0024] Figure 4 It is a fitting curve graph of the power function method for "Maximum takeoff weight - empty aircraft weight" provided by the embodiment of the present invention;

[0025] Figure 5 The effect diagram of the training set of the power function method for "maximum takeoff weight - empty aircraft weight" provided by the embodiment of the present invention;

[0026] Figure 6 The flow chart of the Bayes-LSTM algorithm provided by the embodiment of the present invention;

[0027] Figure 7 The structure diagram of the LSTM neural network provided by the embodiment of the present invention;

[0028] Figure 8 The effect diagram of the training set of the Bayes-LSTM method for "maximum takeoff weight - empty aircraft weight" provided by the embodiment of the present invention;

[0029] Figure 9 The structure diagram of the CNN-LSTM neural network provided by the embodiment of the present invention;

[0030] Figure 10 The effect diagram of the training set of the CNN-LSTM method for "maximum takeoff weight - empty aircraft weight" provided by the embodiment of the present invention;

[0031] Figure 11 The flow chart of the NRBO-XGboost algorithm provided by the embodiment of the present invention;

[0032] Figure 12 The effect diagram of the training set of the NRBO-XGboost method for "maximum takeoff weight - empty aircraft weight" provided by the embodiment of the present invention. Specific embodiments

[0033] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Here, it should be noted that in order to avoid obscuring the present invention due to unnecessary details, only the processing steps closely related to the solution of the present invention are shown in the drawings, while other details less related to the present invention are omitted.

[0034] To make up for the current technical gap, the present invention provides a method for predicting the overall parameters of a light electric helicopter verification machine. Using the GUI technology based on Python, a management system for the overall parameter database of the light electric helicopter verification machine is constructed, realizing the functions of input, query and correction of parameter information of different helicopters at home and abroad. In addition, through the analysis of parameters such as weight, rotor, tail rotor and fuselage in the overall parameter database of the light electric helicopter, a set of simplified calculation methods and neural network models for estimating the overall parameters of the electric helicopter are given, and the conceptual design of the light electric helicopter verification machine is completed.

[0035] As Figure 1As shown in the figure, the present invention provides a method for predicting the overall parameters of a light electric helicopter verification aircraft, comprising the following steps:

[0036] Step 1: Collect parameter data information of different types of helicopters at home and abroad;

[0037] Specifically: Consult books on helicopter parameter data in public materials and helicopter manufacturers at home and abroad (domestic: AVIC Helicopter Co., Ltd., AVIC Harbin Aircraft Industry Group Co., Ltd., etc.; foreign: Bell Helicopter Company, McDonnell Douglas Corporation, Sikorsky Aircraft Corporation, Boeing Company, Robinson Helicopter Company, Mil Moscow Helicopter Plant, Leonardo Company, Airbus Company, Kawasaki Heavy Industries, etc.), and a total of 235 types of helicopter parameter data information are collected. Among them, there are 22 types of ultra-light helicopters (<1 ton), 63 types of small helicopters (1 ton - 3 tons), 62 types of medium-small helicopters (3 tons - 6 tons), 43 types of medium helicopters (6 tons - 10 tons), 33 types of large helicopters (10 tons - 20 tons), and 12 types of giant helicopters (20 tons - 40 tons).

[0038] Step 2: Adopt the Python-based graphical user interface (GUI) technology to construct an overall parameter database management system for the light electric helicopter verification aircraft and input the parameter data information of different types of helicopters collected in Step 1 into the overall parameter database management system for the light electric helicopter verification aircraft;

[0039] Specifically included:

[0040] Step 2.1: First, install the PyCharm software on the Windows system, and then configure the corresponding Python environment. Among them, the hardware environment for development is specifically: operating environment: Windows 10; processor: 11th Gen Intel(R) Core(TM) i5 - 7400@3.00GHz 3000MHz; memory: 16GB;

[0041] Step 2.2: Construct an overall parameter database management system for the light electric helicopter verification aircraft, where the overall parameter database management system for the light electric helicopter verification aircraft includes modules such as selection type, data operation, and data display (such as Figure 2As shown in the figure. This system aims to realize the functions of input, query and correction of helicopter parameters. Among them, the selection types consist of ultra-light helicopters (<1 ton), small helicopters (1 ton - 3 tons), medium-small helicopters (3 tons - 6 tons), medium helicopters (6 tons - 10 tons), large helicopters (10 tons - 20 tons) and giant helicopters (20 tons - 40 tons); data operations consist of data addition, data modification and data deletion; data display consists of parameter range search, model search and search results;

[0042] As Figure 3 As shown in the figure, the helicopter model information module consists of helicopter type and model, basic parameters, other parameters and picture upload module. The helicopter type and model consist of type and model; the basic parameters consist of rotor diameter, rotor disc area, empty weight, maximum takeoff weight, maximum disc loading, tail rotor diameter, fuselage - rotation, fuselage length and tail rotor disc area; the other parameters consist of maximum allowable speed, maximum level flight speed, maximum cruise speed, maximum climb rate, service ceiling, hover ceiling (with ground effect), hover ceiling (without ground effect), range and endurance time;

[0043] Step 2.3: Program, debug and implement the general parameter database management system of the light electric helicopter verification aircraft in the PyCharm compiler. Multiple key modules are imported to support the functional requirements of the platform: First, the os module is used to implement the interaction with the operating system, covering file path operations, file and directory management, and access to environment variables, etc.; Second, the csv module is introduced, which is specifically used for reading and writing CSV files to effectively process and analyze tabular data; Third, the Python standard GUI toolkit tkinter is used to create and manage an intuitive and easy-to-use graphical interface; To further improve the function and aesthetics of the GUI, three sub-modules, namely ttk, filedialog and messagebox, are also imported from tkinter. Among them, the ttk module, as an extension of tkinter, provides more modern and consistent-style GUI components; the filedialog module is used to conveniently open the file selection dialog box; and the messagebox module is used to display dialog boxes such as messages, warnings and error prompts to improve the user interaction experience.

[0044] Step 3: Build a neural network model based on the NRBO-XGboost algorithm, and use the parameter data information of different model helicopters in the general parameter database management system of the light electric helicopter verification aircraft to train the neural network model to establish the corresponding relationship between each parameter;

[0045] Compared with other methods, the fitting effect of the neural network model based on the NRBO-XGboost algorithm is better.

[0046] The following presents a comparative example of four methods (NRBO-XGboost, CNN-LSTM, Bayes-LSTM, power function):

[0047] Step 3.1: Select the parameter data of helicopter models with a maximum takeoff weight within 10 tons as the training set. Among them, there are 77 datasets for "maximum takeoff weight - empty weight", 76 datasets for "maximum takeoff weight - disc loading", 76 datasets for "maximum takeoff weight - rotor diameter", 74 datasets for "maximum takeoff weight - rotor solidity", 69 datasets for "maximum takeoff weight - tip speed", 60 datasets for "rotor diameter - overall length of the aircraft", 71 datasets for "rotor diameter - fuselage length", and 78 datasets for "maximum takeoff weight - tail rotor diameter".

[0048] Step 3.2: On the same training set, simultaneously train the Bayesian - Long Short-Term Memory network (Bayes-LSTM), Convolutional Neural Network - Long Short-Term Memory network (CNN-LSTM), and Newton-Raphson Optimization Algorithm - Extreme Gradient Boosting Algorithm (NRBO-XGboost), and construct a semi-empirical formula based on the power function;

[0049] (1) Power function: The fitting curve of the power function method for "maximum takeoff weight - empty weight" is as Figure 4 shown; the training set effect of the power function method for "maximum takeoff weight - empty weight" is as Figure 5 shown;

[0050] The semi-empirical formula for "maximum takeoff weight - empty weight" constructed using the power function method is:

[0051] G e = 0.8355G 0.9525 ;

[0052] where G e is the empty weight and G is the maximum takeoff weight.

[0053] (2) Bayes-LSTM: The algorithm flow of Bayes-LSTM is as Figure 6 shown. Taking "maximum takeoff weight - empty weight" as an example, when implementing the Bayes-LSTM neural network, the specific steps are as follows:

[0054] ① Divide the dataset: 77 groups for the training set and 14 groups for the test set. Each group of data has 1 input parameter and 1 output parameter;

[0055] ② Normalize the data: Normalize the data of the training set and the test set to [0, 1] to unify the data dimension, suppress the phenomena of gradient disappearance and gradient explosion, and improve the robustness of model training; the normalization formula is:

[0056]

[0057] Wherein, is the i-th data value after normalization, Y i is the i-th data value before normalization, Y min is the minimum value in the training set, Y max is the maximum value in the training set;

[0058] ③ Convert the data format: Convert the normalized data from matrix format to cell array format for easy processing by the LSTM model;

[0059] ④ Bayesian optimization:

[0060] First, determine the hyperparameters to be optimized and their ranges during the Bayesian optimization process (see Table 1);

[0061] Table 1 Hyperparameters to be optimized and their ranges

[0062] Hyperparameters to be optimized Range Number of hidden units in the LSTM layer [40,50] Initial learning rate <![CDATA[[10 -3 ,1]]]> <![CDATA[L2 regularization coefficient]]> <![CDATA[[10 -10 ,10 -2 >

[0063] Then, define the objective function of Bayesian optimization, which evaluates the performance of hyperparameters by training the LSTM network and calculating the root mean square error (RMSE);

[0064] Finally, through the acquisition function, find the optimal combination of hyperparameters, specifically:

[0065] a. The objective function f(x) evaluates the performance of the hyperparameter combination by training the LSTM network and calculating the root mean square error (RMSE). The formula is:

[0066] where x = [x1, x2, x3] is the hyperparameter combination, namely the number of hidden units in the LSTM layer, the initial learning rate, and the L2 regularization coefficient, y

[0067]

[0068] wherein, y i is the i-th true value, is the i-th predicted value, and n is the number of samples;

[0069] b. To model the objective function, Bayesian optimization uses a probabilistic surrogate model to replace the objective function and defines the posterior probability as:

[0070] wherein,

[0071]

[0072] where D = {(x1, f1), (x2, f2), …, (x n , fn )} is the sampled point that has been collected, p(f) is the prior distribution, p(D|f) is the likelihood function, and p(D) is the marginal likelihood;

[0073] c. The Gaussian process is a commonly used probabilistic surrogate model. Assuming that the objective function f(x) follows a Gaussian distribution, we have:

[0074]

[0075] In the formula, m(x) is the mean function, and k(x, x′) is the covariance function.

[0076] d. To select the next hyperparameter combination, the acquisition function is defined as:

[0077]

[0078] In the formula, f(x + ) is the current optimal objective function value.

[0079] ⑤ LSTM network training:

[0080] First, create the LSTM network structure using the optimal hyperparameters: input layer, LSTM layer, ReLU activation layer, fully connected layer, and regression layer;

[0081] Then, set the training parameters (see Table 2);

[0082] Table 2 LSTM training parameters and ranges

[0083] Parameter type Range Optimization algorithm Adam Maximum number of training epochs 1000 times Initial learning rate 0.8558 Learning rate adjustment strategy Piecewise adjustment Learning rate decay period 450 Learning rate decay factor 0.5 <![CDATA[L2 regularization coefficient]]> <![CDATA[2.06×10 -5 > Number of hidden units in the LSTM layer 41

[0084] Finally, use the training set data to train the LSTM network and gradually optimize the objective function. Among them, the LSTM neural network structure is as Figure 7 shown, specifically:

[0085] a. The forget gate determines which information in the cell state needs to be discarded. The output of the forget gate is defined as:

[0086] f t = σ(W f h t-1 + W f x t + b f );

[0087] In the formula, f t is the output of the forget gate, σ(·) is the Sigmoid activation function, W f is the weight matrix, h t-1 is the hidden state at time step t - 1, x t is the input at time step t, and b fis the bias vector;

[0088] b. The input gate determines what new information to add to the cell state and consists of two parts. The formula is:

[0089]

[0090] In the formula, i t determines the required information to be updated into the cell state, W i and W c are weight matrices, b i and b c are bias vectors, C t and C t-1 are the memory cell states at time steps t and t - 1 respectively, and tanh(·) is the hyperbolic tangent activation function;

[0091] c. The output gate determines which state features of the cell to output based on the input sum. The formula is:

[0092]

[0093] In the formula, o t determines the output part of the cell state, W o is the weight matrix, b o is the bias vector, h t is the hidden state at time step t;

[0094] d. The ReLU activation layer can increase the non - linear ability of the network. The formula is:

[0095] a t = max(0, h t );

[0096] In the formula, a t is the output of the ReLU activation function;

[0097] e. The fully - connected layer maps the output of the LSTM layer to the final output dimension. The formula is:

[0098] y t = W y × a t + b y ;

[0099] In the formula, y t is the final predicted value, W y is the weight matrix, b y is the bias vector;

[0100] f. The regression layer is used for regression tasks and calculates the mean squared error (MSE) between the predicted value and the true value. The formula is:

[0101]

[0102] Wherein, y i is the i-th true value, is the i-th predicted value, and n is the number of samples;

[0103] ⑥ Prediction data: Save the trained model and make predictions on the training set and the test set;

[0104] ⑦ Inverse normalization of data: Inverse normalize the prediction results to obtain the predicted values within the original data range;

[0105] ⑧ Model accuracy: Use the coefficient of determination and the mean absolute percentage error to measure the accuracy of the model. The closer the coefficient of determination is to 1, the better the model fitting effect; the smaller the mean absolute percentage error, the higher the model prediction accuracy. The training set effect of the Bayes-LSTM method for "maximum takeoff weight - empty aircraft weight" is as Figure 8 shown.

[0106] (3) CNN-LSTM: The CNN-LSTM neural network structure is as Figure 9 shown. Taking "maximum takeoff weight - empty aircraft weight" as an example, when implementing the CNN-LSTM neural network, the specific steps are as follows:

[0107] ① Divide the data set: 77 groups in the training set, 14 groups in the test set, 1 input parameter and 1 output parameter for each group of data;

[0108] ② Normalize the data: Normalize the data in the training set and the test set to [0, 1] to unify the data dimension, suppress the gradient vanishing and gradient explosion phenomena, and improve the robustness of model training. The formula is:

[0109]

[0110] Wherein, is the i-th data value after normalization, Y i is the i-th data value before normalization, Y min is the minimum value in the training set, Y max is the maximum value in the training set;

[0111] ③ Flatten the data: Flatten the data into 1D data to keep it consistent with the input layer data structure;

[0112] ④ Convert the data format: Convert the flattened data into a cell array format for easy processing by the CNN-LSTM network;

[0113] ⑤ CNN-LSTM model training:

[0114] First, create the CNN-LSTM model structure;

[0115] Then, design the training parameters (see Table 3);

[0116] Table 3 Training Parameters and Ranges

[0117] Parameter type Range Optimization algorithm Adam Maximum number of training epochs 1500 times Initial learning rate 0.001 Learning rate adjustment strategy Piecewise adjustment Learning rate decay period 300 Learning rate decay factor 0.001 <![CDATA[L2 regularization coefficient]]> <![CDATA[10 -4 > Mini-batch size 128 Data shuffling After each epoch

[0118] Finally, use the training set data to train the CNN-LSTM network and gradually optimize the objective function, specifically:

[0119] a. The input layer receives 1-dimensional input data x t ;

[0120] b. The convolutional layer extracts local features of the input data. The convolutional kernel size is 15, the number of convolutional kernels is 64 and 128, and the activation function is ReLU. The formula is:

[0121]

[0122] In the formula, z t is the output of the convolutional layer, W g is the weight matrix of the convolutional layer, b g is the bias vector, * is the convolution operation, and c t is the output after activation of the convolutional layer;

[0123] c. The pooling layer can remove redundant information and simplify the network complexity. The formula is:

[0124] d t = δ(c t ) + b d ;

[0125] In the formula, d t is the output of the pooling layer, δ(·) is the pooling method of the pooling layer, and b d is the bias vector.

[0126] d. The LSTM layer processes sequential data, contains 512 hidden units, and can capture long-term dependencies in the time series. The formula is the same as Bayes-LSTM;

[0127] The relationship, the formula is the same as Bayes-LSTM;

[0128] e. The fully connected layer maps the output of the LSTM layer to the final output dimension. The formula is the same as Bayes-LSTM;

[0129] f. The regression layer is used for regression tasks and calculates the mean square error (MSE) between the predicted value and the true value. The formula is the same as the Bayes-LSTM neural network;

[0130] ⑥ Predict data: Save the trained model and make predictions on the training set and the test set;

[0131] ⑦Inverse normalization of data: Inverse normalize the prediction results to obtain the predicted values within the range of the original data;

[0132] ⑧Model accuracy: Use the coefficient of determination and the mean absolute percentage error to measure the accuracy of the model. The closer the coefficient of determination is to 1, the better the model fitting effect; the smaller the mean absolute percentage error, the higher the model prediction accuracy. The training set effect of the CNN-LSTM method for "maximum takeoff weight - empty aircraft weight" is as Figure 10 shown.

[0133] (4) NRBO-XGboost: The algorithm flow of NRBO-XGboost is as Figure 11 shown. Taking "maximum takeoff weight - empty aircraft weight" as an example, when implementing the NRBO-XGboost algorithm, the specific steps are as follows:

[0134] ①Divide the dataset: 77 groups for the training set and 14 groups for the test set; each group of data has 1 input parameter and 1 output parameter;

[0135] ②Normalize the data: Normalize the data of the training set and the test set to [0, 1] to unify the data dimension, suppress the vanishing gradient and gradient explosion phenomena, and improve the robustness of model training. The formula is:

[0136]

[0137] where, is the i-th data value after normalization, Y i [[ID=?]]is the i-th data value before normalization, Y min is the minimum value in the training set, Y max is the maximum value in the training set;

[0138] ③Transpose the data: Transpose the normalized data to adapt to the input format of the XGBoost model;

[0139] ④NRBO algorithm:

[0140] First, determine the hyperparameters to be optimized and their ranges (see Table 4);

[0141] Table 4 Hyperparameters to be optimized and their ranges

[0142] Hyperparameters to be optimized Range Maximum number of training epochs for the XGBoost model [100,800] Depth of the trees in the XGBoost model [3,10] Learning rate of the XGBoost model <![CDATA[[10 -4 ,10 -1 >

[0143] Then, define the objective function of the NRBO algorithm, which evaluates the performance of the hyperparameters by training the XGBoost model and calculating the prediction error;

[0144] Finally, calculate the fitness value of each individual and find the optimal combination of hyperparameters, specifically: It should be noted that there seems to be a problem with the tag "" in the original text where the content after it is incomplete. I have translated it as best as possible based on the existing information.

[0145] a. Randomly generate an initial population Z = {x1, x2, … x6} with a size of 6. Each individual x i = {x1, x2, x3} represents a set of hyper

[0146] parameter combinations (maximum number of training times, depth of the tree, and learning rate);

[0147] b. The objective function f(x) evaluates the performance of the hyperparameters by training the XGBoost model and calculating the root mean square error (RMSE). The formula is:

[0148]

[0149] where y i is the i-th true value, is the i-th predicted value, and n is the number of samples;

[0150] c. For each individual x i , calculate the fitness value as:

[0151] Fitness(x i ) = f(x i );

[0152] d. Define the Newton-Raphson optimization algorithm as:

[0153]

[0154] where δ is the step size factor, is the gradient of the objective function f(x i );

[0155] Among them, to avoid falling into a local optimal solution, a trap avoidance operation is introduced. The formula is:

[0156] x new = x new + θ·r;

[0157] where θ is the random perturbation factor and r is the random vector;

[0158] e. Update the population Z according to the fitness value, and select the individuals with smaller fitness values to enter the next generation;

[0159] f. Repeat steps b to e until the maximum number of iterations 20 is reached or convergence occurs;

[0160] ⑤ XGBoost model training:

[0161] First, set the model parameters (see Table 5);

[0162] Table 5 XGBoost training parameters and ranges

[0163] Parameter type Range Maximum number of training epochs 800 Depth of the trees 10 Learning rate 0.1

[0164] Then, determine the objective function of the XGBoost model;

[0165] Finally, complete the training of the model by iteratively updating the model parameters and gradually optimizing the objective function. Specifically:

[0166] a. Define the objective function of the XGBoost model as:

[0167]

[0168] In the formula, is the loss function. is the regularization term, used to control the complexity of the model. T is the number of leaf nodes of the tree, w j is the weight of the j-th leaf node, and γ and λ are regularization parameters;

[0169] b. Iteratively update the model parameters through the gradient boosting algorithm. The formula is:

[0170]

[0171] In the formula, is the predicted value after the t-th iteration, and f t (x i ) is the predicted value of the t-th tree.

[0172] c. Assume that the model has a total of M trees, then the final predicted value is:

[0173]

[0174] ⑥ Predict data: Save the trained model and predict the training set and the test set;

[0175] ⑦ Inverse normalize data: Inverse normalize the prediction results to obtain the predicted values in the original data range;

[0176] ⑧ Model accuracy: Use the coefficient of determination and the mean absolute percentage error to measure the accuracy of the model. The closer the coefficient of determination is to 1, the better the model fitting effect; the smaller the mean absolute percentage error, the higher the model prediction accuracy. The training set effect of the NRBO-XGboost method for "maximum takeoff weight - empty aircraft weight" is as Figure 12 shown.

[0177] Step 3.3: Compare the coefficient of determination (R 2 ) and the mean absolute percentage error (MAPE) of the four methods;

[0178] Among them, the coefficient of determination is used to measure the fitting degree between the predicted value and the actual value of the model. Its value range is [0, 1]. The closer the coefficient of determination is to 1, the better the fitting effect of the model. The formula is:

[0179]

[0180] In the formula, is the average value of the true values in the dataset, y i is the i-th true value, is the i-th predicted value, and n is the number of samples;

[0181] The mean absolute percentage error is used to measure the average percentage error between the predicted value and the actual value. The smaller the value, the higher the prediction accuracy of the model. The formula is:

[0182]

[0183] Table 6 Comparison chart of indicators of each method

[0184] Method Coefficient of determination Mean absolute percentage error NRBO-XGboost 0.99995 0.83％ CNN-LSTM 0.9836 8.20％ Bayes-LSTM 0.9787 9.94％ [[ID=৭০]]Power function 0.9766 11.04％

[0185] By comparing the coefficient of determination and the mean absolute percentage error of these four methods (see Table 6), NRBO-XGboost has a good fitting effect. Therefore, the NRBO-XGboost method is selected to establish a mathematical model between the "maximum takeoff weight" and the "empty aircraft weight". The present invention only gives that the NRBO-XGboost is selected for the mathematical model of "maximum takeoff weight - empty aircraft weight", and it can be verified that the NRBO-XGboost has a good fitting effect for the remaining models.

[0186] Step 4: Use the trained neural network model in Step 3 to predict the parameters of the electric helicopter.

[0187] In this embodiment, given that the maximum takeoff weight of the electric helicopter is 930 kg and the rotor diameter is 8.2 m, the NRBO-XGboost algorithm is used to predict other parameters of the electric helicopter (empty aircraft weight, disc loading, rotor diameter, rotor solidity, tip speed, overall length, fuselage length, and tail rotor diameter) (the prediction results are shown in Table 7), providing technical support for the actual manufacturing process of future electric helicopters.

[0188] Table 7 Prediction results of electric helicopter parameters

[0189] Parameter name Predicted value Empty aircraft weight 681 Disk loading 16.48 Rotor diameter 8.21 Rotor solidity 0.0436 Tip speed 182.93 Overall length of the aircraft 9.41 Fuselage length 7.50 Tail rotor diameter 1.31

[0190] It should be noted that the purpose of disclosing the embodiments is to help further understand the present invention. However, those skilled in the art can understand that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection claimed by the present invention shall be subject to the scope defined by the claims.

Claims

1. A general parameter prediction method for a light electric helicopter verification aircraft, characterized in that, Including: Step 1: Collect parameter data information of different models of helicopters; Step 2: Construct a general parameter database management system for a light electric helicopter verification aircraft and input the parameter data information of different models of helicopters collected in Step 1 into the general parameter database management system for the light electric helicopter verification aircraft; Step 3: Construct a neural network model based on the NRBO-XGboost algorithm, and use the parameter data information of different models of helicopters in the general parameter database management system for the light electric helicopter verification aircraft to train the neural network model, and establish the corresponding relationship between parameters; Step 4: Use the trained neural network model in Step 3 to predict the parameters of the electric helicopter.

2. The overall parameter prediction method for the light electric helicopter verification aircraft according to claim 1, characterized in that: In Step 2, a general parameter database management system for a light electric helicopter verification aircraft is constructed using Python-based graphical user interface technology.

3. The overall parameter prediction method for the light electric helicopter verification aircraft according to claim 2, characterized in that: In Step 2, constructing a general parameter database management system for a light electric helicopter verification aircraft using Python-based graphical user interface technology specifically includes: Step 2.1: First, install the PyCharm software on the Windows system, and then configure the corresponding Python environment; Step 2.2: Construct a general parameter database management system for a light electric helicopter verification aircraft; Step 2.3: Program, debug, and implement the general parameter database management system for the light electric helicopter verification aircraft in the PyCharm compiler.

4. The overall parameter prediction method for the light electric helicopter verification aircraft according to claim 3, characterized in that: In Step 2.2, the general parameter database management system for the light electric helicopter verification aircraft includes a selection type, data operation, and data display modules. Among them, the selection type consists of ultra-light helicopters, small helicopters, medium-small helicopters, medium helicopters, large helicopters, and giant helicopters; the data operation consists of data addition, data modification, and data deletion; the data display consists of parameter range search, model search, and search results.

5. The overall parameter prediction method for the light electric helicopter verification aircraft according to claim 1, characterized in that: In Step 3, the training set for training the neural network model includes the following data sets of different models of light electric helicopters: "Maximum takeoff weight - empty aircraft weight", "Maximum takeoff weight - disc loading", "Maximum takeoff weight - rotor diameter", "Maximum takeoff weight - rotor solidity", "Maximum takeoff weight - tip speed", "Rotor diameter - overall length of the aircraft", "Rotor diameter - fuselage length", and "Maximum takeoff weight - tail rotor diameter".

6. The overall parameter prediction method for the light electric helicopter verification aircraft according to claim 5, characterized in that: Using the trained neural network model in Step 3 to predict the parameters of the electric helicopter specifically means: predicting the data of the empty aircraft weight, disc loading, rotor diameter, rotor solidity, tip speed, overall length of the aircraft, fuselage length, and tail rotor diameter of the electric helicopter using the maximum takeoff weight and rotor diameter of the electric helicopter.

Images