Method and system for obtaining performance parameters of an aeroengine and baseline values thereof

Abstract

The application provides an acquisition method and system for performance parameters of an aircraft engine and baseline values of the performance parameters, can automatically acquire real-time space-ground data link messages, classifies the messages according to key fields, reads all parameter fields in the messages and extracts relevant parameters for calculating EGTM and Baseline after determining the type of the messages, finally writes all the read parameters and the calculation results into a database, and the approximator based on the self-improved adaptive GA-ELM has good effects, the calculation precisions of EGTM and Baseline both meet the engine performance monitoring requirements, and the approximator based on the generalized regression network has reached high precision when approximating the deviation values of EGT, N2 and FF.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electric digital data processing and monitoring, in particular to a method and system for obtaining performance parameters and baseline values of an aero-engine. BACKGROUND

[0002] Currently, in the daily engine monitoring work of airlines, the monitoring software provided by the manufacturer must be relied on to calculate the EGTM (Exhaust Gas Temperature Margin), SLOATL (Sea Level Out Air Temperature Limit) and the baseline values of performance parameters. The monitoring software provided by GE (General Electric) is the single-machine version of SAGE (System for the Analysis of Gas turbine Engines) and the network version of RD (Remote Diagnosis) system. The SAGE system was developed too long ago, and GE has stopped technical support. The SAGE message reading is not convenient, the operation speed is slow, and it can only be used for old models, so the software will soon be phased out.

[0003] The server of the RD system is located overseas, and the access speed and data download speed are too slow, and even often interrupted. Therefore, the usability of the RD system in actual monitoring work is low. It is urgent to develop new software for the calculation of EGTM, SLOATL and the baseline values of performance parameters.

[0004] Currently, domestic scholars use the EGTM algorithm of CFM56-3 engine to calculate the EGTM algorithm of CFM56-5C engine, although some achievements have been made, but the precision still cannot meet the engineering application standard, and no application software has been developed for the algorithm, so it has little use value.

[0005] In terms of the calculation of the baseline values of performance parameters, domestic scholars have used Newton-Gauss iteration, artificial intelligence and other algorithms to solve the baseline values. However, due to too few learning samples, the generalization ability and robustness of the baseline value calculation model are insufficient, and when the parameters exceed a certain interval, the precision will be greatly reduced.

[0006] Moreover, these scholars have not developed specific technical solutions for practical engineering applications. SUMMARY

[0007] The first objective of this invention is to provide a method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, aiming to obtain more accurate aero-engine performance parameters and their baseline values.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, characterized by comprising the following steps:

[0010] Message acquisition: Receive and download real-time messages from the engine during takeoff and cruise, identify and determine the engine configuration and its corresponding message type, and store them in the database;

[0011] Parameter filtering involves selecting N types of calculation parameters from the database for calculating the corresponding aero-engine performance parameters or their baseline values, and then normalizing them.

[0012] For each type of calculation parameter, a specific group of data is randomly selected from the database and the specific group of data is randomly divided into four types of samples in an unequal manner: genetic training samples, neural network training samples, test samples, and calculation samples. The first N-1 types of parameters of each type of sample are stored in the first matrix, and the remaining type of parameters are stored in the second matrix.

[0013] Genetic algorithm training involves setting up a temporary ELM for genetic training, which includes an input layer with N-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. The genetic training samples consist of u sets of data. The first matrix corresponding to the data is placed at the input of the temporary ELM, and the second matrix corresponding to the data is placed at the output of the temporary ELM. After determining the parameters of the adaptive GA and performing adaptive GA training, a specific number of individuals with d genes are randomly generated.

[0014] Individual selection involves inputting multiple different test samples into a temporary ELM and repeatedly performing adaptive GA training to obtain the optimal individual corresponding to the optimal fitness value.

[0015] Neural network training involves defining a new ELM for neural network training, where the number of neurons in its input, intermediate, and output layers is equal to that of the temporary ELM. The optimal weights and thresholds corresponding to the best individual are assigned to the new ELM. The number of training samples for the neural network is w sets of data. The first matrix corresponding to the data is placed at the input of the new ELM, and the second matrix corresponding to the data is placed at the output of the new ELM. Neural network training is performed, and an adaptive GA-ELM approximator is obtained after completion.

[0016] Parameter calculation: The sample size of the calculation sample is t sets of data. The first matrix corresponding to it is placed at the input end of the adaptive GA-ELM approximator, and the second matrix corresponding to it is placed at the output end of the adaptive GA-ELM approximator. The adaptive GA-ELM approximator is run, and the training results are stored in the target matrix. The dimension of the target matrix is ​​the same as the dimension of the second matrix.

[0017] The results are obtained by performing inverse normalization on the target matrix to obtain the engine performance parameters.

[0018] Preferably, the performance parameter is the actual value of EGTM, and the baseline value is the EGT baseline value, the FF baseline value, and the N2 baseline value.

[0019] As an improvement, the individual screening step further includes the following steps:

[0020] Step 1: Fitness value acquisition. The test sample randomly selects data of sample size v from the database. The first matrix corresponding to the test sample is placed at the input of the temporary ELM, and the second matrix corresponding to the test sample is placed at the output of the temporary ELM to obtain the fitness value corresponding to each individual.

[0021] Step 2: Select parent and mother individuals using a sequential selection method based on their probability of being selected.

[0022] Step 3, genetic training: Adaptive GA training is performed on the parent and mother individuals to obtain crossover and mutation probabilities. Then, crossover and mutation operations are performed on the individuals in the adaptive GA to generate a new generation of individuals and calculate their fitness values.

[0023] Step 4: Fitness optimization determination. Determine whether to terminate the optimization process based on the determination criteria. If the determination criteria are met, proceed to Step 5; otherwise, repeat Steps 1 to 4. The determination criteria are whether the fitness value reaches a specified value or whether the number of adaptive GA training sessions reaches a specified upper limit.

[0024] Step 5: Individual numerical output. Output the optimal individual corresponding to the optimal fitness value.

[0025] As an improvement, in the parameter selection step, when the required engine performance parameter is ETGM, the selected parameter types include external atmospheric environment parameters, engine operating parameters, engine configuration parameters, and aircraft operating parameters.

[0026] As an improvement, in the calculation parameter screening step, when the required baseline value of the aero-engine performance parameter is the EGT baseline value, FF baseline value, or N2 baseline value, the types of calculation parameters to be screened include external atmospheric environment parameters and engine operating parameters.

[0027] As an improvement, the message acquisition step further includes the following steps:

[0028] Message download: After collecting operational data from various operating conditions, the airborne equipment of the aero-engine is filtered to form real-time air-to-ground data link messages, which are then transmitted to the ground receiving station in real time via the real-time air-to-ground data link.

[0029] Message classification is based on the characteristic data of the real-time air-to-ground data link messages, classifying them into messages under different configurations and different operating states, as well as messages with incorrect formats.

[0030] The message is transferred and stored. Based on the type of the real-time air-to-ground data link message, all its information is extracted into a string, and the string is converted into string type, time type, floating point type, integer type, or boolean type data according to the data type. The converted data is then stored in the database.

[0031] The second objective of this invention is to provide a method for obtaining sea level atmospheric temperature limits, so as to monitor sea level atmospheric temperature limits.

[0032] To achieve the above objectives, the present invention adopts the following technical solution:

[0033] A method for obtaining sea level atmospheric temperature limits is characterized by selecting a large amount of actual takeoff data from the database and then fitting the actual takeoff data with the EGTM using univariate nonlinear regression.

[0034] The third objective of this invention is to provide a method for obtaining a lubricating oil pressure baseline value, which aims to obtain a more accurate lubricating oil pressure baseline value.

[0035] The above objectives are achieved using the following technical solution:

[0036] A method for obtaining a baseline value of lubricating oil pressure, characterized in that it utilizes the N2 and performs fitting using logarithmic regression.

[0037] The fourth objective of this invention is to provide a method for obtaining parameter deviation values, which aims to use the obtained parameter deviation values ​​for real-time monitoring of aero-engine performance status.

[0038] The above objectives are achieved using the following technical solution:

[0039] A method for obtaining parameter deviation values ​​based on the baseline value, characterized by comprising the following steps:

[0040] Parameter filtering involves selecting m types of computational input parameters from the database for calculating parameter deviation values.

[0041] Obtain the original results by using the parameter deviation value reference formula provided by the aero-engine manufacturer to obtain the original parameter deviation value, which is then used as the calculation output parameter;

[0042] Neural training is performed to establish a generalized regression neural network. The computational parameters are put into the input layer of the generalized regression neural network, and the deviation values ​​of the traditional parameters are put into the output layer of the generalized regression neural network. Then, the neural network is trained to obtain a generalized regression neural network approximator.

[0043] The parameters are acquired, and then m types of calculation parameters for calculating parameter deviation values ​​are randomly selected, put into the generalized regression neural network approximator and run to obtain parameter deviation values ​​that can be used for real-time monitoring of the engine performance status.

[0044] The fifth objective of this invention is to provide a system for acquiring aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, which aims to acquire more accurate aero-engine performance parameters and their baseline values.

[0045] To achieve the above objectives, the present invention adopts the following technical solution:

[0046] A system for acquiring aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, characterized in that it includes:

[0047] The message acquisition module is used to receive and download real-time messages from the engine during takeoff and cruise, and to identify and determine the engine configuration and its corresponding message type.

[0048] A database is used to store the categorized real-time messages;

[0049] The parameter filter selects the corresponding n types of calculation parameters based on the type of aero-engine performance parameters to be calculated or the type of their baseline values.

[0050] A normalization processor is used to normalize the calculation parameters;

[0051] The sample partitioning module is used to randomly select a specific group of data from the database for each type of calculation parameter, and randomly divide the specific group of data into four types of samples in an unequal manner. The first n-1 types of parameters of each type of sample are stored in a first matrix, and the remaining type of parameters are stored in a second matrix. The four types of samples are genetic training samples, neural network training samples, test samples, and calculation samples.

[0052] A temporary ELM includes an input layer with n-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. It receives a sample size of u sets of genetic training data. The first matrix corresponding to the data is placed at the input of the temporary ELM, and the second matrix corresponding to the data is placed at the output of the temporary ELM. After determining the parameters of the adaptive GA and performing adaptive GA training, a specific number of individuals with d genes are randomly generated.

[0053] The individual selection module is used to input multiple different test samples into the temporary ELM and repeatedly perform adaptive GA training in order to obtain the optimal individual corresponding to the optimal fitness value.

[0054] An adaptive GA-ELM approximator is obtained by training a neural network using a new ELM whose input, intermediate, and output layers have the same number of neurons as the temporary ELM. This ELM is assigned the optimal weights and thresholds corresponding to the optimal individual and trained on a neural network with w sets of training samples. The adaptive GA-ELM approximator is used to receive t sets of computational samples, train the approximator to obtain the training results, and store them in the target matrix.

[0055] The parameter output module is used to perform inverse normalization on the target matrix and output the aero-engine performance parameters or their reference values.

[0056] The data display module is used to display data in the database, including the aero-engine performance parameters or their benchmark values.

[0057] The above technical solution can be improved as follows:

[0058] As an improvement, the message acquisition module further includes the following sub-modules:

[0059] The message download submodule is used to collect the operating data of the aero-engine airborne equipment under various operating conditions, filter it to form real-time air-to-ground data link messages, and transmit them to the ground receiving station in real time via the real-time air-to-ground data link.

[0060] The message classification submodule is used to classify the real-time air-to-ground data link messages into messages with different configurations and different operating states, as well as messages with incorrect formats, based on their characteristic data.

[0061] The message transfer submodule is used to extract all information of the real-time air-to-ground data link message into a string according to its type, and convert the string into string type, time type, floating point type, integer type, or boolean type data according to the data type, and then store the converted data into the database.

[0062] As a further improvement, a sea level atmospheric temperature limit value acquisition module is also included, which is used to select a large amount of actual takeoff data from the database, and then use the actual takeoff data and the EGTM to perform a univariate nonlinear regression to fit and obtain the sea level atmospheric temperature limit value.

[0063] As a further improvement, a lubricating oil baseline value acquisition module is also included, which acquires the lubricating oil baseline value by utilizing the N2 and fitting it with logarithmic regression.

[0064] As a further improvement, a parameter deviation value acquisition module is also included, which includes a generalized regression neural network approximator that receives and runs m types of calculation parameters for calculating parameter deviation values ​​selected by the parameter filter, and outputs parameter deviation values ​​that can be used for real-time monitoring of the engine performance status.

[0065] As a further improvement, a backup processing module is also included, which is used to detect and complete the missing real-time air-to-ground data link messages.

[0066] Compared with the prior art, the present invention has the following advantages:

[0067] (1) It can operate independently without relying on the technical support of OEM manufacturers, that is, it can calculate the important performance parameters of the engine without using any manufacturer software, which makes the airline's engine monitoring work more autonomous.

[0068] (2) When the user receives the engine message, the system described in this invention can immediately start the data processing mode and calculate engine performance parameters such as EGTM, Baseline, deviation value, and smoothing value. After receiving the message, the user can obtain the engine performance parameters in the shortest time. However, if the manufacturer's system is used, it is necessary to wait for the server to process the data in batches. Due to the time delay, it brings great inconvenience to the monitoring work.

[0069] (3) It solves the problem of international network interruption. Because the manufacturer's monitoring system server is located overseas, the network connection is unstable or even frequently interrupted, which seriously interferes with the monitoring work. The system described in this invention is installed on the user server, and the user can view the engine performance data by connecting through the local area network.

[0070] (4) The system described in this invention can run autonomously in the background and can run automatically on a timed basis without manual operation and intervention. It can automatically read, calculate, and write data, achieving true full automation. At the same time, it has the ability to process messages in batches, which greatly improves the processing speed of messages and the efficiency of monitoring work.

[0071] (5) A backup message reading function is provided. When a real-time air-to-ground data link message is missing, an offline formatted data message can be downloaded through the real-time air-to-ground data link platform. The system described in this invention can read and process the offline formatted message. This function can be used to quickly make up for the data loss caused by the real-time air-to-ground data link message.

[0072] (6) The system described in this invention uses a swarm intelligence algorithm for optimization calculation and an artificial intelligence algorithm (ANN) for calculating engine performance parameters such as EGTM, Baseline, and deviation value. It has fast calculation speed and high calculation accuracy.

[0073] (7) The intelligent algorithm used has been verified by a large number of test samples and has achieved satisfactory calculation accuracy when used for actual engine operation data calculation. The test results show that the method described in this invention has strong anti-noise ability, high robustness and generalization ability, and is suitable for calculating engine performance parameters under different working conditions. Attached Figure Description

[0074] Figure 1 This is a graph showing the linear relationship between EGT and the ambient total temperature.

[0075] Figure 2 A schematic diagram showing the baseline values ​​of exhaust temperature, core engine speed, and fuel flow provided by the aero-engine manufacturer;

[0076] Figure 3 This is a flowchart illustrating the method for obtaining the aero-engine performance parameters and their baseline values ​​as described in Example 1.

[0077] Figure 4 This is a schematic diagram of the specific process of step 100 in Example 1;

[0078] Figure 5 This is a schematic diagram illustrating the specific style of a real-time air-to-ground data link message.

[0079] Figure 6 A schematic diagram of message classification logic;

[0080] Figure 7 This is a schematic diagram of the specific process of step 500 in Example 1;

[0081] Figure 8 This is a schematic diagram of the entire process of Example 1;

[0082] Figure 9 The graphs of EGTM and SLOATL in Example 2 are shown.

[0083] Figure 10 The graphs of OP baseline and N2 in Example 3 are shown.

[0084] Figure 11This is a flowchart illustrating the method for obtaining parameter deviation values ​​as described in Example 4;

[0085] Figure 12 This is a block diagram of the system structure for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, as described in Example 5.

[0086] Figure 13 This is a block diagram of the system structure for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, as described in Example 6. Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0088] As the technical basis for the various embodiments of this application, the relevant technical content of EGT (Exhaust Gas Temperature) and cruise performance parameter baseline value will be described in detail below.

[0089] During takeoff, aircraft engines operate at high power, with their rotors rotating at their highest speeds throughout the flight. At this time, hot-end components must withstand high temperatures, high centrifugal forces, and axial loads. Therefore, the probability of engine failure is higher during takeoff. Statistics show that over 80% of civil aviation accidents and incidents occur during takeoff and landing.

[0090] EGT reflects the temperature of the engine combustion chamber and high-pressure turbine, and is one of the important indicators for judging engine performance status and degradation level. The magnitude of EGT is related to the engine's inherent performance level, as well as throttle lever angle, whether it takes off at full thrust, the engine's rated thrust setting, engine bleed air status, and the external atmospheric environment. Therefore, relying solely on the takeoff EGT value to judge engine performance level without considering other factors will inevitably lead to bias.

[0091] To objectively evaluate engine performance, engine manufacturers, taking into account the above factors, proposed the concept of EGTM (Enterprise Geometric Parameters Time). Engine manufacturers define EGTM as: the difference between the peak value of various engine parameters during takeoff and the engine's certified redline value, where the engine's parameters reach their peak values ​​throughout the entire operating process. Figure 1 As shown. When calculating EGTM, it is necessary to convert the engine parameters to their values ​​under standard conditions.

[0092] EGTM is not only a basis for assessing the degree of engine performance degradation, but also an important basis for engine fleet performance ranking and engine life prediction.

[0093] According to the output results of the manufacturer's monitoring system, the size of EGTM output by the monitoring software was found to be... Figure 1 The calculation methods shown are inconsistent, and the manufacturer refuses to disclose the calculation methods for EGTM and SLOATL to airlines.

[0094] The deviation values ​​of various performance parameters are the basis for evaluating the engine's performance status. The deviation value is the difference between the measured parameter value and the baseline; therefore, the baseline value should be obtained before calculating the deviation value. The relationship between the performance parameters and operating parameters of a newly manufactured or overhauled engine of the same configuration is called the baseline of the aero-engine. Manufacturers provide baseline value diagrams for exhaust temperature, core engine speed, and fuel flow rate, such as... Figure 2 As shown.

[0095] For engines, during cruise, the baseline of various engine parameters is a nonlinear function of the engine operating environment and the corrective fan speed. However, aircraft engine manufacturers have not disclosed the calculation methods for the baseline values ​​of parameters such as N1 (engine low-pressure rotor speed), EGT, FF (Fuel Flow), and OIP (Oil Pressure).

[0096] Example 1

[0097] Embodiment 1 of this application provides a method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM (Genetic Algorithm-Extreme Learning Machine). The obtainable performance parameters include, but are not limited to, the actual value of EGTM, and the obtainable baseline values ​​include, but are not limited to, EGT baseline value, FF baseline value, and N2 (aero-engine high-pressure rotor speed) baseline value. Figure 3 As shown, the steps include the following.

[0098] Step 100, Message Acquisition:

[0099] By receiving and downloading real-time messages from the engine during takeoff and cruise, the engine configuration and its corresponding message type are identified and stored in the database.

[0100] Optionally, in step 100, such as Figure 4 As shown, it also includes the following steps:

[0101] Step 101, Message Download:

[0102] After collecting operational data from various operating conditions, the airborne equipment of the aero-engine is filtered and formed into real-time air-to-ground data link messages, which are then transmitted to the ground receiving station in real time via the real-time air-to-ground data link.

[0103] Specifically, the specific format of the real-time air-to-ground data link message is as follows: Figure 5 The image shows a portion of a correctly formatted real-time air-to-ground data link message. This message is a decoded message, which contains a wider variety of parameters and is easier to read and distinguish from the original message.

[0104] Step 102, Message Classification:

[0105] Based on the characteristic data of the real-time air-to-ground data link messages, they are classified into messages under different configurations and different operating states, as well as messages with incorrect formats.

[0106] Specifically, based on engine operating conditions, real-time air-to-ground data link messages can be divided into two main categories: takeoff messages and cruise messages. For specific aero-engine models, due to iterative improvements, the same aero-engine model can have multiple configurations. Because of these differences in engine configuration, the content and structure of real-time air-to-ground data link messages vary; therefore, real-time air-to-ground data link messages need to be classified according to engine operating conditions and engine configuration. Taking the CFM56-7B engine as an example, its real-time air-to-ground data link messages can be categorized as shown in Table 1.

[0107] Table 1 Summary of Message Types

[0108] Message Name Message Abbreviation Message Number Message Format 916 Configuration Takeoff Message 916 TKO Message 1 TXT 916 Configuration Cruise Message 916 CRZ Message 2 TXT 816 Configuration Takeoff Message 816 TKO Message 3 TXT 816 Configuration Cruise Message 816 CRZ Message 4 TXT 916 Configuration Takeoff Message CSV_916 TKO Message 5 CSV 916 Configuration Cruise Message CSV_916 CRZ Message 6 CSV 816 Configuration Takeoff Message CSV_816 TKO Message 7 CSV 816 Configuration Cruise Message CSV_916 CRZ Message 8 CSV Error Message Error Message 9 TXT

[0109] Specifically, the CFM56-7B engine, based on the characteristics of various message types and the specific content contained within the messages, can... Figure 6 The classification logic shown divides messages into 5 types and proposes corresponding processing methods for different types of messages. Specifically, it first reads the third line of data, then checks if it contains the character "TAKE". If so, it continues reading columns 5-7 of the second line, checking if they contain a space character. If so, it runs the 816TKO procedure, classifying it as an 816 configuration takeoff message; otherwise, it runs the 916TKO procedure, classifying it as a 916 configuration takeoff message. If the read third line of data does not contain the character "TAKE", it checks if it contains the character "STAB". If not, it is considered an incorrect format message and discarded. If so, it continues reading columns 5-7 of the second line, checking if they contain a space character. If so, it runs the 816CRZ procedure, classifying it as an 816 configuration cruise message; otherwise, it runs the 916CRZ procedure, classifying it as a 916 configuration cruise message.

[0110] Step 103, Message Transfer:

[0111] Based on the type of the real-time air-to-ground data link message, all its information is extracted into a string, and the string is converted into string type, time type, floating point type, integer type, or boolean type data according to the data type, and the converted data is stored in the database.

[0112] Specifically, after the transformed data has been written, the original message is closed and cut and pasted into a dedicated folder named "Processed". If the message being read is an Error format message, it is discarded and stored in a dedicated folder named "Reject".

[0113] Taking the CFM6-7B engine as an example, the data that the engine's data reading module needs to read can be divided into takeoff / climb data and cruise data, as shown in Tables 2 and 3 respectively.

[0114] Table 2 Data Readings During Takeoff / Climb Phase

[0115]

[0116]

[0117]

[0118] Table 3 Cruise Status Reading Data

[0119]

[0120]

[0121] Step 200, Calculate parameter filtering:

[0122] The database is used to select n types of calculation parameters for calculating the corresponding aero-engine performance parameters or their baseline values, and these parameters are then normalized.

[0123] Specifically, when the required engine performance parameter is ETGM, since there are many types of parameters in the real-time air-to-ground data link message, most of which are unrelated to the calculation of EGTM, if these unrelated parameters are substituted into the calculation of EGTM, the calculation accuracy of EGTM will be seriously affected. Therefore, the selection of calculation parameters is particularly important.

[0124] Based on engine principles and technical data provided by the manufacturer, the parameters required for calculating EGTM include external atmospheric parameters, aircraft parameters, engine operating parameters, and engine configuration parameters. The selected influencing factors are summarized in Table 4 below.

[0125] Table 4. Influence factors after EGTM screening

[0126]

[0127]

[0128] Because the data in Table 4 have different dimensions and vastly different orders of magnitude, this will significantly affect the computational accuracy of the Extreme Learning Machine. Therefore, it is necessary to remove the dimensions from each parameter in Table 4 and proportionally convert all values ​​to values ​​between [0,1] or [-1,1]. The normalization formula is shown in Equation 1:

[0129]

[0130] In the formula, x′ represents the normalized parameters, x max The maximum value of the representative parameter is estimated. The minimum values ​​of the representative parameters are shown in Table 4. Based on the actual operating conditions of the CFM56-7B engine and combined with the principles of mathematical statistics, the preset values ​​of each parameter in Table 4 are shown in Table 5.

[0131] Table 5. Parameters selected for EGTM calculation

[0132]

[0133] The parameters are normalized according to Equations 1 to 5 and Table 5. The normalization formulas and methods for each parameter are shown in Equations 2 to 9.

[0134]

[0135]

[0136]

[0137] In addition, when the required baseline values ​​for the aero-engine performance parameters are EGT baseline values, FF baseline values, or N2 baseline values, the types of calculation parameters selected include external atmospheric environment parameters and engine operating parameters.

[0138] Specifically, baseline calculation requires cruise parameters, which must be read from real-time air-to-ground data link cruise messages. The structures and parameter types of the 816 and 916 cruise messages are largely the same, but there are minor differences. Not all parameters need to participate in the baseline calculation. A parameter filter is used to filter the baseline's influencing factors from numerous parameters, eliminating interfering parameters. Passing the filtered parameters to the baseline calculation model yields a more accurate baseline value. After filtering, it was found that the baselines of the three cruise parameters EGT, FF, and N2 are all determined by four common parameters, as shown in Table 6.

[0139] Table 6 Baseline Impact Factors

[0140]

[0141] Because the four influencing parameters of the baseline have inconsistent units and significantly different orders of magnitude, directly inputting the original values ​​of these four parameters into the Extreme Learning Machine (ELM) computational model would greatly affect the accuracy of the baseline calculation. Before inputting the parameters into the ELM, it is necessary to remove the units of each parameter and convert their values ​​to values ​​between 0 and 1. This is more conducive to improving the accuracy of the baseline calculation.

[0142] The parameters in Table 6 are normalized, and the formulas for normalizing each parameter are shown in Equations 10 to 13.

[0143]

[0144]

[0145] Step 300, Sample Splitting:

[0146] For each type of calculation parameter, a specific group of data is randomly selected from the database. The specific group of data is then randomly divided into four types of samples: genetic training samples, neural network training samples, test samples, and calculation samples. The first N-1 types of parameters for each type of sample are stored in the first matrix, and the remaining type of parameters are stored in the second matrix.

[0147] Specifically, when the cruise parameter to be obtained is EGTM, according to Table 5, the value of N is 12, then the first matrix is ​​as shown in Equation 14, and the second matrix is ​​as shown in Equation 15.

[0148]

[0149] Specifically, when it is necessary to obtain the baseline value, according to Table 6, the value of N is 4, then the first matrix is ​​as shown in Equation 16, and the second matrix is ​​as shown in Equation 17.

[0150]

[0151] Step 400, Genetic Algorithm Training:

[0152] Set up a temporary ELM (Extreme Learning Machine) for genetic training, which includes an input layer with N-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. The genetic training samples consist of u sets of data. The first matrix corresponding to the data is placed at the input of the temporary ELM, and the second matrix corresponding to the data is placed at the output of the temporary ELM. After determining the parameters of the adaptive GA (Genetic Algorithm) and performing adaptive GA training, a specific number of individuals with d genes are randomly generated.

[0153] Specifically, the number of individuals generated is 100, and the formula for calculating d is shown in Equation 18:

[0154] d = NUM input +NUM input ×NUM hid +NUM hid +NUM hid ×NUM output +NUM output ……(18);

[0155] In the formula, NUM input NUM represents the number of neurons in the temporary ELM input layer. hid NUM represents the number of neurons in the temporary ELM hidden layer. output This represents the number of neurons in the temporary ELM output layer.

[0156] Step 500, Individual Screening:

[0157] Multiple sets of different test samples are input into the temporary ELM to repeatedly perform adaptive GA training, and the optimal individual corresponding to the optimal fitness value is obtained.

[0158] Optionally, such as Figure 7 As shown, step 500 also includes the following steps.

[0159] Step 501, Obtaining fitness value:

[0160] The test sample is a dataset with a sample size of v randomly selected from the database. The first matrix corresponding to each sample is placed at the input of the temporary ELM, and the second matrix corresponding to each sample is placed at the output of the temporary ELM. The fitness value for each individual is obtained by means of the following formula:

[0161]

[0162] In the formula, fit(x) i ) represents the fitness value of the i-th individual.

[0163] Specifically, the sample size v is 30.

[0164] Step 502, Screening of parent and child individuals:

[0165] The sequential selection method is used to select parent and mother individuals based on their probability of being selected.

[0166] Specifically, in traditional genetic algorithms, the probability of an individual being selected is directly determined by its fitness value. The formula for calculating the selection probability in this traditional genetic algorithm is shown in Equation 19:

[0167]

[0168] In the formula, P i Let x be the probability that the i-th individual is selected. i Let be the fitness value of the i-th individual, and n be the total number of individuals. From Equation 5.4.1.3-1, if the fitness value of the i-th individual is 0, then P... i If the value is 0, the individual will be permanently eliminated and will not be able to produce offspring. This will affect the diversity of individuals in the algorithm and lead to the phenomenon of "precocious convergence".

[0169] To avoid this phenomenon, the genetic algorithm training in this embodiment proposes a sequential selection strategy, also known as the sequential selection method. It first sorts individuals in descending order according to their fitness values, and then defines the selection probability of the individual with the highest fitness value as P. max Then the probability of the i-th individual making a choice is P. i The calculation formula is shown in Equation 20 below:

[0170]

[0171] In the formula, n represents the total number of individuals. The P value shown in Equation 20 is adopted. i The calculation formula effectively ensures "gene" diversity and avoids premature convergence. Premature convergence in the GA algorithm refers to the algorithm converging too early during iterative optimization, even though the converged value is not the final optimal solution. Premature convergence renders subsequent iterations ineffective, wasting time without improving accuracy. However, by employing the sequential selection strategy shown in Equation 20, individuals with low fitness or zero fitness are prevented from being "eliminated," giving them a chance to iterate and avoiding premature convergence.

[0172] Step 503, Genetic Training:

[0173] The parent and mother individuals are trained using adaptive GA to obtain crossover and mutation probabilities. Then, crossover and mutation operations are performed on the individuals in the adaptive GA to generate a new generation of individuals and calculate their fitness values.

[0174] Specifically, the crossover probability P c and the probability of mutation P m It has a direct impact on the optimization process of genetic algorithms. For example, P c The larger the fitness value, the faster new individuals are generated; however, individuals with high fitness values ​​are also more likely to be destroyed. Conversely, P... c If the value is too small, the search speed will be slow, thus losing the ability to find the optimal solution. m With the same effect, if the value is too small, it will be difficult to generate new individuals; if it is too large, the genetic algorithm will become a random search algorithm.

[0175] For different optimization problems, repeated trials are needed to determine the specific P. c and P m Value, and the optimal P c and P m The value of P is usually not easy to obtain. And in order to make P c and P m As the fitness value changes, an adaptive formula is proposed, as shown in Equations 21 and 22.

[0176]

[0177]

[0178] In Equations 21 and 22, fit max The maximum fitness value for all; fit avg The average fitness value for all individuals; fit c The fitness value is the larger of the two individuals in the crossover; m Let c1, c2, c3, and c4 be the fitness values ​​of the mutated individuals; c1, c2, c3, and c4 are all constants. If the algorithm gets stuck in a local optimum, the fitness of individuals in the population tends to be consistent. P is increased by equations 21 and 22. c and P m This leads to the generation of more new individuals. Similarly, when the fitness of the group is relatively dispersed, it causes P to... c and P m Reducing the value of P improves search accuracy. After improving the crossover and mutation probabilities using formulas 21 and 22, it is no longer necessary to calculate P separately. c and P m The value of P, but rather makes P c and P m It improves itself as it iterates, constantly adjusting its fitness to enhance the search accuracy of the algorithm before it gets trapped in local optima, and increases gene diversity to escape local optima when it is about to get trapped in them.

[0179] Step 504, Fitness Optimization Determination:

[0180] Determine whether to terminate the optimization process based on the judgment criteria. If the judgment criteria are met, proceed to step 505; otherwise, repeat steps 501 to 504.

[0181] The judgment criteria are whether the fitness value reaches a specified value, or whether the number of adaptive GA training sessions reaches a specified upper limit.

[0182] Step 505, Optimal individual value output:

[0183] Output the optimal individual corresponding to the optimal fitness value.

[0184] Step 600, Neural Network Training:

[0185] Define a new ELM for neural network training, whose input layer, intermediate layer, and output layer all have the same number of neurons as the temporary ELM. Assign the optimal weights and optimal thresholds corresponding to the best individual to the new ELM. The number of training samples for the neural network is w sets of data. The first matrix corresponding to the data is placed at the input of the new ELM, and the second matrix corresponding to the data is placed at the output of the new ELM. Perform neural network training, and obtain an adaptive GA-ELM approximator after training.

[0186] Specifically, the sample size w is 5000. The criteria for determining whether the neural network training can be completed are whether the given upper limit of the number of iterations has been reached, or whether the average error of the engine performance parameters obtained in step 800, compared with the EGTM value calculated by the manufacturer's monitoring system, is within a specified range.

[0187] Step 700, Parameter Calculation:

[0188] The number of samples for the calculation sample is t sets of data. The first matrix corresponding to the data is placed at the input end of the adaptive GA-ELM approximator, and the second matrix corresponding to the data is placed at the output end of the adaptive GA-ELM approximator. The adaptive GA-ELM approximator is run, and the training results are stored in the target matrix. The dimension of the target matrix is ​​the same as the dimension of the second matrix.

[0189] Specifically, the sample size t is 1000.

[0190] Step 800, Result Acquisition:

[0191] The results are obtained by performing inverse normalization on the target matrix to obtain the engine performance parameters.

[0192] Specifically, the inverse normalization formula is shown in Equation 23:

[0193] x=x′(x max -x min )+x min ……(twenty three);

[0194] x is stored in the parameter matrix and used to compare it with the EGTM value or baseline value calculated by the manufacturer's monitoring system to calculate the average error β, which is shown in Equation 24:

[0195]

[0196] For example, for ETGM, if the value of β is less than 1°C, it is considered that the required accuracy for monitoring is basically met; if the value of β is greater than 1°C, steps 200 to 800 need to be repeated. Figure 8 As shown, this continues until the value of β is less than 1℃.

[0197] Example 2

[0198] Embodiment 2 of this application provides a method for obtaining sea level atmospheric temperature limits. This method involves selecting a large amount of actual takeoff data from the database and then fitting the actual takeoff data with the EGTM using a univariate nonlinear regression. The EGTM value can be obtained using Embodiment 1.

[0199] Specifically, the airport's outside temperature (Airport OAT) is converted to sea level temperature. If the converted sea level temperature (SLOAT) is higher than the sea level temperature (SLOATL), then rated thrust cannot be used for takeoff; otherwise, the engine thrust (EGT) will exceed the redline, causing engine damage. The temperature conversion formula is shown in Equation 25.

[0200] SLOAT=Airport OAT+0.065Airport ALT……(25);

[0201] In the formula, Airport ALT represents the altitude of the departure airport, in meters (m). Using a large amount of actual takeoff data, a univariate nonlinear regression method is employed to fit SLOATL. The value of SLOATL is only affected by one parameter, EGTM. Since the SLOATL function is a univariate function, a univariate nonlinear regression method is used to obtain the fitting formula. The fitting results are as follows: Figure 9 As shown.

[0202] A good fit was achieved using a logarithmic function, and the fitting formula is shown in Equation 26:

[0203] R 2 =0.973

[0204] SLOATL=8.4808(EGTM)+10.458……(26);

[0205] In Equation 26, R 2This represents the residual; the closer this value is to 1, the more accurate the fit. Using univariate quadratic polynomial regression yields better fitting results, with residual R0... 2 =1 indicates that a perfect fit has been achieved, as shown in Equation 27:

[0206] R 2 =1

[0207] SLOATL = 0.0003EGTM 2 +0.2944MEGT+30.006……(27).

[0208] Example 3

[0209] Embodiment 3 of this application provides a method for obtaining a baseline value of lubricating oil pressure, which utilizes N2 and performs logarithmic regression fitting to obtain the value. The value of N2 can be obtained using Embodiment 1.

[0210] Specifically, the calculation of the OP Baseline differs from the Baselines of other parameters in that the value of the OP Baseline is not affected by parameters such as altitude, total temperature, low-pressure rotor speed, and Mach number. According to engine principles and the data provided by the manufacturer, the baseline value of lubricating oil pressure OP is only affected by N2, and OP and N2 basically satisfy a linear correlation.

[0211] An independent computational model was established for the OP Baseline. Linear regression was used to obtain a linear formula, but its accuracy was found to need improvement. Using the logarithmic function in nonlinear regression, the resulting formula is shown in Equation 26.

[0212] R 2 =0.9977

[0213] OP baseline=97.746ln(N2)-379.7……(26);

[0214] In the formula, OP baseline is the baseline value of lubricating oil pressure, and R 2 R represents the residual between the values ​​calculated by the fitted formula and the manufacturer's calculations in the sample. 2 The closer the value is to 1, the smaller the error. This formula R... 2 =0.9977, indicating an almost perfect fit. Verified using actual operating parameters and formula 26, the maximum relative error among 350 test samples was 0.28%, which meets the requirements for engine performance monitoring.

[0215] Since the OP baseline and N2 essentially satisfy a logarithmic function relationship, they are plotted together in a single coordinate system, with N2 as the independent variable x and the OP baseline as the dependent variable y. The graph is shown below. Figure 10As shown.

[0216] Example 4

[0217] The parameters such as EGT Baseline, FF Baseline, N2 Baseline, and OIP Baseline obtained in Example 3 cannot be used as a direct basis for evaluating engine performance status in actual engine monitoring. This is because the Baseline only serves as an important intermediate parameter. The Baseline characterizes the relationship between the performance parameters and operating condition parameters of the same engine configuration after a new factory release or major overhaul, and is a reference benchmark for engine cruise parameters.

[0218] What needs close attention in daily operations is not the baseline values ​​of each parameter, but rather the deviation values ​​calculated based on the baselines, namely ΔEGT, ΔFF, ΔN2, and ΔOIP. Different engine manufacturers provide reference formulas for these deviation values ​​for different engine models, but these formulas have some problems:

[0219] (1) Insufficient universality. Different models require different calculation formulas. Even when the engine model changes, the original calculation formula will become invalid.

[0220] (2) Insufficient accuracy. Foreign manufacturers have always regarded the calculations related to monitoring parameters as trade secrets, and the calculation formulas they provide are only theoretical formulas for reference and often do not have practical application value.

[0221] For the reasons mentioned above, Embodiment 4 of this application provides a method for obtaining parameter deviation values, such as... Figure 11 As shown, it includes the following steps:

[0222] Step 10, Parameter Filtering:

[0223] Sixteen types of input parameters for calculating parameter deviation values ​​were selected from the database.

[0224] Specifically, the input parameters required for calculating the parameter deviation value are shown in Table 7.

[0225] Table 7 Input parameters for calculating the deviation of cruise parameters

[0226]

[0227]

[0228] Step 20, obtain the original result:

[0229] The original parameter deviation values ​​are obtained by referring to the parameter deviation values ​​provided by the aero-engine manufacturer, and used as the calculation output parameters.

[0230] Specifically, the calculation output parameters required for calculating the parameter deviation value are shown in Table 8.

[0231]

[0232] Step 30, Neurotraining:

[0233] A generalized regression neural network is established. The computational parameters are put into the input layer of the generalized regression neural network, and the deviation values ​​of the traditional parameters are put into the output layer of the generalized regression neural network. Then, the neural network is trained to obtain a generalized regression neural network approximator.

[0234] Step 40, Parameter Acquisition:

[0235] Specifically, 16 types of calculation input parameters are randomly selected for calculating parameter deviation values, and then fed into the generalized regression neural network approximator and run to obtain parameter deviation values ​​that can be used for real-time monitoring of the engine performance status.

[0236] Example 5

[0237] This application provides a system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, such as... Figure 12 As shown, the system for acquiring aero-engine performance parameters and their baseline values ​​includes a message acquisition module 1, a database 2, a parameter filter 3, a normalization processor 4, a sample division module 5, a temporary ELM 6, an individual screening module 7, an adaptive GA-ELM approximator 8, a parameter output module 9, a data display module 10, and a backup processing module 11.

[0238] The message acquisition module 1 is used to receive and download real-time messages from the engine during takeoff and cruise, and to identify and determine the engine configuration and its corresponding message type. For example, Figure 11 As shown, the message acquisition module 1 also includes a message download submodule 110, a message classification submodule 120, and a message transfer submodule 130.

[0239] Specifically, the message download submodule 110 is used to execute step 101 in Embodiment 1, the message classification submodule 120 is used to execute step 102 in Embodiment 1, and the message transfer submodule 130 is used to execute step 103 in Embodiment 1.

[0240] Database 2 is used to store the classified real-time messages. Parameter filter 3 filters out the corresponding n types of calculation parameters based on the type of aero-engine performance parameter to be calculated or its baseline value type. Normalization processor 4 is used to normalize the calculation parameters. Parameter filter 3 and normalization processor 4 work together to execute step 200 in Embodiment 1.

[0241] The sample partitioning module 5 is used to randomly select a specific group of data from the database 2 for each type of calculation parameter, and randomly divide the specific group of data into four types of samples in an unequal manner. It also stores the first n-1 types of parameters of each type of sample in a first matrix and the remaining type of parameters in a second matrix. The four types of samples are genetic training samples, neural network training samples, test samples, and calculation samples. The sample partitioning module 5 is mainly used to execute step 300 in Embodiment 1.

[0242] The temporary ELM6 includes an input layer with n-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. It receives u sets of genetic training data, with a first matrix corresponding to each set placed at the input of the temporary ELM6 and a second matrix corresponding to each set at the output. After determining the parameters of the adaptive GA and performing adaptive GA training, it randomly generates a specific number of individuals with d genes. The temporary ELM6 is mainly used to execute step 400 in Example 1.

[0243] Individual selection module 7 is used to input multiple different test samples into temporary ELM6 for repeated adaptive GA training to obtain the optimal individual corresponding to the optimal fitness value. Individual selection module 7 is mainly used to execute step 500 in Example 1.

[0244] The adaptive GA-ELM approximator 8 is obtained from step 600 in Embodiment 1. It is used to receive calculation samples with a sample size of t sets of data, train them to obtain training results, and store them in the target matrix.

[0245] The parameter output module 9 is used to perform inverse normalization processing on the target matrix and output the aero-engine performance parameters or their reference values.

[0246] Because the data storage module is installed on a shared server, it is very difficult for users to access the data, and directly reading data from database 2 is also challenging. Therefore, the data display module has the following functions:

[0247] (1) Users can connect to the shared server using their personal computers and retrieve the parameters they need from the data storage module;

[0248] (2) The call response speed is fast, and the data in the real-time air-to-ground data link message can be viewed in real time;

[0249] (3) Users can freely choose the name of the data they want to view and freely choose the time period for viewing the data;

[0250] (4) After the user selects data, the display module can display the data in the form of images in chronological order;

[0251] (5) After the user selects the parameters and time period to be viewed, the selected data can be output in the form of a spreadsheet through the data display module 10, which makes it convenient for the user to analyze the data.

[0252] The data display module 10 is used to display the data in the database 2, including the aero-engine performance parameters or their benchmark values.

[0253] Because spatiotemporal data link messages are often lost due to network interruptions, server crashes, and other factors, and once real-time spatiotemporal data link messages are lost, the missing data cannot be recovered, this greatly affects monitoring work. Engine monitoring software based on real-time spatiotemporal data link messages requires manual input of data through a GUI interface when such messages are missing. This method is inefficient, significantly increases the user's workload, and severely impacts the real-time performance of engine monitoring.

[0254] To address the aforementioned issues, the aero-engine performance parameters and their baseline values ​​acquisition system in Embodiment 5 of this application further includes a backup processing module 11, used to detect and complete any missing real-time air-to-ground data link messages.

[0255] This backup processing module 11 has the following functions:

[0256] (1) When real-time air-to-ground data link data is found to be missing, first locate the missing aircraft number, the time range of the missing data, the flight number, engine type, and the missing message type (take-off message / cruise message);

[0257] (2) Based on the relevant information in (1), download the corresponding offline formatted data message. Taking the CFM56-7B engine as an example, the message is divided into four categories: 816 cruise message, 816 takeoff message, 916 cruise message, and 916 takeoff message.

[0258] (3) Based on the different specific content of the messages in (2), automatically classify the messages, read the relevant data and perform calculations;

[0259] (4) Write the reading results and calculation results into database 2;

[0260] (5) When calling data from database 2, relevant monitoring information is obtained by filtering information such as aircraft number, flight time, flight mode, and parameter type.

[0261] Example 6

[0262] Based on Example 5, such as Figure 13As shown, the system for obtaining the performance parameters and baseline values ​​of the aero-engine also includes a sea-level atmospheric temperature limit calculation module, which is used to execute Embodiment 2; it also includes a lubricating oil baseline value acquisition module 13, which is used to execute Embodiment 3; and it also includes a parameter deviation value acquisition module 14, which is used to execute Embodiment 4.

[0263] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-graded storage medium, such as a read-only memory, a disk, or an optical disk.

[0264] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, characterized in that, Includes the following steps: Message acquisition: Receive and download real-time messages from the engine during takeoff and cruise, identify and determine the engine configuration and its corresponding message type, and then store the data in the real-time messages into the database. Parameter filtering involves selecting N types of calculation parameters from the database for calculating the corresponding aero-engine performance parameters or their baseline values, and then normalizing them. For each type of calculation parameter, a specific group of data is randomly selected from the database and the specific group of data is randomly divided into four types of samples in an unequal manner: genetic training samples, neural network training samples, test samples, and calculation samples. The first N-1 types of parameters of each type of sample are stored in the first matrix, and the remaining type of parameters are stored in the second matrix. Genetic algorithm training involves setting up a temporary ELM for genetic training, which includes an input layer with N-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. The genetic training samples consist of u sets of data. The first matrix corresponding to the data is placed at the input of the temporary ELM, and the second matrix corresponding to the data is placed at the output of the temporary ELM. After determining the parameters of the adaptive GA and performing adaptive GA training, a specific number of individuals with d genes are randomly generated. Individual selection involves inputting multiple different test samples into a temporary ELM and repeatedly performing adaptive GA training to obtain the optimal individual corresponding to the optimal fitness value. Neural network training involves defining a new ELM for neural network training, where the number of neurons in its input, intermediate, and output layers is equal to that of the temporary ELM. The optimal weights and thresholds corresponding to the best individual are assigned to the new ELM. The number of training samples for the neural network is w sets of data. The first matrix corresponding to the data is placed at the input of the new ELM, and the second matrix corresponding to the data is placed at the output of the new ELM. Neural network training is performed, and an adaptive GA-ELM approximator is obtained after completion. Parameter calculation: The sample size of the calculation sample is t sets of data. The first matrix corresponding to it is placed at the input end of the adaptive GA-ELM approximator, and the second matrix corresponding to it is placed at the output end of the adaptive GA-ELM approximator. The adaptive GA-ELM approximator is run, and the training results are stored in the target matrix. The dimension of the target matrix is ​​the same as the dimension of the second matrix. The results are obtained by performing inverse normalization on the target matrix to obtain the engine performance parameters.

2. The method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 1, characterized in that, The performance parameters are the actual values ​​of EGTM and N2, and the baseline values ​​are the EGT baseline value, FF baseline value, and N2 baseline value.

3. The method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 1 or 2, characterized in that, The individual screening step also includes the following steps: Step 1: Fitness value acquisition. The test sample randomly selects data of sample size v from the database. The first matrix corresponding to the test sample is placed at the input of the temporary ELM, and the second matrix corresponding to the test sample is placed at the output of the temporary ELM to obtain the fitness value corresponding to each individual. Step 2: Select parent and mother individuals using a sequential selection method based on their probability of being selected. Step 3, genetic training: Adaptive GA training is performed on the parent and mother individuals to obtain crossover and mutation probabilities. Then, crossover and mutation operations are performed on the individuals in the adaptive GA to generate a new generation of individuals and calculate their fitness values. Step 4: Fitness optimization determination. Determine whether to terminate the optimization process based on the determination criteria. If the determination criteria are met, proceed to Step 5; otherwise, repeat Steps 1 to 4. The determination criteria are whether the fitness value reaches a specified value or whether the number of adaptive GA training sessions reaches a specified upper limit. Step 5: Individual numerical output. Output the optimal individual corresponding to the optimal fitness value.

4. The method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 3, characterized in that, In the parameter selection step, when the required engine performance parameter is ETGM, the selected parameter types include external atmospheric environment parameters, engine operating parameters, engine configuration parameters, and aircraft operating parameters.

5. The method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 3, characterized in that, In the parameter selection step, when the required baseline value for the aero-engine performance parameter is an EGT baseline value, an FF baseline value, or an N2 baseline value, the selected parameter types include external atmospheric environment parameters and engine operating parameters.

6. The method for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 3, characterized in that, The message acquisition step also includes the following steps: Message download: After collecting operational data from various operating conditions, the airborne equipment of the aero-engine is filtered to form real-time air-to-ground data link messages, which are then transmitted to the ground receiving station in real time via the real-time air-to-ground data link. Message classification is based on the characteristic data of the real-time air-to-ground data link messages, classifying them into messages under different configurations and different operating states, as well as messages with incorrect formats. The message is transferred and stored. Based on the type of the real-time air-to-ground data link message, all its information is extracted into a string, and the string is converted into string type, time type, floating point type, integer type, or boolean type data according to the data type. The converted data is then stored in the database.

7. A method for obtaining the sea-level atmospheric temperature limit, characterized in that, It is obtained by selecting a large amount of actual takeoff data from the database and then fitting the actual takeoff data with the EGTM described in any one of claims 2-6 using univariate nonlinear regression.

8. A method for obtaining a baseline value of lubricating oil pressure, characterized in that, It is obtained by using N2 as described in any one of claims 2-6 and fitting it with logarithmic regression.

9. A method for obtaining the parameter deviation value of a baseline value, characterized in that, The baseline value is obtained based on any one of claims 2-6, including the following steps: Parameter filtering involves selecting m types of computational input parameters from the database for calculating parameter deviation values. Obtain the original results by using the parameter deviation value reference formula provided by the aero-engine manufacturer to obtain the original parameter deviation value, which is then used as the calculation output parameter; Neural training is performed to establish a generalized regression neural network. The computational parameters are put into the input layer of the generalized regression neural network, and the traditional parameter deviation values ​​are put into the output layer of the generalized regression neural network. Then, the neural network is trained to obtain a generalized regression neural network approximator. The parameters are acquired, and then m types of calculation parameters for calculating parameter deviation values ​​are randomly selected, put into the generalized regression neural network approximator and run to obtain parameter deviation values ​​that can be used for real-time monitoring of the engine performance status.

10. A system for acquiring aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM, characterized in that, include: The message acquisition module is used to receive and download real-time messages from the engine during takeoff and cruise, and to identify and determine the engine configuration and its corresponding message type. A database is used to store the data extracted from the classified real-time messages, as well as the acquired aero-engine performance parameters and their baseline values. The parameter filter selects the corresponding n types of calculation parameters based on the type of flight performance parameter to be calculated or the type of its baseline value. A normalization processor is used to normalize the calculation parameters; The sample partitioning module is used to randomly select a specific group of data from the database for each type of calculation parameter, and randomly divide the specific group of data into four types of samples in an unequal manner. The first n-1 types of parameters of each type of sample are stored in a first matrix, and the remaining type of parameters are stored in a second matrix. The four types of samples are genetic training samples, neural network training samples, test samples, and calculation samples. A temporary ELM includes an input layer with n-1 neurons, d intermediate layers for internal nonlinear summation calculations, and an output layer with 1 neuron. It receives a sample size of u sets of genetic training data. The first matrix corresponding to the data is placed at the input of the temporary ELM, and the second matrix corresponding to the data is placed at the output of the temporary ELM. After determining the parameters of the adaptive GA and performing adaptive GA training, a specific number of individuals with d genes are randomly generated. The individual selection module is used to input multiple different test samples into the temporary ELM and repeatedly perform adaptive GA training in order to obtain the optimal individual corresponding to the optimal fitness value. An adaptive GA-ELM approximator is obtained by training a neural network using a new ELM whose input, intermediate, and output layers have the same number of neurons as the temporary ELM. This ELM is assigned the optimal weights and thresholds corresponding to the optimal individual and trained on a neural network with w sets of training samples. The adaptive GA-ELM approximator is used to receive t sets of computational samples, train the approximator to obtain the training results, and store them in the target matrix. The parameter output module is used to perform inverse normalization on the target matrix and output the aero-engine performance parameters or their reference values ​​to be stored in the database. The data display module is used to display the aero-engine performance parameters and / or their benchmark values ​​in the database.

11. The system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 10, characterized in that, The message acquisition module also includes the following sub-modules: The message download submodule is used to collect the operating data of the aero-engine airborne equipment under various operating conditions, filter it to form real-time air-to-ground data link messages, and transmit them to the ground receiving station in real time via the real-time air-to-ground data link. The message classification submodule is used to classify the real-time air-to-ground data link messages into messages with different configurations and different operating states, as well as messages with incorrect formats, based on their characteristic data. The message transfer submodule is used to extract all information of the real-time air-to-ground data link message into a string according to its type, and convert the string into string type, time type, floating point type, integer type, or boolean type data according to the data type, and then store the converted data into the database.

12. The system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 10, characterized in that, It also includes a sea level atmospheric temperature limit acquisition module, which selects a large amount of actual takeoff data from the database and uses the actual takeoff data and EGTM to perform a univariate nonlinear regression to fit the data and obtain the sea level atmospheric temperature limit.

13. The system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 10, characterized in that, It also includes a lubricating oil baseline value acquisition module, which obtains the lubricating oil baseline value by using N2 and fitting with logarithmic regression.

14. The system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to claim 10, characterized in that, It also includes a parameter deviation value acquisition module, which includes a generalized regression neural network approximator. This approximator receives and runs m types of calculation parameters selected by the parameter filter for calculating parameter deviation values, and outputs parameter deviation values ​​that can be used to monitor the performance status of the aero-engine in real time.

15. The system for obtaining aero-engine performance parameters and their baseline values ​​based on adaptive GA-ELM according to any one of claims 10-14, characterized in that, It also includes a backup processing module, which is used to read formatted batch data from the database, find and complete the missing data in the real-time air-to-ground data link message.

Images