Aero-engine and propeller full working condition real-time matching method and system

Abstract

The application discloses an aero-engine and propeller full-working-condition real-time matching method and system, which dynamically calculates optimal control instructions (including engine rotating speed and throttle opening) according to real-time changing flight states (including flight height, flight speed and power), and immediately executes the control instructions. The application creatively discovers and applies an extremely simplified three-tree random forest model, combines a unique power pre-calculation and online interpolation strategy, and for the first time realizes high-precision and high-real-time optimal matching control of an aero-propeller system under full working conditions on a low-cost and low-power airborne ECU, and achieves the technical effect of significantly improving fuel economy.

Description

Technical Field

[0001] This invention relates to the field of aero-engine control technology, and particularly to the control system therein and its electronic digital data processing, specifically to a method and system for real-time matching of aero-engine and propeller under all operating conditions. Background Technology

[0002] The propulsion system, consisting of an aero-engine and a propeller (referred to as "propeller-engine" for convenience in this invention), is the main power source for medium-altitude long-endurance unmanned aerial vehicles (UAVs). Its core performance lies in achieving efficient matching between engine output power and propeller absorbed power under all flight conditions (including takeoff, climb, and cruise). However, there is a strong coupling relationship between the characteristics of the engine and the propeller, and both are highly sensitive to parameters such as flight altitude, speed, and power requirements, resulting in complex nonlinear characteristics of the entire propulsion system.

[0003] Currently, traditional propeller-engine matching methods have the following limitations:

[0004] 1) Limitations of operating conditions: Existing research focuses on specific flight conditions (such as cruise at fixed altitude and speed), only considering engine performance characteristics or propeller aerodynamic performance characteristics, and there is insufficient research on the matching of large-scale full-condition operations from sea level to high altitude and from low speed to high speed, which makes it difficult to meet the needs of complex flight missions of UAVs.

[0005] 2) Limitations of the method: The lookup table method, which is widely used in engineering, is simple to calculate, but it requires storing massive amounts of data to cover all operating conditions, which places a huge demand on the storage resources of the airborne engine control unit (ECU). At the same time, for points not directly stored in the table, the interpolation accuracy is limited, making it difficult to accurately describe the nonlinear characteristics of the system, resulting in a trade-off between control accuracy and computational efficiency.

[0006] 3) Model limitations: In recent years, although some studies have introduced data-driven methods (such as CNN, LSTM and other algorithm models) into the field of aero-engines, these models are usually complex in structure and have a large amount of computation. They have high hardware requirements during training and prediction and are difficult to run in real time in resource-constrained airborne ECUs.

[0007] Chinese patent CN 114357089B discloses an evaluation method for a propeller and piston engine matching scheme, including the following steps: characteristic curves of the propeller and piston engine to be evaluated; determining the operating point parameters of the UAV; calculating the ideal value W1 and the actual value W2 of constant thrust fuel consumption; calculating the ideal thrust value T1 and the actual thrust value T2; obtaining the ideal value P of the engine's maximum available power. max1 And the actual value of the engine's maximum available power Pmax2 ; Receive the matching score.

[0008] However, this technology is only an offline evaluation method used in the UAV design phase to statically and retrospectively assess the compatibility between a particular engine and a particular propeller, providing a matching score. Essentially, it is a design tool used to assist in selection; it does not participate in the real-time control of the aircraft and cannot run on the ECU.

[0009] Chinese patent document CN 116931434A discloses a modeling method for matching control parameters of an unmanned aerial vehicle (UAV), engine, and propeller, including: Step 1, calculating propeller performance; Step 2, calculating engine high-altitude power characteristics; Step 3, calculating the additional shaft power requirement extracted from the engine by the aircraft equipment; Step 4, simulating the thrust requirement of the UAV at various flight altitudes and level flight speeds; Step 5, establishing a matching control parameter model for the UAV, engine, and propeller; Step 6, correcting the calculated fuselage thrust value, propeller performance data, and engine ground and high-altitude power.

[0010] However, this technology is still a high-precision offline modeling method. Through a large number of experiments in wind tunnels, high-altitude test benches, etc., it aims to establish a physical model that is as accurate as possible to describe the system characteristics. Its essence is a high-fidelity simulation model, mainly used for early simulation and design verification. Its model is complex and computationally intensive, making it difficult to deploy in real time in low-simulation ECUs.

[0011] Therefore, there is an urgent need in this field for a propeller-generator full-condition matching method that can simultaneously meet the requirements of high precision, low complexity, and strong real-time performance. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for real-time matching of aero-engines and propellers under all operating conditions, so as to solve the deficiencies of traditional methods in terms of adaptability under all operating conditions, control accuracy and embedded real-time computing.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] This invention provides a method for real-time matching of aero-engines and propellers under all operating conditions, comprising the following steps:

[0015] S1: With the goal of optimizing propeller thrust and fuel economy respectively, and with power balance and speed balance as constraints, while taking into account the characteristics of both propeller aerodynamic performance and engine performance, and considering the influence of external parameters such as flight altitude, flight speed and aircraft power requirements, a matching model and database covering all operating conditions of takeoff, climb and cruise are constructed.

[0016] S2: Using the database constructed in step S1 as training samples, with flight altitude, flight speed and power as input parameters, and engine throttle opening and engine speed as output parameters, the algorithm model is trained to establish a mapping relationship from input parameters to output parameters, and a training model is obtained.

[0017] S3: The training model obtained in step S2 is deployed in the engine's onboard ECU through the control system. Based on the current flight phase and real-time sensor data, the required engine throttle opening and engine speed are calculated in real time, and control commands are sent for execution.

[0018] Preferably, the algorithm model in step S2 is selected from quadratic formula method, convolutional neural network, long short-term memory network, radial basis function neural network or random forest.

[0019] More preferably, the algorithm model in step S2 is the multi-tree random forest algorithm.

[0020] More preferably, in step S2, the calculation of each output parameter adopts the three-tree random forest algorithm. In each tree, the segmentation condition values ​​of the three parameters, flight altitude, flight speed and power, are first set respectively. Then, according to the relationship between the current value of each parameter and the size range of the corresponding segmentation condition value, the corresponding training branch is selected to perform random forest algorithm prediction to obtain the output parameters. Finally, the output parameters predicted by the three trees are averaged to obtain the predicted final output parameters.

[0021] Preferably, the segmentation condition value is determined according to the minimum MSE segmentation rule shown in the following formula:

[0022] ;

[0023] In the formula, The index representing the feature, i.e. the feature used for splitting; Representation of features The dividing point on; For the first The actual value of each sample; This represents the set of samples contained in the left child node after the split; This represents the set of samples contained in the right child node after the split; The average of the predicted values ​​of all samples within the left child node; The average of the predicted values ​​of all samples within the right-hand child node.

[0024] Preferably, the algorithm model in step S2 has two working modes: a "maximum thrust model" and an "optimal economy model." The former is used to predict the output parameters during takeoff and climb, while the latter is used to predict the output parameters during cruise. The optimal economy model is characterized by a fuel economy index, defined as follows:

[0025] ;

[0026] In the formula, F represents the fuel economy index, η represents the propeller aerodynamic efficiency, and sfc represents the specific fuel consumption rate.

[0027] More preferably, step S2 further includes first setting the change amounts of the three parameters of flight altitude, flight speed and power respectively, taking the current input of the three parameters of flight altitude, flight speed and power as the center, generating an input matrix covering the surrounding working conditions according to their respective change amounts, and pre-computing the corresponding output parameter output matrix based on the input parameter combinations included in the matrix through the three-tree random forest algorithm.

[0028] Most preferably, in step S3, the current flight phase is first determined, and then it is checked whether the new input parameters fall within the range of the pre-calculated input matrix. If they do, the output parameters are directly calculated from the output matrix by linear interpolation and used directly as the final output parameters. If they do not fall within the range, the final output parameters are calculated in real time using the three-tree random forest algorithm.

[0029] Secondly, the present invention also provides a real-time matching system for aero-engines and propellers under all operating conditions, the system being used to execute the above-mentioned real-time matching method under all operating conditions.

[0030] Preferably, the system includes:

[0031] The data acquisition and modeling module is used to acquire test data of the engine and propeller, and to build matching models and databases;

[0032] The algorithm model processing module is used to train, store, and run the algorithm model to obtain the trained model;

[0033] The interpolation module is used to pre-compute the input parameters and store the output matrix, and to perform linear interpolation calculations at runtime.

[0034] The ECU control execution module receives real-time sensor data of the current flight phase and flight status parameters, calls the aforementioned algorithm model processing module and interpolation module, and outputs control commands for the final output parameters to be executed by the engine.

[0035] The present invention has the following beneficial effects:

[0036] The core function of this invention lies in providing an online real-time control method. The goal of this invention is to directly embed it into the engine ECU, dynamically calculating the optimal control commands (including engine speed n and throttle opening ε) within milliseconds based on real-time changes in flight status (including flight altitude H, flight speed V, and power P), and executing them immediately. This invention provides a real-time control system for aircraft, forming part of the aircraft's "brain," aiming to solve the "real-time control" problem. This invention transcends existing technologies that rely on "post-event evaluation" and "offline modeling," providing a technical solution for "online real-time optimal control," resulting in a qualitative leap.

[0037] The solution provided by this invention enables real-time acquisition of high-precision data. Both traditional lookup table methods and the high-precision physical models provided in existing technology CN 116931434A face the impossible triangle of "precision-efficiency-storage": high precision inevitably comes with high complexity, affecting efficiency and storage difficulty, making real-time operation or rapid response impossible. Through extensive simulation comparisons, this invention reveals that in the specific problem of propeller-engine matching, a random forest using only 3 decision trees can reduce model complexity by two orders of magnitude (to about 3% of a 100-tree model) with minimal accuracy loss (R² error <2%). This extremely simplified "3-tree" configuration is the cornerstone of this invention's solution, enabling online deployment and real-time computation.

[0038] This invention preferably employs a random forest model containing only three decision trees, deployed in the ECU of an aero-engine, to map optimal ε and n based on real-time acquired H, V, and P. This invention can train this simplified model to replace traditional lookup table methods or complex physical models. Furthermore, this invention can combine this simplified model with power pre-calculation and linear interpolation to further improve real-time performance.

[0039] The invention demonstrates excellent performance, with an optimal single-match calculation response time as low as 0.015 milliseconds. This response time data proves that the proposed solution meets the millisecond-level real-time requirements of flight control, providing crucial evidence for its transition from theoretical to engineering applications. At an altitude of 5000 meters and a speed range of 140-260 km / h, the maximum thrust mode matching strategy of this invention improves overall fuel economy by 15.29%, while the optimal fuel economy mode matching strategy improves overall fuel economy by 15.52%. In a 28-hour long-endurance mission for a certain type of UAV, the adoption of this invention's technical solution reduced total fuel consumption by 8% (saving 45.7 kg of fuel).

[0040] The model of this invention has high accuracy and low complexity. A horizontal comparison with Quadratic Formula Method (QFM), Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), and Radial Basis Function Network (RBFN) algorithms shows that the 3-Tree-RF model of this invention significantly outperforms other models in fitting accuracy (R²>95%), while having the lowest model complexity. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the propeller-engine matching data model one obtained by the real-time matching method for all operating conditions of the aero-engine and propeller of the present invention.

[0042] Figure 2 This is a schematic diagram of the trained model 2 obtained by the real-time matching method for aero-engines and propellers under all operating conditions of the present invention.

[0043] Figure 3 The results are from the online actual operation test of the real-time matching system for aero-engines and propellers under all operating conditions of this invention.

[0044] Figure 4 This is a simulation comparison diagram of the real-time matching method for aero-engines and propellers under all operating conditions of the present invention and the traditional matching method;

[0045] Figure 5(a) is a comparison of the residuals of engine speed obtained by different algorithms used in the real-time matching method of aero-engine and propeller under all operating conditions of the present invention. Figure 5(b) is a comparison of the residuals of throttle opening obtained by different algorithms used in the real-time matching method of aero-engine and propeller under all operating conditions of the present invention.

[0046] Figure 6 This is a comparison chart of the results between different algorithms used in the real-time matching method for aero-engines and propellers under all operating conditions of this invention. Detailed Implementation

[0047] The following detailed description provides further details through specific embodiments. However, it should be noted that the embodiments described below are merely for illustrating the content of the invention and do not represent that the invention is limited to the described embodiments. Therefore, non-essential improvements and adjustments made to the implementation schemes by those skilled in the art based on the above-described invention still fall within the protection scope of the invention, and the scope of protection of the appended claims shall prevail.

[0048] This invention is applicable to various aircraft or flying vehicles, especially those requiring long-duration flight, such as unmanned aerial vehicles (UAVs) for extended medium-altitude flight. The aircraft engine is the "heart" that provides propulsion and support for the aircraft / flying vehicle. Major types include aircraft piston engines and aircraft turbine engines. This invention is particularly applicable to aircraft piston engines with propellers, where the propeller is the core component that converts engine power from rotational force into thrust / pull.

[0049] In this invention, "operating condition" broadly refers to the combination of all external environment and internal system states experienced by an aircraft / aircraft during its mission cycle, including atmospheric conditions, flight status, power / energy requirements, onboard system loads, and other multi-dimensional parameters. In this invention, operating condition specifically refers to the flight phases of an aircraft / aircraft—takeoff, climb, and cruise—and the combination of its flight status parameters (altitude H, flight speed V, and power P).

[0050] The propeller thrust T referred to in this invention is the component of the resultant force in the flight direction of the combined force of the propeller blades continuously pushing against the air and the air's reaction force on the blades; it is the force that allows the aircraft / aircraft to move forward / hover. Propeller thrust optimization refers to maximizing the propeller thrust T to its maximum value. max This constitutes a matching pattern target for the takeoff and climb phases. Propeller thrust T and its maximum thrust T0 max All numerical values ​​can be obtained through the propeller-engine matching model in stage S1. This model is based on the engine performance characteristics and propeller aerodynamic characteristics, with power and speed balance between the engine and propeller as constraints, flight altitude, flight speed, and aircraft power requirements as input conditions, and propeller thrust optimization as the calculation objective, outputting the maximum propeller thrust.

[0051] The fuel economy referred to in this invention refers to the economic index or economic cost of the combined fuel consumption of the engine and propeller. This value is obtained through the propeller-engine matching model in stage S1. That is, based on the engine performance characteristics and propeller aerodynamic performance characteristics, with the power balance and speed balance of the engine and propeller as constraints, and the flight altitude, flight speed and aircraft power requirements as input conditions, the optimal fuel economy calculation objective is to output the maximum fuel economy. In this invention, the ratio of propeller aerodynamic efficiency η to engine specific fuel consumption rate sfc is used as the fuel economy index F, which is calculated according to the following formula (1):

[0052] (1)

[0053] In formula (1), F represents the fuel economy index (calculated value; unit (kW·h) / g), η represents the propeller aerodynamic efficiency (dimensionless), and sfc represents the specific fuel consumption rate (obtained from the database of Model 1, unit g / (kW·h)). η is calculated according to the following formula (2):

[0054] (2)

[0055] In formula (2), C T J represents the propeller thrust coefficient (calculated according to formula (3)), J represents the propeller advance ratio (calculated according to formula (4)), and C represents the propeller thrust coefficient (calculated according to formula (4)). P Represents the propeller power coefficient (calculated according to formula (5)):

[0056] (3)

[0057] (4)

[0058] (5)

[0059] In formula (3), T represents propeller thrust (obtained from the database of Model 1, unit N), and ρ is air density (air density at the current flight altitude, unit kg / m³). 3 n1 is the propeller speed (obtained from the database of Model 1, in rps). This refers to the propeller diameter (a fixed value corresponding to a specific propeller specification, in meters).

[0060] In formula (4), V is the flight speed (input parameter, unit m / s), n1 is the propeller speed (obtained from the database of Model 1, unit rps), and D is the propeller diameter (a fixed value corresponding to the specific propeller specification, unit m).

[0061] In formula (5), N e Power absorbed by the propeller (ideally equal to the engine output power P) e Consistent, this output power can be obtained from the database of Model 1, divided by 1000 (representing the unit adjustment from kW to W), ρ is the air density (air density at the current flight altitude, unit kg / m³). 3 n1 is the propeller speed (obtained from the database of Model 1, in rps), and D is the propeller diameter (a fixed value corresponding to the specific propeller specification, in meters).

[0062] During takeoff and climb, the power system employs a maximum thrust matching strategy; during cruise, it switches to an optimal economy matching strategy. This process allows for simultaneous and rapid adjustment to the optimal required engine speed and throttle opening (the mode of this invention). In contrast, the traditional single-path matching mode of throttle opening and engine speed requires adjusting the throttle opening first, followed by adjusting the propeller speed to achieve the thrust or economy requirements. Furthermore, due to the lack of an optimal data matching process, its thrust or economy cannot reach the optimal level (traditional mode).

[0063] The linear interpolation method described in this invention is a simple numerical interpolation method. Its core is to construct a straight line using two known data points to estimate the unknown value at any position between the two points. The basic principle is to assume that two known points (x0, y0) and (x1, y1) have a linear relationship (i.e., a constant rate of change). When the unknown point x lies between x0 and x1, its corresponding y value is derived through the equation of the straight line; essentially, it proportionally distributes the difference. This linear interpolation calculation method is well-known to those skilled in the art.

[0064] Example 1

[0065] This invention provides a method for real-time matching of aero-engines and propellers under all operating conditions, comprising the following steps:

[0066] Step S1: Data Foundation and Model Building (Offline Phase)

[0067] like Figure 1 As shown, the system first obtains a basic model, including a propeller model and an engine model, based on engine high-altitude simulation test data and propeller wind tunnel test data, and then constructs a propeller-engine coupled model that combines physical mechanisms with experimental data. Figure 1The model shown is Model 1. Its matching logic mainly includes two parts: an integrated propeller-engine matching model and a data-driven model. The integrated propeller-engine matching model consists of an engine model, a propeller model, and an atmospheric environment model. It aims to optimize propeller thrust and fuel economy, with power balance and speed balance as constraints. It also considers the characteristics of both propeller aerodynamics and engine performance, and takes into account external parameters such as flight altitude, flight speed, and aircraft power requirements. This constructs a matching model and database covering all operating conditions from takeoff to climb and cruise. Propeller aerodynamic characteristics are the relationship between propeller aerodynamic efficiency and thrust coefficient with power coefficient and approach ratio, which can be obtained through experiments or calculations. This invention obtains propeller aerodynamic characteristics through experiments. Engine performance characteristics are the relationship between engine power and fuel consumption rate with engine throttle opening and speed, which can be obtained through experiments or theoretical calculations. This invention obtains engine performance characteristics through experiments. The matching model yields propeller-engine matching data, whose main parameters include flight altitude H, flight speed V, power P, engine throttle opening ε, engine specific fuel consumption rate sfc, propeller thrust T, engine speed n, propeller aerodynamic efficiency η, and combined propeller-engine fuel economy F. Due to the large volume and high dimensionality of the parameter data, and the nonlinear relationships between the parameters, a data-driven model is employed for processing. Ultimately, a multidimensional, nonlinear propeller-engine matching database covering the entire flight envelope is generated.

[0068] In the engine model, a systematic study of the altitude characteristics of a certain type of single-stage turbocharged aero-piston engine was conducted on a controlled high-altitude simulation test bench. The test bench simulated an atmospheric environment at altitudes of 0-10000m, and steady-state operating point data of the engine were tested within a speed range of (2000-5800) rpm and a load range of 5%-100%. Universal characteristics of the engine at different altitudes were constructed. This dataset quantifies the characteristics of engine power, specific fuel consumption rate, and other parameters as a function of flight altitude, providing accurate input for subsequent propeller-engine matching under all operating conditions.

[0069] In the propeller model, the present invention conducted a full-scale wind tunnel test on the propeller in a 4.5×5 meter wind tunnel. The test used a high-precision torque meter, speed sensor, and environmental parameter acquisition system to measure the propeller's thrust, torque, efficiency and other performance parameters under different advance ratio combinations within the airspeed range simulating actual flight conditions.

[0070] Based on the aforementioned performance matching process, and referencing the flight profiles of multiple medium-altitude long-endurance UAVs equipped with this engine and propeller, the flight altitude range is (0-10000) m, the maximum thrust matching flight speed range is (0-80) km / h, the optimal fuel economy matching flight speed range is (160-60) km / h, and the power range is (30-00) kW. According to the maximum thrust and optimal fuel economy matching strategy, approximately 2000 sets of data were obtained by taking altitude, speed, and power intervals of 1000m, 30km / h, and 10kW, respectively. Each set of data includes flight altitude, speed, required power, throttle opening, and engine speed. The overall matching data volume is enormous and cannot be directly stored in the ECU for effective data retrieval, and the data in each dimension exhibits significant non-linear characteristics.

[0071] Step S2: Training and Deployment of the Core Prediction Model (Offline Phase)

[0072] To address the challenges of real-time database queries and the large volume of data, a Random Forest (RF) model is employed to learn and replace the paddle-matching database obtained in step S1, resulting in... Figure 2 Model 2 is shown.

[0073] This invention uses the random forest prediction model shown in the following formula (6):

[0074] (6)

[0075] In the formula, The number of decision trees in the random forest; For the first Decision trees for input samples The predicted value; For the random forest model, the input samples The final predicted value.

[0076] In the model, the goal is to minimize the mean squared error (MSE) to improve regression accuracy. MSE represents the mean squared error, which measures the sum of squared errors between the predicted and actual values ​​of the child nodes after splitting. It can be calculated by the following formula (7):

[0077] (7)

[0078] In the formula, The total number of samples; For the first The actual value of each sample For the first The predicted value for each sample.

[0079] The selection rules for random forests depend on the splitting rules of each decision tree. The splitting rules of each decision tree are based on feature selection and the selection of the optimal splitting point, and the final result is obtained by averaging all decision trees. The regression tree splitting criterion used in the model of this invention is to minimize the MSE. The minimization of MSE is calculated according to the following formula (8):

[0080] (8)

[0081] In the formula, The index representing the feature, i.e. the feature used for splitting; Representation of features The cut-off point (or threshold) on the data; This represents the set of samples contained in the left child node after the split; This represents the set of samples contained in the right child node after the split; The average of the predicted values ​​of all samples within the left child node; The average of the predicted values ​​of all samples within the right-hand child node.

[0082] This invention, verified through extensive experiments, determines the number of decision trees in the random forest to be three. This three-tree random forest (3-Tree-RF) model can operate in two modes: maximum thrust mode and economy mode. The former is used for the takeoff and climb phases of the aircraft / aircraft to train and predict the output parameters, while the latter is used for the cruise phase of the aircraft / aircraft to train and predict the output parameters. In each mode, the corresponding engine speed n and throttle opening ε parameters are output respectively.

[0083] Step S3: System Input and Mode Selection (Online Operation Phase)

[0084] Further as Figure 2 As shown, during the online operation of the trained system (Model 2), the system first determines the input parameters for operation: flight altitude H, power P, and flight speed V. Then, based on the current operating phase of the aircraft / aircraft, it selects either the maximum thrust mode or the economy mode. Then, using the three-tree random forest algorithm, it calculates and outputs the corresponding throttle opening ε and engine speed n, respectively, through the pre-trained speed model and throttle model of this invention.

[0085] Taking the engine speed model as an example, such as Figure 2 As shown, the rotational speed calculation model consists of three decision trees (tree 1, tree 2, and tree 3). Each tree performs branching reasoning based on the judgment of segmentation condition values ​​(segmentation height Hx, segmentation power Px, and segmentation flight speed Vx).

[0086] Taking tree 2 as an example: First, determine whether "current altitude H ≥ segmentation altitude Hx". If it is satisfied, then branch to determine "power P ≥ segmentation power Px" and "flight speed V ≥ segmentation flight speed Vx". In the branch of "V ≥ Vx", it further determines "P ≥ Px", then "H ≥ Hx", and so on, until each branch has determined all segmentation condition values. Finally, each branch will point to a specific engine speed n output (such as speed 1, speed 2, speed 3). The prediction results of the three trees (for the speed model, it is the three engine speed n values; for the throttle model, it is the three throttle opening ε values) are averaged to obtain the final prediction output of the random forest.

[0087] The segmentation condition values ​​(Hx, Px, Vx) are determined according to the segmentation rules in step S2 and will change with the measured and predicted values.

[0088] Figure 2 The dashed line on the right indicates that the same calculation logic is not fully displayed. That is, the throttle opening ε and engine speed n in the maximum inference mode, and the throttle opening ε and engine speed n in the optimal economy mode are all calculated according to the same logic.

[0089] Step S4: Control Execution

[0090] Ultimately, the selected output parameters, engine speed n and throttle opening ε, are sent to the ECU to drive the engine and propeller actuator, completing the closed-loop control of the power system and achieving optimal matching between the propeller and engine under all operating conditions.

[0091] Example 2

[0092] This invention provides a real-time matching method for aero-engines and propellers under all operating conditions. Steps S1, S2, and S4 are the same as in Example 1, except that step S2-1 replaces the original step S3, and a pre-judgment step S4-0 is added to step S4.

[0093] Step S2-1: Fast path determination and matrix pre-calculation

[0094] To overcome the latency that may be caused by multi-level decision-making in the RF model, the system starts the fast path in parallel.

[0095] like Figure 2 As shown, this invention uses a full-factor combination pre-calculation method to generate the pre-calculated output matrix. The system uses the current input (H, P, V) as the center and generates an input matrix covering the surrounding conditions based on the changes (Hδ, Pδ, Vδ). This input matrix contains all possible parameter combinations, such as the altitude, power, and flight speed matrices used in this embodiment:

[0096] Height: H - 2Hδ, H - Hδ, H, H + Hδ, H + 2Hδ

[0097] Power: P - 2Pδ, P - Pδ, P, P + Pδ, P + 2Pδ

[0098] Flight speed: V - 2Vδ, V - Vδ, V, V + Vδ, V + 2Vδ

[0099] The variables (Hδ, Pδ, Vδ) can be selected based on the empirical data collected in specific flight projects to determine the surrounding conditions that need to be covered, and can be optimized and adjusted based on training results.

[0100] Based on the combination of these input matrices, the engine speed n and throttle opening ε data corresponding to the collected data are used to construct the speed matrix and throttle opening matrix, which serve as the pre-calculated output matrix of this invention, providing a fast and comprehensive input scenario for subsequent model training.

[0101] Step S4-0: Perform the judgment

[0102] When the next control cycle arrives, the system checks whether the new input parameters (H', P', V') fall within the range of the pre-calculated input matrix in step S2-1.

[0103] Scenario A: If the new input falls within the range of the input matrix, step S3 is skipped. The system directly calculates the precise engine speed n and throttle opening ε from the pre-stored output matrices (speed matrix and throttle opening matrix) using linear interpolation, and uses these as the final output (output parameters: engine speed n and throttle opening ε). This path has an extremely fast response speed to handle rapid responses during real-time operation.

[0104] Case B: If the new input does not fall within the range of the input matrix, then continue to output the engine speed n and throttle opening ε through step S3 as described in Example 1.

[0105] Example 3: Effect Verification

[0106] Table 1 shows the fuel economy index F of the UAV at an altitude of 5000 meters and a speed of 140-260 km / h after adopting the technical solution of the present invention.

[0107] Table 1: Strategy matching performance of the UAV of the present invention at an altitude of 5000 meters and a speed of 140-260 km / h.

[0108]

[0109] As shown in Table 1, at a flight altitude of 5000m, as the flight speed increases from 140km / h to 260km / h, the fuel economy index F under the maximum thrust matching strategy increases from 0.0327 (kW·h) / g to 0.0377 (kW·h) / g, a relative increase of 15.29%; under the optimal fuel economy matching strategy, the fuel economy index F increases from 0.0335 (kW·h) / g to 0.0387 (kW·h) / g, a relative increase of 15.52%. This indicates that increasing flight speed can effectively improve the fuel economy index F.

[0110] This invention applies trained and optimized matching model code to the ECU (Engine Control Unit) of a certain type of aircraft piston engine and propeller integrated controller. The selected ECU chip is a 32-bit domestically produced microcontroller, manufactured using an embedded flash memory process node, exhibiting extremely high reliability in harsh operating environments. Key chip parameters include a 5-core 32-bit RISC processor, with each core featuring integrated scalar DSP processing capabilities, supporting single / double precision floating-point operations, a 300MHz operating frequency, and an integer arithmetic capability of 2.7 DMIPS / MHz. Testing has verified that the response time for acquiring engine throttle opening and speed is within 0.015ms. Figure 3 The specific test results of the real-time matching method for aero-engines and propellers under all operating conditions of the present invention are shown.

[0111] This invention compares and analyzes the fuel consumption of traditional matching strategies and the optimized matching strategy of this invention. The invention uses a typical 28-hour long-endurance flight mission of a certain type of medium-altitude long-endurance UAV. The flight process of this mission is as follows: the UAV rapidly climbs to the initial altitude according to the flight plan; it flies to the mission area at a relatively high cruising airspeed; it descends to perform the mission; after the mission, to improve return efficiency, the UAV climbs to an altitude of 8000m and returns at high speed. In the optimized matching strategy of this invention, the power system adopts a maximum thrust matching strategy during takeoff and climb; during the cruise phase, it switches to the optimal economy matching strategy. This process can simultaneously and quickly adjust to the optimal required engine speed and required throttle opening. In contrast, the traditional single-path matching mode of throttle opening and engine speed requires adjusting the throttle opening first, and then adjusting the propeller speed to achieve the thrust or economy requirements. Furthermore, due to the lack of an optimal data matching process, its thrust or economy cannot reach the optimal level. Figure 4 The fuel consumption variation curves using the aforementioned segmented matching strategy were compared with the corresponding results using the traditional single-path matching mode for throttle opening and engine speed. Based on a complete 28-hour flight mission profile and cumulative fuel consumption calculations for each stage, the total fuel consumption using the optimal matching strategy of this invention was 525.2 kg, while the total consumption using the traditional single-path matching mode for throttle opening and engine speed was 570.9 kg. The optimal matching strategy significantly reduced fuel consumption by 45.7 kg, a relative reduction of 8.00%.

[0112] In the data processing stage, preliminary training and data cleaning were performed using sample data, followed by the formal training process according to the set parameters. The cleaned dataset was divided into training, validation, and test sets at a ratio of 70%, 15%, and 15%, respectively. The Box-Behnken response surface methodology was used to train the parameters of various models, including Quadratic Formula Method (QFM), Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), Radial Basis Function Network (RBFN), and Random Forest (RF), and the optimal parameters for each model were selected for configuration. Figure 5(a) shows the residual comparison of engine speed, and Figure 5(b) shows the residual comparison of throttle opening. The results show that the distribution of the residuals of engine speed and throttle opening in the RF model is concentrated in a narrow range centered at 0 (the closer the residual is to 0, the better the model fit), demonstrating a high degree of fit to the target value, far exceeding the control accuracy and stability of the QFM, CNN, LSTM, and RBFN models. The results show that QFM, CNN, LSTM, and RBFN models suffer from underfitting when fitting multidimensional nonlinear sample data, while the RF model learns from samples to better fit actual working conditions and has significant advantages in fitting multidimensional nonlinear systems. For the propeller-generator nonlinear matching problem, the RF algorithm is superior when the sample data is sufficient and covers multiple nonlinear features with strong nonlinear coupling, or when the data contains some measurement noise.

[0113] The quadratic formula method (QFM), convolutional neural network (CNN), long short-term memory network (LSTM), radial basis function network (RBFN), and random forest (RF) with different numbers of decision trees (100, 15, 5, 3, 2, and 1) were used to model engine speed and throttle opening. Figure 6 This visually demonstrates the prediction results of different machine learning models for engine speed and throttle opening under the maximum thrust matching strategy. The results are based on the mean absolute error (MAE) (lower is better), root mean square error (RMSE) (lower is better), and coefficient of determination (R²). 2(The closer to 1, the better). It can be seen that the MAE and RMSE values ​​of QFM, CNN, LSTM, and RBFN models are all higher than those of RF, and their R² values ​​are all lower than those of RF. This indicates that the RF model has a significantly better learning ability when dealing with multidimensional nonlinear data than the other models. It can also be seen that the 3-tree configuration compresses the model complexity to about 3% of that of the 100-tree model (only 20% of that of the 15-tree model), significantly reducing the model complexity. At the same time, it accurately avoids the risk of underfitting (R² is improved by 1.27% and MAE is reduced by 9.75% compared to the 2-tree model), and compared to the more complex models with 5 trees or more, it only comes at the cost of a slight decrease in model performance (R² error fluctuation range within 2%), achieving model simplification. A horizontal comparison of QFM, CNN, LSTM, RBFN and RF models shows that the RF model has a significantly better learning ability when dealing with multidimensional nonlinear data. A vertical comparison of RF models with different numbers of decision trees (100, 15, 5, 3, 2 and 1) shows that RF-3 not only demonstrates excellent fitting ability in effectively capturing complex nonlinear relationships in dynamical systems, but also has the comprehensive advantages of simplified model structure and high computational efficiency.

[0114] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. An aero-engine and propeller full working condition real-time matching method, characterized in that, The method comprises the following steps: S1: constructing a matching model and database covering all working conditions of take-off, climb and cruise, with the optimization of propeller thrust and fuel economy as the target respectively; S2: taking the database constructed in step S1 as the training sample, taking the flight height, flight speed and power as the input parameters, taking the engine throttle opening and engine speed as the output parameters, training through the algorithm model to establish the mapping relationship from the input parameters to the output parameters, and obtaining the training model; S3: deploying the training model obtained in step S2 in the engine on-board ECU through the control system, calculating the required engine throttle opening and engine speed in real time according to the current flight phase and real-time sensing data, and sending the control instruction for execution; In step S2, the calculation of each output parameter adopts the three-tree random forest algorithm, in each tree, the segmentation condition values of the flight height, flight speed and power are set respectively, then according to the size interval relationship between the current values of the parameters and the corresponding segmentation condition values, the corresponding training branch is selected to predict the output parameters by the random forest algorithm, and finally the predicted output parameters of the three trees are averaged to obtain the final predicted output parameters.

2. The method of claim 1, wherein: The segmentation condition values are determined according to the minimum MSE splitting rule shown in the following formula: ; where, denotes the index of the feature, i.e. the feature used for splitting; denotes the feature on which the split point is determined; is the actual value for the th sample; denotes the set of samples contained in the left child node after splitting; denotes the set of samples contained in the right child node after splitting; is the average of all sample prediction values within the left child node; is the average of all sample prediction values within the right child node.

3. The method of claim 2, wherein: The training model in step S2 has two working modes of "maximum thrust model" and "best economy model", the former is used to predict the output parameters in the take-off and climb phases, and the latter is used to predict the output parameters in the cruise phase; where the best economy model is characterized by a fuel economy indicator, defined as follows: where F represents the fuel economy indicator, η represents the propeller aerodynamic efficiency, and sfc represents the specific fuel consumption.

4. The method of claim 3, wherein: In step S2, the variation of the flight height, flight speed and power is set respectively, the input matrix covering the surrounding working conditions is generated based on the current input of the flight height, flight speed and power, and the output matrix of the corresponding output parameters is constructed by pre-computing based on the input parameter combination included in the input matrix through the three-tree random forest algorithm.

5. The method of claim 4, wherein: In step S3, the current flight phase is first determined, then it is checked whether the new input parameters fall within the range of the pre-computed input matrix, if they do, the output parameters are calculated by linear interpolation from the output matrix and directly used as the final output parameters; If not, the final output parameters are calculated in real time by using the three-tree random forest algorithm.

6. An aero-engine and propeller all-condition real-time matching system, characterized in that, The system is used to execute the real-time matching method of the aero-engine and propeller in all working conditions according to any one of claims 1-5.

7. The aero-engine and propeller all-condition real-time matching system according to claim 6, characterized in that: The system comprises: a data acquisition and modeling module for acquiring test data of the engine and propeller, and constructing a matching model and database; an algorithm model processing module for training, storing and running the algorithm model to obtain a training model; an interpolation module for pre-computing the input parameters and storing the output matrix, and performing linear interpolation calculation during running; an ECU control execution module for receiving real-time sensing data of the current flight phase and flight state parameters, calling the above algorithm model processing module and interpolation module, and outputting the control instruction of the final output parameters for execution by the engine.

Images